Skip to main content

Full text of "HP Journal 1987-02"

See other formats


© Copr. 1949-1998 Hewlett-Packard Co. 


February 1987 Volume 38 • Number 2 


4 A New Family of Precise, Reliable, and Versatile Fiber Optic Measurement Instru- 
ments, by Michael Fleischer-Reumann They're tor single-mode and mullimode appli- 
cations in the first, second, and third wavelength windows. 

5 A Color-Coding Scheme tor Fiber Optic Instruments and Accessories 

6 Stable LED Sources for a Wide Range of Applications, by Michael Fleischer-Reumann 
Three models provide power at 850, 1300, and 1550 nm. 

8 An Accurate Two-Channel Optical Average Power Meter, by Horst Schweikardt 
Accuracy is as high as ±0.15 dB. Resolution is 1 pW. 

1Q Optical Power Meter Firmware Development, by Bernhard Flade and Michael Goder 
Objectives included a friendly operating concept and effective support for the hardware 

6 Detectors for Optical Power Measurements, by Josef Becker Silicon is best for short 
wavelengths only. Germanium has broader bandwidth. 

Precision Optical Heads for 850 to 1 700 and 450 to 1 020 Nanometers, by Hans Huning, 
Emmerich Mi/Her, Siegmar Schmidt, and Michael Fleischer-Reumann On-board calibra- 
tion data and a precision optical interface contribute to accurate measurements. 

25 Optical Power Splitter 

A High-Precision Optical Connector for Optical Test and Instrumentation, by Wilhelm 

Radermacher Key characteristics are reliability, long lifetime, repeatability, temperature 
stability, and low insertion toss. 

O A Design Approach for a Programmable Optical Attenuator, by Bernd Maisenbacher, 
O I Siegmar Schmidt, and Michael Schlicker A fiberless design makes the long-wavelength 
model suitable for both single-mode and mullimode applications. 

A Programmable Fiber Optic Switch, by Michael Fleischer-Reumann Its main features 
are good repeatability and low insertion loss. 

Quality Microwave Measurement of Packaged Active Devices, by Glenn E. Elmore and 

Louis J. Salz A special fixture makes de-embedded measurements possible 

47 HP 8510 Software Signal Processing 


3 In this Issue 
3 What's Ahead 
37 Authors 

Editor. Richard P Dolan • Associate Editor Business Manager Kennotn A Snaw • Assistant Editor. Nancy R Teatei • An Director. Photographer Arvid A Danielson 
Support Supervisor. Susan E Wright • Administrative Services. Typography, Anne S LoPresti • European Production Supervisor. Michael Zandwijken 


© Copr. 1949-1998 Hewlett-Packard Co. 

© Hewlett-Packard Company 1987 Printed in U S A 

In this Iss 

If "fiber optics wasn't a household word before, it s about to become one. 
In the U.S.A., where several long-distance telephone companies compete, 
we re now seeing television commercials promising clearer, quieter com- 
munications thanks to fiber optic transmission. This is indicative of the ex- 
plosive growth rate of fiber optic communication links in the last few years. 
Applications have expanded from multimode propagation at a wavelength 
of 850 nanometers to single-mode (more suitable for long distances) at 1 300 
and 1550 nanometers. To meet the need for versatile, precise instruments 
to test fiber optic receivers, transmitters, and components at these 

wavelengths. HP designers have come up with a new family of fiber optic test instruments (page 
4). The family includes an optical average power meter (pages 8 and 12) that works with either 
of two optical heads (page 22), depending on the wavelength. The optical heads have individual 
calibration data stored in ROM and a high-precision optical interface. Two models of variable 
optical attenuator (page 31) cover the three wavelength windows; one model is distinguished by 
its usability in both single-mode and multimode applications. Three models of optical source (page 
6) provide highly stable optical power for testing components and receivers. An optical switch 
(page 36) provides flexibility in building test setups. Because test instruments have to be connected 
and disconnected hundreds of times during their lifetimes, the new family is equipped with a 
special connector (page 28) designed for high stability, repeatability, and lifetime. A crucial design 
consideration for the optical heads was the choice of a detector to convert optical power to electric 
current. The article on page 1 6 presents a survey of detector characteristics that illuminates the 
reasoning behind the choice of different detectors for the two optical heads. 

Accurate, repeatable device and component measurements become more difficult at frequencies 
in the microwave range. As a result, microwave designers have always had a problem getting 
circuits designed using measured data to work as expected. HP microwave designers have now 
addressed one aspect of this problem — the need for a standard fixture, capable of accurate 
calibration and high repeatability, for measuring transistors in a variety of package styles. Their 
solution uses the error-correction capabilities of the HP 8510 Microwave Network Analyzer, and 
consists of a specially designed transistor fixture and some software for an HP 9000 workstation. 
The article on page 39 explains the theory and applications of the HP 8501 4A Active Device 
Measurements Pac. 


The detector assembly of the HP 81520A Optical Head. 

What's Ahead 

In March, we continue our series of issues on the new HP Precision Architecture. There'll be 
an article on the first two products based on the architecture — the HP 9000 Model 840 and HP 
3000 Series 930 Computers. The terminal controller for the Series 930, the test system for both 
products, and compiler performance issues will also be covered. 

Tr, e MP Jou'fMil encou'agoi technical uvscu5s*>n ol the topics pftrtantoa in tecem atKles ano will publ'Sh tenets eapected 10 be 01 intetest to out teaduts lettets most bo bi'Wl una ate subtecl 
to Mil no Lettets should be addressed lo EO'lot Hewlett Pacliata Journal MOO Milty«w Avenue Palo Alto. CA 94304 USA 

■R. P. Dolan 

© Copr. 1949-1998 Hewlett-Packard Co. 


A New Family of Precise, Reliable, and 
Versatile Fiber Optic Measurement 

The family members are an average power meter, two 
optical heads, three LED sources, two optical attenuators, 
and an optical switch. 

by Michael Fleischer-Reumann 

communication techniques, it has taken only a short 
time for this fast-developing market to reach a 30% 
annual growth rate and an installed base of 500.000 km of 
fiber core in the U.S.A. in 1985. The shift from multimode 
first-window (a = 850 nm) components to single-mode 
second-window (1.3 jim) and third-window (1.55 (im) com- 
ponents occurred earlier than expected, as soon as reliable 
and reasonably priced components became available. The 
end of this rapid development is not yet in sight. 

Of course, the more that fiber optic links come into use 
replacing old copper communications links or opening new 
paths, and the more megabytes of information are transmit- 
ted, the more dependent we become on the reliability and 
serviceability of these systems. Systems and components 
manufacturers and telecommunications companies are de- 
manding accurate, precise, and reliable measurement 
equipment to ensure this high reliability. 

A new family of Hewlett-Packard fiber optic measure- 

ment instruments takes some new approaches to achieve 
the high goals of reliability and honest accuracy that cus- 
tomers expect and need. The family includes the HP 8152A 
Optical Average Power Meter (see article, page 8), which 
is used with the HP 81520A and 81 521 B Optical Heads 
(see article, page 22), the HP 81 54B LED Sources (see article, 
page 6). the HP 8158B Optical Attenuators (see article, page 
31). and the HP 81 59 A Optical Switch (see article, page 36). 

The fiberless technique used in the HP 8158B Option 
002 Optical Attenuator makes it, to our knowledge, the 
only variable optical attenuator usable in both single-mode 
and multimode systems. The HP 81520A and 81 521 B Op- 
tical Heads for the HP 8152A Power Meter feature indi- 
vidual wavelength calibration stored in EEPROM and a 
specially designed high-precision optical interface. 

All of the instruments have HP-IB (IEEE 488/IEC 625) 
capability for computer-controlled operation in produc- 
tion, R&D, or maintenance. 

Fig. I. HP's new family of pro- 
grammable, fully specified, fiber 
optic (est instruments includes 
(clockwise from top left) the HP 
815BB Optical Attenuator, the HP 
8154B LED Source, the HP 81 52 A 
Optical Average Power Meter, the 
HP 81 59 A Optical Switch, and the 
HP 81521B Optical Head (two 
shown in foreground). 


© Copr. 1949-1998 Hewlett-Packard Co. 

A Color-Coding Scheme for Fiber 
Optic Instruments and Accessories 

Imagine you are working m R&D or a production environment 
and you are dealing with both 850-nm MM and 1300-nm SM 
components You want to select the right accessones for your 
measurement setup (let's say patchcord cables, lens and con- 
nector adapters, a splitter or attenuators) Many accessories 
and measurement instruments are only usable for a special 
wavelength or tiber core diameter. HP marks its accessories with 
a user-friendly color code to show the purpose of each device 
clearly For example, m Fig. 1. the HP 81050BL Lens Adapter s 
color code reads as follows: 

■ Red/orange second and third wavelength windows (1 3 >im 
and 1.55 >im) 

■ Green/blue: 50 to 62 /im core diameter 
The complete code is listed below 


Core Diameter: 








1st window 
2nd window 
3rd window 




SI = step-Index 

Gl = graded-index 

SM = single-mcde 

MM = multimode. 

Fig. 1. HP 81050BL Lens Adapter, showing color coding 

Measurement Standards Needed 

The more communication networks expand, the more 
systems of different manufacturers have to work together. 
Here two problems can arise, both related to standardiza- 
tion issues. 

First, it is important that everyone measure and specify 
according to the same standards. In developing the new 
family of fiber optic instruments. HP has worked exten- 
sively with standards laboratories such as NfiS in the 
U.S.A. and PTB in Germany, and has installed its own 
state-of-the-art fiber optic standards laboratory. The goal 
has been to increase not only the relative accuracy of the 
instruments (many measurements in fiber optics are purely 
relative), but especially to increase their absolute accuracy, 
particularly that of the HP 8154B LED Source and the HP 
81 52 A Power Meter. The HP 81520A and 81 52 IB Optical 
Heads' flexible optical interface is an advantage in this 
effort, because it allows calibration with either parallel or 
divergent beams. 

Second, there is a need for standardization in other areas, 
from seemingly trivial things like connectors to more 
sophisticated measurement procedures. 

Because of the number of different fibers, wavelengths, 
and connectors in use, HP has adopted a color-code ap- 
proach to help the user work with HP equipment and acces- 
sories (see box. this page). 


To the family of instruments introduced in this article, 
many people contributed either directly as R&D engineers 
and project leaders, marketing engineers, or production 
and QA personnel, or indirectly, by offering their ideas as 
valuable inputs. I would especially like to mention Peter 
Aue. fiber optics section manager, and Christian Hentschel, 
who is a member of several committees that are involved 
with laser safety or standardization of measurement 
methods, tools, and terms. 

© Copr. 1949-1998 Hewlett-Packard Co. 


Stable LED Sources for a Wide Range 
of Applications 

by Michael Fleischer-Reumann 

ponents like connectors, splitters, patchcord cables, 
attenuators, and other devices usually requires a 
stable light source. The absolute value of its power output 
is not critical. Stability is the main feature. For fewer prob- 
lems from interference or modal noise, a light-emitting 
diode (LED) is preferred over a laser diode. 

The new HP 8154B Optical Sources (Fig. 1) provide stable 
optical power for testing fiber optic components. The HP 
8154B Option 001 provides - 17 dBm (20 /xW) at 850 nm, 
the HP 8154B Option 002 provides -20 dBm (10 /iW) at 
1300 nm. and the HP 81 54B Option 003 provides -23 dBm 
(5/xW) at 1550 nm. Short-term stability is better than 0.02 
dB within a ±2°C window for one hour. Stability for one 
year over the entire operating range of 0°C to 55°C is bettor 
than 0.3 dB. The HP 8154A/B/C output is CW, but can be 
externally modulated at frequencies up to 1 MHz. Each 
source also has a built-in 270-Hz square-wave generator. 
All of its functions can be controlled via the HP-IB (IEEE 
488/IEC 625). 

Factors Influencing Stability 

Two major factors influence the stability of an optical 
source. The first, of course, is the stability of the light that 
is generated by the LED and launched into the internal 
fiber available at the connector on the front panel. The 
second factor, which is usually ignored, is the stability of 
the mechanism used for coupling the light into the fiber 
that the user connects to the light source. The user is gen- 
erally interested in the amount and stability of the power 
at the end of the connected fiber. This means that a precise, 
reliable, very stable connector with low temperature depen- 
dence is required. 

Fig. 1. The HP 8 7 548 LED Sources provide stable optical 
power at 850 nm /Option 001). 1300 nm (Option 002). and 
1550 nm (Option 003) 

The Diamond* HMS-10/HP connector used in the HP 
8154B and other HP fiber optic instruments, patchcords, 
and other accessories (see article, page 28) meets these 
requirements. The mechanics of this connector are inde- 
pendent of whether it is used with a 50/125-/xm or 62/125- 
graded-index multimode fiber or a 9/1 25-fim step-index 
single-mode fiber, This means that without loss of stability 
one can couple light into a variety of fibers. Only the abso- 
lute power changes, for example from -20 dBm into a 
50/125-/xm fiber to -36 dBm into a 9/125-/im single-mode 
fiber for the long-wavelength sources (HP 8154B Options 
002 and 003). 

The HMS-10/HP connector offers high stability and re- 
peatability and tolerates a large number of mating cycles. 
Another major advantage of the connector and the HP 8154B 
front-panel design is that the connector is easily cleaned. 
This is essential for reliability of the optical contact. No 
tools for opening the instrument are required. Simply un- 
screwing the outside part of the connector gives direct ac- 
cess to the instrument's inside connector for cleaning. 

LED Stability 

Fig. 2 shows the LED output power as a function of 
temperature relative to the power at 25°C, for a current of 
'led =100 mA. The temperature coefficient (TC) for short- 
wavelength devices (850-nm) is rather small, about -0.02 
dB/°C, which is not far from the design goal. It also does 
not vary much between individual devices, so that a typical 
amount of compensation is possible. This is provided in 
the HP 8154B Option 001. The TC of long-wavelength de- 
vices (1300-nm and 1550-nm) is much larger, about - 0.05 
dB/°C. which is far from the desired specification. It also 

3 • • 


-4 ■ ■ 

Fig. 2. Typical temperature coefficient of LED output power 
as a function of temperature for 1^0 = 100 mA. 


© Copr. 1949-1998 Hewlett-Packard Co. 

varies considerably from one device to another. 

To create a stable long-wavelength source, there is no 
other way than to operate the LED at a controlled temper- 
ature. For this reason the LEDs in the HP 8154B Options 
002 and 003 are mounted on a thermoelectric Peltier cooler 
and are regulated to about 25°C. so that the desired specifi- 
cations are achieved over the full temperature range of 0°C 
to 55°C. 

Holding the LED chip at a constant temperature, how- 
ever, does not result in a zero TC because of two effects. 
First, there is a small but well-known TC of the front-panel 
HMS-10/HP connector. Second, the coupling efficiency 
from the LED into the fiber changes with temperature. Both 
effects together result in a positive TC of about 0.01 to 0.02 
dB/°C. Since these effects are of the order of magnitude of 
the design goal, very well-known, and without large vari- 
ations between individual devices, temperature compensa- 
tion is possible. In the HP 8154B Options 002 and 003 it 
is implemented by measuring the ambient temperature 
with a negative temperature coefficient (NTC) sensor {see 
block diagram. Fig. 3). This sensor is outside the LED case, 
and supplements the primary NTC sensor which is inside 
the case, mounted on the Peltier cooler. The outside sen- 
sor's output is used to change the input value to the tem- 
perature regulator slightly, so that the LED chip tempera- 
ture is not constant, but is a function of the ambient tem- 
perature. This results in excellent stability for the overall 

In addition to these techniques, a five-turn mandrel wrap 
is used inside the instrument to achieve nearly ideal optical 
conditions (i.e., an equilibrium mode distribution) at the 
front-panel fiber connector. 

Absolute Output Power and Its Calibration 

Although many applications don't depend on knowledge 
of the absolute value of the output power, its calibration 
and the related accuracy are physical measurement prob- 
lems of general interest. 

Fig. 4 shows the typical power distribution of a 1300-nm 
LED and the responsivity of a germanium detector diode 
as functions of wavelength. The detector is a type com- 
monly used by measurement instruments like the HP 
81521B Optical Head. 

-r + 1 

800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 
Wavelength (nm) 

Fig. 4. Typical cower distribution ot a 1300-nm LED and the 
responsivity ol a germanium detector diode 

The absolute amount of power represented by the LED 
curve in Fig. 4 is 

Pled = X* P ' x > dx 


and the current output of the detector in use is 

Iu= /P(X)r(X)dX (2) 

where r(X) is the responsivity of the detector. There are 
two ways to make a correct measurement of the absolute 
power of this LED. 

First, and this is the way the HP standards laboratory 
calibrates all HP 8154B LED Sources, one can use a 
wavelength independent detector, that is, r(X) = 1. like a 
thermopile or a pyrometer. The second way is to do an 
integration according to equation 2. This is not a simple 
user-friendly method, although r(\) can be read out of the 
HP 8152A Optical Power Meter via the HP-IB and each 
LED can be measured with an optical spectrum analyzer. 

Hewlett-Packard measures all LEDs in the factory. The 
wavelengths X, and where P ou , = '/iPoiit max are measured 
(see Fig. 4). and FWHM = X 2 - X, and CWL = X, + 
VzFWHM are computed and supplied to the customer on 
a rear-panel label on each source. Note that CWL is not 













Hz . 







Fig. 3. HP 8I54B Option 002/003 
block diagram 


© Copr. 1949-1998 Hewlett-Packard Co. 

necessarily the wavelength of maximum peak output 

If the HP 8154B is used in a stimulus/response test setup 
with the HP 8152A Optica] Power Meter, and if the value 
of CWL supplied with the HP 8154B is entered into the 
HP 8152A as the wavelength of the source, the absolute 
power measurement error (depending on the symmetry of 
the LED's power distribution and the linearity of the detec- 
tor's responsivity) is less than 1 to 2% (0.05 to 0.01 dB] 
compared with a measurement done in our standards lab- 
oratory. The error comes from the inaccuracy of the math- 
ematical model, which assumes a Gaussian LED power 
distribution. This error is well below the tolerances in con- 
nector loss and is therefore negligible. 

Versatile Modulation Capabilities 

The HP 81 54B LED Sources are not only CW light sources 
but also have internal and external modulation capabilities, 
easily selectable by a keystroke on the front panel or a 

simple HP-IB command. The internal 270-Hz modulator is 
useful if you have a large amount of attenuation in a system 
and your detector is not sensitive enough, so that you have 
to deal with a lock-in amplifier (a trigger signal is available 
at the front panel). External modulation with TTL signals 
up to 1 MHz is possible. Also, the light output can be 
disabled by a keystroke. This is helpful, for example, when 
zeroing an optical detector without a shutter (like the HP 
81521B Optical Head), so that no optical disconnection is 


Wilhelm Radermacher did the electronic design of the 
HP 8145B Option 002. Manfred Wacker. test supervisor in 
fiber optic production, made the necessary changes in de- 
sign and environmental testing for the HP 8154B Option 
003. Michael Goder evaluated the 850-nm LED devices and 
designed the compensation methods for the HP 8154B Op- 
tion 001. Rudi Vozdecky did the mechanical design. 

