Skip to main content

Full text of "HP Journal 1987-07"

See other formats


© Copr. 1949-1998 Hewlett-Packard Co. 


July 1987 Volume 38 • Number 7 


4 Dedicated Display Monitors Digital Radio Patterns, by David J. Haworth, John R. 
Pottinger, and Murdo J. McKissock An end-on-view, or constellation, ol the waveforms 
making up a digital radio signal provides a graphic display olthe effects of several impairments 

11 Automated Timing Jitter Testing 

IO Constellation Measurement: A Tool for Evaluating Digital Radio, by Murdo J. McKissock 
O Statistical analysis of displayed constellation clusters helps determine digital radio 

1q A Digital Radio Noise and Interference Test Set, by Geoffrey Waters Establishing 
\2 accurate carrier-to-noise ratios and interference levels is a necessary part of measuring 
bit error rate. 

22 Noise Crest Factor Enhancement 

23 Noise Bandwidth Measurement 

Microprocessor-Enhanced Performance in an Analog Power Meter, by Anthony 

Lymer A custom thermal converter and an autozero circuit are key design components. 

An Accurate Wideband Noise Generator and a High-Stability Reference Source, by 

Dayananda K. Rasaratnam These modules make it possible to adjust the injected noise 
level automatically to maintain a desired carrier-to-noise ratio. 

33 General-Purpose Wideband Thick-Film Hybrid Amplifier 

Automated Radio Testing Shortens Test Time and Enhances Accuracy, by John A. 

Duff This system can reduce the time required to measure flat fade performance by 60 
to 90 percent. 

38 A Reusable Screen Forms Package 


3 In this Issue 
3 What's Ahead 
18 Authors 

Editor Richard P Dolan • Associate Editor Business Manager. Kenneth A Shaw • Assistant Editor Nancy R Teatei • Art Director Photographer ArvKJ A Oanielson 
Support Supervisor. Susan E Wngnl • Administrative Services, Typography Anne S LoPresti • European Production Supervisor. Michael Zandwt|Ken 


© Copr. 1949-1998 Hewlett-Packard Co. 

C Hewlett-Packard Company 1987 Printed in U S A 

In this Issue 

Hl^P^rSm Behind many of the microwave antenna towers one can see riding the 
I tops of ridges and tall buildings are digital radio systems, that is. systems 
^^Si I in which the information is imposed on the microwave carrier signal by digital 

I modulation techniques. Two instruments for testing digital radios are the 
W ^Vflj I sub J ec,s of ,nis issue. 

jjk' One important measure of the performance of a digital radio is the bit 

^^£&J^^^M error ra,e - or BER What portion of the data bits being transmitted can be 
HnMfn^BB expected to be corrupted by noise? One in a million? One in ten million? 

Rhh^*— The answer depends, as one might expect, on typical radio conditions like 
fading and interference. If the carrier power fades by as little as 10%, the BER can get ten times 
worse. The HP 3708A Noise and Interference Test Set is designed to inject precise amounts of 
noise and interference into a digital radio so the effect on the radio's BER can be measured. The 
design and theory of operation of this instrument are presented in the papers on pages 19, 26, 
and 30, and on page 36 is a description of an automatic test system for BER tests of digital radios 
in production. Among the engineering challenges faced by the HP 3708A's designers were how 
to maintain a precise carrier-to-noise power ratio when the carrier power is changing, how to 
guarantee that the noise has not only the required power level but also high enough peaks to 
approximate real noise, and how to get both fast response and high accuracy in microwave power 

There are many different digital modulation techniques in use today. They have names like 
QPSK, 49QPR, and 64QAM, where the Q stands for quadrature. The modulation signal has two 
components, called in-phase (I) and quadrature (Q), each of which can assume only a finite 
number of values and will have exactly one of these values when the receiver samples the signal. 
In 64QAM, for example, which means 64-state quadrature amplitude modulation, I and Q can 
each take on eight values (8 *8 = 64). If you were to plot the I and Q values received in such a 
system, the plot should look like a regular 8x8 grid of 64 dots. Because of noise and other 
impairments, an l-versus-Q plot for a real digital radio more often consists of clusters of dots, and 
the grid may be distorted. The nice thing is that you can tell a lot about how the radio is performing 
by looking at such a plot, which is called a constellation diagram. There's one on the cover, and 
you'll find others in the papers on pages 4 and 13, which describe the design of the HP 3709A 
Constellation Display. This special-purpose display adapts to various digital modulation schemes 
at the flip of a switch, and not only displays the constellation diagram but also computes statistics 
of the dot clusters and various distortion parameters. Among the design challenges were how to 
minimize timing jitter in the sampling circuit and how to deal with display distortion, which can 
mimic distortion in the digital radio. 

-Ft. P. Dolan 

What's Ahead 

The design and development of the HP-18C and HP-28C, HP's latest handheld calculators for 
business and technical professionals, are the subjects of the August issue. 

The HP Jouevi! encourages tecnn^et discussion at me (op.cs presented <n recent ort-cies and *■« pubt'sn letters expected to De of interest to our readers Letters must be Dfiol ond are sutiect 
to edging Letters should be addressed to Editor Hewlett PscHard Journal. 3200 HQMtW Avenue. Palo Alto CA 94304 USA 

© Copr. 1949-1998 Hewlett-Packard Co. 


Dedicated Display Monitors Digital Radio 

One way of displaying the complex waveforms generated 
in digital radio systems is the constellation display, a 
method that allows rapid visual evaluation of a system's 

by David J. Haworth, John R. Pottinger, and Murdo J. McKissock 

is a specially engineered two-channel sampling os- 
cilloscope lor monitoring the eye and constellation 
patterns in a digital radio. Although called digital radio 
because the input is digital, much of the circuitry is analog 
and the design of a digital radio is an exercise in high-per- 
formance analog and digital design. The waveforms are 
random multilevel wideband signals, and require a high- 
performance sampling oscilloscope to display them. The 
HP 3709A design goals were high performance, low cost, 
ease of use, quality, and manufacturability. 

Because the instrument is dedicated to one application, 
it has some novel features such as user-selectable graticules 
and dedicated constellation measurements. Many of the 
controls have been simplified for easier use. The time base 
autoranges to display two complete cycles of the input 
waveform. The usual trigger and trigger mode controls are 
not necessary because the instrument triggers from a clock 
waveform supplied by the radio. The samplers have been 
optimized for use on random signals and do not have the 
dot response problems of earlier sampling oscilloscopes. 

A choice of internally generated graticules can be 
selected by a rear-panel switch to correspond to commonly 
used modulation schemes: QPSK, 9QPR, 16QAM, 49QPR, 
and 64QAM (see references 1 through 4 for discussions of 
these modulation methods and other digital radio funda- 
mentals). For 16QAM, a 4x4 graticule is displayed and 
for 49QPR, a 7 x 7 graticule is displayed. Alphanumeric 

information is also presented to display instrument status, 
self-test messages, and results of the dedicated measure- 
ments. A hard copy of the constellation can be output by 
the HP 3709A on an HP Thinkjet Printer via the HP-IB 
(IEEE 488/IEC 625). See Fig. 9 on page 17 for a sample 

The two input channels are labeled I (in-phase) and Q 
(quadrature) instead of Channel A and Channel B because 
this is how the typical customer refers to the two radio 

There are four modes: constellation, I eye, Q eye, and 
measure, selected directly by dedicated keys. In eye mode 
the HP 3709A operates as a conventional oscilloscope with 
volts indicated on the Y axis and time base sweep on the 
X axis. A typical eye diagram is shown in Fig. 2. Constel- 
lation mode is what some oscilloscopes label A versus B. 
This mode displays a sampled instant of the I signal on 
the X axis and the Q signal on the Y axis. In this mode 
there is no time base sweep, the same point on the 
waveform is sampled every time, and timing jitter perfor- 
mance must be good. 

The displayed I and Q signals are controlled by a group 
of dedicated controls whose sensitivity is displayed on the 
CRT. The samplers are calibrated even' three minutes to 
eliminate gain and offset errors. A delay control allows the 
user to move the sampling instant to the position of 
maximum eye opening. The two data inputs are normally 
sampled at the same instant, but two additional delays can 


©Copr. 1949-1998 Hewlett-Packard Co. 

Fig. 2. Eye diagram viewed on an HP 3709A Looking grainy 
because the display is sampled, the eye is in the center with 
a bright line marker. 

be enabled to move the sampling instants independently. 
A trigger edge select key completes the user controls. 

Fig. 3 shows the baseband signals and symbol clock for 
a typical modulation scheme. 16QAM. The signals are sam- 
pled at instants defined by the symbol clock generated 
within the digital radio receiver. The I and Q signals have 
discrete values when the active edge of the clock occurs. 
The eye has four discrete levels at the correct sampling 
instant for 1 6QAM signals. An ideal constellation is a single 
point for each modulation state and the whole pattern is 
aligned to the graticule. However, the typical constellation 
(Fig. 4) shows clusters instead of points because of imper- 
fections in the radio. Similarly, the eye diagram is not a 
single line but many overlapping lines generated by the 
distortion of (he data flowing through the radio. 

Usually there are impairments causing small amounts of 
noise and geometric distortions of the constellation. The 

Time - 

Fig. 3. Typical baseband and symbol clock waveforms tor 
a I6QAM radio system. The two channels, I and Q, have lour 
possible states: 0, 1,2, and 3 The waveforms are filtered to 
conserve bandwidth, so they look more sinusoidal than 
square. At the valid edge of the clock (leading edge) the I 
and 0 data will be at one of the lour levels. 

MEASURE key takes 600 samples from the constellation, 
calculates values for several common impairments, and 
displays these as numbers on the CRT (see article on page 
13). The sampled data is also available over the HP-IB for 
further analysis by a computer. 