An Accurate Two-Channel Optical Average 
Power Meter 

by Horst Schweikardt 

IN COMBINATION WITH the HP 81520A or 81521B Op- 
tical Head, the HP 8152A Optical Average Power Meter 
(Fig. 1) is useful for both absolute and relative power 
measurements over a wavelength range of 450 to 1700 nm. 
Two optical inputs are available for power measurements 
on two channels or for power ratio measurements. The two 
channels are useful for such applications as checking the 
insertion loss or attenuation of optical connectors or cables. 
For automatic test system use. the power meter is program- 
mable via the HP-IB (IEEE 488/IEC 625). A flexible optical 
interface connects quickly and easily to all common optical 

The HP 8152A is designed for both single-mode and 
multimode applications. With the HP 81521B Optical 
Head, it measures power levels between + 3 dBm and - 70 
dBm with 10-pW resolution. ±0.15-dB accuracy is achieved 
between + 3 dBm and - 50 dBm over the temperature range 
of 0 to 40°C. Accuracy derating is an additional ±0.2 dB 
at 55°C. At 25±5°C, the accuracy is ±0.05 dB between + 3 
dBm and -60 dBm. To extend the measurement range be- 
yond + 3 dBm, additional attenuation filters can be inserted. 

A high-performance optical splitter, the HP 81000BS. is 
a useful companion instrument. It provides a 1:10 power 
split ratio. The splitter works with fiber core diameters 
between 9 /*m and 85 pm, and is mode independent within 


This article describes the hardware design of the HP 
81 52 A mainframe. The HP 8125A firmware is described 
in the article on page 12 and the optical head is the subject 
of the article on page 22. The main functions of the main- 
frame are to provide the interface for two optical heads 
and to serve as the user interface. 

Head Interface 

As the head interface, the mainframe delivers the power 
for the head electronics (5 lines) and the Peltier current for 
temperature stabilization of the detector chip (2 lines). Nine 
control lines are used for analog feedback of the chip tem- 
perature, feedback of head status, data and mode control, 
range control, head on/off information, and the read/write 
signal for the head EEPROM. There are two lines for signal 
and signal ground, for a total of 18 pins at each head input 

User Interface 

On the front panel are LEDs for units indication of abso- 
lute average power (mW, /xW, nW, pW). and to indicate 
an average power measurement relative to 1 mW (dBm) or 
relative to a user-defined reference value (dB). Switches 
are provided for setting the reference power and for setting 
the wavelength equal to the wavelength of the source. The 


© Copr. 1949-1998 Hewlett-Packard Co. 

Fig. 1. The programmable, two-channel HP 81 52 A Optical 
Average Power Meier has accuracy as high as =0 15 dB for 
both absolute and relative power measurements It operates 
with the HP 81 520 A Optical Head from 450 to 950 nm and 
with the HP 81 521 B Optical Head from 850 to 1700 nm. 

selected wavelength must be within the wavelength range 
of the connected head. Wavelength resolution is 1 nm. The 
user-defined reference power can be set in watts (mW. /xW. 
nW. pW) or in dBm (-199.99 dBm to +199.99 dBm). A 
calibration factor can be set in dB ( - 199.99 dB to + 199.99 
dB). The CLR button clears the parameters to default values 
(X = 1300 nm, calibration factor = 0.00 clB. and reference 
power = 0.00 dBm or 1000 mW). 

The warning HEAD in the display indicates that no head 
is connected to the selected channel. CAL#o indicates that 
the calibration factor is not 0.00 dB for the selected channel. 
REF#() indicates that the reference value is not 0 dBm when 
making relative measurements on the selected channel. 
Overflow and underflow indications are also provided. 

In measurement mode, the HP 81 52 A displays the abso- 

lute or relative power at the selected channel. A or B: 
P (dBm) = Measured Power + Calibration Factor 


P (dB) = Measured Power + Calibration Factor + 
Reference Value 

Alternatively, it displays the measured power at channel 
B relative to the power at channel A according to the refer- 
ence setting and the specific channel settings for calibration 
factor and wavelength. The user can select autoranging. a 
filter (2 Hz or 8 Hz), and zeroing to compensate for offset 
errors in each range. 

The XDCR OUT (transducer output) connector provides 
an analog head output signal with head dependent 
bandwidth and an output impedance of 600 ohms. 1999 
display counts equals 2.00V into an open circuit, indepen- 
dent of the wavelength setting. This makes it easy to build 
a synchronous detection system by using an external chop- 
per and a lock-in amplifier with the standard head (HP 
81521B) and mainframe (HP 8152A). All modes and param- 
eters are programmable via the HP-IB connector on the rear 
panel. HP-IB address information is displayed when the 
LCL button is pressed. The default address. 22. results in 
a display of A22. 

Analog Section 

Fig. 2 is a simplified block diagram of the HP 8152A 
mainframe showing one of the two channels. Fig. 3 is a 
block diagram of the analog section. The head input signal 
V| N is fed to an input amplifier of gain Gl via the input 
switch. If no head is connected, the input switch is con- 
nected to signal ground. The output VI of the input 
amplifier and the output of the offset DAC (0 volts before 
activation of the zero routine) are summed and fed to a 
variable-gain amplifier of gain C2(\) to compensate for the 
wavelength dependence of the detector signal. The output 
V2 of the \ DAC is buffered and sent to the front-panel 
XDCR OUT connector. After the mainframe offset and gain 
adjust switch. V2 goes to a fixed x4 amplifier and then to 

Analog 3 



Digital B y 


Analog Pari 

Microprocessor Interface 


4 ► 


4 ► 





Fig. 2 Simplified block diagram ol 
one channel ol the HP 81 52 A Op- 
tical Average Power Meter. 

© Copr. 1949-1998 Hewlett-Packard Co. 


one of two filters. Depending on the amount of averaging 
desired, either the 8-Hzor the 2-Hz filter can be selected. 

The filter output signal goes to a CMOS switch, which 
selects one out of eight input signals (four for each channel). 
The switch output is shifted down by 3.901V to adapt the 
level to the requirements of the ADC. Reference voltages 
for the offset DAC and the gain and offset adjustment cir- 
cuits for the ADC are taken from a reference IC. which 
delivers 10.00V. 

For self-test and troubleshooting purposes, the analog 
outputs of the offset DACs and the attenuated head input 
signals are fed to inputs of the CMOS switch and thus can 
be measured with the ADC. For a visual indication of the 
power input level, the 8-Hz filter output controls an LED 
array on the front panel that acts as a trendmeter. 

850 935 1020 1105 1190 1275 1360 1445 1550 1615 1700 
A (nm) 


The responsivity of a detector is the ratio of output cur- 
rent to absorbed optical power and is a function of 
wavelength. For the germanium detector used in the HP 
81521B Optical Head, Fig. 4 shows a typical curve of re- 
sponsivity versus wavelength. If the responsivity at 1300 
nm is defined to be 0 dB, typical relative responsivity is 
-5.5 dB at 850 nm,-4 dB at 1700 nm. and +1.3 dB at 
1520 nm. To compensate for this variation, the HP 8152A 
mainframe should have a wavelength calibration range 
from - 6 dB to +2 dB. and a calibration point at 1300 nm 
and 0.00 dB. 

These requirements are met by a variable-gain amplifier — a 
multiplying DAC — in the signal path of the mainframe. 
The input signal of the X DAC is related to the optical input 
power P,,,,, by the equation 

VI = G0xGlxR,xrfX)xP apl 

where CO is total head gain. R, is the feedback resistor of 
the transimpedance amplifier. Gl is the input amplifier 

Fig. 4. Typical responsivity ol Ihe detectoi in the HP 81 521 B 
Optical Head. 

gain (1, 10, or 100), and r(X) is the responsivity of the 
detector in A/W. The output of the X DAC is given by 

V2 = G2xVl 

which is independent of wavelength and equal to KP ap , if 
C2(X) = r(X tA1 .)/r(X). 

Each head has individual G2 values stored in an EE- 
PROM. When a head is connected to Ihe mainframe, these 
values are read into the mainframe RAM for wavelengths 
from 850 nm to 1700 nm in 10-nm steps for the HP 81521B. 
The wavelength resolution of the mainframe is 1 nm. and 
all values for wavelength settings not on the valid lO-iim 
grid are calculated by linear interpolation. 

The resolution of the X DAC should be <0.01 dB for all 
wavelength settings. This means that the minimum number 
of X-DAC counts is 434. For a total compensation range of 
8 dB. which corresponds to a gain factor G2 = 10~ 08 = 

Multiplexer ; 




Signal ' n P ul Amplifier 
v Switch 






Fig. 3. Block diagram of the 
analog signal path 


©Copr. 1949-1998 Hewlett-Packard Co. 

0.1585. the maximum number of counts must be greater 
than 434/0.1585 = 2739. Therefore, a 12-bit DAC is neces- 
sary. G2 at the calibration point (1300 nm) then should be 
10 -08 = 0.2512. The actual value used is 0.2500 and a 
postamplifier with a gain of 4 reestablishes VI. The specifi- 
cations of the operational amplifier limit the maximum 
dynamic input signal to 12.5V. so at the calibration point, 
a full-scale signal of V E = 12.5 xlO' 02 = 7.887V is al- 
lowed. The actual full-scale voltage of the input signal at 
the calibration point is + 8.000V. Thus the actual upper 
limit of the compensation range is +1.94 dB (i.e.. 10 log 
12.5/8) and the actual lower limit is - 6.02 dB (i.e.. 10 log 

Analog-to-Digital Converter 

For the ADC, better than 0.01-dB resolution at 1/10 of 
full scale is needed. This leads to n > 434/0.1 = 4340. The 
choice, therefore, was a 13-bit ADC. the AD7550 from 
Analog Devices (or equivalent). For the AD7550, the follow- 
ing equation holds: 

n = A, N x 4096/V FS + 4096, 

where n is the number of ADC counts, A, N is the analog 
input voltage, and V FS is the full-scale input voltage. Thus 
n = 0 gives A w = -V FS . and n = 4096 gives A m = 0. 
For convenience, the transfer function of the ADC should 
be 1 count per millivolt of V K . From the equation 

dn/dA, N = 4096/V FS = 1/1 mV 

we get Vfs 

= + 4.096V. Therefore we 

have the following 



A IN (V) 

ADC Counts 

Display Counts 



+ 1 













0.000 -3.901 195 0 

-0.195 -4.096 0 -49 

< -0.195 underflow -1 

This is why V K is shifted down by 3.901V as explained 

Offset Compensation 

For offset compensation, or zero adjustment, there were 
two requirements. First, the resolution had to be less than 
or equal to 1 display count at minimum responsivity ( — 6 
dB from the calibration point). The full-scale value of VI 
is 2.000V. so 1 display count equals dVl = 1 mV. 

Fig. 5 shows the head offset compensation circuitry. 
From equation 2 in Fig. 5. dA = 0 and dX = 1 gives 

ndVl = VgupxR/Ri. 

The second requirement was that the compensation 
range be at least 10% of full scale at maximum responsivity 
(1.94 dB from the calibration point). Here the full-scale 
value of VI is 12.5V, and therefore 10% of full-scale equals 
V M «= 1.3V. From Fig. 5, equation 1, with A = 0 and X = 
1 , we get 

2V M = Vke F xR/R1. 

Combining these results gives 

n > 2V M /dVl = 2600 < 2 ,z . 

Thus the offset DAC is also a 12-bit DAC. With a = 4096. 
dVl = l mV. and Vrq, = 10.00V. R/Rl = 0.4096. With R 
= 10 kfi, we get Rl = 24.3 kil and V M > 2.057V. Thus 
the offset compensation range is >16% of full scale and 
the resolution is <1 mV. 


The project team consisted of )osef Becker (detector 
evaluation). Bernhard Flade and Michael Coder (software), 
Hans Huniiig (head hardware), and Rainer Kggert (mechan- 
i( . 1 1 design). Michael Fleisi liei-Keiim.inn was the project 

© Copr. 1949-1998 Hewlett-Packard Co. 


Optical Power Meter Firmware 

by Bernhard Flade and Michael Goder 

8 152 A Optical Power Meter firmware (which the 
design team thought of as software, since it was 
not yet installed in ROM) were as follows: 

■ To provide a human interface with a friendly operating 
concept, that is. it should be easy to program via the 
HP-IB. the front panel should be self-explanatory, and 
the operating modes should offer great flexibility. 

■ To provide effective software support for the hardware 
designers, without interfering with the software develop- 

■ To complete the software with no known bugs in a very 
short time. 

The software also had to meet a number of requirements. 
These included: 

■ Ability to work in a "dumb" mainframe until intelligence 
is provided by connected heads. 

■ Data processing in ranges not known by the mainframe 
until a head is connected in the active channel. 

■ Results in either linear or logarithmic units. Single-chan- 
nel or ratio measurements with no or little difference in 
processing speed. 

■ Independence of the channel specific parameters for one 
channel from those of the other channel during ratio 

■ Production support for adjustments, troubleshooting and 
individual head calibration. 

■ Extensive self-test capabilities. 

The HP 8152A processor system includes a Motorola 
6809A microprocessor, a Texas Instruments 9914A HP-IB 
controller, an Intel 8279A keyboard and display controller, 
and the following memory: 
32K bytes of ROM 
8K bytes of static RAM 
8K bytes of I/O RAM 
16K bytes configurable as: 

8K bytes of ROM + 8K bytes of static RAM or 
8K bytes of ROM + 8K bytes of EEROM or 
16K bytes of ROM, etc. 

The software development tools consisted of an HP 
64000 Logic Development System and C, SPL. and assem- 
bly programming languages. SPL6809 is a dedicated com- 
piler for the 6809 processor, developed for the HP 1630A 
Logic Analyzer project and later used in the HP 81 75 A 
Data Generator development project. SPL6809 runs on the 
HP 64000 system. 

Meeting the Objectives 

Once the first definition of the members of the new fiber 
optic instrument family was done, a complete external ref- 
erence specification (ERS) was worked out, using as many 

inputs as possible from anyone knowledgeable in fiber 
optic measurements. Subsequently, a complete simulation 
of the front-panel functions, the resulting displays, and the 
signal processing was written on an HP 9000 Series 200 
Computer. Working with this simulation, the last refine- 
ment of the ERS was done. From that time, the ERS was 
not altered in any detail. 

A by-product of the simulation, in conjunction with a 
mini-HP-IB software driver, was the ability to access any 
hardware device directly. This turned out to be a very 
powerful tool. It could perform any front-panel function 
needed for testing or modifying any latch or DAC, or for 
reading out ADC values. Since no working front panel was 
available in the early project stages, this capability was 
invaluable. The only effort required to solve any hardware 
demand was to write a small sequence of direct-access 
commands and insert it after a simulated key recognition 
or replace a simulated data input with it. This software 
was so useful that it was still being used through the first 
environmental test and until the lab prototype was com- 

To save time and effort, software was leveraged from 
other projects as much as possible. For example, the operat- 
ing system and the HP-IB kernel software were leveraged 
from the HP 8175A Data Generator project. The software 
was divided into independent modules that could be de- 

Fig. 1. Layer model of the HP 8752 'A software. Interactions 
between modules are always in the radial direction. 


© Copr. 1949-1998 Hewlett-Packard Co. 

veloped without interfering with one another. Newly writ- 
ten software was shared among the members of the new 
instrument family as much as possible. Time was also saved 
by decoupling the hardware and software designs as long 
and as much as possible, and by working with good and 
easy-to-use software tools. 

Using C 

The motivation for using C was to get some experience 
to see how this language mighl meet our needs, and of 
course, the everlasting dream of universal, portable, and 
ready-to-use software packages. The results were different, 
depending on the assignment. 

The tasks written in C are the keyboard and display man- 
agement, the output data formatter, and the lin'log/lin con- 
versions. The data formatter and the conversions were writ- 
ten, tested, and debugged on an HP-UX system and trans- 
ferred to the HP 64000 after completion. 

Writing a keyboard and display driver is an instrument 
specific and especially hardware-oriented task. It needs 
on-line debugging in the emulator environment with 
single-stepping, memory examination, and program trac- 
ing. The advantage of being freed from memory allocation, 
memory organization, problems of parameter passing, and 
other tasks during program development is paid for dearly 
in the debugging phase, because following all the steps the 
compiler did for you is a very tedious and bothersome job. 

On the other hand, writing a conversion program is an 
instrument independent task, and commonly needs no de- 
bugging on the emulation level. Debugging this kind of 
software is more of a high-level job. where commands like 
WRITE. WRITELN. READ, or READLN from Pascal or a short 
test program allow easy tracing of program flow and data 
manipulation. In such cases, if you can remain in the C 
environment, we feel that C is a good tool for increasing 
the efficiency of software development. 


Execution time 
for typical OS activities 
Interrupt processing: 
Signal a semaphore: 
Wait for a semaphore: 

Pascal OS 

4297 bytes code 
1 2533 bytes library 

6.5 ms 
1.7 ms 
955 fis 


3 109 bytes 

465 /xs 

These results caused us to abandon plans to write a Pascal 
software system for the fiber optic family 

Tesl Wort. Aroa 

Address = START 

While Address 

Store Data end 
Address Into 
Work Aim 

Restore Data trom 
Work Area 

DSC = Invalid 

Using SPL 

During our search for a suitable operating system for the 
fiber optic instruments, we determined that the one used 
in the HP 81 75 A Data Generator, which is the same as the 
one in the HP 1630A Logic Analyzer, would meet our needs 
in a nearly itleal way. In addition, since the processor 
hardware is similar, adapting this system for the fiber optic 
instruments promised to be an easy task. 

This operating system was written in SPL. SPL programs 
have the readability of Pascal programs. Also. SPL provides 
data structures and a modular program structure very simi- 
lar to Pascal, and alluws easy assembly code embedding 
(if the need for it is unavoidable). In conjunction with an 
optimizer, which is also available to reduce the amount of 
final code, it seemed to be an ideal tool, if only it had not 
been dedicated to a single processor (6809). 

To get away from this processor dependence, and be- 
cause of the resemblance with Pascal, we rewrote this 
operating system in Pascal with a very small amount of 
assembly code. A comparison between the SPL version and 
the Pascal version produced the following results: 

DSC ■ Valid 

Test RAM 




RAM Error 

Restore Data 
trom Work Area 

DSC = Invalid 

Address = 
Addreia + 1 

Fig. 2. RAM tesl algorithm The test is not destructive al- 
though it overwrites the entire RAM with test patterns. 

© Copr. 1949-1998 Hewlett-Packard Co. 


Using the Assembler 

From our experience in assembly language programming 
in earlier projects we knew lhal il is best to avoid assembly 
language. The portion of assembly code in the HP 8152A 
software is less than 1%, It is only used for interfacing SPL 
with C and vice versa and for direct accessing of the proces- 
sor's condition code register. 

Structure of the Software 

The heart of the HP 81 52 A software is the operating 
system leveraged from the HP 8175A and modified for the 
needs of the fiber optic instruments. The attributes of this 
operating system are control of nine competitive processes 
(multitasking), priority-controlled preemption by interrupt 
or signal and wait mechanism, no dynamic priorities, and 
no special task communication. 

The tasks are: 

■ Idler process. Does the measurement if there is nothing 
else to do. 

■ Command processing. Synchronizes and schedules any 
activity or command initiated by an interrupt, 

■ Keybord management. Recognizes keystrokes, performs 
the autovernier function, and sends commands accord- 
ing to any keystroke. 

■ HP-IB I/O. Handles communication via the HP-IB and 
activates the command interpreter. 

■ HP-IB talker initialization. Loads the HP-IB output buffer 
with the instrument messages sent by the HP-IB I/O pro- 
cess if the instrument is addressed to talk. 

■ Measurement result processing. Works up measurement 
results and provides results display. 

■ Hardware interrupt handler. Recognizes hardware inter- 
rupts and sends commands accordingly. 