Hardware Design 

A block diagram of the instrument is shown in Fig. 5. 
The sample-and-hold circuit is similar to that used in con- 
ventional HP sampling oscilloscopes except that it is reset 
to zero between each sample. Digital radio waveforms are 
random multilevel signals and viewing these normally re- 
quires careful adjustment of a sampling oscilloscope to 
avoid excessive noise caused by intersymbol interference. 
By resetting the circuit between samples, the HP 3709A is 
free of intersymbol interference by design, and therefore 
is easier to use. 

The sample-and-hold circuit consists of a four-diode 
sampling gate and holding capacitor followed by a stretcher 
circuit (Fig. 6). The input signal is sampled by switching 
on diodes Dl to D4 for about 1 ns. The hold capacitor Cl 
is charged to about 10% of the voltage appearing at the 
input to the gate. The charge on Cl decays rapidly, but the 
impulse on the base of Ql causes an output waveform at 
its collector proportional in amplitude to the input voltage 
but stretched in time over several microseconds. The first 
half cycle of this waveform is integrated and the output of 
Ul is a voltage proportional to the input voltage. By varying 
the time switch Si is closed, the gain can be varied. An 
autocalibration loop is used to maintain unity gain over a 
wide temperature range. 

The waveforms in Fig. 7 show the circuit's operation. 
The input signal is a four-level pseudorandom binary se- 
quence (PRBS) generator which is representative of a digital 
radio signal. The upper trace is the output of the impulse 
generator and is a damped resonance proportional in 
amplitude to the input signal but fixed in frequency. The 
lower trace is the integrator output. 

Time A is when the sample is taken. The signal is then 
integrated from A to B, during which time Si is closed. At 
time B the output of the integrator is equal to the input 

Fig. 4. Constellation display on an HP 3709A with measure- 
ments of closure, lock angle, and quadrature angle 

© Copr. 1949-1998 Hewlett-Packard Co. 



I Signal 

O Signal 



Multiplexer sM LHI 


Graphics and 


Fig. 5. Block diagram of HP 3709A Constellation Display 

vollage and SI opens, a data valid signal becomes active, 
and the signal is displayed on screen. The next step resets 
(he integrator to zero volts by closing S2. This happens 
between times C and D. and the sampler is then ready for 
the next sample. 

In common with many high-frequency analog circuits, 
the sample-and-hold circuit suffers from drift of gain and 
offset. Rather than correct all of this in the circuit, a micro- 
processor-controlled autocalibration loop and a four-chan- 
nel digital-to-analog converter (DAC) perform the correc- 
tion. Gain drift is caused mainly by the sample pulse width 
varying with temperature and is corrected by varying the 
integration time. Offset drift is caused by diode forward 
voltage variations in the diode bridge sampling gate and 
changes in sample pulse shape. Offset correction is applied 
by varying the bias to the diode bridge. These corrections 
occur every three minutes unless autocalibration is dis- 

abled via the HP-IB. 

Time Base Modes 

The time base operates in three modes: constellation, 
eye diagram, and calibration mode. The principal elements 
of the time base circuitry are shown in Fig. 8. A clock input 
edge is selected every 20 jxs. delayed by a precise time, 
and output as the I and Q strobe to the samplers. In eye 
diagram mode an X-axis time sweep is also generated. The 
design goal was a CRT display with minimal time base 
jitter. Since the X axis is quantized to 200 increments over 
two symbols (two complete cycles), the time between sam- 
ples is 125 ps at the maximum clock rate of 80 MHz. The 
jitter specification was set at 30 ps. In practice, the jitter 
is less (see box on page 11). 

Eye Diagram. This mode is the most complex mode, and 
is similar to conventional sampling oscilloscope operation. 
There is no time base sweep control. The time base is 
adjusted automatically to make I he display width equal to 
the width of two symbols for any clock in the range from 
1 MHz to 80 MHz. Automatic width control is convenient 
for comparing eye diagrams. Since they are all displayed 
the same width, it is easier to pick out impairments. A 
two-symbol width allows one eye to be placed in the center 
of the screen, with a half symbol displayed on each side. 
Two ranges of automatic control are used: 1 MHz to 10 
MHz and 8 MHz to 80 MHz. Autoranging handles the range 
selection. A voltage proportional to the clock input fre- 
quency is used to control the gain of the I-and-Q delay (and 
the I delay and Q delay) one-shot circuits. This voltage also 
selects either divide-by-2 or divide-by-1 6 operation via the 
autorange circuit. 

The synchronizer uses two 13 flip-flops in cascade, both 
clocked by the divided input clock. The variable divider 
ensures that all clock inputs are at least 200 ns apart, which 
ensures that the first flip-flop has settled even if the data 
and clock change together. The second flip-flop then 
achieves jitter-free synchronization because its D input is 
stable long before the second clock edge occurs (see Fig. 9). 

V Impulse 

Generator Fig. 7. Sampler waveforms lor a 4-level PRBS input signal 

Upper trace is output of impulse generator and lower trace 
Fig. 6. Sampler schematic i s output ol integrator 


© Copr. 1949-1998 Hewlett-Packard Co. 

Delay Sweep 


Fig. 8. Time base block diagram 

In Fig. 8, the I-and-Q delay allows the user to adjust the 
sampling instant of both strobes together over a range of 
one symbol. The I delay and Q delay allow each strobe to 
be delayed by up to one symbol period. The delay range 
is independent of the clock rate. The I-and-Q delay is swept 
by the time base sweep, which has 201 discrete levels. This 
allows the delay to be stepped progressively through two 
symbols in 1% steps. The I-and-Q delay function is per- 
formed by a monostable (one-shot) circuit designed to have 
very low timing jitter. This circuit is described later. 

The time base sweep is generated by an 8-bit counter 
driving a 10-bit DAC with states 0 to 54 blanked. This 
allows 10-bit linearity, approximately 0.1%: the two least- 
significant bits are unused. 

In eye diagram mode, a sweep is also required to drive 
the CRT display. A conventional sampling oscilloscope 
uses the time base sweep, which requires the sweep to be 
stable for the duration of the I-and-Q delay, the I (or Q) 
delay, the sample time, and the display time. The sample 
rate can be increased if two pipelined sweeps are generated. 
The first sweep only needs to be stable for the I-and-Q 
delay time, and the second sweep only needs to be stable 
while a point is displayed on screen. This allows one point 
to be displayed while the time base increments to the next 
point. The X-axis sweep is conveniently generated by latch- 
ing the time base count and using a second 10-bit DAC. 
Constellation. In constellation mode the 8-bit time base 
sweep counter is fixed at the center of the sweep. The 
display shows the I and Q signals on the X and Y axes. 
The sample times can still be manually adjusted by the 
three delay controls. 

Calibration. In calibration mode, the microprocessor routes 
the 50-kHz clock directly to the I and Q strobes, so that 
sampler calibration can be performed even if no clock input 
is present. 

Low-Jitter One-Shot Circuit 

The sampling instant on the HP 3709A is varied with 
the I-and-Q-delay one-shot circuit (Fig. 10). A discrete cir- 
cuit was designed because available integrated circuits did 
not have the required jitter performance. 

The one-shot delay time is given by: 

t = (Cl+C2)(V Slart -V Slf ,p-iR)/i 

For frequencies below 10 MHz, Cl and C2 are used; above 
10 MHz, only Cl is used. V, siar) is set by the user front-panel 
control. Vstop is the sweep voltage in eye diagram mode, 
or a constant value in constellation mode. R represents the 
effect of Rl and R2. which are added to damp the resonant 
circuit formed by the timing capacitor and the printed cir- 
cuit board trace inductance. For Cl. a resonant frequency 
of about 100 MHz is observed. If undamped, the start-up 
transient would distort the ramp. The current i is supplied 
by Q3. 

The circuit functions as follows. The voltage V, at the 
collector of Q3 is clamped to V stnrt initially. When the 
flip-flop is triggered, Ql turns off and current supplied by 


FF1 D 

FF1 Q and 
FF2 D 

FF2 Q 

i r 

- Can be up lo 50 ns. 

Fig. 9. Time base synchronizer waveforms 

© Copr. 1949-1998 Hewlett-Packard Co. 



Voltage ' 

Voltage from 
Voltage / 

I and O 




^ — Wv — * 

Voltage-to-Current Converter 

Fig. 10. Low-jitter one-shot cir- 

Q3 causes V, to ramp down. When V, is more negative than 
the sweep reference voltage, the flip-flop is reset and Ql 
turns on. 

The discharging current i is accurately controlled by the 
voltage-to-current converter operational amplifier. The 
amplifier is decoupled from the high-speed ramp by the 
cascode transistor Q3. This configuration allows an accu- 
rate current to be maintained despite the high discharge 
rate. The current source is controlled over a decade range. 
An additional decade range is provided by switching in 
the timing capacitor C2 using Q2. 

The performance of such a circuit cannot be evaluated 
by observing the ramp directly at V,. Any probe degrades 
the jitter performance, stray capacitance upsets calibration, 
and stray inductance causes ringing. Indirect means must 
be used. The best method is to use the time base to drive 
the sampler, and sample a sine wave. The zero crossing of 
a sine wave has a defined dV/dt which allows easy conver- 
sion from volts to picoseconds. For example, a IV peak. 
80-MHz sine wave has a dV/dt = 0.503 mV/ps at the zero 

Using a 10-bit analog-to-digital converter after the sam- 
pler gives quantization steps of 2 mV for a ±1V input, 
allowing a time resolution of 2 ps. Linearity is measured 
by sampling a signal running at twice the clock frequency. 
Two symbols then contain four cycles or nine zero cross- 
ings. Linearity is measured by observing the displacement 
of the zero crossing. Using this principle, a reliable mea- 
surement of jitter and linearity can be achieved in a com- 
puter-based test system. 

Fig. 1 la shows the linearity and jitter performance of the 
time base one-shot circuit. Fig. lib shows a residual jitter 
of approximately 9.5 ps. This value is derived as follows: 
n 80 MHz = 12.500 ps/cycle 









Fig. 11. Jitter performance on sine wave, (a) 80 MHz. 1V 
p-p (b) Above trace at increased sensitivity, jitter =9.5 ps. 


© Copr. 1949-1998 Hewlett-Packard Co. 

■ 1% of cycle = 125 ps. which corresponds to a displace- 
ment of 16.5 mm 

■ Dot peak-to-peak displacement = 5 mm 

■ Dot rms displacement ^S'4 = 1.25 mm 

■ Rms jitter =125 x 1.25/16.5 = 9.5 ps. 

However, at lower clock input levels this value increases 
because of input buffer noise. 

Analog Signal Processing 

After being sampled, the signal is amplified to a suitable 
level for driving the display. Because the signal is now 
lower in frequency, this can be done with high-speed op 
amps. Perhaps the most interesting feature is the hardware 
multiplexing done to perform several tasks concurrently. 
The hardware is configured to one of four principal func- 
tions bv the microprocessor. The various signal paths are 
(Fig. 12): 

1. Display the incoming signal 

2. Display the graticule and text from the microprocessor 
graphics generator 

3. Measure the setting of the front-panel gain control, dis- 
played on screen as mVdiv 

4. Measure and correct sampler gain and offset. 

The first three functions are implemented with FET 
analog switches and appear concurrent to the user. The 
display continuously switches between incoming I and Q 
data at a 50-kHz rate, the graticule drawing is updated at 
a 40-Hz rate, and a dc calibration voltage is routed through 
the front-panel gain control twice per second to measure 
the setting and thus calculate the instrument sensitivity. 

Sampler offset and gain correction normally occurs every 
three minutes. A three-level dc calibration signal is routed 
through the samplc-and-hold circuit with the front-panel 
gain control bypassed, leaving only fixed gain stages. The 
offset is adjusted to zero volts and the gain is adjusted to 
the value held in ROM. 

Resampling and Data Conversion. Because the output of 
the sample-and-hold circuit is only valid for 10 /is. it is 
resampled using a slower sample-and-hold circuit to give 
sufficient time for analog-to-digital (A-to-D) conversion. 
This conversion is performed by the microprocessor and 
the DAC using a successive approximation algorithm. The 
same DAC is used for both graticule generation and mea- 
surements, thus reducing errors between the signal on the 
screen referenced to the graticule and the signal measured 
by the processor. 

Vector Graphics Generator 

The need for a vector graphics generator was identified 
early in the design of the HP 3709A Constellation Display. 
To meet project design goals for low cost and high reliabil- 
ity, circuit complexity had to be minimized. Investigation 
showed that existing designs offered excellent display qual- 
ity and plenty of features, but exceeded the target cost by 
an order of magnitude. Ruthless trimming of unnecessary 
display features resulted in an efficient design which has 
a low component count and is easy to manufacture. 

The instrument display is an electrostatic CRT module 
with XYZ vector drive inputs. Graphics are produced by 
the vector graphic, or calligraphic, technique often used 
with this type of display. This technique builds a CRT 
display pattern by modulating the X and Y deflection to 
draw individual lines and characters. In contrast to a raster 
scan display, the time needed to refresh the display de- 
pends on the number and length of the individual lines 
(vectors). The technique is used here to generate full-screen 
graphics in a small fraction of the display time, so that 
more display time is available for analog data. 

The design minimizes cost and board area by tight cou- 
pling between the graphics hardware and the main micro- 
processor. Two DACs are required. With the addition of 
precision comparators, these DACs are also used to perform 






(From X 




To Resampler 
and Microprocessor 

Time Base Sweep V 
(Eye Mode Only) / 

Fig. 12. Analog signal processing Four functions are performed 1) inpul signal displayed on 
CRT (S1 = S2=1 and S3 = 0), 2) graticule is displayed on CRT, 3) front panel gain potentiometer 
is measured (S1 =S2=S3 = 0), and 4) sampler is calibrated (S1 =0 and S2 = S3= 1). 

© Copr. 1949-1998 Hewlett-Packard Co. 


Digital Section 

Analog Section 

Character Segment 

X Data 
Latch Counter 


Bus > 

Graphics Control 
State Machine 

X Output 

Y Output 

Fig. 13. Graphics generator block diagram 

analog-to-digital conversion for measurement of constella- 
tion samples. By reusing the microprocessor and DACs. 
the marginal cost of the graphics generator is reduced to 
only 160 cm 2 of printed circuit board area. 
Display Refresh. The graphics display is continuously re- 
freshed at a rate of 40 Hz. Although this rate is unusually 
low. it is sufficient to prevent display flicker with a P39 
long-persistence CRT phosphor. 

Each refresh cycle starts with a microprocessor interrupt 
from the real-time clock. This activates the firmware 
routine that drives the graphics generator. The display is 
stored internally as a list of primitive operations which 
represent horizontal lines, vertical lines, hidden moves, or 
ASCII characters. Line and move operations are sent di- 
rectly to the graphics generator hardware while ASCII 
characters are interpreted by the firmware, using table 
lookup to generate a sequence of line segments for each 

The display refresh cycle typically uses 5% to 15% of 
the available display time, and a similar fraction of micro- 
processor time. Execution speed of the firmware routine 
is important, so it is coded in assembly language. The inner 
loops for display list output and character interpretation 
contain only six microprocessor instructions. This is pos- 
sible because simultaneous design helped select the best 
trade-off between hardware and firmware complexity. 

Although described as a single operation, display refresh 
is divided into four phases to even out processor loading 
and minimize latency for other real-time processes. Two 
phases are allocated to parts of the graticule and the other 
two phases are allocated to display text. 

Graphics Hardware. The graphics generator is divided into 
analog and digital sections, with a state machine to provide 
control signals. Fig. 13 shows the block diagram. 

The analog section generates X and Y signals to drive 
the display. When drawing a line, these signals must ramp 
smoothly between the start and end points. If a line is at 


+ lnput^ |^\q 2 

P Output 



Fig. 14. Differential limiter circuit 


©Copr. 1949-1998 Hewlett-Packard Co. 

Automated Timing Jitter Testing 

Automated testing allows measurement of parameters on every 
production instrument In many cases mere is virtually no manual 
method oi performing the test in an acceptable time A good 
example of this is timing jitter testing A computer-driven auto- 
matic test is the only means of Quantifying the amount of timing 
jitter in the HP 3709A Constellation Display 

The test system's clock input generator is set to the HP 3709A's 
specified minimum clock input level. This is the condition at which 
most timing jitter is likely to be seen A sinusoidal signal is then 
fed into the I and Q channel inputs from the test system's signal 
input generator When the HP 3709A is set to eye mode, two 
cycles of the signal can be seen However, the lest system is 
ignorant of the phase of the displayed waveform. 

Because the I and Q position controls, which correspond to 
vertical position controls in eye mode, cannot be switched out, 
a measurement of the offset is required to be used as a reference 
for calculating the eye position of zero crossing points and peak 
points, etc This is accomplished by switching the signal input 
generator off (i.e., to its minimum amplitude of approximately 
-90 dBm, which is approximately 50 dB below the HP 3709A 
noise floor) and measuring 1000 samples at Eye Position = 
0.0000 Sample Periods. 

From these 1000 samples, the mean is calculated to give a 
reference voltage V, e , By adjusting the phase of the signal input 
generator using its minimum phase increment of 0 1 degree, the 
signal can be set up so that the zero crossing point occurs at 
Eye Position = 0.0000 Sample Periods. This is the point of 
maximum slew rale where the worst timing jitter will be observed 
By measuring 1000 samples al this point , a distribution of voltage 
values will be obtained Using the 1000 samples, the rms value 
of the voltage at the zero crossing point can be estimated This 
rms voltage corresponds to noise at the zero crossing point and 
has to be converted into the time domain before the liming jitter 
can be calculated. To calculate Ihe liming |iiter from the noise 
al Ihe zero crossing point, the peak value and frequency of Ihe 
incoming signal must be known. The peak value is measured by 
changing the eye position to the peak position of Ihe waveform 
and measuring the mean value in volts. From this value V le , Is 
subtracted to give V^m* The frequency is known 

Using Ihe following formula for the instantaneous voltage V, of 
a sine wave, the rms noise can be converted to litter in the time 

V, = Vfp,* x sin ml 

where id = frequency in radians/s and t = 1/o> x sin " ' (v7V pea J 

Timing jitter = T, = 1/ui x sin"' (ViWVpaw) seconds 

The results obtained during the development phase of the 
project and the results of the first production instruments were 
stored on disc The analysis of the test results calculated Ihe 
lowest and highest timing jitter values, the mean timing jitter, and 
the standard deviation at each frequency This ensured that there 
were no hidden specification problems Fig. 1 shows a typical 
distribution of timing |ilter test results. 

David Robertson 

Production Engineer 
Queensferry Telecommunications Division 

J i tier . i 1 

Obser vat . ons 



0 5 10 15 20 25 30 35 40 45 

Percentage of s pec • < i c at '■ on 
Customer spec=100i 

Total observations = 19785 
Fig. 1. Timing jitter test results. 

© Copr. 1949-1998 Hewlett-Packard Co. 


an angle to the axes, the X and Y ramp rates must be pro- 
portional to the X and Y components of the line so that 
both signals reach the endpoint at the same time. In a 
general-purpose vector display the ramp generators are pre- 
cision circuits since proportionality errors of only 1% are 
easily visible on a long line. Fortunately it was possible to 
limit graphics to horizontal lines, vertical lines, and short 
line segments at 45 degrees to the axes. This made it pos- 
sible to use the simple ramp generator shown in the dia- 

The analog section operates in three states: idle, hold, 
and ramp. In the idle state the output of each channel tracks 
the input. The error signals are small, so the limiters act 
as simple amplifiers. Overall each channel behaves as a 
unity-gain buffer with single-pole frequency compensa- 
tion. In the hold state the FET switches isolate the output 
integrators. This allows the inputs to be changed to a new 
graphics position while the outputs remain at the old po- 
sition. In the ramp state there is a large error between the 
input and output of one or both channels. The limiter 
supplies a constant current to the output integrator that 
causes the output to ramp in the direction of decreasing 
error. As the error becomes small, the limiter moves into 
its linear region. Each channel has a pair of comparators 
which detect position errors that exceed the limiter 
threshold. For a visible line, the display spot is enabled 
from the start of the ramp state until the position error 
signal becomes false, indicating that both channels have 
returned to the idle state. 