■ Hardware I/O. Driver routines manage data transfer to 
and from the hardware. 

■ Soft timer. Provides timer functions for time-outs, repeti- 
tive actions, etc. 

Fig. 1 (page 12) shows the layered nature of the software. 

A self-test procedure runs at power-on or whenever a 
TST? command is received via the HP-IB. The self-test in- 
cludes testing RAM, the keyboard and display, the device 
bus, the offset and \ DACs, the filters, the ADC. and the 
optical heads if present. A SYSTEM FAILURE bit and an error 
number Exxx shown on the display indicate that a defect 
has been detected while testing the HP 8152A. This error 
number can be interrogated by a controller or thrown away 
by pressing a key or sending any command to the HP8152A. 

RAM Area 

Data Address 

16 Bits 

Work Area — ^y, 


8 Bits 



Data Secure 
Code (DSC) 

32 Bits 

Slate of DSC = Valid or Invalid 


Fig. 3. The RAM test is bytewise. A work area is used to save 
the data at each RAM location during testing. 

In the latter case the displayed error number is transferred 
into a LAST ERROR register which can be interrogated later. 
A ;i-digil error number is used to indicate the error state. 
The coding is as follows: 

1 Specific Error Cude 


Error Indicator f- Channel Selector: 1 = Channel A 

2 = Channel B 

3 = Channels A 


For example, -E 1 50 means that zero calibration failed in 
the -50 dBm range of channel A. 

An error track in the HP 8152A is used to save the com- 
plete hardware status of the defective state. A valuable 
feature for production and service is that this hardware 
status can be restored by sending a special HP-IB command, 
even if the error stale has been left. Once this command is 
sent, the HP 8152A holds the defective hardware state as 
long as no key is pressed. Pressing a key causes the HP 
81 52 A to do a power-on restart. 

A special algorithm, shown in Fig. 2. is used in the RAM 
test. The test is not destructive, although the complete RAM 
is overwritten by test patterns, including the stack area. 
This is accomplished by a bytewise test: a work area (Fig. 
3) is used to save the data and its corresponding address 
while testing each RAM location. A data secure code (DSC) 
is used to indicate that data in the work area is either valid 
or invalid. If there is a valid data byte in this area at the 
beginning of the test, it is restored before testing the RAM 
from address START. This could occur, for example, if the 
HP 8152A is powered off while a test pattern is being writ- 
ten into a RAM location. 


The HP 8152A has a self-calibration procedure, which 
eliminates undesirable offset voltages caused by the optical 
head and/or other active and passive components in the 
mainframe. The calibration routine is activated by pressing 
the ZERO key or by the ZER 1 command on the HP-IB. After 
activation, the HP 8152A is either in autozero or single-zero 
mode. This is determined by the attached optical head. 
Autozero means that the calibration is repeated automati- 
cally after a time determined by the optical head. Single- 
zero means that the function has to be activated every time 
the user wants to calibrate the instrument. 

The HP 8152A is capable of eliminating an offset of one 
quarter of full scale. The offset is measured in each range 
and the compensation values are stored in the internal 
memory. Each time the measurement range changes, the 
corresponding compensation values are applied to the 
offset DACs of both channels. 

When the calibration is initiated while an optical head 
with internal shutter is connected, the shutter moves into 
the disable position so that incident light is blocked. This 
causes the offset to be the only signal remaining in the 
system. If an optical head without shutter is applied, the 
user has to darken the source to avoid an erroneous abort 
of the calibration. The offset compensation is done by 


© Copr. 1949-1998 Hewlett-Packard Co. 




Signal Path . 

Wavelength (») 



ADC | 

Fig. 4. Slock diagram tor offset compensation 

measuring the offset voltage twice with different settings 
of the offset DAC. The compensation value programmed 
into the offset DAC is the complement of the measured and 
calculated offset voltage. 

The offset voltage is calculated by the following equation 
(see Fig. 4): 

v = v DAC + v oll 

where V = voltage measured by the ADC on the signal 
path, V UA(; = voltage of the offset DAC. and V 0( , = offset 
voltage to be compensated. 

The first measurement is done with the offset DAC set 
to its maximum output voltage. That leads to the first com- 
pensation value: 

- V. 

The second measurement is done with the offset DAC 
set to V 0ml|jl . This leads to the final compensation value: 

The compensation error is less than 1 display count. 

Mainframe Control by the Heads 

Mainframe hardware control and parameter limiting de- 
pend on what optical-to-electrical transducer is connected 
to the mainframe. All of the head dependencies (center 
wavelength, range limits, etc.) are stored in a nonvolatile 
memory in every head. Problems of data loss protection (a 
head connection is never a controlled power-on to the head 
electronics because of contact bouncing, sequence of pin 
connections from mechanical tolerances, etc.) are solved 
by using a special write adapter w hen programm ing the 
head EEROM and then pulling the write enable input of 
the EEROM to V, , during normal operation. 

Each time a head is connected to the HP 8 152 A this data 
is read out and stored in a RAM copy of the head memory. 
After testing for a complete data transfer without errors, 
all the information the mainframe needs to modify its con- 
trol software is transferred. In case of erroneous or invalid 
head data all specials are set to defaults, and an error mes- 
sage is initiated and sent to the display and the HP-IB. 

The defaults are selected so that, despite the erroneous 
condition, the measurement can continue, although it may- 
be out of specification. 

Measurement Modes 

The HP 8152A provides two different measurement 
modes, continuous and single. The single mode is available 
only under remote control. In this case, measurements are 
only done when a GROUP EXECUTE TRIGGER or the TRG 
command is received via the HP-IB. 

This is a useful feature for synchronizing measurements 
by the HP 8152A with other operations within a measure- 
ment system. Another important feature is single mode in 
conjunction with autoranging. This ensures that erroneous 
measurement results such as overrange and underrange 
during ranging are suppressed and not output over the 

© Copr. 1949-1998 Hewlett-Packard Co. 


Detectors for Optical Power Measurements 

by Josef Becker 

ALTHOUGH MOST OF TODAY'S intelligent optical 
power meters are found in fiber optic measurement 
setups, they are not generic fiber optic instruments. 
Design and characterization of an optical power meter de- 
tector are more easily understood if one considers that its 
parents are the photometer for chemical analysis and I he 
radiation detector in physics labs or national bureaus of 

In a fiber optic communications link, the detector has to 
respond to fast and weak digital signals, so its speed and 
noise performance are optimized, while linearity and sta- 
bility are of secondary interest. A typical detector for this 
application is the avalanche photodiode, mounted straight 
onto the fiber end. 

A power meter, on the other hand, must give accurate 
information about the beam intensity radiating onto its 
detector from a fiber end. an LED, a laser diode, or any sort 
of freely propagating light beam. If the beam intensity is 
not constant with time, it is generally accepted that the 
power meter reads Ihe average intensity. From this we see 
that, for a detector in a power meter, wide dynamic range 
with good linearity, low aging and temperature depen- 
dence, and spectral response are of primary interest, while 
speed is of no concern. For these reasons, typical detectors 
are large-area photodiodes (2-mm to 10-mm diameter), ther- 
mopiles, or pyroelectric detectors. 

Detector Physics 

All of the detectors mentioned above for optical radiation 
(ultraviolet, visible, infrared) are either tiny thermometers 
or make use of the photoeffect. 

Thermal detectors measure the heat generated when radi- 
ation power is absorbed (AT = IK for P r = 250 jiW). The 
small temperature rise can be converted into a dc voltage 
by a thermopile, that is. a series array of 10 to 50 thin-film 
thermoelements (Fig. la). In a bolometer. Ihe change of 
resistance of a gold film or a thermistor is measured. In a 
pyroelectric detector, a capacitor with temperature depen- 
dent electret material as dielectric produces a measurable 
charge/discharge current into a high-impedance load when 
it is struck by alternating or pulsed radiation. To measure 

(a) (b) 

Fig. 1. Two types of thermal detectors, (a) Thermopile, lb) 
Pyroelectric detector 

constant light intensity, this type of detector needs a chop- 
per (Fig. lb). 

A black coaling is applied to Ihe thermal detector's sur- 
face to achieve a broad spectral range of constant sensitivity 
(Fig. 2). Spectral response is limited, and can be tailored 
by the window material in front of the detector. 

Quantum detectors for measurement purposes are photo- 
diodes, photoresistors. and photocathodes in photomulti- 
plier tubes. Photocathodes are favorable in the ultraviolet 
and for very large-area detectors, while in the visible and 
near infrared parts of the spectrum, silicon, germanium, or 
ternary (GalnAs) photodiodes are the first choice. 

In a quantum detector, an electron-hole pair is generated 
for every radiation energy quantum (hf) absorbed in Ihe 
depletion layer (Fig. 3). If all quanta striking the detector 
surface were exploited, the ideal response (dashed line in 
Fig. 2) would be 

where A is in /tm. 

t| = — —l—; is called the quantum efficiency. 

r ideal( A ) 

Two effects cause tj to be less than 100%: surface reflection 
(some 30%) and wavelength dependence of Ihe absorption 

List of Symbols 


2 998 x I0' 0 m/s = vacuum speed of light 


6.626 x 10 _34 Ws 2 = Planck's constanl 


1.38 x 10" M Ws/K = Boltzmann's conslant 


1 602 x 10 19 As = elementary charge 

frequency (Hz) 

pholocurrenl (A) 


responsivity (A/W) 


wavelength (^m) 


quantum efficiency 

c D 

diode (junction) capacitance 


reverse saturation current of a diode 


numerical aperture = n sin « where a is the off-axis 

angle where the beam intensity has fallen to 5% ( - 1 3 dB) 


rms noise equivalent radiation power 


radiation power 


diode parallel impedance 

R 5 

series (spreading) resistance of a diode 


temperature (K) 

qV aG 

bandgap energy of a semiconductor 


© Copr. 1949-1998 Hewlett-Packard Co. 



a (/im) 

t tt 

Visible Fiber 
Light Optic 


Fig. 2. Spectral responses of silicon and germanium photo- 
diodes. S20 photocathode with MgF? window, and thermopile 
with CaF 2 window 

depth. Only carriers generated in the depletion layer d 
contribute t<> photocurrenl i,, in the external circuit. At the 
short-wavelength end of the useful spectrum, most of the 
energy is absorbed in the front layer of the diode, where 
minority carriers are quickly lost by recombination. 

At the long-wavelength end. there is a sharp cutoff when 
the quantum energy hf falls below the bandgap qV HG of 
the diode material. In Si. with V HI; = 1.2V. A, = 1.05 jim. 
In Ge. V BG = 0.7V and \ r = 1.8 (im. Photons of longer 
wavelength cannot provide the energy to release an electron 
into the conduction band For them, the diode is a transpar- 
ent window. 

Spectral response is temperature dependent. Technolog- 
ical progress in the last three years has made it possible to 
design photodiodes for fiber optic applications with zero 
temperature coefficient at the wavelength of interest. In a 
power meter, however, where a broader spectral range is 
needed. Fig. 4 shows a clear demand for constant detector 


Fig. 3. Two types ol quantum detectors (a) Germanium 
photodiode (b) Silicon pin diode, d is the depletion layer 

Homogeneity and Angular Response 

As a result mainly of passivation and anlireflection coat- 
ing imperfections, photocurrent will vary to some extent 
if a photodiode is scanned with a small light spot. It is 
common practice to adjust the scanning beam diameter 
and step size to 110 of the active detector diameter. Then, 
with Si diodes of 1-cm diameter, inhomogeneities in the 
detector response of less than ^0.5% can be expected over 
the central area or at least 70% of the active diameter. This 
causes no problems. But with Ge diodes, inhomogeneities 
of several percent are normally found (Fig. 5). Recently 
introduced SiO passivation on Ge can reduce the problem 
to ±0.5% or less. To reduce the influence of local in- 
homogeneities to a negligible level, the light beam should 
cover at least 50% of the detector diameter. 




o ■ 


UDT (1981)- 
Si pin Diodes 

HP 4204 -(1981 

Centronic (1984) 
AMC (1984) 

1 ■ 

^ 0.5 • 


-0.5 - ■ 

GalnAs pin Diode 

.7 .8 .9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 
A (em) 

Fig. 4. Temperature coefficient of the responsivity of silicon 
(top) and long-wavelength (bottom) photodiodes. 

© Copr. 1949-1998 Hewlett-Packard Co. 


Fig. 5. Responswty map of a 5-mm diameter Ge photodiode 
with polyimide passivation, showing inhomogeneities of ±3% 

Attention to the detector properties discussed so far is 
sufficient for accurate measurements of the intensity of a 
collimated (parallel) beam that is perpendicular to the de- 
tector surface. If divergent radiation out of a fiber is coupled 
to the detector, we have to consider the detector's respon- 
sivity for oblique rays. Surface reflection will increase with 
growing aperture angle according to Fresnel's formulas. On 
the other hand, the absorption path in the detector crystal 
is longer, thus giving better absorption efficiency. Both 
effects cancel to some extent. Therefore, in a Si diode at 
850 nm, responsivity remains constant within r0.3% for 
incident angles from 0 to ±15°, and is down 1% at ±20°. 
With Ge diodes, coating layers may introduce large errors 
(Fig. 6) and there is a distinct dependence on wavelength. 
However, technological progress in detector surface finish- 
ing within the last year has made it possible to reduce this 
effect to well below ±1%. 

Optical Interface 

To couple optical fibers to a detector head of the HP 
8152A Optical Average Power Meter, a collimating lens 
adapter is used, eliminating possible inaccuracies caused 
by slanting rays. The required aperture of the lens — or of 
the detector itself, if no lens is used — depends on the angu- 
lar radiation pattern (far field) of the optical fiber. Fig. 7 
shows the collecting efficiency for a typical 50/125-/xm 
graded-index fiber. 

The simplest optical collimator, a planoconvex lens, per- 

forms nearly ideally with monomode fibers; a lens with 
NA = 0.2 is sufficient to collect all the light emitted from 
the fiber end. With multimode fibers, Fig. 7 shows that the 
collimator should have a useful aperture of 0.4 to 0.5. How- 
ever, spherical aberration of a planoconvex lens produces 
marginal rays at NA = 0.4 that are inclined 7 C to the axis. 
Therefore, the accuracy problem shown in Fig. 6 is only 
partly solved by the simple planoconvex lens. 

Two collimators for higher apertures are shown in Fig. 
8. The molded aspherical lens and the two ground spherical 
lenses have comparable price and aberration performance. 
The latter is preferred because of looser optical manufactur- 
ing tolerances, and because of smaller refraction angles for 
edge rays. This reduces variations in the spectral transmis- 
sion of the lens coating with incident angle. 

Photodiode Circuit 

The irradiated pn or pin structure of Fig. 3 is described 
by the equivalent circuit of Fig. 9a. The diode symbol rep- 
resents the pn junction's v-i characteristic, in parallel with 
the junction capacitance C I3 . 

i D = I s (exp^?-l) 

is the spreading resistance in the outer layer (10 to 100 
ohms). R| is the external load. Fig. 9b shows the v-i charac- 
teristics of this circuit. Irradiation, like thermal motion, 
generates electron-hole pairs that appear as reverse current 
L, driven by the inner diffusion voltage across the depletion 

For high-speed applications, an additional reverse bias 
voltage is applied in series with R| to reduce diode capaci- 
tance and speed up carrier collection, at the expense of 
added dark current noise. For wide dynamic range down 
to the lowest possible noise level, no bias should be 
applied. The diode then operates in the photovoltaic mode 
(V„ 3 0). 

100°o - • 

0.2 0.3 0.4 0.S 

Fig. 7. Percentage of light reaching the detector from a 
graded-index fiber with NA = 0 2. « = lens or detector di- 
ameter, d = distance of lens or detector from fiber end. NA 
= sin tan ' (&2d). 

Fig. 6. Responsivity of a coated Ge photodiode varies up to 
±5% with wavelength and beam incidence angle. 


© Copr. 1949-1998 Hewlett-Packard Co. 

In this mode, neglecting the voltage drop across K,. the 
short-circuit current fl^j = i for Rj = 0) increases linearly 
with incident radiation power P r . that is, i&on = i p - With 
a high-impedance load. \ a = i,„ and the detector voltage 
varies logarithmically with light intensity: 

— = 26 mV x In -it at room temperature. 

, A 

V V V 



D > 


' < 


Photographic exposure meters make use of this operating 
mode. For solar energy conversion. R, is chosen to make 
vi a maximum at a given light level. 

In optical power meters, photodiodes are loaded with a 
virtual short by a transimpedance amplifier to provide a 
linear response to light intensity. V D is always very small, 
so we can represent the diode by its small-signal parame- 
ters. R D and C n . where 

Rd" = (dV^)v„.o = ("kT Is6XP ^T ) Vu = o = kT 

i„ = 600 

R,+R,= 1 Ml: Solar Energy 


Rm = 


26 mV 

From analysis of the linearized circuit (Fig, 10). we learn 
that diode dynamic resistance R„ is the key parameter that 
determines the noise level and dc error of the detector unit. 
Diode noise is essentially shot noise, not thermal (|ohnson) 
noise. But a short-circuited diode generates an amount of 
shot noise equal to the thermal noise produced by a resistor 
of magnitude R D . 

Response linearity at high light-power levels is limited 
by R,: 

'short - 'p >U ~ ' P ~l* y 

ex P aM^-i 

To extend the range of linear response of a Ge diode from 
a few milliwatts to several tens of milliwatts, it has been 
proposed to measure high intensities with the detector 
slightly biased (about - IV). The argument of the exponen- 
tial is thereby shifted far to the negative, thus making the 
diode current term i D negligible. 

i D is strongly dependent on temperature and bandgap, 

I. = Jf - - Constant x T ' exp ( - ^ V BG j 

Fig. 9. (a; Photodiode equivalent circuit, (b) v-i characteris- 
tics ol a silicon photodiode. 

Typical R„ values at room temperature are 500 Mil and 
5 kil for Si and Ge diodes, respectively, of 5-mm diameter. 
From this, we calculate thai with bandgap voltages of 1.2V 
and 0.7V. respectively, the Ge detector would have to be 
cooled down to approximately -60°C to perform as well 
as its silicon counterpart does at +23°C, or room temper- 
ature. Practical results are even worse for the Ge diode 
because of surface effects that have been neglected in this 

DC Stability and Noise Level 

The low end of the useful dynamic range is reached at 
a radiation power level that produces an output signal V„ 
(Fig. 10) equal to either dc drift or to the rms noise voltage 
produced at the output of the amplifier. 

WithR,, = 100 kil in Fig. 10 (Ge, 5-mm diameter, - 10°C). 

Fig. 8. Two collimators lor higher apertures (a) Molded 
aspherical lens (b) Two ground spherical lenses. 

Fig. 10. Linearized photodiode with transimpedance 
amplifier . including dc and noise error equivalent sources. 

© Copr. 1949-1998 Hewlett-Packard Co. 


R, = 50 Mil. and an OPA111BM operational amplifier, the 
drift over 0 to 55°C is 



drift < 55K(1 mV/K)(1 + ) + (1 pA)(R,)(2 — +2 "Tir 

V o.infi = 27.56 mV + (50 /xV)(8 + 0.16) = 28 mV max. 

(14 mV is typical.) 

Offset voltage is the predominant error source with the Ge 
detector. If the responsivity is r(X) = 0.5A/W. this corre- 
sponds to a drift equivalent power of 

28 mV 1 
50 Mil ' 0.5AAV 

= 1.12 nW maximum. 