At the heart of the circuit is a differential limiter (Fig. 
14). This venerable circuit is one of the simplest designs 
for an op amp input stage. Ql supplies a constant current 
to the differential pair Q2 and Q3. Q3's collector feeds the 
current mirror formed by Q4 and Q5 and the output is 
added at Q2's collector to give an output current propor- 
tional to the differential input voltage. Input voltages of 
only 100 mV are sufficient to divert all current through 
one side of the differential pair, providing the limiting 

The complete circuit for each channel can be likened to 
an operational amplifier: the integrator amplifier corre- 
sponds to the output stage while the integrator capacitor 
corresponds to the frequency compensation capacitance. 
The main function of the circuit, the constant rate output 
ramp, corresponds to one of the least desirable features of 
an operational amplifier — output slew rate limiting. In this 
circuit, slew rate limiting is realized with a precision that 
is not currently available in commercial integrated circuits. 

The digital section latches the next graphics position for 
a line or move operation. For a line segment (ASCII charac- 
ter), it must generate the next position by adding relative 
segment offsets to the current position. These operations 
and the idle-hold-ramp-idle sequence for the analog section 
are controlled by the state machine. The control functions 
could be performed by the microprocessor, but perfor- 
mance would suffer. The state machine, implemented with 
a single 20-pin programmable logic device, is an extremely 
cost-effective solution. 

Display Graticule 

An unusual feature of the HP 3709A is the user-selectable 

graticule. Depending on the modulation scheme, a constel- 
lation display can have clusters arranged in two, three, 
four, seven, or eight parallel rows and columns. Instead of 
a fixed display grid, a different graticule is provided for 
each modulation scheme. These graticules are simpler than 
a single general-purpose graticule and provide an unam- 
biguous reference position for constellation measurements. 

Since the graticule is drawn on the CRT, display imper- 
fections affect the measured constellation and the graticule 
equally. The operator can use the graticule to resolve small 
levels of impairment on screen, eliminating any error 
caused by the CRT. This is important since some constel- 
lation impairments are similar to common CRT imperfec- 
tions. For example, traveling-wave-tube amplitude com- 
pression looks very much like CRT barrel distortion. 

The graticule generator provides drive signals to the vec- 
tor display module via a multiplexer, which selects be- 
tween the sampled analog constellation signals and the 
graticule. In addition to drawing horizontal and vertical 
graticule lines, the graticule generator generates display 
text for the graticule calibration factor, measurement re- 
sults, and error messages. 

Measurement Verification 

The measurements performed by the HP 3709A are com- 
plex and involve a combination of hardware and firmware. 
The three main stages are: sampling and displaying data 
(hardware), A-to-D conversion (hardware and firmware), 
and statistics accumulation and analysis (firmware). 

The hardware system of the HP 3709A can be checked 
by applying input signals and measuring accuracy, flatness, 
noise, et cetera over the HP-IB. However, the dedicated 
measurements of closure, lock, and quad angle cannot eas- 
ily be verified in this way. To allow firmware verification, 
the HP 3709A can accept data over the HP-IB that simulates 
sampled data from a constellation. A verification program 
was written on an HP 9000 Series 200 Computer to generate 
constellations with a variety of impairments for all modula- 
tion schemes covered by the HP 3709A. This program 
checks that the statistics accumulation agrees exactly and 
the analyses (closure, lock, and quad angles) agree within 
acceptable limits. 

HP-IB Firmware Verification 

A Series 200 BASIC program was written to exercise all 
the HP-IB commands separately and in sequence. The test 
program was written around the HP 3709A HP-IB External 
Reference Specification. This document defines all aspects 
of the HP-IB behavior of the instrument and is used as a 
basis for customer documentation. The program is written 
as a number of modules, each designed to exercise one 
HP-IB function as thoroughly as possible. Commands with 
a finite set of parameters are tested for each permissible 
parameter, while those with infinitely variable parameters 
are tested at key values (e.g., zero, limits, and just beyond 
limits) to test error detection and recovery. The modules 
are executed in a fixed sequence. 

The long-term benefit of this approach to interface testing 
is having a test program to run on any future revisions of 
HP 3 709 A firmware. 


©Copr. 1949-1998 Hewlett-Packard Co. 

Product Design 

During the design of the HP 3709A. ease of manufacture 
was a prime goal. Some of the points considered were: 

1. Keep the number of printed circuit boards lo a mini- 
mum The HP 3709A is made up of only seven printed 
circuit boards. 

2. Where possible, choose components that can be inserted 
in the printed circuit boards by machine to minimize 
hand loading. 50% of the components in the instrument 
are inserted automatically. 

3. Fewer printed circuit boards means that fewer cables 
axe required to interconnect them. There are 13 cables 
in the HP 3709A: five coaxial, five ribbon, and three 
simple wire looms. By mounting the processor board 
against the rear panel we were able to mount the the 
HP-IB socket and address switch directly on the proces- 
sor board. The electrical cable from the line power 
switch was eliminated by putting the switch on the line 
input board at the rear of the instrument and operating 
it via a flexible mechanical cable just like a bicycle brake 
cable. This is purchased as part of the line switch assem- 
bly. The advantages of this mechanical cable are that 
the hazardous line voltages are kept in one small area, 
and the 50/60-Hz line is not carried through the cabinet 
with its possible screening problems. This lype of cable 
linkage is less sensitive to mechanical variances than a 
rigid link would be. 

4. The preset adjustments and test pins should all be acces- 
sible with minimum effort. This was achieved without 
too much trouble. The only difficulty was providing 
easy access to the HP 1340 A Display Module's X/Y gain, 
position, and alignment controls. By mounting them 
along the top edge of the keyboard, we made them acces- 
sible through the spare mounting holes in the top of the 
front frame casting. 


Boyd Williamson did the product definitions for the HP 
3709A and was project manager through the early design 
phase. Peter Roubaud did the operating system and HP-IB 
firmware. Ross Maclsaac was responsible for the analog 
gain switching and resampler. Arthur Thornton did the 
product design and David Robertson was responsible for 
test software. 


1. H. Walker. "Modulation Schemes and Digital Radio Growth." 
Microwaves 6- RF, Vol. 26, no. 2, February 1987, p. 75. 

2. H. Walker, "Gauging Errors Set Digital Radio Quality." Micro- 
waves & RF. Vol. 26. no. 4. April 1987. p. 89. 

3. A Revieiv of Digital Radio Principles and Measurements. Hew- 
lett-Packard Publication No. 5954-7941 Icontains parts of refer- 
ences 1 and 2), 

4. K. Feher. Digital Communications: Microwave Applications, 
ISBN 0-13-214080-2. 

Constellation Measurement: A Tool for 
Evaluating Digital Radio 

by Murdo J. McKissock 

tool for alignment and fault diagnosis of digital 
radios, and an important indicator of the radio's per- 
formance margin. The HP 3709A Constellation Display is 
the first commercial instrument that provides the capability 
to make quantitative measurements of a constellation. 

Measurement data is provided at three different levels: 
raw samples, statistical accumulations, and analysis results. 
The raw samples represent individual constellation points 
and typically might be used to reproduce a constellation 
on a plotter, or to measure nonconstellation signals. The 
statistical accumulations provide information on the posi- 
tion, size, and orientation of each display cluster. This 
forms the basic data for analysis. The analysis results 
evaluate some common constellation impairments: relative 
rms cluster size, angle of rotation, and angle of quadrature. 
Using an external HP-IB (IEEE 488/IEC 625) controller it is 
possible to use the statistical accumulations to perform 



Q Axis 
+ 512 



I Axis 

► 512 


Fig. 1. Constellation measurement scale. 

© Copr. 1949-1998 Hewlett-Packard Co. 


further analysis to measure other constellation impair- 
ments such as amplitude compression, amplitude-to-phase 
conversion, and phase noise. 

Raw Constellation Samples 

The real-time display is generated from input samples 
taken at regular intervals. These analog samples drive the 
display directly. A small proportion of the samples are 
converted to digital form during a constellation measure- 
ment. Each sample has two 10-bit digital values, one each 
for the in-phase (I) and quadrature (Q) modulation compo- 
nents. Sample values are returned as pairs of integers using 
the measurement scale illustrated in Fig. 1. A typical mea- 
surement might take 1000 digitized samples and plot them 
to produce a hard copy of the constellation (Fig. 2). 

Statistical Accumulations 

Instead of measuring and outputting raw samples, the 
HP 3709A can process samples internally. For each cluster 
in the constellation, the instrument accumulates six values: 

Number of samples in the cluster: a* 

Sum of the I values: 

Sum of the Q values: 

x i|k 


Sum of the squares of the I values: "V xf jk 

Sum of the squares of the Q values: 


Sum of the products of the land Q values: ^ XijkYiik 

where {x i(k , y 1(k ) is the kth sample in the cluster with I and 
Q indices i and j. 

This is similar to the statistics accumulation occurring 
in many pocket calculators, but there can be up to 64 sets 
of 6 registers each, one for each cluster in the modulation 

From the register contents, it is easy lo compute the mean 
position, rms size, and correlation coefficient between the 
I and Q values for each cluster. Fig. 3 shows a plot from a 
computer program which uses this information to generate 
ellipses representing Gaussian probability contours. 

The individual samples are assigned to a particular "sam- 
ple bucket" or set of registers using slicing boundaries mid- 
way between the constellation graticule lines (Fig. 4). Ob- 
viously, the position and overall size of the constellation 
must be adjusted so there is just one cluster in each of the 
sample buckets. 

When the accumulation is complete, the contents of the 
registers can be output loan external controller or analyzed 
by the HP 3709A to obtain measures of constellation im- 

Analysis Results 

The HP 3709A computes four analysis results: 

■ I and Q constellation closure measures the relative rms 
cluster size in each direction. Constellation closure = 
[rms cluster size]/[0.5(cluster separation)] x 100% (see 
Fig. 5). 

■ Lock error measures the angle of rotation between the 
graticule and the cluster lines. This is related to receiver 
carrier phase lock error in a digital radio. In Fig. 6. lock 
error = (0, + 0 2 )/2. 

■ Quad error measures the quadrature error between the 
cluster lines. Quad error = 0 2 -0, (see Fig. 6). 


Fig. 2. Typical constellation plot (1000 samples). 


1 J 

1 — 1 

| ! 

i — 


Fig. 3. Constellation cluster ellipses. 


l Copr. 1949-1998 Hewlett-Packard Co. 

The results are determined using two separate analysis 

■ The cms cluster size is estimated, computed as an average 
over all clusters in the constellation. 

■ The position, spacing, and slant of the cluster lines are 
estimated using a least mean squared error (LMSE) esti- 
mator to compute the parameters of a suitable model. 
The rms cluster size is obtained by assuming that the 

distribution of points in each cluster is the same. The usual 
equation for standard deviation is modified to take into 
account the different mean positions of the individual clus- 
ters in the constellation. This gives the equations: 

4 = £ I [ | yf* - 4(| * ) 2 ] 

where M is the number of levels in the modulation scheme. 
Hence. M J is the number of clusters. N is the total number 
of samples. That is: 

M M 

i-t |-i 

The factor (N -M 2 ), sometimes called the number of de- 
grees of freedom, is used instead of N so that the estimate 
of rms cluster size is unbiased. It represents the total 
number of samples less the number of cluster mean position 
values used in the computation. The equations for s„ and 
s v are designed to give equal weight to all samples to 
minimize the unavoidable errors caused by random vari- 
ations in the distribution of samples. 

The rms constellation closure for each axis is computed 



1 si i 





— * 






* — i 
* — 


.f.. « 

= 1 M 

= 2 Li 

= 3 1 « 

= 4 

Fig. 4. Constellation slicing boundaries. 

©Copr. 1949-1998 Hewlett-I 

by dividing the rms cluster size s x or s v by one-half the clus- 
ter spacing p xx or p^ obtained from the LMSE estimator 

1 closure = 2 (s^p^J Q closure = 2 (tyfcyyj 

The term eye closure is often used to express the reduc- 
tion in the opening of an eye diagram. There is a difference 
between eye and constellation closure (see Figs. 5 and 7). 
Constellation closure measures only the size of individual 
clusters, while eye closure also includes the effect of over- 
lapping clusters in the eye diagram. A feature of the constel- 
lation display is that it is possible to resolve separately 
impairments that cause clusters to overlap in the eye dia- 
gram such as lock and quad errors. 

The position, spacing, and slant of the cluster lines are 
estimated using a modeling technique. It is possible to 
estimate, for example, the spacing from the mean positions 
of a few clusters. However, this ignores most of the available 
data so that random errors (measurement variance) in the 
result will be much larger than necessary. Even the average 
of all cluster spacings effectively uses only the outer clus- 

Instead, a model constellation is defined. All clusters of 
the model lie on equally spaced parallel lines. The param- 
eters of the model are the I and Q offsets, I and Q spacings, 
and I and Q slants. The parameters are computed such that 
the mean-squared error between the model and the mea- 
sured constellation samples is minimized. Fig. 8 shows a 
constellation with mainly linear impairments. Not only is 
the constellation rotated and out of square, it is slightly 
offset from the graticule. The best-fit model is shown by 
the superimposed lines. 

The model defines a linear relationship between the I 
and Q cluster indices and the cluster positions. As a result, 
the parameters that minimize mean squared error are easily 
computed from the constellation statistics. The model posi- 
tion of cluster i.j is given by: 

cluster separation 

Fig. 5. Constellation closure parameters. 



y« = I 1 I Pxv+ I ) "l - 1 Pyy + Py 

The parameters are represented as vectors: 

"■[•?] '-[?] 

The values are computed by solving matrix equations: 
AP X = B, AP y = B y 


Fig. 6. Lock and quad error parameters. 

Fig. 7. Constellation and eye diagrams. 

Lock and quad errors are computed from the model param- 

Lock, error = >/ 2 [tan~ 1 (p xy /p yy ) - tair'tp^/p^)] 

Quad error = - tan^'Cp^y/p^) - tan "'(Pyx/Pxx) 

The model gives results that make good use of a limited 
number of constellation samples. In addition, it neatly solves 
the question "What happens if the clusters do not lie on 
equally spaced parallel lines?" The model supplies a solu- 
tion that will approximate the true constellation. It cannot 
be an exact fit if the constellation does not fit the linear 
model. However, the results are at least consistent, and 

Fig. 8. Constellation best-fit 


©Copr. 1949-1998 Hewlett-Packard Co. 

have low measurement variance. To determine the form 
and extent of nonlinear impairments, it is necessary to 
perform additional analysis in an external controller. 

Notice in Fig. 8 that two of the corner clusters are slightly 
compressed towards the center of the constellation. The 
model gives more weight to the other fourteen clusters, so 
the lines do not pass through the centers of these corner 

Stand-Alone Operation 

All constellation measurement functions can be per- 
formed under HP-IB control. An external HP-IB controller 
can be programmed to collect constellation data for remote 
monitoring applications or to allow more extensive 

During stand-alone operation, access to the HP 3709A"s 

measurement functions is provided by the MEASURE ke\ 
on the front panel. When this key is pressed, the instrument 
accumulates statistics and displays the analysis results. 
This allows the operator to resolve small changes in the 
constellation which are often difficult to detect by eye. If 
an HP Thinkjet Printer is attached to the instrument, press- 
ing the PRINT key will produce a report (Fig. 9) containing 
a copy of the constellation plus the analysis results. This 
can be filed to provide a permanent record, allowing a 
carrier organization to monitor long-term changes in the 
performance of its radio links. 


Peter Roubaud developed the firmware for the HP 3709A 
HP-IB interface, including the PRINT function. 

HP3"7®3R Constel 1 £at ion Displ ay 





Modulai ion 
Scaling, I axis 
Scaling, 0 a«u 
I/O Delay 
Closure, I 
Closure, 0 
Lock Angle Error 
Quad Angle Error 

1 60AM 

220 mU/div 

233 nv/div 


14.4 H 

13.8 X 

0.0 ' 

0.1 ' 


Fig. 9. HP 3709A constellation re- 
port printed on a Think/el Printer 

© Copr. 1949-1998 Hewlett-Packard Co. 



July 1987 

4 "Constellation Display 
John R. Pottinger 

John Pottinger holds a 
1970 BSc degree in elec- 
tronics from North Slalts 
Polytechnic Institute and a 
1985 MSc degree from 
Heriot-Watt University With 
HP since 1 978, he contrib- 
uted to the development ol 
the HP 3724A, HP 3725A, 
and HP 3726A Baseband 
Analyzers before working on analog and digital 
radio measurements and constellation analysis He 
did the time base circuitry for the HP 3709A Con- 
stellation Display and is named coinventor on a pat- 
ent application related to constellation analysis 
Born in Reading, England. John now lives in Dun- 
fermline, Scotland He's married and has two chil- 
dren His outside interests include gardening and 
mountain climbing, especially in challenging winter 

Murdo J. McKissock 

Author's biography appears elsewhere in this 

David J. Haworth 

A proiecl manager at the 
Queensferry Telecommuni- 
cations Division, David 
Haworth has been with HP 
since 1972. He was re- 
sponsible lor the develop- 
ment of the HP 3709A Con- 
stellation Display, the HP 
371 7 A 70-MHz Modulator/ 
Demodulator, and the HP 
3756A 90-MHz Switch. He earned a BSc degree 
in eleciiunics from Salford University in 1964 
and is a specialist in high-lrequency analog de- 
sign, ampliliers. and oscillators. David Is married 
and has two children He likes photography and 
shares with Murdo McKissock an interest m Munro 

26 Analog Power Meter \ 

13 Constellation Measurement ! 

Murdo J. McKissock 

W With HP since 1981, Murdo 
McKissock worked on mi- 
croprocessor and graphics 
hardware, constellation 
analysis, and instrument 
control firmware for the HP 
3709A Constellation Dis- 
play He is named inventor 
lor one patent application 
/ ana coinventor for 3 sec- 
ond application. Both are related to constellation 
measurement. Murdo is a graduate ol the Univer- 
sity of Manchester. I nstitute of Science and Tech- 
nology (BSc electronics 1 981 ). A resident of South 
Queensferry, Scotland, he enjoys bicycling, archery, 
cross-country skiing, and go. Another pastime is 
Munro bagging . (This means that he climbs Scot- 
tish mountains over 3,000 feet. Sir Hugh Munro 
published a table of such peaks in 1891.) 

1 9 — Digital Radio Test Set 

Geoffrey Waters 

With HP's Queensferry 
Telecommunications Divi- 
sion since 1980. Geoff 
Waters is the R&D section 
manager responsible for 
developing instruments lor 
testing broadband trans- 
mission systems. He was 
proiect manager lor the HP 
3708A Noise and Interfer- 
ence Test Set. Before coming to HP he led an en- 
gineering group at Marconi Communication Sys- 
tems, Ltd. that developed equipment for micro- 
wave radio links and satellite earth stations. Bom 
in Sunderland, England, he earned a BSc degree 
in electrical engineering from the University of New- 
castle upon Tyne in 1966. He has wntlen several 
papers on diverse topics and is Interested m the 
evolution of broadband fiber-optic-based telecom- 
munication systems. Geoff and his wife and three 
children are residents of Edinburgh, Scotland His 
favorite leisure activity is field archaeology, espe- 
cially discovering unrecorded prehistoric sites He 
also sings in a choral society and is learning to play 
traditional music on the piano accordion 

Anthony Lymer 

I A 1 975 graduate of the Uni- 
I versity College of North 

Wales. Tony Lymer holds a 
I BSc degree in electrical 
I and electronic engineer- 
ing. He was a researcher at 
££\ f the University of Bath be- 
^fe* > lore coming to HP in 1 982 
He developed Hie rms-10- 
dc converter lor the HP 
3708A Test Set and more recently has worked on 
gate array design. He's coauthor of five articles on 
mobile radio modulation techniques and phase- 
locked loops and is a member of the Institution of 
Electronic and Radio Engineers. Born in Chelms- 
ford. England, Tony Is now a residenl of Edinburgh, 
Scotland, Cross-country skiing and hiking head his 
list of leisure activities. 