Zeroing the instrument after warm-up will depress this 
error below the fluctuations caused by flicker noise. 

If any leakage current has a chance to reach the summing 
node, it might increase the effective bias current from fem- 
toamps to nanoamps. Careful guarding on the printed cir- 
cuit board will prevent this. 

The noise voltage at the output is (for low frequencies): 

<& = (i; D 

InA)Rf + V 

With the same circuit elements as above, and substituting 
6 times the rms value for the peak-to-peak noise of the 
"white" resistive sources, peak-to-peak noise at the output 
measured over 0.1 to 10 Hz comes out as 

+ (2.5^V) 2 (501) 2 

= (7.64pA) 2 (50 Mil)' 1 + (1.25 mV) 2 

(12 fA)" (50Mil)- 

= (1.253 raV)* max. 
(0.60 mV is typical.) 

The peak-to-peaic noise output of 1.25 mV corresponds to 
an input signal fluctuation of 


p » lulls' few) = 50 t,w 

(24 pW is typical.) 

Noise equivalent power (NEP) is usually measured with a 
Chopping arrangement at 1-kHz chopping frequency and 
1-Hz bandwidth. The circuit of Fig. 10 then delivers 

V 2 „orm 5 r 4kT 4kT ( fA Yl 

( 7 ^J (501)2 =( 2a54 x7Hz) 

The rms noise of 20.54 /xV corresponds lo a noise equivalent 
power of 

NEP = 

20.54 mV 

50 Mil 
= - 90.9 dBm typical. 

The noise levels calculated above are in full agreement 
with measured results. They reveal that NEP specifications 
may be misleading when the real resolution of an optical 
dc power meter has to be determined. The performance of 
nonchopped long-wave detectors with large-area diodes 
having low R D is limited by amplifier flicker noise voltage. 

The circuit of Fig. 10 will resolve radiation powers down 
to the detector NEP. if I„dR ( is the predominant noise term 



1.2 pW 
0.8 pW 

0.4 pW 4- 


0.12 pW 
80 (W 

-100 |<V. 


--50 (.VsiN. 

-30 /iV \ 
-20 \ 

f=30 Hz 
Af=1 Hz 

^^S. r(A)=0.5 A W 


-10 /iV 


1.2- - 

T3 1 - " 

/ 0.6- 

-- 3 ,.V 
-2 /iV 

— . F ^^-^"^ 

^"^^■0.2- - 

-1 /iV 

-4 — - 

o- - 

v„„.„ 20* 

1 1 1 1 1 1 1 

50k 100k 200k 500k iM 2M 5M 

1 1 

10M 20M 

Diode Resistance R D (1!) 

Fig. 11. Noise equivalent power 
(NEP I and noise figure F of photo- 
diode detector circuits. 


© Copr. 1949-1998 Hewlett-Packard Co. 



Real Value 
(Never Known) 

Error Sources 



Standards Office 

Primary Standard: 
ESR with Thermopile 

• Physical Standard 
Realization Uncertainty 

• Measurement Variance 

• P r =10^W. T. ml) =23-C 

Total Uncertainty at 

A = 850 ran: 
A = 1300 nm: 

t 1.7% 

Standards Lab 

Transfer Standard: 
ESR Pryoetectrtc Of 
Si or Ge Diode 

Aging of Transfer 


Tolerance of 




P, = 10/iW.T, nlc , = 23X 


Without Working Standard: Traceabllity- 

Production or 
Service Station 

Worfong Standard 
Si or G* Diode 



> Value 

• Aging and TC ot 
Working Standard 

• Wavelength and Beam 
Profile Variations 

• P, = 10 jiW. T=1S to 35'C 

Traceabllity - 

:2.7% - 
:4.2% • 

• Aging. TC. Linearity 
Noise. Quantization, 
Beam Geometry. 
Wavelength Dependence 

• P,=1 pW to 2 mW 
T =0 10 55 a C 

For P, 
For P. 

A=850 nm P, 
X = 1300 nm P, 

Uncertainty - 

100 pW: t 6.2% - 
10 nW: * 7.7% - 

Uncertainty - 

100 pW. t 5.2% 
10 nW: ± 6% 

Fig. 12. Measurement uncertainty is the sum ot instrument inaccuracy and calibration uncer- 
tainty (traceabllity) 

in the rms noise formula. This can be achieved, even with 
large-area Ge detectors, using a chopper. Fig. 1 1 shows the 
NEP of the whole circuit and of the detector alone (with 
an ideal noise-free circuit) as a function of diode impedance 
Ru. The modulation (chopper) frequency is 30 Hz. Noise 
figure, F. is the ratio of total NEP to detector NEP. 

Calibration and Standards 

The total uncertainty of a value displayed by a measuring 
instrument is composed of the instrument's inaccuracy 
plus the absolute uncertainty of its calibration. 

Instrument inaccuracy includes variations of its trans- 
ducer gain with intensity and other input signal param- 
eters, temperature dependence, aging, noise, and errors in- 
troduced by digitizing and processing. 

Calibration uncertainty, often called traceability. usually 
refers to a specified signal level at fixed environmental 
conditions. Fig. 12 explains its meaning for radiation power 
measurements. Listed numbers are tolerance limits, or 3<r 
statistical uncertainties, which correspond to a 99.7% con- 
fidence level. 

One step in the calibration chain — the working stan- 
dard — can be eliminated if all instruments are (re)cali- 
brated in the manufacturer's standards laboratory. This pro- 
cedure, used within HP, reduces the traceability and the 
uncertainty specifications of the instrument to ±1.7% and 
±5.2% for 850-nm wavelength and to ±2.5% and ±6.0% 
for 1300-nm wavelength. 

Lowest possible uncertainty is achieved if the power 
meter itself is used as a secondary (transfer) standard and 
has a calibration certificate from a national bureau of stan- 
dards. The uncertainty margins in Fig. 12 then shrink from 
±6.2% and ±7.7% for 850 and 1300 nm, respectively, to 
±4.6% and ±5.2% over the full specified operating range. 

Primary standards for radiation power measurements 
today are blackened absorber discs with a thin-film heater 
network added for electrical simulation of the radiation 

power. This element is coupled to a thermopile or a 
pyroelectric crystal to detect equality of absorbed radiation 
and electrical heating power. Sophisticated but bulky and 
slow-reacting constructions, at high power levels, are esti- 
mated to yield uncertainties down to 0.2%. Practical cali- 
bration procedures at national standards laboratories re- 
quire one or more copying steps, so the calibration certifi- 
cate comes out with about ±1% uncertainty, based on a 
95% confidence level. 


Today's fiber optic links use one or more of the three 
near infrared bands around wavelengths of 850. 1300. and 
1550 nm. A universal power meter transducer with 
adequate sensitivity and spectral range for these applica- 
tions can be built with a large-area Ge detector. 

For applications in the short-wavelength band only 
fim), a silicon diode offers broader dynamic range. 

A drawback of both detectors in broadband applications 
is the considerable and nonlinear variation of their respon- 
sivity with wavelength. For instance, when measuring the 
total radiation of an LED source at 1300 nm ±50 nm. the 
spectral response curve of Ge and GalnAs may introduce 
inaccuracies of several percent. A thermal detector, with 
its inherently flat response, seems preferable. However, the 
sensitivity and dynamic range of these detectors are 100 
to 10,000 times worse than those of the quantum detectors. 


I would like to thank Milan Cedilnik of our standards 
lab for his help with the equipment and software and his 
reliable Inputs during these evaluations. 

© Copr. 1949-1998 Hewlett-Packard Co. 


Precision Optical Heads for 850 to 
1700 and 450 to 1020 Nanometers 

by Hans Huning, Emmerich Muller. Siegmar Schmidt, and Michael Fleischer-Reumann 

Optical Power Meier. Depending on the wavelength 
of the optical source, either the HP 81521B. for 850 
to 1700 nanometers, or the HP 81520A, for 450 to 1020 
nanometers, can be used. 

HP 81 521 B Optical Head 

The HP 81 52 IB Optical Head is characterized by a wide 
dynamic range for average power measurements from + 3 
dBm to -80 rlBm. wide spectral responsivity individually 
measured in 10-nm steps from 850 nm to 1700 nm and the 
values stored in each optical head, high stability over a 
temperature range from 0°C to 40°C, a well-cooled detector 
chip for operation up to 55°C. a noise floor well below - 70 
dBm, user-friendly optical interfacing, and availability of 
a large variety of optical adapters and other accessories 
including a filter holder, a beam splitter, and a bare fiber 

The head consists of three main functional blocks, as 
shown in Fig. 1. The optical detector device includes a 





Range Control 

Output ^ 


~T" 0.3V/>W W 

3V nW 




♦ Cooler Current 

Cooler |fl 

-Cooler Current 


Temperature 1 

Sensor 1 

Reference ^ , 

for -10°C 

PI Regulator 


y Current 

Current W»™V 

Power On Off 
Data In Out 
Write Enable 
Read Enable 






Fig. 1. Block diagram oi the HP81521B Optical Head The 
HP 8 1520 A is similar. 

two-stage Peltier cooler, a temperature sensing device, and 
a transimpedance amplifier. A Wheatstone bridge for tem- 
perature measurement is followed by a PI (proportional 
integral) regulator and a current limiter. An EEPROM stores 
head specific data and calibration factors. The cooler cur- 
rent control and the EEPROM are included in the head to 
guarantee maximum flexibility in future heads. 

The germanium photodiode has an active area of 5-mm 
diameter. When struck by incident light it generates a cur- 
rent proportional to the absolute optical power with a con- 
version factor of typically 0.6A/W at 1300 nm. It operates 
in the photoconductive mode, as described in the article 
on page 16. 

The transimpedance amplifier. Fig. 2. converts this cur- 
rent into an equivalent voltage depending on the selected 
feedback resistor. The amplifier chosen for this application 
is the OPAlllBM. which exhibits very small input offset 
voltage drift (maximum 1 /tV/K). 

Depending on the mainframe gain ( x 100. x 10, x 1 ), the 
head has to deliver a full-scale input voltage of 0.08V, 0.8V, 
or 8V to the mainframe. 

The following table shows the head gain, feedback resis- 
tor (R|). mainframe gain, and switch positions (Fig. 2) for 
each input power range. Total gain is the product of head 
gain and mainframe gain. 




Fig. 2. Transimpedance amplifier of the HP 875276 Optical 
Head R,. the effective feedback resistance, depends on the 
settings of K1 and K2. 


© Copr. 1949-1998 Hewlett-Packard Co. 




Head Gain 
















500 kfl 





-30 dBm 

500 kft 





-40 dBm 

500 kil 






50 Mil 





With the internal gain adjust control, each head is indi- 
vidually adjusted so that its gain is 0.400V >W at -20 
dBm. In the lowest range (full scale = -47 dBm = 19.99 
nW) the feedback resistor is 50 Mfl and the photocurrent 
at - 80 dBm is 6 pA. At -47 dBm the photocurrent isl2nA. 
Surface leakage currents caused by contaminated surfaces 
will produce offset errors, so guarding is used to prevent 
this kind of failure. 

HP 81 521 B Detector Temperature Control 

As described in the article on page 16. the absolute tem- 
perature is the dominant parameter influencing the noise 
behavior and parallel impedance of a Ge detector. The heat 
pumping capacity of the two-stage Peltier cooler required 
a trade-off between achieving the lowest possible chip tem- 
perature and a maximum ambient operating temperature 
of 55°C. The operating chip temperature selected is - 10°G. 

The temperature regulator consists of two parts, a driver 
and a regulator. The driver is installed in the mainframe 
and supplies the head with a current of +1.5A maximum 
for cooling or -0.8A for heating. The mainframe also has 
a slew rate limiter to increase the lifetime of the cooler 

The dominant time constant of the regulator, which con- 


50 100 
Time (s) 

Fig. 3. The dominant time constant ol the temperature reg- 
ulator is fixed by the thermal coupling between the cooler 
and the thermal sensor (NTC) This diagram shows typical 
sensor response to a step change in cooler current at 40°C 

1.5 -r 


- os- 

io 20 

Time (S) 


Fig. 4. Typical temperature regulator settling behavior at 

sists of the resistor bridge, the PI regulator, and the current 
limiter. is fixed by the thermal coupling between the cooler 
stage and the thermal sensor, which has a negative temper- 
ature coefficient (NTC). This time constant is typically 25 
seconds at 25"C ambient temperature. Fig. 3 shows typical 
behavior at 40°C ambient temperature. The whole system 
is optimized for minimum overshoot and ringing, and 
stabilizes the chip temperature within a few hundredths 
of a degree Celsius. 

Settling behavior at 25°C is shown in Fig. 4. Typical 
cooler current and offset are shown in Fig. 5 as functions 
of the ambient temperature. 

HP 81 520 A Optical Head 

The HP 81520A Optical Head is designed for optical 
average power measurements at shorter wavelengths, in- 
cluding Ihe spectral window from 450 nm to 1020 nm. A 
wide dynamic range from + 10 dBm down to - 100 dBm 
for average power measurements, spectral responsivity in- 
dividually measured in 10-nm steps from 450 nm to 1020 
nm and the values stored in each head, high stability over 
8 temperature range from 0°C to +55°C, and a noise floor 
well below - 90 dBm are features of this head. 

The detector device of the HP 81520A, in contrast to the 
HP 81521B, is a silicon chip. However, it has the same 
active area of 5-mm diameter. When struck by incident 
lighl, it generates a current proportional to the absolute 

1.5 -r 

1.0 - - 



0.5 - - Current 


-\ h 

- - 1 

- -0 





0 10 20 30 40 50 60 

Ambient Temperature (°C) 

Fig. 5. Typical cooler current and ollset as functions ol am- 
bient temperature. 

© Copr. 1949-1998 Hewlett-Packard Co. 


AI 3 Oj Chip 

r > 

Si Detector 

Si Detector 

optical power with a conversion factor of typically 0.5A/W 
at H50 DDL It operates in the photneonductive mode. 

The HP 81520A Optical Head has the same three main 
functional blocks as the HP 81 521 B. Fig. 6 shows the Irans- 
impedance amplifier of the HP 81520A. 

Depending on the mainframe gain ( x 100, xio, XI) the 
head has to deliver a full scale input voltage of 0.08V, 0.8V, 
or 8V at the mainframe input. The following table shows 
the head gain, feedback resistor (R,), mainframe gain, and 
switch positions (Fig. 6) for each input power range. Total 
gain is the product of head gain and mainframe gain. 




Head Gain 


K2 K3 K4 


+ 10dBm 

1.01 M 








1.01 ki2 







- lOdBm 

100 kil 








100 kil 

50 V/mW 























10 Mil 























With the internal gain adjust control, each head is indi- 
vidually adjusted so that its gain is 0.04V/W at - 20 dBm. 
Because of the large tolerance of the 1-Gil chip resistor 
(±10%), it is also necessary to adjust this head in the lowest 
two ranges to get the specified accuracy. 

To reach a resolution of - 100 dBm (that is, 0.1 pW or 
10~ 13 watt) with acceptable stability and noise floor, the 
detector chip and other parts are enclosed in a hermetically 
sealed TO-8 housing filled with dry argon. To achieve a 
constant, controlled temperature of 20°C, the package in- 
cludes a one-stage Peltier cooler and a temperature sensing 
resistor (NTG). Also in the package are the detecting and 
amplifying devices that are sensitive to temperature. These 
include the silicon detector chip, whose dynamic resis- 

Fig. 7. HP 81520A Optical Head detector assembly 

tance changes with temperature, the 10''-ohm chip resistor 
(R f ), and the operational amplifier, whose offset voltage 
drifts with temperature. The chip resistor is also sensitive 
to contamination and humidity, and is protected from these 
influences by the package. 

The one-stage thermoelectric cooler is soldered onto a 
standard TO-8 header at 125°C using In-Sn solder paste. 

1 Gl! 



o n 

Hermetically Sealed 
Detector Device 
(Including Cooler 
and Thermistor) 

H, • 
10 MS! 




100 k!! 

1.01 kSi 

-20 dBm 

O Output 

-50 dBm 

^7 Gain Adjust \f 

Fig. 6. Transimpedance amplifier 
of the HP 81520A Optical Head. 
The effective R, is a combination 
of the four resistors so labeled, de- 
pending on the switch positions 


© Copr. 1949-1998 Hewlett-Packard Co. 

Optical Power Splitter 

The HP 81000AS'BS Optical Power Sorters are used in com- 
bination with the HP 8152A Optical Power «/eter to measure 
insertion loss and attenuation of passive optical components or 
to control power levels 

The HP 81000AS'BS power splitters are three- port devices 
They have one fiber connector input, one fiber connector output 
and one parallel beam output to the optical head ot the HP 
8152A A conventional optical beamsplirting technique is used 
A high-precision optical system with the same obiective lens as 
m the HP 8158B Optical Attenuator coilimates the divergent light 
ot the input fiber to an expanded parallel beam and retocuses 
it into the output fiber. A beamsplitter plate, located between the 
two lens systems, reflects one part of the light intensity to the 
photodiode of the optical head The beamsplitter is low-angled 
to the optical axis to reduce the polarization sensitivity of its 
reflection factor to less than ±4% Physical beamsplittmg ensures 
that there is no mode selectivity in mullimode applications 

Fig 1 shows the optical system of the splitter The light is not 
fiber guided and so different fiber types with numerical aperture 
NA less than or equal to 0.3 can be used The HP 81000AS is 
designed for multimode applications and accepts fiber core di- 
ameters from 50 /im to 100 jtm The HP 81000BS is designed 
for single-mode and multimode applications and accepts fiber 

core diameters from 9 pin to 100 /im The HP 8 1000 AS covers 
the first fiber optic window that is. the wavelength range from 
600 nm to 1200 nm and Ihe HP 81000BS covers the second 
and third fiber optic windows the wavelength range from 1200 
lo 1630 ran 

The optical splitters combine the advantages of low insertion 
loss, stable splitting ratio, and good environmental characteris- 
tics The typical insertion loss, including the two Diamond* HMS- 
10'HP connectors is 1 dB tor mullimode and 2.5 dB for single- 
mode operation The splitting ratio depends on the insertion loss 
of Ihe fiber input to fiber output coupling and is about 1 0 dB for 
single-mode and 12 dB for multimode Thermal stability of the 
splitting ratio within a =2°C temperature window is better than 
+ 0.06 dB tor smgle-mode and +0 01 dB for multimode 


Special thanks to Peter Klement. who was responsible lor op- 
tical materials engineering, and lo Ramer Eggert. who did ihe 
excellent mechanical design 

Siegmar Schmidt 

Development Engineer 
Bdblingen Instruments Division 


HP 81000AS BS 


To HP 8I52A 

The lop of the Peltier cooler serves as a mounting base for 
all Ihe temperature stabilized detector parts (Fig. 7). A 
thick-film hybrid on an ALO, ceramic substrate is glued 
to the cooler with heat-conductive epoxy. The hybrid con- 
tains the op amp IC chip, the 1-Gil chip resistor, and the 
thermal sensor (NTC chip), which are attached with con- 
ductive epoxy. All of these chips are connected with 
wedge-wedge bonds to the substrate or In ihe TO-8 header 

To achieve more component mounting space, a second 
floor is generated by mounting the detector chip carrier 
hybrid on a ceramic heat stud. This design provides a very 
good temperature stabilized environment for Ihe electrical 
and optical components, which are responsible for the high 
stability and sensitivity of this detector assembly. To 
achieve the MTBF design goal, the whole assembly is her- 
metically sealed with a laser-welded window cap under 
an argon atmosphere. 