30~ Noise Generator and Reference : 

Dayananda K. Rasaratnam 

?Daya Rasaratnam was 
born in Colombo, Sri Lanka 
and studied violin at the 
Royal College of Music, 
London (ARCM diploma 
1 975) before continuing his 
education in electrical and 
electronic engineering He 
received a BSc degree 
from the University of Bir- 
mingham in 1 978 and joined HP the same year. An 
R&D engineer, he has contributed to the develop- 
ment of the HP 3724A Baseband Analyzer and the 
HP 3708A Test Set. At the same time, he was work- 
ing toward an MSc degree in digital techniques 
from Heriot-Waft University and completed work lor 
nis degree in 1984. A resident of South Queens- 
ferry. Daya is married and has a young daughter. 
Music is an important outside interest. He partici- 
pates in several orchestral concerts each year and 
plays duets with his wife for church meetings. 

36 Automated Radio Testing: 

John A. Duff 

With HP since 1983. John 
Duff contributed to the de- 
velopment of the HP 3708S 
Measurement System and 
is currently working on a 
system for monitoring data 
lines. His professional spe- 
cialty is workstation-based 
instrument control systems. 
He was born in Earley, 
Berkshire, England and educated at the University 
of Southampton (BSc electronic engineering 
1 983) . He now lives in Edinburgh, Scotland , is ac- 
tive In scouting, and is studying for an MBA degree 
John likes all sports, especially white water canoe- 
ing, swimming, golf, and squash. 


©Copr. 1949-1998 Hewlett-Packard Co. 

A Digital Radio Noise and Interference 
Test Set 

This instrument facilitates the measurement of the bit error 
ratio (BER) for a digital communication system under 
simulated path fade conditions. A desired C/N or CI I ratio 
can be established and maintained in the presence of 
received radio signal variations. 

by Geoffrey Waters 

SINCE THE LATE 1970s, an increasing proportion 
of long-haul telecommunications link equipment 
has used digital modulation techniques and this 
trend is expected to continue. Because of the nature of the 
error-generating mechanism in digital transmission, the ac- 
curate evaluation of a digital radio in terms of BER (bit 
error rate) versus C/N (carrier-to-noise) ratio requires a 
higher degree of measurement accuracy and repeatability 
than is required to evaluate an analog radio. Furthermore, 
the crest factor of the noise causing the errors assumes a 
more vital role. These factors initiated the development of 
the HP 3708A Noise and Interference Test Set (Fig. 1). 

Establishing a C N or C/N„ Ratio 

The BER is usually measured over the range of C/N ratios 
encountered during hostile propagation conditions. The 
required C/N ratio is established either by varying the re- 
ceived signal level (RSL) by means of an attenuator (varying 
C) or by injecting additive noise into the receiver IF (varying 
N). Frequently in the laboratory, factory, or field, a variable 
RF attenuator is inserted into the microwave receiver input 
waveguide to attenuate the RSL. The thermal noise from 
the receivers front end then defines the C/N„ (carrier-to- 
noise-density) ratio according to: 

C/N„ = RSL + 174 - F at 17°C 

where F is the noise figure of the receiver in dB and 174 
dBm/Hz is the thermal noise floor at 17°C. 

For example, assume the receiver noise figure is 6 dB 
and RSL is - 70 dBm for a 10~ 6 BER. then C/N D = 98 
dB-Hz. This ratio remains unchanged throughout the re- 
ceiver RF. IF, and predetection circuits because signal and 
noise are amplified equally. The typical unfaded C/N 0 ratio 
would be about 40 dB higher, or 138 dB-Hz. and the BER 
would revert to a typical residual value of much less than 
10 _IU . A 40-dB attenuator can be used to vary the C/N„ 
ratio from its unfaded value, allowing a BER curve to be 
plotted as a function of C/N D ratio down to the receiver's 
threshold. Additional information on establishing a C/N or 
C/N„ ratio is given in reference 1. 

Unlike ON,,, the C/N ratio in the receiver varies at points 
throughout the RF/IF chain and depends on the appropriate 
noise bandwidth. That is: 

C/N = C/N„ - 10 log B„ 

where B,, is the noise bandwidth at the point of interest. 
In terms of RSL: 

C/N = RSL + 174 - F - 10 log B t . 

Using this equation, the C/N ratio can be related to the 

(conlmued on page 21 1 


© Copr. 1949-1998 Hewlett-Packard Co. 

Carrier Noise Ratio (dB) 

Fig. 2. Probability of symbol error versus ON ratio for finite noise crest factor c for N-phase 
PSK (phase shift keying) data transmission and (b) QAM (quadrature amplitude modulation) 

data transmission 


© Copr. 1949-1998 Hewlett-Packard Co. 



Down I 



Convener 1 


HP 3764A 
HP 3782A B 
HP 3789A B 

IF Filter IF Amplifier 

Alternative access 
points to which 
HP 3708A can be 

Radio Receiver 



HP 3764A 
HP 3781 A B 
HP 3789A B 

Fig. 3. Typical test configuration 
using HP 3708 A Test Set. 

RSL for the purposes of flat fade simulation. This method 
relies on knowing the noise bandwidth and noise figure 
for every radio under test. The sensitivity of the method 
to these values is unacceptably high, since BER can change 
by more than one order of magnitude for a C/N change of 
less than 0.5 dB. 

The traditional method of fade simulation uses a power 
meter to check the effective C/N ratio in the receiver's IF 
strip or to measure the RSL at the RF stage. In the latter 
case, the levels are low and the nominal RSL is measured. 
The attenuator accuracy is relied on to set the desired RSL. 
assuming that the incoming signal level does not vary. 

The RF attenuator method is time-consuming and incon- 
venient. In field tests, variations in RSL make it difficult 
or impossible to fade the RF signal as desired and make 
accurate BER measurements, particularly when working 
near receiver thresholds where a small scintillation fade 
can cause loss of synchronization. 

Sometimes the inaccessibility of the waveguide at- 
tenuator makes the measurement slow, and it is difficult 
lo automate. Matching problems and inherent attenuator 
inaccuracy at microwave frequencies reduce the reliability 
and repeatability of the measurements and increase the 
probability of operator error. There are loo many uncon- 
trolled variables for repeatable measurements. 

Another method, the additive noise method, cannot 
check the overall fade margin but gives an accurate analysis 
of the C/N penalties caused by individual impairments in 
the radio. It is difficult and sometimes impossible to obtain 
this using the traditional method. The additive noise 
method establishes a C/N, C/N„, or E,/N 0 (energy/bit-to- 
noise-density) ratio by injecting relatively high-level noise 
into the receiver's demodulator or IF al normal RSL. The 
wideband noise spectral density of a noise generator is 
filtered to the desired bandwidth and supplied via an at- 
tenuator and amplifier combination to a network where it 
is combined with the IF carrier. The C/N ratio is established 
in the known noise bandwidth of the filter. 

Fig. 4. Injected noise level versus variation in IF signal power 
(a) Variation = 10 dBls. (b) Variation = 50 dB/s. The top 
waveform in each plot is the sinusoidally varying carrier level 
and the bottom waveform is the averaged noise level 

© Copr. 1949-1998 Hewlett-Packard Co. 


Noise Crest Factor Enhancement 

Crest factor is by definition, the ratio of peak value to rms 
value The noise crest factor at the noise output or the IF output 
of the HP 3708A is defined as the ratio of the hard-limiting voltage 
level to the rms value of the noise voltage It has been shown 1 
thai the crest factor of noise used for determining bit error rate 
(BER) versus carrier-to-noise (C/N) ratio curves can significantly 
affect the accuracy of these curves for low BERs. Although a 
crest factor greater than 15 dB is sufficient to achieve good 
accuracy at BERs down to 1CT 6 . it may sometimes be necessary 
to observe digital radio operation at significantly lower BERs. 

The HP 3708A control algorithm in C/N mode makes it possible 
to have several different combinations of fine attenuator and step 
attenuator settings with a given carrier level and C/N ratio. This 
property can be exploited to enhance the noise crest factor at 
the IF output by exchanging attenuation between the fine at- 
tenuator and the step attenuators to reduce the level at the input 
to the final pair of amplifiers in the noise chain. The fine attenuator 
can. by this method, be set with a resolution of 1 PB to any part 
of its dynamic range provided that the resulting noise level at 
this output is s-10 dBm and - 152 dBm/Hz." The highest 
crest factor that can be so obtained is typically 25 dB 

The following general procedure lets the user establish a de- 
sired noise crest factor between 15 dB and 25 dB with a given 
carrier level and C/N ratio. Let the desired crest factor be k dB. 
where k is an integer such that I5«k«25. Then: 
1 Start with the desired C/N ratio as the current entry 

2. Enter desired C/N ratio incremented by &12 dB 

3. Enter desired C/N ratio incremented by x dB, where x = 20 — k. 

4. Reenter desired C/N ratio (if x is not zero). 

"Value in dBm is Mter dependent. 


I. I Young and G Waters. "Practical Error Probability Estimation for Digital Radio 
Systems m the Presence ot Interference and Noise ot Finite Crest Factor, and the 
Prediction ot Residual Error Rata, IEE International Coherence on Measurements tor 
Telecommunication Transmission Systems— MTTS 85. IEE. London. November 27-28. 