Fig. 1. Optica/ system ot Ihe HP 
81000AS/BS Optical Power Split- 
ters with Ihe HP 81 520 A and HP 
81 521 B Optical Heads 


To give Ihe system Ihe flexibility to accept different 
heads, head specific data is included in the head and not 
in the mainframe. 

Each head contains an electrically erasable programma- 
ble read-only memory (EEPROM) lhal has a capacity of 
2K x 8 bits, although only 2K x 2 bits are used. The amount 
of memory used represents a compromise between short 
reading time and a minimum of head-mainframe connec- 
tions. This memory is split into 512 head bytes of 4 x 2 bits 

Fig. 8 shows the memory map. The control bytes include 
checksum bytes to detect errors and prevent data manipu- 
lation, a data valid byte, and a compatibility identifier in- 
dicating which mainframes Ihe head is compatible with. 
There are pointers to several start locations, and head spe- 
cific data consisting of Ihe following: 

■ Head specials: zero repetition times, chopper wail lime, 
shutter wait limes, etc. 

■ Range table: gain factors for the head and Ihe mainframe 

© Copr. 1949-1998 Hewlett-Packard Co. 


Head Byte 

X=0 or 1 

EEPROM Organization 

Numbers: o 


Control Bytes— »-|-«- Pointer - 
12 3 4 

. Head. , 

Specific Data 

« A Calibration Data »j 

850 nm 1300 nm 

1700 nm 











Fig. 8. Organization of the 
EEPROM containing head spe- 
cific data in the HP 81 520 A and 
81521B Optical Heads. 

m Calibration identifier: head serial number, date of cali- 
bration, calibration channel, etc. 

■ Calibration description: start wavelength, center wave- 
length, stop wavelength, number of calibration points, 
distance between two calibration points, field length 

■ Calibration data: individually measured gain factors de- 
pending on the wavelength (see article, page 8). 


Each head is calibrated in a homogeneous, parallel beam 
with a spot of 2.5-mm diameter at - 20 dBm. Spectral re- 
spnnsivity is measured with a high-resolution mono- 
chromator in 10-nm steps using a pyroelectric detector as 
a standard. 

Precision Optical Interface 

The HP 81520A and 81521 B Optical Heads are precision 
optical-to-electrical converters. To maintain their high 
overall accuracy, a precise optical interface had to be estab- 
lished between the detector surface and the point at which 


f~ Adapter 




Reflected Beam 
Direct Beam 

Fig. 9. Diagram ol the precision optical interlace lor the HP 
81520A and 8752)8 Optical Heads, showing detector tilt 

the power of the incoming light is to be measured. 

The first design goal was to gather onto the detector a 
certain (and constant with temperature, mechanical toler- 
ances, and other parameters) amount of power from the 
user's light emitting device, regardless of the nature of this 
light, whether it be spatial or fiber guided. In the latter 
case, the power has to be independent of fiber type and 
core diameter, and connector type, if any. With spatial light 
of unknown beam diameter, the "certain amount" men- 
t ioned above is 100% of all the light it is possible to capture. 

The second design goal was to minimize the interference 
between the measurement instrument and the system to 
be measured. This means that neither the optical head nor 
the optical interface should produce reflections back into 
the system or device under test. This system or device 
might be a light source in a parallel beam system on an 
optical bench, which is typically rather uncritical, or the 
end of a fiber connected to a laser diode, which, especially 
in a single-mode system, can be very sensitive. 

The third goal was to make accurate absolute calibration 
possible in spite of the wavelength dependence of the cho- 
sen detectors (Ge and Si pin diodes). 

The best trade-off between the desired sensitivity and 
versatility was a 5-mm diameter for the pin diode detectors. 
But once these semiconductor detector types were chosen, 
it was necessary to deal with their high reflection factor of 
about 20%. 








300 fim 

Fig. 1 0. Reflection-reducing inner surface design of the con- 
nector adapter 


© Copr. 1949-1998 Hewlett-Packard Co. 

In terms of calibration this is no problem, because the 
light that hits the detector surface and is reflected does not 
generate electron hole pairs inside the detector or a voltage 
at the output of the optical head, and can be calibrated out 
as long as the amount is constant and the light never returns 
to the detector surface again. For calibration with spatial 
light in a parallel beam system, we only had to prevent re- 
flections back into the light source or any optical compo- 
nent, which was easy to achieve by slightly tilting the de- 
tector, as can be seen in Fig. 9. 

When dealing with fibers, a category that includes the 
main applications of this instrument, some different re- 
quirements arise. First, as described in the article on page 
16. a collimating lens system between the fiber end and 
the detector surface is necessary. This is to ensure that 
nearly 100% of the light reaches the detector with reason- 
able distance between fiber and detector, and that the il- 
luminated spot on the detector surface is independent of 
fiber type and core diameter. The collimating lens system 
also ensures the same conditions — angle and spot diame- 
ter — in actual service as during calibration, either in-house 
or at the German PTB or other standards laboratory. 

Another requirement in a fiber optic system is that the 
optical system be closed, so that no ambient light reaches 
the detector. This includes the requirement that the 20% 
of the incident light that is reflected from the detector sur- 
face never return to the surface again in any nonnegligible 
amount. Surrounding parts have to be highly absorbing, 
and return to the detector must only be possible after sev- 
eral reflections to reduce the returning power to a negligible 
value. Coating with optical absorbing paint is one way to 
achieve this, and again, angling the detector is another. By 
these methods, not only is instrument accuracy improved 
by avoiding internal back reflections to the detector, but 
also reflection back into a connected fiber is totally avoided, 
so interference between the device under test and the mea- 
surement instrument is no longer a problem. 

The last remaining location of reflections is the front 
surface of the customer's connector, lis reflection coeffi- 
cient is unknown and totally out of the instrument de- 
signer's control. The tilting angle would have to be larger 
than 13° for a 3.5-mm diameter connector to ensure that 
reflected light misses the connector. Such a large angle is 
not practicable because it reduces the efficiency of the de- 
tector in terms of conversion factor (A/W). It also decreases 











Fig. 1 1 . Nonideal Iransmitlance of a coated lens Reflections 
are about 4% lor an uncoated lens, 0.5% lor coated. 

Fig. 12. D oss/b/e back reflection into the fiber 

the usable parallel beam diameter and thereby increases 
the sensitivity to mechanical tolerances. This is because: 

detf = d n cos Q. 

where d u is the diameter of the detector and u is the tilting 

With reasonable angles (<5°) the only way to avoid pos- 
sible inaccuracies is to shade the (unknown and reflecting) 
front surface of the connector with a specially shaped, ab- 
sorptive painted surface inside each connector adapter |see 
Fig. 10). 

Uncoated glass lenses usually reflect about 4% of the 
light at each surface (Fig. 11), so that an optical nonreflec- 
tive coating is necessary. This can reduce reflections well 
below 0.5%, so the power reflected back into the fiber is 
small enough not to be a problem. 

Fig. 12 shows a simple approach for estimating the per- 
centage of power that might return into the fiber. Let's 
assume the typical multimode case with a fiber-core diam- 
eter D = 50 (im and a distance between the fiber and the 
front surface of the lens system of s = 2.5 mm (not the 
focal length). Then the cone angle for possible back reflec- 
tions into the fiber core, according to the formula arctan 
(0.5D/2s), is about 0.28°. With the NA of this fiber being 
0.2, the 95% power radiation angle is 11.5°. Assuming a 
Gaussian power distribution, less than 3.5% of I hi; emitted 
power has to be taken into account. Since the lens only 
reflects 0.5% of this power, direct back reflection is really 
negligible (<200 ppm). 

Nevertheless, although negligible in terms of back reflec- 
tion, this reflection has to be taken into account for the 
measured power, because 0.5% reflection at each surface 
means 99% transmittance for a single lens, or 0.04-dB at- 
tenuation. This value is not negligible. Therefore, each lens 
is delivered with a calibration factor, which easily can be 
entered into the HP 81 52 A Optical Power Meter mainframe. 
Because of possible variations of the nonreflective coating, 
the transmittance of each lens is measured individually 
and is engraved as a calibration factor on the lens housing 
(for example. -0.04 dB for a single-mode HP 81010BL). 


Special thanks to Josef Becker, who evaluated the detec- 
tor devices, to Rainer Eggerl and Rudi Vozdeoky for the 
mechanical design, and to Milan Cedelnik for the excellent 
accuracy of the optical standards. 

© Copr. 1949-1998 Hewlett-Packard Co. 


A High-Precision Optical Connector for 
Optical Test and Instrumentation 

by Wilhelm Radermacher 

tant part or liber optic test instrumentation and sys- 
tems. The performance of fiber optic instruments 
and systems depends to a large extent on the performance 
of the optical connectors. In fact, in many cases the specifi- 
cations of instruments depend more on the connectors than 
on other factors. For example, the stability of a source de- 
pends more on the stability of the connection than on the 
stability of the emitting device if the circuitry is designed 

High-performance fiber optic connectors are used not 
only in a direct connector-to-connector scheme likea utility 
connector, but also as an interface to instrumentation con- 
taining bulk optical modules. What are the requirements 
for a high-performance optical connector for instrumenta- 
tion and why is a high-performance connector different 
from a utility connector? 

Utility connectors are used in large quantities in optical 
links. Therefore, they should be small and inexpensive, 
and have low insertion loss. It is an advantage if they are 
field-installable, especially in single-mode applications. 
High-performance connectors, on the other hand, require 
careful optical alignment which cannot be done in the field 
and is expensive. 

The key characteristics of an optical connector for in- 
strumentation are: 

High reliability and long lifetime. Utility connectors are 
specified for 500 to 1000 cycles. In an optical communi- 
cations link, this is more than sufficient, because connec- 
tions occur only during manufacturing, installation, and 
occasional maintenance. In instrumentation, especially 
in a production environment, connecting cycles occur 
very often, perhaps as many as 100 times per day on one 

1.0 t 
0 2 

Fiber In Contact 



20 40 
Temperature (*C) 

Fig. 1. Temperature sensitivity ol a connector without phys- 
ical contact At approximately 35°. the liber has expanded 
enough to make contact. 

instrument. Lifetimes of 500 or 1000 cycles are not ac- 
ceptable in this application. 

■ Repeatability of insertion loss on successive connecting 
cycles. Although the overall insertion loss of a connec- 
tion can be calibrated out in many applications, it is 
essential that a measurement that is repeated lead to the 
same result each time. 

■ Temperature stability of the insertion loss must be excel- 
lent to ensure repeatable measurements, especially dur- 
ing environmental testing of modules and components. 

■ Overall insertion loss is important to maintain the power 
budget of a measurement. Since 1-dB maximum inser- 
tion loss is a common specification for good single-mode 
connectors this is seldom a problem. 

Need for Physical Fiber Contact 

The insertion loss of an optical connector is caused by 
two main factors. One is misalignment of the fiber cores. 
The use of good high-quality parts with tight tolerances 
and good workmanship in the connector production pro- 
cess can reduce the influence of core misalignment to a 
minimum. The other main influence on the insertion loss 
is the spacing between the ends of the glass fibers if they 
are not in physical contact. At the wavelengths of coherent 
infrared light used for information transfer in the fiber, 
standing waves are caused by distances between fiber ends 
of more than one tenth of the wavelength, which can be 
as little as 100 nm. Minimal changes in this distance, which 
are mostly caused by ambient temperature changes or 
mechanical stress in the connectors, will cause large vari- 
ations in insertion loss (see Fig. 1). 

The only solution to this problem is a connector scheme 
that assures physical contact between the fiber ends under 
all circumstances. Only a physical-contact connector 
makes it possible to reach low insertion loss under the 











Loss (dB) 








0 to 60 C (dB) 

Repeatability (dB) 




Fig. 2. Specifications and measured values for the main per- 
formance characteristics ol Diamond* HMS- 1 01 HP connec- 
tors. Top number in each box is for single-mode operation, 
bottom number is lor multimode. 


© Copr. 1949-1998 Hewlett-Packard Co. 

following conditions: 

■ Changes of ambient temperature of the connection. This 
is very important if one connector is built into an instru- 
ment and exposed to the internal temperature rise caused 
by power dissipation. 

■ Variations of insertion loss between connector pairs if a 
large number of connector combinations are tested. 

A number of precautions are necessary to ensure physical 

■ The front face of the connector must be polished at a 
very precise 90° angle to the connector ferrule. 

■ The diameter of the mating surfaces must be small 
enough to ensure that resilient deformation of the mate- 
rial leads to physical contact, and must be large enough 
to ensure that no permanent deformation of the material 
occurs. A well-defined spring force can ensure physical 
contact and prevent permanent damage. 

■ A loose connector sleeve bushing is needed to prevent 
mechanical stresses in the connector ferrule and bushing 
when the connector is tightened. 

The Diamond' HMS-10/HP Connector 

The high-precision, high-performance connector used in 
the new family of fiber optic, instruments described in this 

Diamond HMS-10 HP vs. Reference 

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 
Insertion Loss (dB) 

Diamond HMS-10 HP vs Reference 

50 t 

Inserlion Loss (dB) 

Fig. 3. Typical distributions ol measured Diamond' HMS- 10/ 
HP insertion loss. 

issue is the Diamond HMS-10 HP. developed jointly by 
Diamond and HP. and assembled by HP. This connector 
meets all of the requirements discussed above. Also, since 
many of the instruments described in this issue are in- 
tended for use in single-mode as well as multimode appli- 
cations, it is necessary that the connector meet the require- 
ments of both applications and that single-mode and multi- 
mode connectors are fully compatible. The Diamond HMS- 
10VHP also meets this requirement. 

Both the connector ferrule and the connector bushing 
are made of tungsten carbide. Tungsten carbide is a very 
hard material and it was expected that in spite of the very 
tight tolerances involved it would wear less than any other 
material. Extensive lifetime tests of more than 10.000 cycles 
have shown that the connector has a very long lifetime. 
All connectors tested were well within specifications after 
these tests. 

Connector Characterization 

A major problem that every manufacturer of optical con- 
nectors faces is the definition of a worst-case specification 
of the insertion loss and other parameters. This is because 
only a small number of connectors are available and can 
be characterized at any time, since most connectors from 
previous production lots are at customer locations and not 
available. The insertion loss is a characteristic of a particu- 
lar connector pair, not of one connector only. The only 
way to guarantee quality is to make measurements on a 
limited sample base using statistical methods. 

The margin between the average measured insertion loss 
and the maximum specification depends on the variance 
and the desired probability of meeting the specifications. 
It was our goal to reach a probability of 99.7% that any 
HMS-10/HP connector would meet the specified insertion 
loss measured with any other HMS-10/HP connector. As- 
suming that the measured values fit a Gaussian distribu- 
tion, this means that desired maximum specification is 
equal to the average value plus three standard deviations. 

Fig. 2 shows the main performance characteristics of the 
HMS-10/HP connector, comparing measured values with 
maximum specifications. Fig. 3 shows the distribution of 
insertion loss measurements and compares it with a Gaus- 
sian distribution. Fig. 4 shows the temperature sensitivity 
of a typical HMS-10/HP connector. 




0C 60 C 

Fig. 4. Temperature sensitivity ol a typical Diamond' HMS- 
Wi 'HP connector The two curves are the results ol cycling 
up and down 

© Copr. 1949-1998 Hewlett-Packard Co. 



Fig. 5. Principal liber core alignment procedure. 

Core-to-Core Alignment 

Although the characteristics of single-mode optical fibers 
have been improved over the last few years, eccentricity 
of the core can lead to excess insertion loss unless a core-to- 
core alignment is done. The production process of the 
HMS-10/HP includes this alignment. 

The principal method of core alignment is as follows. 
The hole in the ferrule, which will eventually hold the 
fiber, has a diameter of nominally 135 fim. It is. therefore, 
quite easy to insert the fiber, which has a cladding diameter 
of nominally 125 fim. Another advantage of this connector 
is that only one type of ferrule is used for different cladding 
diameters (they can vary approximately ±2 /xm even within 
one batch of fibers). The fiber is centered in the hole by 
applying force on the front face of the connector with a 
special tool, which stamps a circular V-groove into the 
front face of the nickel-silver insert of the tungsten carbide 
ferrule (Fig. 5). After this, however, the core of the fiber 
may be slightly out of center. 

The eccentricity of the fiber core is then measured using 
a TV camera and visible light. An adjustment tool then 
stamps a larger V-groove into a segment of the previously 
stamped groove and moves the fiber such than the core is 
exactly centered (Fig. 6). 

The area inside the circular V-groove is later polished 
and used as the contact area. This ensures a reproducible 
diameter and consequently a reproducible size of the con- 
tact area. 




Fig. 6. Fine liber core alignment procedure. 


© Copr. 1949-1998 Hewlett-Packard Co. 

Design Approach for a Programmable 
Optical Attenuator 

by Bernd Maisenbacher. Siegmar Schmidt, and Michael Schlicker 

peater spacing, requirements on optical fiber 
transmission systems have become more strin- 
gent and tolerances have become smaller. Better test equip- 
ment, especially on the optical side, is needed to measure 
performance and ensure that specifications are met. 

At the same time, the demand for high-performance 
transmission equipment grows rapidly. Production capa- 
cities must be built upand test procedures generated. Short- 
er testing times are needed to keep up with demand. One 
strategy for achieving these goals has been the increasing 
use of computer control and automation. Computer-aided 
testing and HP-IB (IEEE 488/IEC 625) programmable test 
equipment are solutions to the problem of long test times, 

In general, electrical test instrumentation can be operated 
under computer control. This trend, however, has not yet 
penetrated the area of optical measurement equipment to 
the same extent. In the case of optical attenuators, there is 
quite a variety of manual attenuators available, but only a 
handful of programmable models for single-mode applica- 
tions. This availability is important for high-performance 
transmission systems, which are now almost exclusively 
implemented with single-mode technology. 

The new HP 815811 Option 002 Optical Attenuator (Fig. 
1) is believed to be the first variable optical attenuator 
suitable for both single-mode and multimode applications. 
It handles all fiber core diameters from 9 to 100 ftm and 
is calibrated at both 1300 nm and 1550 nm. For other 
wavelengths between 1200 nm and 1650 nm. it automati- 
cally calculates adjustment factors. 

The HP 81 58B Option 001 Optical Attenuator is designed 
for the wavelength range from 600 to 1200 nm and is cali- 

Fig. 1. The programmable HP 8158B Option 002 Optical 
Attenuator is designed lor multimode and single-mode appli- 
cations and is calibrated at 1300 and 1550 nm Another 
model, the HP 8158B Option 001, is calibrated at 700, 850. 
and 950 nm. 

brated at 850 nm It handles fiber core diameters from 50 
to 100 /xm. 

The maximum attenuation range of both attenuators is 
60 dB. Resolution is 0.01 dB and typical insertion loss is 
1.0 dB for multimode and 2.0 dB for single-mode. Precise 
calibration and a digital display ensure repeatability within 
0.04 dB over a temperature range of 0 to 55°C. 

The HP 8158B has high-precision Diamond* HMS-10 HP 
connectors (see article, page 28) and is fully programmable 
via the HP-IB. 

Optical System 

Commercially available fiber optic attenuators, both pro- 
grammable and manual, use a range of techniques fur 
achieving the optical attenuation. Many use a technique 
based on some angular, lateral, or axial displacement be- 
tween the two fiber ends. Others use some sort of filter or 
dispersive element. The HP 8158B uses two filter wheels 
and various bulk optical components. 

As shown in Fig. 2, a high-precision optical system col- 
limates the light of the input to an expanded parallel beam 
and refocuses it onto the output fiber. Fixed filters, a reflec- 
tion prism, and a continuously variable 0-to-10-dB circular 
filter are located between the two lens systems. Fig. 3 shows 
the optical block. 

There are five fixed filters in 10-dB steps from 10 to 50 
ilB lo cover the attenuation range up to 60 dB. These filters 
are coated with a metallic neutral density layer, which 
reduces the wavelength dependency of the optical density 
of the filters. 

All optical surfaces are coated with multilayer broadband 
antireflection coatings lo prevent interference modulation 
and to achieve low insertion loss. The filters are angled to 
the optical axis to prevent back reflection into the fibers. 
The expanded-beam technology with attenuation fillers en- 
sures that there is no mode selectivity in the case of mul- 
timode fiber applications, and reduces the sensitivity of 
the instrument's performance to dust particles. In addition, 
the attenuator optical blocks are produced and mechani- 
cally adjusted in a flow box and are sealed under this same 
condition to ensure a high level of dust-free air inside the 

Both filters are aligned in the optical path, and each has 
its own drive motor and optical encoder. 

The digital design of the motor control circuits guaran- 
tees fast realignment (or any change in attenuation setting. 
Realignment time is typically 50 ms, so short measurement 
cycles can be achieved with automatic test systems. 

Automated Calibration with Wavelength Correction 

Linearity of attenuation characteristics is achieved by 


© Copr. 1949-1998 Hewlett-Packard Co. 

individual calibration in production. Using an HP 9000 
Series 200 Computer, an automatic calibration program is 
run on each unit. The attenuation of the fixed filters for a 
given setting is measured at 1300 and 1550 nm for the HP 
8158B Option 002 and at 700, 850, and 950 nm for the HP 
8158B Option 001. Within the same measurement cycle, the 
dynamic attenuation of the circular filter is also measured. 

The measured data is processed by the computer and 
transferred via the HP-IB to an EEPROM in the attenuator. 
Any errors caused by the manufacturing tolerances of the 
optical filters are calibrated out. In addition, the wave- 
length characteristics for the metallic filter coatings are 
stored in the EEPROM. This permits the microprocessor 
to calculate correction factors for each wavelength/attenu- 
ation combination. As a result, automatic correction is per- 
formed for the wavelength and attenuation settings made 
by the user. 

Connector Considerations 

In a sense, this approach to designing an optical at- 
tenuator can be described as an optical feedthrough connec- 
tor with attenuating filters located along the optical path. 
With single-mode operation selected, the optical coupling 
is approximately characterized by a Gaussian distribution 

Fig. 4 shows how the coupling loss varies with misalign- 
ment of two fibers. To minimize connector loss, the fiber 
ends must be exactly positioned in the X, Y, and Z axes. 
The HP 81 58B uses connector adapters that can be precisely 
adjusted at the focal point of the lens system with better 
than 0.2-/xm repeatability. To maintain this precision over 
the instrument's operating life, a hard metal bushing and 
a hard metal Diamond HMS-10/HP connector are used for 
each part. This approach minimizes wear and resultant 

Loss resulting from off-axis connector coupling can be 
evaluated quantitatively. Two Gaussian beams can be 
coupled without loss when they have identical beam 
shapes. This means that the propagation axes must coincide 
and that the spot radii must be the same. These conditions 
must be fulfilled in all directions perpendicular to the prop- 

Fig. 3. The optical block of the HP 8758S Optical Attenuator 
has two connectors, each with a lens system. 

agation axis. If these conditions are not fulfilled, the loss 
can be calculated as follows: 1,2 

v kx =ioio g { k,-exp f-k 2 [^ir_4_ + -4-1+ 

2 Wot* W ( ,2x 

^(wL + wl)-^1)} 

8 V. J W £ lx J J J 

K _ 2W olx W o2x 

I '., \ '. 4AZ; 

V< W oi* + W 2 2s ) 2 + 

W? x = W 2 lx (1 + S x ) 


© Copr. 1949-1998 Hewlett-Packard Co. 

_ 217 

S« = 2 


W olx 

w o2> 


spot radius after optical imaging in the X direction 
spot radius of the fiber in the X direction 
separation of the spot from the fiber 
misalignment of the optical axes of the two fibers 
(X component) 

angular misalignment between the fibers in the 
X direction. 

The above formulas also apply to the Y direction, so that 
the total loss then can be determined bv the formula: 

= v kl( + v ky . 


Both servo loops are closed in one of the two micropro- 
cessors of the HP 8158B (Fig. 5). The other microprocessor 
controls the display and the keyboard, handles the HP-IB, 
and calculates and transmits the positions for the motor 
controller via the device bus. 

Microprocessor. The heart of the digital motor controller 
is an 8-bit CMOS microprocessor with 128-byte internal 
RAM and a 16-bit internal timer. The firmware is resident 
in a standard EPROM. 

Pulse Width Modulator and Motor Driver. To extract the 

3 4 5 6 7 
Fiber Otlset (;.m) 

Fig. 4. Plot ol coupling loss versus misalignment ol two tioers 
follows a Gaussian distnbution 

best possible performance at minimum cost, the digitaJ-to- 
analog converter for the motor driver is a simple pulse 
width modulator and the motor driver itself consists of an 
inexpensive transistor full bridge. Therefore, the HP 8158B 
needs only a unipolar 15V power supply for the motors 
and a 5V supply for the digital logic. Fig. 6 shows the 
design of the motor driver circuitry. 

The control processor sets the PVVM register. A 7-bit 
comparator compares the content of the PWM register with 
a 7-bit free-running counter. Whenever the PWM register 
value is higher than the value of the free-running counter 

Device Bus Interface 

Pulse Width Modulator 

Device Bus 
(Data Bus 

Data Bus 














CPU and 

Inlt'f nnl 







4 MHj 

Fig. 5. HP 8158B digital motor controller diagram. 


© Copr. 1949-1998 Hewlett-Packard Co. 

(he Iransistor bridge is enabled. The high-order bit of the 
PWM register determines the turn direction for the motors. 
The period of the PWM is 31.25 kHz, high enough to pre- 
vent audible noise. 

Position Sensing System. The optical filter wheels are fixed 
on the front of the motor shaft, so no mismatch between 
the filter position and the position sensing system is pos- 
sible. For the variable filter wheel, a 1024-track incremental 
encoder is used, and for the fixed filter wheel, a 512-track 
incremental encoder is used. Both encoders are standard 
HP products. To get better resolution, the encoders are 
driven in quadrature mode. In this mode, the effective res- 
olution is 4096 (2048| tracks per revolution. This technique 
is explained in the HP Optoelectronic Designer's Catalog. 
The complete logic for the encoder driver including the 
microprocessor interface is integrated in a gate-array-based 

Device Bus Interface. The interface between the two micro- 
processors consists of three 8-bit latches. Two latches are 
used by the main microprocessor to transfer a command 
and if necessary a bit value to the motor controller. The 
third latch indicates the status of the motor controller and 
is readable by the main microprocessor. It is also used to 
acknowledge receipt of a command on the device bus. The 
latches are software debounced. 


The software is designed to reduce the hardware require- 
ments wherever possible. For this reason, a very simple 
device bus interface and an inexpensive motor driver inter- 
face were possible. The test hardware was also reduced by 
intelligent software design. 

Self-Test. The self-test is divided into two sections. The 
first section verifies the internal RAM after a power-on 
reset. If this test fails, the motor controller tries to go into 
a defined EXIT condition. Otherwise, the second section of 
the self-test checks the internal timer of the CPU. the timer 
compare register of the CPU, the external EPROM, the 
PWM. motor, and encoder of the fixed filter wheel, the 
PWM. motor, and encoder of the variable filter wheel, and 
the initial position logic: of both encoders. 

II an error occurs during the self-test, the motor controller 

indicates the error and transmits the errors by way of the 
status latch to the main microprocessor to aid effective 
troubleshooting, In this case the motors are disabled to 
prevent damage to the optical system. 
Initialization of the Position Sensing System. Since both 
encoders are incremental encoders, the HP 8158B does not 
know its position at power-on. Therefore, the software en- 
ables the itiitial channel of the encoders, and turns the 
motors for a maximum of one revolution. When the index 
pulse occurs, the 16-bit position counter is cleared and an 
interrupt tells the motor controller that the encoders are 

Synchronization. After completion of the self-test, or after 
an error, the motor controller waits for synchronization 
with the main processor. The main processor starts the 
synchronization with a special POWER ON keyword in the 
command register and the value register. The motor con- 
troller acknowledges receipt of this keyword by a specific 
flag combination in the status register. The main processor 
then sends the TRANSFER FINISH command, which is 
answered by the READY flag from the motor cc-ntroller. After 
this synchronization procedure, the main processor can 
transfer commands and values to the motor controller. 
Transfer of a New Command. A command transfer is simi- 
lar to synchronization. Normally, the main processor sends 
TRANSFER FINISH to the command register. The motor con- 
troller polls the contents of the command register, waiting 
for another command. When it finds a stable command in 
the command register, the motor controller fetches the 
value register. To acknowledge receipt of the command, it 
sends BUSY by way of the status register. The main proces- 
sor then sends TRANSFER FINISH again and the motor con- 
troller sets the BUSY flag to READY when it completes execu- 
tion of the command. 

Motor Control Algorithm 

Since the motor has practically no friction, the device 
under control is an extremely unstable third-order system. 
The transfer function is given by equation 1. 

F(s) = 

s(s + io 1 )(s + a» 2 ) ' 

+ 15V 


'pWM - 

PWM Register Value 
= 0.25 fis 


-t=32 ,iS- 


4 MHz 


A A B 



f=31.25 kHz 














Right Turn 



Lett Turn 

Fig. 6. Motor drive circuitry. 


© Copr. 1949-1998 Hewlett-Packard Co. 



PD Controller 



Implemented | 

Btidge Driver 

DC 1024 (512) Filter 
Motor Tracks Wheel 

— *0 — ^ 

1 Real Position 12(11) Bit 


3 Channels 


Irom Incremental Encoder 

Fig. 7. Motor servo loop 

where G is the gain, to, is the mechanical pole frequency 
(2 to 8 Hz), and u>, is the electrical pole frequency (16.5 

Different control algorithms were investigated to find a 
solution that could guarantee a step accuracy of ±1 step 
of the encoders, a short settling time, and an overshoot less 
than 1%. One of these algorithms, the deadbeat controller.' 1 
is very fast and simple, but extremely sensitive to parameter 
changes. If the temperature changes over a 50" range, the 
position error resulting from the corresponding change in 
the motor resistance is 25%. 

Another algorithm investigated, the state-variable con- 
troller,'' is an accurate algorithm, but requires one division, 
six multiplications, and four additions per sample and 
motor. This needs higher performance than an 8-bit CPU 
can offer. 

A third algorithm, the standard PD controller, is an accu- 
rate solution.'' but it tends towards some instabilities on a 
third-order system, such as large overshoot and long set- 
tling times. For the CPU it is not a problem, requiring only 
two multiplications and one addition per sample and 
motor. The HP 8158B uses a standard PD controller, but 
the software changes the controller gain depending on the 
actual position. Equation 2 shows the transfer function of 
an ideal PD controller, while equation 3 gives the corre- 
sponding uutput of a PD controller for a time discrete solu- 

F R (s) = C R (l + s) (2) 

y H |n) = G K x„ + u, H X "" X " 1 (3| 

* Rumple 


F K (s) = PD controller transfer function 
s = Laplace operator 
G K = Gain of PD controller 
(i) R = Compensation pole frequency of PD con- 

y K (n) = Output ofPDcontrolleratsamplingtimen 
x„ = InputofPDcontroIleratsamplingtimen 
= Delta position. 

Every 1.048 ms, the CPU timer produces a timer inter- 
rupt. On each interrupt, the interrupt service routine TIMER 
alternately calls the motor controller routine for the vari- 
able filler wheel or the motor controller routine for the 
fixed filter wheel. These motor controller routines. MOTOR 
motor positions and calculate the delta position by forming 

the difference between the target position and the real po- 
sition. Depending on these delta positions, the control al- 
gorithm selects the PD controller gain. For small delta po- 
sitions, the PD controller shows aperiodic behavior, so the 
system response is free of overshoot. Otherwise, the con- 
troller works with low velocity feedback. This would nor- 
mally produce a large overshoot but a fast rise time. If the 
delta position reaches the band with the aperiodic be- 
havior, the higher velocity feedback brakes the motor and 
prevents a large overshoot. 

Fig. 7 shows the schematic of the motor servo loop. 

To minimize the dead time between sampling and 
stimulating the system, the motor controller precalculates 
the terms involving only the preceding sample. 


We wish to thank Hans loachim Ziegler and Erhard |anz 
from Boblingen Instrument Division materials engineering 
for their help in the specification and introduction of the 
encoders and motors of the HP 8158B. Also, we thank Peter 
Klement. materials engineer for all of the optical compo- 
nents. Rudolf Vozdecky. who was responsible for mechan- 
ical design of the HP 8158B. and Markus Dieckmann, pro- 
duction engineer, who helped us with the calibration soft- 


1 . M. Saruwatari and K. Nowata. "Semiconductor I jiser to Single- 
Mode Fiber Coupler." Applied Optics. Vol. 18. 1979, p. 1847. 

2. P. Lecoy and H. Richler, Benndiungder Transmission von e/lip- 
lischen Mikni/insen zur Oplimierung der Kopplung zivisc.hen 
Hcdb/eitorlnser and Monomodefaser. Deutsche and Bundesposl 
FTZ. 1979, 

:t. H.P. Becker. "Entwurf und Realisierung digitaler Regler mil 
Mikroprozessoren." Clekfrnnik, no. 5. March 9, 1684. 

4. O. Foellinger. Lineare Abtuslsysterne, Oldenburg Verlag. 1 974. 

5. D.C. Tribolet. K.A. Regas. and T.J. Halpenny. 'The HP 75S0A 
X-Y Servo; State-of-lhe-Arl Performance on a Budgel," Hewlett- 
Packard faurnal, Vol. 36, no. 4. April 1985. 


© Copr. 1949-1998 Hewlett-Packard Co. 

A Programmable Fiber Optic Switch 

by Michael Fleischer-Reumann 

TThe HP 8159A Optical Switch is a fiber optic switch 
designed to simplify measurement systems in pro- 
duction and R&D environments. The main feature 
of the switch is its good repeatability, which means that 
once its three optical paths are characterized in terms of 
insertion loss, many reliable measurements can be per- 
formed without having to do a recalibration cycle. 

The HP 8159A is designed for applications at 850 and 
1300 nm with 50-/xm graded-index fibers. It has two inputs, 
A and B, and two outputs, C and D. Three optical paths are 
possible, as shown symbolically on the front panel (Fig. 
1): AC, BO, or AD. 

The switch has repeatability of 0.2 dB, low insertion loss. 
HP-IB programmability. and high-precision Diamond" 
HMS-10/HP connectors that can be taken apart for cleaning. 


ia n < 


Source A 


nr oiss 



Source B 


Fig. 1 . The HP 81 59 A Optical Switch is designed tor applica- 
tions at 850 and 1300 nm with 50-u/n graded-index libers. 


Fig. 3. Testing a receiving element with dillerent sources 

modules. The manufacturer's own transmitter is driven 
with the output of a BER (bit error rate) test set, and the 
electrical output of the receiver under test is the input for 
(he BER test set. The purpose of the programmable at- 
tenuator is to find the power level where the BER begins 
to increase over specified limits. 

Since transmitter modules often are not very stable in 
terms of output power (not temperature stabilized), the 
source might be drifting. Therefore, after finding the attenu- 
ation setting where the BER begins to increase, output 
power of the transmitter is to be measured. This is easily 
done by simply switching the transmitter output to a power 
meter (e.g., HP 8152A) using the HP 8159A Optical Switch. 
In this application, a power splitter like the HP 81000BS 
would also be helpful. 

In Fig. 3, the reaction of a receiving element (for example, 
a pin diode or a receiver module) to different types of 
sources (a laser and an LED or sources of different wave- 
lengths) is being tested. The HP 8159A is used to switch 
the two sources. 

Fig. 4 shows a test setup for a completely optical device 
(for example, an attenuator). The HP 8159A makes it easy 
to measure power with and without this device in the op- 
tical path. 

Typical Applications 

Fig. 2 shows a test setup that might be used by a manufac- 
turer of fiber optic links to test the sensitivity of receiver 

Bit Error Rate 
Test Set 


HP 81S9A 


HP 81588 





(To Take Reference) 



HP 81521B 

HP 8152A 

Power Meter 


Power Meter 

Fig. 4. Test setup lor an optical device such as an attenuator 


The mechanical design was done by Rudi Vozdecky. 
Michael Coder wrote the firmware and Wilhelm Rader- 
macher did the hardware design. Special thanks to Erhard 
Janz from materials engineering, for his help with the op- 
tical switching module. 

Fig. 2. Receiver sensitivity test setup. 


© Copr. 1949-1998 Hewlett-Packard Co. 