Dayananda K. Rasaratnam 

Development Engineer 
Oueensferry Telecommunications Division 

Finite Crest Factor Noise 

The injected noise used in C/N tesling must fulfill certain 
criteria: it must have sufficient bandwidth, a defined spec- 
tral density, and the correct statistics. The last requirement 
is frequently ignored. 

The crest factor of a waveform is the ratio of the peak 
voltage to the rms value. An ideal Gaussian noise source 
process has infinitely high peaks. In other words, there is 
a finite chance that a peak as high as 7u will occur, even 
though this probability is only 1 in 10 12 . In practice the 
maximum noise output is constrained by power supply 
limitations. Gaussian noise with an infinite crest factor 
always produces a small but finite BER, no matter how 
large the C/N ratio is. However, when the crest factor is 
finite, reducing the noise below a certain threshold (or 
equivalently. raising the C/N ratio above a certain limit] 
makes the BER zero. 

The theoretical probability of various symbol error rates 
has been computed for N-phase PSK (phase shift keying) 
and QAM (quadrature amplitude modulation) systems for 
various finite crest factors. The results are shown in Fig. 
2. The effect of noise crest factor on the error rate can be 
seen together with the bound in C/N ratio discussed above 
where the error rate falls to zero. The crest factor should 
be high enough to allow accurate measurement at the 
appropriate BER. If this is not the case, the BER-versus-C/N- 
ratio plot for the radio under test can deviate significantly 
from the theoretical curve, and different noise sources will 
produce different deviations. A critical factor in the design 
of the HP 3708A was to maximize the noise crest factor. 
Over the dynamic range of the instrument, typical values 
of 15 dB to 25 dB are achievable. 

HP 3708A Principles of Operation 

Two methods of establishing a C/N ratio have been de- 
scribed. The additive noise or noise injection method is 
used by the HP 3708A (see Fig. 3). Noise of known spectral 
density is injected into the IF section of the receiver under 
lest to establish the desired carrier dependent ratio (C/N, 

Power Me 


IF Input ^ 

Input ' 

(I Input) 






HP 3708 A 

Noise Generator Section 



Bandpass E 





HP 370BA 
Power Meter 

6-dB Loss 

15-dB Loss , 



C I or C N 

6-dB Loss 

70.' 140-MHz. 


I or N 

> IF Output 




' Output 

Fig. 5. Block diagram ot HP 


l Copr. 1949-1998 Hewlett-Packard Co. 

Noise Bandwidth Measurement 

The HP 3708A can measure !he equivalent noise bandwidth 
of bandpass filters centered at 70 MHz or 1 40 MHz This measure- 
ment is automatic except for one connection change Other filters 
in the range of 10 to 200 MHz can be automatically measured 
after the insertion loss at the titter center frequency is entered 
into the HP 3708A 

The filter noise bandwidth (NBW) is the width in Hz of a rectan- 
gular filter that gives the same total noise power output as the 
actual filter, and has the same level of output noise density at 
the center frequency An NBW of 10 MHz might be expressed 
as 70 dB-Hz where NBW (dB) = 10 log NBW (Hz) An NBW oi 
20 MHz would pass twice the power and would be given as 73 

The HP 3708A measurement sequence starts with a gain or 
loss measurement of the filter under test at the band center 
frequency (70 MHz or 140 MHz) Using the gain or loss value 
and the input noise density, the HP 3708A calculates the output 
noise density at the center frequency. The total output power 
divided by this output noise density gives the NBW, which is 
then displayed on the front panel. 

Fig. 1 shows a noise bandwidth calculation for a 70-MHz filter 
In Fig. 1 , the filter output noise density is 1 mW/Hz at 70 MHz 




70-MHz Bandpass Filter 

(Device Under Test) 
Loss = 3 dB at 70 MHz 





Noise Density 
= 2/iW Hz 

and so a bandwidth of 1 MHz at this noise density must be 
postulated to account for the output power of 1 watt The equiva- 
lent noise bandwidth is therefore 1 MHz in this case the input 
noise density is 2 /iW/Hz since there is a 3-dB loss at the band 

Fig. 1 also illustrates a method of measunng noise power den- 
sity. When the output power. NBW. and loss of the filter are 
known, the input noise density can be calculated For example, 
since the output power is 1 watt, the noise bandwidth is 1 MHz. 
and the loss at 70 MHz is 3 dB. the constant-level input noise 
density extending over the test-filter window must be 2 >iW/Hz. 

The shape of the filter response is not critical. The filter and 
power meter represent a black box This method is used to 
measure the output noise density of the HP 3708A in production 
test. Periodic calibration of the filter and power meter combination 
to find the relation between input noise density and power meter 
reading is carried out by a computer-controlled routine using a 
signal generator, a power meter, and numerical integration 

/an M. Matthews 

Development Engineer 
Queensferry Telecommunications Division 


Noise Density 
N„ at 70 MHz 
= 1/iW Hz 

Power = 1 wait 

70 MHz 

Noise Bandwidth = Total Power Power per Hz 

= P N. = 1 MHz 

Fig. 1 . Derivation ot noise band- 
width NBW from total output power 
and output noise density at 70 

C/N„, or E|/N 0 ). The receiver IF power is measured at the 
point of noise injection, and the noise density is adjusted 
automatically under microprocessor control to maintain 
the required carrier dependent ratio in the presence of re- 
ceived signal level variations. The typical instrument re- 
sponse time is 10 ms. which enables BER testing to be 
performed in the field in the presence of rapid changes in 
RSL. Fig. 4 shows the injected noise tracking sinusoidal 
variations in IF signal power at 50 dB/s and lOdB/s. Because 
the carrier power is measured at the point of noise injection, 
the required carrier dependent ratio is maintained at any 
point chosen for noise injection throughout the IF chain. 

The block diagram of the instrument consists of a power 
meter, a noise source, an injection assembly, and a micro- 
processor (Fig. 5). To this is added a 0-dBm reference 
source, an HP-IB (IEEE 488/IEC 625) port, and a power 

The user's radio signal is passed through the injection 
assembly via the HP 3708A's IF input and output ports. 
The injection assembly has 0-dB loss and negligible trans- 
mission impairments to carrier powers as high as 5 dBm. 

Noise injection is controlled by a switched attenuator 

and a continuously variable pin diode attenuator in the 
noise source. When a requested C/N ratio is established, 
the microprocessor sets the switched attenuator so that the 
pin diode attenuator is in the middle of its range. This 
allows a rapid variation of the noise level over ±5 dB 
without operating the switched attenuator, which ensures 
fast response to carrier variations. In some special applica- 
tions this noise tracking facility is not required and a track 
inhibit mode of operation is available. 

The HP 3708A can be used to test TDMA (time division 
multiple access) burst mode systems. The carrier power is 
measured with an external burst mode power meter and 
entered into the HP 3708A using the ENTER C data entry 
key in track inhibit mode. 

Special Features 

Three ratios are available on the HP 3708A: carrier-to- 
noise ratio, carrier-to-noise-density ratio, and energy/bit-to- 
noise-density ratio. The last two ratios are independent of 
bandwidth and are particularly useful in R&D work for 
comparing the efficiency of different radio systems. These 
measurements require an accurately known noise density 

© Copr. 1949-1998 Hewlett-Packard Co. 


for injection into the radio's IF section. The concept of 
Ei/N,, ratio results in a figure of merit for comparing systems 
that emphasizes the minimization of the total transmitted 
energy required to convey a given amount of data. E b is 
obtained by dividing the carrier power C by the bit rate f|,. 
Therefore, E,,/N H = (C/f b )(l/N Q ), or in dB. E h /N„ = C/N 0 - 
10 log f b . 

After measuring the IF carrier power, the HP 3708A must 
know the transmission bit rate of the system under test to 
inject the appropriate noise density to establish a required 
E b /N„. This can be entered via the data entry keys on the 
front panel If the receiver noise bandwidth is B fi Hz and 
the total noise power measured in this bandwidth is N 
watts, then N„ = N/B e watts/Hz. Hence, E b /N a = (C/N) 
(B e /f b ), or in dB. E b /N„ = C/N - 10 log f b /B 0 . Therefore, if 
the noise bandwidth equals the bit rate, E b /N„ = C/N. 

The measurement of the BER of a system at a particular 
C/N ratio must be related to a specified system bandwidth 
(e.g.. the noise bandwidth of practical receiver filters) or 
to the theoretical minimum Nyquist bandwidth (symbol 
rate bandwidth). Therefore, a system bandwidth key is as- 
sociated with the C/N data entry key on the HP 3708A. 
This allows noise of appropriate noise density to be injected 
at a convenient point in the radio's IF section to establish 
a required C/N ratio at another point (e.g., downstream in 
the bandwidth of the baseband filters at the regenerator 

The HP 3708A operates at common microwave receiver 
IF bands from 10.7 MHz to 140 MHz. The injected noise 
is band limited in the range of 10 MHz to 200 MHz by a 
choice of four internal filters or by connecting an external 
filter of the operator's choosing. To ensure that the injected 
noise density is calibrated, the noise bandwidth of each 

filter in each instrument is individually measured and 
stored as a soft constant. 

Because the HP 3708A contains an accurate noise source, 
an IF power meter, and a microprocessor, the instrument 
can be configured to measure the noise bandwidth of an 
external filter in seconds, thus avoiding the lengthy nu- 
merical integration procedures normally used (see "Noise 
Bandwidth Measurement," page 23). To increase the instru- 
ment's flexibility, both its internal power meter and its 
noise source are accessible from the front panel for more 
general-purpose applications. 

Additive Interference Testing 

The HP 3708A offers two distinct facilities for interfer- 
ence testing. Aimed at different applications, both facilities 
have broadband inputs with a frequency range of 10 MHz 
to 200 MHz. 