12 ~ Power Meter Firmware; 

72~ Optical Heads ' 


Michael Godei 

A native of Wese* am Rhein 

0 received his Diptom In- 
mieur in 1979 He pmed 
; BbWngen Instruments 
vision tne same year ana 
I nas developed HP-IB solt- 
I ware lor me HP 8152A 
I Power Meier . the HP 81 58A 
Attenuator, and tne HP 8154B LED Source He's 
coauthor of a 1985 HP Journal article on me HP 
B151A Power Meter Michael is married and lives 
in Gaulelden He enjoys skiing, photography, and 

Johannes Hunmg 

| WWi HP S'Ace 1982 Hans 
rung nas cont/ibuted to 
oe-<e*opmer-t of the hp 
it APower Meter anc the 
8150A Optcai S<jnal 
I Source He also developed 
. ^^^m* a pulseu-ta:-'. 
j/jW J -A ~HP8l512AOp- 

■H^BHBBHB he earned his Diplc 
gen«ur Irom the Renisch Westfalisch Techn<sche 
Hochschule at Aachen and then pined HP Hans 
ana his wile ana two chiiaren are residents ol Her- 
renberg He likes bicycling with his lamily, playing 
chess, and llymg gliders 

4 — New Fiber Optic Family I 
6 ZZ LED Sources ' 

Michael Fleischer-Reumann 

I With HP since 1980 
^mj^^^^^^^, Michael Fleischer-Reumann 
is an R&D project manager 
— at the Boblingen Instru- 

ments Division He was 
I proiect manager or proiecl 
I leader tor several ol the 
liber optic products de- 
I scribed in this issue and 
I continues 10 manage work 
on other liber optic products Earlier he contrib- 
uted to hardware design lor the HP 8112A Pulse 
Generator He's named Inventor on a patent relaied 
lo pulBe generator timing and is author or coaulhor 
ol several HP Journal amc les Michael was born in 
Essen and attendea the Runt University of 
Bochum. from which he received his Diplom In- 
gemeur He and his wile are residents ol Nulringen 
and ne teaches electronics at a Stuttgart college 
His outside interests include backpacking, hiking 
kayaking, playing guitar, ana almost all kinds ol 

Bernhard Flade 

Bernhard Flade firsi 
— worked at HP from 1 967 to 
1970 as an apprentice 
electrician He returned to 
the company in 1981 after 
earning a degree in com- 
munication engineering 
from the University ol 
Karlsruhe He has contrib- 
uted to firmware develop- 
ment lor the HP 81 50A Optical Signal Source, the 
HP 8 1 52A Power Meter, the HP8154A LED Source 
and the HP 8158A Attenuator He's coauthor of a 
January 1985 HP Journal article Born in Dresden, 
Bernhard is a resident of Boblingen. is married, and 
has Iwo children He's a member ol the German 
and the international associations ol ski instructors 
and a member of a German technical reiiel organi- 
zation When he s not skiing, he may be lound rid- 
ing Ins Italian motorcycle or working with his home 

16 — Detectors: 

Siegmar Schmidt 

| With HP since 1984 Sieg- 
1 mar Schmidt has a degree 
m physics from the Fned- 
nch Schiller University at 
Jena He s an optics 
specialist and worked in 
that held before coming to 
I HP He contributed to the 
development ol the HP 
8158A/B Attenuators and 
Ihe HP BlOOOAS'BS Optical Power Splitters He's 
Ihe author ol a Laser Focus article on the design 
ollheHP8158B Bom in Jena. Thunngen, Siegmar 
and his wile and Iwo sons live in Wildberg in the 
Black Forest region His hobbies include lishing 
wmdsurling and swimming 

8 = Optical Power Meter : 

Horst Schweikardt 

I An R&D proiect leader at 
I the Boblingen Instruments 
I Division. Horst Schweikardl 
I has been with HP since 
1 1972 In addition to leading 
| work on Ihe HP 8152A Op- 
tical Power Meter and Ihe 
HP 81521B Optical Head, 
i he has headed proiect 
iHWMHH teams for the hp 214B 
Pulse Generator and other products He received 
his Diplom Ingenieur in electrical engineering Irom 
Ihe University of Stuttgart in 1972 A native ol 
Baden-Wurttemberg. he was born in Schwabisch 
Gmund He and his wile and daughter now live In 
Herrenberg His hobbies Include bowling, moun- 
taineering, and photography 

Joset Becker 

With HP since 1979. Jo 
_ Becker Oesignea the de- 
| tecior heads lor the HP 
8152A Power Meter and 
I was responsible lor the de- 
ugnoflheHP8180A Data 
p Generator output ampli- 
ers Before coming to HP 
e was a biomedical en- 
I gmeer at the University ot 
Stuttgart He holds a 1 969 Diplom Ingenieur m elec- 
trical engineering from the University ot Karlsruhe 
His work on a programmable pulse generator is the 
subiect ol a pateni and he's the author of over 25 
papers on electronic circuits and automated 
analyses lot clinical laboratories He also lectures 
on electrical engineering at a local university Jo 
was bom in Hoengen, Aachen, lives in Sindei- 
Imgen, and has three sons He enpys aslronomy 
and amateur radio 

Emmerich Muller 

With HP since 1981. 
Emmerich Muller is respon- 
sible for hybrid and liber 
''f optic prototypes at the 
^ Boblingen Instruments Di- 
vision He was bom in Vil- 
li lingen and attended the En- 
—1 gmeenng School In 

Furtwangen (Black Forest), 
from which he received his 
Diplom Ingenieur (FH) He s a coauthor of several 
HP Journal articles ana a member ol the Interna- 
tional Society lor Hybrid Microelectronics Emmerich 
ana his wile and two children live in Gartringen His 
outside interests include cycling and all kinds ol 

© Copr. 1949-1998 Hewlett-Packard Co. 


28 Optical Connector \ 

Wilhelm Radermacher 

Wilh HP's Bbrjlmgen instru- 
ment Division since 1985 
Wilhelrn Radermacher 
worked on Ihe Diamond" 
HMS 1 0/HP Connector and 
has contributed lo ihe de- 
velopment ol Ihe HP 8 1 54 B 
LED Source and the HP 
J 8159A Optical Switch. Be- 
I IB^^'^^B lore coming to HP he de- 
signed microwave integrated circuits lor satellite 
applications His Diplom Ingenieur (FH) was 
awarded m 1 978 by the Julich Engineering School 
and he earned his Diplom Ingenieur in electronic 
communications from Ihe Rheimsch Weslfalisch 
Technische Hochschule at Aachen in 1982. 
Wilhelm was born near Cologne and lives in Smdel- 
Imgen Newly married, he enjoys sailing 

31 "Optical Attenuator : 

Michael Schlicker 

Michael Schlicker was born near Saarbriicken and 
studied at the Engineering School ol Saarland and 
at the University ol Metz in France. He |Oined HP 
m 1 982 and alter working as a materials engineer, 
he contributed to I he development ol the HP 8 1 58B 
Optical Attenuator and other fiber optic products. 
He has recently left the company Michael and his 
wife and Iwo children live in Wildberg, on Ihe edge 
of ihe Black Forest. He spends much of his spare 
time building a house 

Bernd Maisenbacher 

I Bernd Maisenbacher was 
I proiecl leader lor Ihe HP 
8158B Atlenuator and is 
now a project manager with 
I responsibility lor fiber optic 
laboratory instrumentation, 
including cable and adapter 
I accessories, at the Bob- 
i lingen Instruments Division 
I He received his Diplom In- 
genieur from Ihe University of Stuttgart and has been 
with HP since 1981 He's named inventor on Iwo 
patent applications related to fiber optics Born in 
Pforzheim, Bernd is now a resident ol Schdmberg 
in Ihe Black Forest region. His outside interests in- 
clude skiing, dancing, and stamp collecting 

36 — Fiber Optic Switch: 

Michael Fleischer-Reumann 

Author's biography appears elsewhere in this 
seel ion 

39 De-Embedding Active Devices. 

Louis J. Salz 

Lou Salz was born in Water- 
loo, Iowa and completed 
work lot his BS degree in 
computer engineering Irom 
Iowa Slate University in 
1981 He started at HP Ihe 
same year lor Ihe Micro- 
wave Technology Division. 
Recently he has developed 
computer-aided test soft- 

ware tor HP network analyzers. He has also written 
software for o'her computer-aided testing tools to 
support R&D and manufacturing efforts His pro- 
fessional interests include real-time operating sys- 
tems and automated test equipment A resident ol 
Santa Rosa. California, he enioys bicycling, back- 
packing, and skiing 

Glenn E. Elmore 

I Glenn Elmore is an R&D 
engineer at the Network 
Measurements Division 
who has been with HP 
since 1972 He has worked 
on various test sets for the 
HP 8510 family of network 
analyzers and on the HP 
8620 and HP 8350 larnilies 
of sweep oscillaiors His 
work on test set tirmware lor Ihe HP 851 OA is the 
subject ol a patent and he is coauthor ol a 1982 
HP Journal article on the HP 83500 series plug-ms 
lor Ihe HP 8350A Sweep Oscillator. A California na- 
tive, Glenn was born m Sebaslopol and now lives 
in nearby Sanla Rosa He's married and has two 
children. An amateur radio operator (N6GN), his 
interests include packel and microwave radio 



© Copr. 1949-1998 Hewlett-Packard Co. 

Quality Microwave Measurement of 
Packaged Active Devices 

A special fixture, the HP 8510 Microwave Network Analyzer, 
and the concept of de-embedding provide a solution to a 
formerly difficult measurement problem. 

by Glenn E. Elmore and Louis J. Salz 

COMPONENTS AND DEVICES become more dif- 
ficult to measure directly at microwave frequencies. 
Although automatic network analyzers such as the 
HP 851U can make direct measurements when used with 
calibration standards that have the same connector type as 
the device under test, many times the device cannot be 
connected directly to the calibration plane of the analyzer. 
This is the case with packaged transistors. In the past there 
has been no uniform way to measure such devices to pro- 
vide useful and accurate data that can be verified by mea- 
surements made at different times and places by different 
operators. As a result, circuits designed using measured 
data did not always operate as expected. This was for two 
primary reasons: the measurements were not sufficiently 
accurate because of instrumentation errors or limitations, 
and the device environments (the fixtures) were not alike 
and were often different from the application. 

With the advent of the HP 8510 Microwave Network 
Analyzer, instrumentation to make accurate and rapid 
error-corrected measurements is available. The need for a 
standard fixture in which to make packaged transistor mea- 
surements lor a variety of package styles has become appar- 
ent. Accurate calibration of such a fixture is necessary to 
provide repeatable and accurate device data. 

The HP 85014A Active Device Measurements Pac re- 
sponds to this need by adapting a suitable fixture for op- 
timum operation with the HP 8510 and providing data 
output formats and archiving capabilities to meet the needs 
of those involved in measuring and using packaged active 
devices in the microwave region. The fixture allows the 
measurement of the two most common package styles 
through the use of specific fixture inserts for each style of 
p.ii kage. Ai i urate data is provided through tbe 1186 "I the 
precision coaxial calibration standards available with the 
HP 8510. together with careful fixture characterization and 
a technique called de-embedding. De-embedding combines 
the information a network analyzer obtains from measuring 
known standards at the analyzer's calibration plane with 
known fixture characteristics to allow fully error-corrected 
measurements right at the desired measurement location 
inside the fixture. 

The HP 85014A Active Device Measurements Pac is a soft- 
ware and hardware applications product designed to har- 
ness the speed, power, and accuracy of the HP 8510 (or 
the measurement of active devices mounted in the HP 
85041A Transistor Test Fixture, Fig. 1 (next page] shows 
the active device measurement system block diagram. 


Conventional calibration standards could be used to 
allow fixtured device measurements with the HP 8510. For 
packaged device measurement, such standards would have 
to be device-like, that is, small packaged opens, shorts, and 
loads for each type of device to be measured. They should 
be of quality similar to available coaxial standards. Such 
standards would cause additional expense and require par- 
ticular care to use and protect if accuracy were to be main- 
tained. To avoid these difficulties, a different approach 
was taken. 

To allow the HP 8510 to make error-corrected measure- 
ments without using in-fixture calibration standards, a pro- 
cess called de-embedding is provided. After the HP 
8510 is calibrated with precision coaxial standards, the 
software modifies the HP 8510 error-correction process. The 
analyzer then operates in a normal manner, as if a conven- 
tional calibration had been performed at the desired device 
measurement plane within the fixture. 

After measurement, the software acquires the fully error- 
corrected data from the analyzer and provides a variety of 
output formats that are specific to active device measure- 
ment and design. The resulting data can also be archived 
in formats compatible with computer circuit analysis pro- 
grams. This ability provides a vital connection between 
real device measurement and application product design 
or transistor fabrication process engineering. Examples of 
some of the available output formats are shown in Fig. 2. 

Distributed Processing 

Like many of today's newer instruments, the HP 8510 
performs a complex measurement process internally. It is 
therefore necessary to use new methods to interface the 
complex system software within the instrument with addi- 
tional external software. The approach taken with the HP 
85014A Active Device Measurement Pac was to use the 
capabilities of the HP 8510 whenever possible. This ap- 
proach has resulted in a number of advantages. 

The first and most obvious advantage is vastly improved 
performance of the overall system. By taking advantage of 
the HP 8510's ability to perform high-speed measurements 
and error correction in real time, the HP 8501 4A software 
is able to eliminate the need for any postprocessing of the 
data before fully corrected s-parameter data can be dis- 
played. Tin; HP 85D14A software only jumps into the mea- 
surement process when the instrument cannot perform the 
needed operations. The modification of the HP 85 Ill's in- 
ternal error correction is an example of this approach. 

© Copr. 1949-1998 Hewlett-Packard Co. 


HP 2225A 
ThinkJet Printer 

HP 7470A 
Graphics Plotter 

SSfiHSP - r .: ii ; ; -; jgBH 
1 JQaaoaDaaQaafi' B □□□□ 
aao a a a o 
t ' ' ,j eg o a a □ 

HP 8340A 




Supply Output 

HP 871 7B 
Bias Supply 



Port 1 

RF Input 

Port 2 

HP 8514A15A 
Test Set 

Port 1 

Port 2 


HP 85132A 
Test Port Return 
Cable (7-mm) 

Fig. 1. Block diagram ol a typical system for transistor measurements using the HP 85041 A 
Transistor Test Fixture and the HP 8501 4 A Active Device Measurements Pac 


© Copr. 1949-1998 Hewlett-Packard Co. 

The HP 85014A uses the HP 8510's internal error-correc- 
tion algorithms to measure coaxial standards and produce 
an error-correction data set that removes the systematic 
errors in the HP 8510. The HP 8501 4 A software then inter- 
venes by reading these error terms from the HP 8510 and 
modifying them to include modeled errors introduced by 
the HP 85041 A test fixture. These modified error terms are 
then returned to the HP 8510 so that the analyzer can per- 
form a fully error-corrected measurement using its built-in 
algorithms and high-speed processing capabilities. The HP 
8501 4 A software from this point on only directs the mea- 
surement process within the HP 8510 and reads data from 
the instrument after the data has been corrected. 

A second advantage of the distributed processing ap- 
proach is the reduced complexity of the HP 85014A soft- 
ware package relative to what would have resulted if more 
of the measurement process had been included in the ex- 
ternal computer. All measurement parameters are set 
within the measurement package and sent to the HP 8510 
to be checked for validity. They are then adjusted to values 
the HP 8510 can use. 

Another benefit is consistency between the HP 8510 and 
the HP 85014A. Since the measurement pac uses much of 
the analyzer's system software directly, the feature set of 
the particular HP 8510 configuration in use is directly ap- 
parent to the user of the measurement pac. 

Measurement Pac Contributions 

Contributions and benefits of the HP 85014A Active De- 

vice Measurement Pac include the following: 

■ Improved HP 85041 A Test Fixture designed for more 
accurate, repeatable. and verifiable measurements. Im- 
provements in fixture repeatability allow precise, con- 
trolled measurements to be made to 18 CHz. By control- 
ling fixturing variations the quality of the measured data 
is improved. Reduction of fixture losses also allows more 
accurate measurement of devices with high reflection. 
Another result is improved measurement verification, 
assuring that measurements made are correct. 

■ More accurate transistor measurement because of cali- 
bration by de-embedding. The use of de-embedding al- 
lows precision coaxial standards to be used for cali- 
bration. Such standards, along with a carefully control- 
led and verifiable fixture, yield precision measurements. 
This allows leveraging available high-quality coaxial 
standards to provide the best available measurement 

■ Common verifiable measurement environment for in- 
dustry-wide standardization of transistor measurements. 
The controlled and modeled characteristics of the fixture 
and verification device permit rapid verification of the 
entire measurement system. Data taken with one pac can 
be verified and directly compared with data taken at 
other times and places to allow meaningful analysis. 
This allows device manufacturers and users to communi- 
cate more effectively. Better correlation between man- 
ufacturers' data sheets and user designs may be obtained. 

■ Real-time fully error-corrected display of measured data. 

0.} wilt 

H«ton 1 U 

o«ft» FET M n|c*oti -jm 

SWT 2 lX« 

5io*» i6 QMj 

• 1 00 . OOO0 

|7 !f..M»fi<l 

41150. 0000 

4650. OQOO 

14. IB 

fiDio vex 





















4. OS 











-37 . 60 






























It. 15 


-37. M 



















30.61 1 





















































10'.. 0 



4 790. OOOO 













41, B3 



1 .391 








1 07 , 1 














13). 5 



























119. » 


• 9.52 



1 . 391 




,479 -101.9 


-107. 1 
-107. 5 
-101. ) 


85. 6 




• 40.0 

Fig. 2. The HP 80514 A measurement software offers a variety of output formats. It provides 
data specific to the design and analysis of transistor amplifiers and oscillators and more 
conventional s. h, y, and z device parameters 


© Copr. 1949-1998 Hewlett-Packard Co. 

Since the desired device data is presented in real time 
with the errors of the fixture removed from the data, the 
effects of varying device parameters such as bias or tem- 
perature can be directly observed. This direct observa- 
tion, along with the human ability to synthesize informa- 
tion, allows fundamentally new information about de- 
vice performance to be obtained. This is of importance 
in, for example, the design of low-phase-noise oscillators 
where knowledge of parameter changes as a function of 
bias may be useful in determining optimum resonator 
design or coupling. 

Output formats appropriate to active device measure- 
ments. Although the HP 8510 Microwave Network 
Analyzer can, by itself, provide several different formats 
of data output, active devices are frequently charac- 
terized in additional ways. The active device measure- 
ment pac provides amplifier parameters including uni- 
lateral, transducer, conjugately matched, and Mason's 
gain values. Mismatch gain, group delay, and amplifier 
stability are also available. In addition, termination pa- 
rameters, useful for amplifier and oscillator design, are 
provided. These include optimum source and load termi- 
nations (to provide maximum available gain) and 1/s n 


Error Adapter 






and I/822 for oscillator design. Unformatted output data 
can be stored for later retrieval and comparison by the 
measurement pac or stored in a format compatible with 
circuit analysis and optimization programs. 

Automatic Network Analyzer Error Correction 

Normally, an automatic network analyzer makes mea- 
surements by first going through a process called calibra- 
tion. This process consists of measuring a number of known 
devices called calibration standards. 

Based on the resulting data and a model of how the 
characteristics of the microwave hardware in the measure- 
ment system contribute to errors in measurement, called 
an error model or error adapter, the analyzer corrects future 
measurements to remove the systematic measurement er- 
rors. Device measurements are then made as though an 
errorless network analyzer were located right at the location 

Fixtured Measurement Problem 












Including Fixture Errors in Network Analyzer Error Adapter 









Automatic *9^^BJB 

Network S 


Fig. 3. For one-port error correction when the measurement 
plane is the same as the calibration plane, known calibration 
standards are measured first (top) and error-term values lor 
an error adapter are automatically computed by the automatic 
network analyzer When theDUTis connected /bottom), these 
error-term values are used by the analyzer to produce error- 
corrected data 

Error Adapter 


Fig. 4. In many cases the device to be measured is not 
located at the calibration plane Rather, it is separated by a 
fixture consisting of adapters, transitions, or other kinds of 
connecting networks If these fixture effects are included with 
the original error terms obtained from calibration and the re- 
sults returned to the network analyzer, direct measurement 
of devices embedded within the fixture can be achieved. This 
is termed a de-embedded measurement. 


© Copr. 1949-1998 Hewlett-Packard Co. 

where the calibration standards were connected. 

Fig. 3 shows the network analyzer calibration and mea- 
surement process for a one-port test device. The calibration 
process provides values for an error model of the network 
analyzer hardware characteristics. Imperfections in con- 
nectors and the analyzer's directional device are included 
here. The error-correction software in the automatic net- 
work analyzer then adjusts the measured data to remove 
the effects of these system imperfections. 

For an error-corrected one-port measurement, the net- 
work analyzer solves the following equation. Since each 
term has both magnitude and phase (or alternatively, real 
and imaginary) components, the solution is performed in 
the complex domain. 

(Si lm - e df ) 

where s,„ = 

s llm 


actual s-parameter of the device being 

measured (uncorrected) value 
error caused by the directional device in 

the system 
transmission characteristics of the 


error attributable to source match of the 
system hardware. 

Calibration for measurement of two-port devices requires 
additional calibration standards and a more complex error 
model. Commonly an open, a short, and a load on each 
port, as well as a through connection are used. Nine addi- 
tional error terms are used for a total of twelve error terms 
in the error model. 

For a description of software signal processing in the HP 
8510. including error correction, see the box on page 47. 

De-embedding Concepts 

Conventional calibration and measurement provides the 
desired data as long as standards that can be attached right 

at the measurement point are available. For many micro- 
wave measurements, calibration cannot be performed at 
the precise location where data is desired. This is the situ- 
ation addressed by de-embedding. De-embedding consists 
of modifying the normal error-correction process of the 
automatic network analyzer so that the errors introduced 
by the hardware between the point at which calibration 
was performed and the desired measurement location can 
be removed by the error-correction process of the analyzer. 

Referring to Fig. 4 and "Automatic Network Analyzer 
Error Correction" above, the necessary error term modifica- 
tion for de-embedding a one-port measurement can be de- 
scribed. In Fig. 4. F,,,F 2 , and F 22 .F, 2 represent the fixture 
reflection and transmission s-parameters for port 1 and 
port 2. respectively. To allow the network analyzer to make 
the de-embedded measurements, new error terms, e'. must 
be calculated: 

e'df = e d ( + 

KfF n ) 
(l-e 5( F n ) 

(e rf F 12 F 21 ) 
(1-e,,F n ) 

e'si = F 22 + 

(e 5l F 12 F 2l ) 
n-e sf F„) 

Here e^ is set to the product of the forward and reverse 
terms of the combined network to put the error matrix into 
the normalized form required by the error analyzer. 

These new error terms must be put into the network 
analyzer for use by its error-correction algorithm. Real-time 
display of the de-embedded measurement is the result. In 
the two-port measurement case, nine additional terms must 
be calculated. 

For de-embedded measurements to be possible, errors 
caused by the intervening hardware must be known at the 
time of measurement. 

The HP 85041 A Transistor Test Fixture is designed to 
play the role of accurately known hardware to make de-em- 

© Copr. 1949-1998 Hewlett-Packard Co. 



Transistor Leads 


Fig. 7. Measurement of s„ of a 
0 5-fim gallium arsenide field ef- 
fect transistor (left) A measure- 
ment without de-embedding in- 
cludes losses and reflections in- 
troduced by the entire fixture, 
(center) The measurement with 
the effects of errors caused by the 
fixture body halves removed, 
(right) Errors caused by the entire 
fixture — body halves, leads, and 
parasitics — removed by de-em- 

bedding possible. The fixture is shown in Figs. 5 and 6. 

Fig. 7 shows the results, as first the fixture effects and 
then the transistor leads themselves are removed from the 
measured data by de-embedding. The device is a 0.5-fim 
gate length GaAs FET. 

Fixture Characterization for Accurate De-embedding 

The function of a fixture is to provide a convenient con- 
nection to the measurement system. Often a fixture is used 
to provide information about how a device will perform 
in a proposed application. In such a situation, it is impor- 
tant that a fixtured measurement be nondestructive of the 
device under test and relatable to the application. 

A fixture used for network measurement should transmit 
energy from the network analyzer to the device under test 
with minimum loss and reflection. If the characteristic im- 
pedance of the device to be measured is greatly different 
from that of the network analyzer, it may be necessary for 

the fixture to provide impedance transformation. In any 
case the fixture must efficiently couple the device to the 
analyzer while perturbing the measurement as little as pos- 

For an error-corrected measurement, careful hardware 
characterization is necessary to allow the errors to be re- 
moved from the measurement, leaving only device data. 
In any measurement, it is desirable to understand not only 
the numerical value of errors introduced by the measure- 
ment process, but also the fundamental causes of errors 
and how they relate to the application as well as (he mea- 

This characterization can be performed in a number of 
ways. The most common way is through the conventional 
calibration process already described. Various types of 
standards can be used with a variety of techniques to quan- 
tify the errors in the system hardware including the fixture. 
An alternative is to calibrate (characterize) the system 

Straight Lead. 
Good Contact 
on Pedestal 

Port 1 



Bent Lead. 
Poor Contact 
on Pedestal 

Port 2 

7-mm Center 

Fig. 6. (a) Dielectric rods ensure well-defined connection to the through leads in the HP 85041 A 
fixture. Common leads are grounded very close to the transistor package by sandwiching them 
between the top and bottom halves of the fixture insert. Inserts can be changed to accommodate 
different package styles, (b) One body half, showing dielectric rods, center conductor, and 

conductive elastomer. 


© Copr. 1949-1998 Hewlett-Packard Co. 

Supponmg Supporting 

Capacitance Discontinuity Capacitance 

Trough Trough 
Region Region 

hardware excluding the fixture with the conventional pro- 
cess and available precision standards. The fixture can then 
be separately characterized and the two characterizations 
combined to allow error-corrected measurements inside 
the fixture. This alternative is termed de-embedding. 

In a sense, the fixture offsets the calibration standards 
used to characterize the rest of the microwave hardware 
just before the time measurements are to be made. For the 
measurement to be accurate, the description (model) of this 
offsetting network must reflect reality at the time of mea- 
surement. The fixture essentially becomes part of the cali- 
bration standards used to arrive at the device measurement. 
For high-quality measurements, whether de-embedded or 
not. the fixture must be well-controlled and huve known, 
repeatable characteristics. 

The fixture itself can be characterized for de-embedding 
in more than one way. Direct measurement is one obvious 
method. However, direct measurement of the fixture im- 
plies the availability of calibration standards that might be 
used to characterize the entire system in a more conven- 
tional manner not requiring de-embedding. Also, this ap- 
proach requires either that the measurements always be at 
the same frequencies as the original fixture measurements 
or that some kind of interpolation be performed to provide 
values based on the original data points. Such interpolation 
generally requires smooth, well-behaved performance be- 
tween the original data points. 

Another method is to use a model of the fixture to calcu- 
late fixture parameters. This model can be an equivalent 
circuit model or an analysis based on physical dimensions. 
In any case the modeling must produce a sufficiently accu- 
rate representation of the fixture's actual performance at 
the time it is used for device measurement. Representing 
a fixture in this way has the advantage of automatically 
providing values at arbitrary frequencies within the valid 
range. Additionally, the modeling process may help to un- 
cover discrepancies between intended fixture operation 
and actual performance, which simple measurement may 
not reveal. This is of great importance in understanding 

Fig. 8. The final fixture model has 
good correlation with the physical 
fixture and its measured electrical 
characteristics. (TL indicates trans- 
mission line with impedance, 
length, and loss ) 

the overall measurement process and applying the results 
to any final application. 

The second approach, that of using a circuit model to 
represent the fixture, was used with the HP 85041 A Tran- 
sistor Test Fixture. 

Fixture Development and Characterization 

At the outset of the project, a commercially available 
fixture was selected. This fixture provided a degree of de- 
vice package flexibility and manufacturing precision im- 
portant for the project. As can be seen in Fig. 5, the fixture 
consists of symmetrical body halves between which is 
sandwiched an insert. The transistor or other device to be 
tested rests in the insert.which is tailored for the particular 
device package style. The through leads of the device con- 
tact the coaxial center conductor of the body halves. These 
halves effectively act as adapters between 7-mm coaxial 
connectors and the device package leads. 

The common leads of the device are sandwiched between 
the top and bottom halves of the insert to provide a low-im- 
pedance ground connection. Good contact to the through 
leads is assured by spring-loaded dielectric rods in the lid. 
Each rod holds a through lead against the bottom of a trough 
in the end of the coaxial center conductor (Fig. 6a). 

The top portion of the insert and the top part of the 
coaxial body half are hinged. This allows access for device 
insertion and removal by simply opening the ltd. 

Very soon after we set out to characterize (model) the 
fixture, the real-time measurement ability of the HP 8510 
revealed that there were some fixture nonrepeatabilities. 
The connections made by the lid to the fixture body were 
not consistent and could result in considerable reflection 
and loss. To combat this problem, a cylindrical conductive 
elastomer was installed in grooves in the four mating sur- 
faces of the lid to act as microwave gaskets. This signifi- 
cantly improved the fixture repeatability and also reduced 
insertion loss. 

Fig. 6b shows one body half without insert. The conduc- 
tive elastomer is visible in the lid. 

© Copr. 1949-1998 Hewlett-Packard Co. 


To begin the modeling process, the fixture was consid- 
ered in smaller sections rather than as a whole. One half 
of the body of the fixture, without insert, was examined 
first. Special coaxial devices that attached at the plane of 
the insert were available. Although these devices were not 
in themselves of sufficient quality and design to charac- 
terize the fixture adequately, they did provide a starting 
point. Problems not only with the devices but with provid- 
ing reliable connection to the fixture were obvious and 
gave insight about how a better characterization might be 
performed. Best results were finally achieved by measuring 
a short made from thin shim brass. The flexibility of this 
material allowed good contact to be made to the ends of 
the coaxial center and outer conductors. Repeatability of 
measurement not only for a given fixture half but also from 
half to half was ascertained. This was possible since the 
entire fixture is symmetrical. These measurements indi- 
cated that differences between shim-shorted half fixtures 
were at least 40 dB down (less than 1%). 

Once a good-quality short and high fixture repeatability 
were obtained, as evidenced by smooth, near-unity reflec- 
tion coefficient across the entire 45-MHz-to-18-GHz range, 
further measurements were made. An open-circuit mea- 
surement was made using one of the special coaxial devices 
as a shield for the open. This shielding was necessary to 
prevent radiation from the open end of the fixture. A two- 
port measurement was made of the two body halves con- 
nected together with no insert. A small shim was placed 
between the center conductors to assure good contact. The 
outer conductor contact was already assured by the conduc- 
tive elastomer. The data from these measurements was read 
by a computer and a circuit analysis and optimization pro- 
gram was used. This program allowed a tentative circuit 
model of the fixture half to be developed and provided 
circuit values that best fit the measured data for all three 
measurements. Initial topology for the circuit model came 
from physical examination of the fixture. The coaxial struc- 
ture was represented as series-connected lossy transmis- 

sion lines with discontinuities at their junctures rep- 
resented by a fringing capacitance, which was allowed to 
be frequency dependent. 

The model for the open half-fixture includes an unknown 
fringing capacitance at its end in addition to internal 
capacitances which were common to all three models. Ini- 
tial values, as well as limits for these values, were obtained 
through physical examination of the fixture. For example, 
the overall physical length of the body half and the loca- 
tions of discontinuities resulting from supporting beads, 
the slotted center conductor, and the dielectric were easily 
measured. Allowing the optimizer to select values for line 
impedance, loss, and discontinuity capacitance yielded a 
circuit model that is a good fit to all three sets of measured 
data. Examining and optimizing a portion of the complete 
fixture instead of the entire fixture as a whole speeded up 
the process and gave results that more nearly matched 
physical reality. Solutions that fit the data but were not 
physically accurate were avoided. Additionally, greater un- 
derstanding of Ihe nature of the fixture was obtained. 

Final results for modeling two connected fixture halves 
yielded a model that fits the measured data within better 
than -40 dB (1%) for all measurements. 

Along the way it was possible to use the current best 
model to perform a de-embedded measurement of the two 
fixture halves connected together without insert. This, 
along with measurement of the shim short, gave a direct 
indication of the quality of the model, since residue or 
error in the model was directly visible in real time. The 
through connection looked very nearly like a zero-length, 
lossless, reflectionless transmission line. Similarly, the 
short appeared as a near-unity reflection at 180° on the 
Smith chart over the entire frequency range. 

Corresponding calibrations performed with the special 
coaxial standards mentioned above were typically 15 dB 
worse than this. Thus, the effort lo characterize and under- 
stand the fixture resulted in a considerable improvement 
in measurement accuracy and, perhaps as important, a 


R£F 1 Unit* B* • 

3 V. 18 <■* 

25^--^ ( 









stpbt 2 CHi 

flEF a Unt*« 
.5 Un 1 

1.6 GHi nir 

Fig. 9. Modeling and de-embedding along with improvements in the fixture result in a significant 
improvement in transistor measurements. Improvements in trace smoothness over previous 
measurements 4 (black trace) are evident with the HP 8501 4A (color trace). The same packaged 
GaAs FET with the same bias conditions is being measured in both examples. The transistor, 
which is small in terms of measurement wavelengths and has a relatively simple structure, can 
be expected to have a smoothly varying response as a function of frequency. 


©Copr. 1949-1998 Hewlett-Packard Co. 

commensurate improvement in understanding of the fix- 
ture's operation, its limitations, and areas where additional 
work might provide maximum return. 

Once a good model for the body was obtained (residue 
more than 40 dB below unity reflection for both short and 
through connection), the insert and transistor characteris- 
tics and parasitics were arrived at in a sim ila r iterative 
manner, but this time using special transistor-like stan- 

dards inside the inserts. Here again the process involved 
using measurements of the best characteristics of available 
standards and using an analysis/optimization program to 
generate a circuit and values to fit. As before, this iterative 
process yielded new understanding of the operation, which 
led to modification of the measurement process and reop- 
timization of the circuit model topology and values. 
Fig. 8 shows the final model used for the transistor test 

HP 8510 Software Signal Processing 

Digital signal processing in the HP 8510 Microwave Network 
Analyzer (Fig 1 ) begins at the outpui ot the synchronous detector 
pair, which provides ihe real (X) and imaginary (Y) pans of Ihe 
test and reference signals Offset, gain, and quadrature errors 
are corrected for Doth of the IF/detector chains before the test 
vector is ratioed against the reference vector The result is an 
unprocessed s-parameter stored into the raw array If requested 
by the HP 8510 user, subsequent data points taken at the same 
frequency are averaged together using a stable averaging 
technique, thus enhancing the HP 851 0's dynamic range 

While the raw array is continually filled under control of the 
data acquisiiion software, the data processing software concur- 
rently removes data from the raw array and controls additional 
signal processing Using error coefficients that model the micro- 
wave measurement hardware, the data is further corrected 
through a set of vector math operations. This corrected data can 
be converted from Ihe frequency domain to the time domain 
using the chirp z-transform technique. Storage into a data array 
allows quick response to the user when making format or trace 
math changes. 

Data can be stored into memory and used in vector computa- 
tions with data from a second device Comparison is ac- 
complished through vector division, subtraction, or simultaneous 

display of data and memory. 

The vector data is reformatted into magnitude, phase, group 
delay, or other formats It is stored into the format array, which 
provides convenient access for scale and offset changes. Scaled 
data is stored into a display list, from which the display generator 
hardware repetitively creates a plot on the CRT for a flicker-tree 

Input and output access is provided to all the arrays via the 
HP-IB (IEEE 488/IEC 625) S-parameters can be obtained from 
the data array Direct plotter output is from the format array 

The user can trade off the daia update rate against the number 
of data points by selecting resolutions from 50 to 400 points. 

A multitasking software architecture provides the fastest pos- 
sible update rate by allowing data processing to take place when 
the data acquisition software is not busy. Overlying command 
and control tasks interleave data processing with acquisition 
cycles for two-port error correction and dual-channel display 

Michael Neering 
Project Manager 
Network Measurements Division 


i i 


I ► 



o >f 


o — __: 1 

' — ► 

□ at In 
























Fig. 1. HP 8510 Automatic Net- 
work Analyzer software signal 

© Copr. 1949-1998 Hewlett-Packard Co. 



Application of the Data 

Since the transistor or other active device is not generally 
mounted in the application as it is in the fixture, some 
correcting network may be desirable to correlate fixture- 
measured device parameters with those of an application. 
A correlating network can be obtained using techniques 
similar to those described above. Such networks depend 
upon the application. For example, in a microstrip applica- 
tion, the board thickness and common lead grounding tech- 
niques can affect device operation. Once the differences 
between the test fixture and the application are known, 
these differences can be included with fixture-measured 
data to predict the performance of the device in the appli- 
cation accurately. Such fixture/application differences are 
of most concern at higher frequencies and for devices with 
impedances greatly different from 5011. 

The de-embedded measurements that result from the 
models developed have proven themselves to be consider- 
ably better than any available before. Fig. 9 demonstrates 
the results of fixture improvements and de-embedding 
calibration techniques. 

Even without accounting for differences in mounting 
technique in the application, data from the HP 85014A 
Active Device Measurements Pac has been used to achieve 
finished amplifier results very close to computer predicted 
values at 10 GHz. Constructed amplifiers have been mea- 
sured with results within a few tenths of a dB of those 
predicted by fixture-measured data and a circuit analysis 
program. It is believed that the dominant sources of vari- 
ation in measurements are operator technique and package 
characteristics. Bent leads, variations in plating, and the 
positioning of the device in the fixture are critical param- 
eters, particularly above 12 GHz. This is to be expected 
when one considers that a few thousandths of an inch 
variation in package position can easily cause several de- 
grees of error in the phase of a reflection measurement. 


For the HP 85041 A Transistor Test Fixture, the verifica- 
tion standard is a small planar cross similar in dimension 
to a packaged transistor. Using such a device allows the 
fixture's common lead characteristics to be verified also. 
This verification or check device is mechanically simple 

and provides a good way to verify system performance. 
Although this device is a good conductor, it is not a perfect 
short from the point of view of the measurement planes, 
having both length and loss. To make the verification pro- 
cess easier, a special model was developed to de-embed 
the check device's nonideal characteristics from the verifi- 
cation measurement. Doing this effectively normalizes the 
check device to appear very nearly like an ideal short, 
although it in fact has some reactance and loss. The actual 
value of the measurement is not of importance, since the 
goal is only to verify that the proper data is obtained and 
that a calibration is good. This normalization by de-embed- 
ding the measurement calibration using the check short 
normal izer model and measuring the check short allows 
quick and easy verification of the system. The transmission 
characteristics of the fixture are effectively verified by the 
two one-port reflection measurements, since the stimulus 
signal must travel from the connectors to the center and 
back for such measurements. Although use of a single stan- 
dard does not completely characterize (verify) a two-port 
fixture, experience has shown that a broadband measure- 
ment of the check device provides a high degree of certainty 
of fixture performance. 


We would like to thank the many people who were in- 
volved in bringing this product to fruition. Thanks go to 
those in all areas who helped to take an idea from prototype 
to product, including those in industrial design, new prod- 
uct introduction, production engineering, manufacturing, 
marketing, and product support. In particular, we thank 
Kevin Kerwin for his help developing the feature set, and 
Jeff Meyer for his help during fixture characterization and 
software model generation. 


1. R.F. Bauer and P. Penfield. Jr.. "De-embedding and Unterminat- 
ing," IEEE Transactions on Microwave Theory and Techniques, 
Vol. MTT-22, March 1974. pp. 282-288. 

2. G. Elmore, "De-embedded Measurements Using the HP 8510 
Microwave Network Analyzer," Hewlett-Packard RF and Micro- 
wave Symposium. 1985. 

3. G. Elmore, "De-embed Device Data with a Network Analyzer," 
Microwaves and RF, November 1985, pp. 144-146. 

4. R. Lane, R. Pollard, M. Maury, and J. Fitzpatrick, "Broadband 
Fixture Characterizes Any Packaged Transistor." Microwave Jour- 
nal. October 1982. p. 104. 

Hewlett-Packard Company, 3200 Hillview 
Avenue, Palo Alto. California 94304 


February 1987 Volume 38 • Number 2 

Technical Information from the Laboratories ot 
Hewlett-Packard Company 

Hewlett-Packard Company. 3200 Hillvtew Avenue 
Palo Alio. California 94304 USA 
Hewlett-Packard Central Mailing Department 
P O Bon 529. Startbaan 16 
1 1B0 AM Amstelveen. The Netherlands 
awa-Hewlett- Packard Ltd Sugmami-Ku Tokyo 168 
Hewlett-Packard (Canada] Lid 
ay Onve. Mississauga Ontario L4V 1MB Ca 

Bulk Rate 
U.S. Postage 


LAUREL 20707 


' A f ^% m To subscribe change your address or delete your name ttom our mailing l«t. send your request to Hewlett-Packard 

f\ |_y U ntOO. Journal. 3200 Hdiview Avenue Palo Alto. CA 94304 USA Include your okj address label it any Allow 60 days 


© Copr. 1949-1998 Hewlett-Packard Co.