Co-channel and adjacent channel interference effects in- 
fluence the design of radio hardware and are of interest to 
the frequency planner and regulatory agencies. These ef- 
fects can have a predominating influence on the BER per- 
formance of a radio. It is common practice to measure the 
C/N threshold degradation that occurs in the presence of 
interference over a range of co-channel and adjacent chan- 
nel frequencies. Fig. 6 shows a typical co-channel C/I (car- 
rier-to-interference) plot for an 8-phase PSK radio. The deg- 
radation in threshold can be seen clearly. 

To simplify this measurement, the HP 3708A has an 
auxiliary inlerferer input on its rear panel with a fixed loss 
to the HP 3708A's IF output port of typically 15 dB. The 
required C/N ratio is established via the keyboard in the 
normal manner and the C/I ratio is determined by the level 
presented to the auxiliary interferer input. In this mode of 

16 18 20 22 24 26 

C/N Ratio (dB) 

Fig. 6. Typical co-channel camer-to-interlerence ratio plot 
for an 8-phase PSK radio. 

160 -- 


SS ioo- 




0 1 I 1 I h- 

-0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 

Deviation (dB) 

Fig. 7. ON ratio repeatability lor HP 3708A measurements 
at 70^20 MHz. 


©Copr. 1949-1998 Hewlett-Packard Co. 

operation the established ON ratio is independent of vari- 
ations in received carrier power, but the Ol ratio is not. 

An external source of interference (modulated or unmod- 
ulated) can be used or the internal O-dBm calibrator (nor- 
mally at 70 MHz and 140 MHz) can be substituted. 

Because of a growing interest in deterministic testing of 
QAM radio systems (rather than statistical testing with 
noise), the OI mode of the HP 3708A allows the injection 
of a noise-free additive sinusoidal tone. In this mode of 
operation the automatic leveling algorithm maintains the 
OI ratio in the presence of received signal level variations. 
The interferer is connected to the input port on the instru- 
ment's front panel. 

The residual BER of a radio system at nominal received 
signal level is practically impossible to measure because 
the mean time between errors can be from tens to hundreds 
of hours. The C/l ratio required to establish a specific BER 
depends on the amount of intersymbol interference and 
residual noise present around the phase states in the con- 
stellation display. The estimation of this C/l ratio can be 
used to predict the residual, or dribble. BER for a radio 
system. This technique is described in reference 2. 

Measurement of residual BER can now be made in min- 
utes as opposed to the three days previously required to 
wait for 30 errors (100 Mbits/s. BER = 10""). Measurement 

440- - 

400- - 
360- - 
320 - 
280- - 


S 24°- - 



jj 200- - 


160 - - 

80 - 

40- - 

-0.04 -0.03 -0.02 -0.01 0.0 0.01 0.02 0.03 0.04 
Devialion (dB) 

Fig. 8. Linearity ol internal HP 3708A power meter over 1240 

of residual BER is now feasible in the field and test time 
can be saved in production, allowing repeated tests to be 
made for evaluation of adjustments. 

Self Calibration Enhances Accurate Operation 

When measuring the fade margins of a microwave radio 
system, the accuracy with which the ON ratio can be 
specified is of vital importance. This is primarily because 
the slope of the BER-versus-ON-ratio curve is so steep that 
less than a 0.5-dB change in ON ratio can result in one 
order of magnitude variation in BER. Hence, the HP 3708A 
uses a powerful 16-bit microprocessor to implement 
sophisticated calibration procedures that enable a ON or 
OI ratio to be established to a typical accuracy of 0.1 dB 
at room temperatures. 

Each signal path in the HP 3708A and the noise 
bandwidths of the noise defining filters are characterized 
by individually measured software constants. This calibra- 
tion data is held in nonvolatile memory- The exact values 
are determined for each instrument during production test 
to achieve maximum accuracy. Thirty-three soft constants 
are stored in each HP 3708A. These constants are: 

1. RF path gains and losses 

2. Filter noise bandwidths 

3. Measured attenuator incremental attenuation values. 
A temperature sensor ensures that an autocalibration 

cycle is initiated should the noise source temperature 
change by 3°C or more. 

The power meter can be calibrated using the accurate 
internal 0-dBm source described in the article on page 30. 
and subsequent measurements are referred to this level. 

The HP 3708A specifications include test station mea- 
surement uncertainties and guarantee traceability to inter- 
national standards. ON ratio accuracy and repeatability 
are measured by a substitution method. A precision cali- 
brated attenuator is used to attenuate the carrier signal 
until its power is within ±1 dB of the noise being generated 
by the HP 3708A under lest at a specific ON ratio. The 
noise and attenuated carrier power levels are then mea- 
sured separately on the same range of an HP 438A Power 
Meter connected to the IF output port of the HP 3708A. 
The error in the ON ratio is computed from the measured 
results and known calibration data for the attenuator. For 
a typical production HP 3708A. ON ratio accuracy was 
measured at carrier powers of 5, 1, -2. -5. -10. -20. 
and -40 dBm and at ON ratios of 0. 10. 20, 30. and 40 
dB. The test was repeated automatically 416 times and 
consisted of 26 combinations of carrier power and ON 
ratio. The ON ratio repeatability is shown in Fig. 7. For 
16 out of the 26 combinations the standard deviation of 
the ON ratio accuracy was less than 0.01 dB. The standard 
deviation of the mean of the 26 combinations was 0.02 dB. 
The ON ratio accuracy was ±0.06 dB with 99.7% confi- 
dence. These measurements were made with noise gener- 
ated via the 70 ± 20-MHz internal filter. 

Power meter linearity was measured al 31 different 
power levels from 6 to - 45 dBm and the lest was repeated 
40 times, resulting in 1240 measurements. The results are 
shown in Fig. 8. The power meter is linear to ±0.03 dB 
with a 99.7% confidence level. 

© Copr. 1949-1998 Hewlett-Packard Co. 



I would like to thank the following people for their en- 
thusiasm and contributions to the instrument. Brian Wood- 
roffe developed the firmware, HP-IB circuitry, and micro- 
processor circuitry. David Stockton, whose ideas are to be 
found in many parts of the instrument, designed the C/N 
switch in the power meter, and developed the noise filter 
board. Daya Rasaratnam designed the power supply and 
Ian Matthews developed the noise injection assembly and 

power meter attenuator. James Gentles was the produc- 
tion engineer. Harry Elder was responsible for the product 


1. G. Waters, "Digital Radio Measurements," Telecommunication 
Measurements, Analysis and Instrumentation, K. Feher, editor. 
Prentice-Hall, 1986. 

2. I.K. Compston, Methods for Estimating Residual BER in Digital 
Radio. Hewlett-Packard Company, Publication No. 5954-2036. 

Microprocessor-Enhanced Performance in 
an Analog Power Meter 

by Anthony Lymer 

SET can be used to test microwave point-to-point 
radios. It simulates flat (frequency independent) fad- 
ing and co-channel and adjacent channel interference by 
injecting band-limited noise or interference into the IF 
stages of the radio. The instrument continuously monitors 
the incoming carrier level and adjusts the impairment level 
to maintain an accurate carrier-to-noise (C/N) ratio or car- 
rier-to-interference (C/I) ratio. This feature is particularly 
convenient when live radio traffic is being used as the test 
signal to measure fade margin. The radio signal may vary 
because of multipath interference effects. A 0.5-dB change 
in signal level might, in the absence of a tracking C/N test 
set, lead to an order of magnitude change in bit error rate 
(BER) for the link, making repeatable measurements dif- 
ficult to obtain. 

A special feature of the HP 3708A, its tracking capability, 
demands a very fast response time from the power meter 
that measures the incoming carrier level and the impair- 
ment level. A settling time of 3 ms has been achieved, so 

the instrument response time is limited by the rate at which 
the processor samples the carrier and updates the noise 
output, and not by the power meter. A custom thermal 
converter IC manufactured by HP 1 is used to achieve both 
short settling time and true rms response. The latter is 
important when dealing with white Gaussian noise and 
with both sinusoidal and modulated digital radio signals. 

Since carrier power and noise power are not usually 
measured at the same levels, the linearity of the power 
meter over its range of 5 dBm to —45 dBm is of critical 
importance. A very simple but powerful enhancement 
technique employing a microprocessor solves the problem 
of producing the very accurate attenuators necessary to 
meet the stringent linearity requirements of C/N ratio test- 
ing. Operator conveniences such as automatic sensor zero- 
ing with temperature changes, the capability of calibrating 
the power meter against the built-in 0-dBm reference oscil- 
lator at the touch of a button, variable-time-constant averag- 
ing, and calculation of filter noise bandwidths are easy and 
cheap to provide. HP-IB control eases the problems of test- 

From Noise 
-9 dBm, max.) 



5, 10, 10, 10 OB 


From Carrier 

(5 dBm to -45 dBm) 



Noise Source 


and Control 


Fig. 1. Block diagram ol the 
HP3708A's internal power meter. 


© Copr. 1949-1998 Hewlett-Packard Co. 

ing the instrument, so it is no longer an optional extra. 
These were major influences on the design of the power 
meter section of the HP 3708A. 

Oual-Purpose Power Meter 

A block diagram of the power meter is shown in Fig. 1. 
The first functional block is a pin diode switch (Fig. 2) 
which selects either the carrier at the IF input or the noise 
from the final amplifier in the noise chain. This is followed 
by a processor-controlled step attenuator with one 5-dB 
and three 10-dB sections. These values were chosen to 
reduce the need to change range with cyclically varying 
signals. Such signals occasionally occur in microwave 
radio links. The basic rms converter operates over a 12-dB 
range without loss of accuracy. With a 5-dB difference be- 
tween ranges, a signal must vary by at least 7 dB to require 
a change in range (Fig. 3). If only 10-dB ranges were used, 
a 1-dB peak-to-peak fade at either end of a range would be 
enough to produce switching relay chatter, causing unneces- 
sary wear. 

To increase the signal level to a suitable power for the 
thermal converter, a 50-dB fixed-gain amplifier is provided. 
The final stage is electronically switchable so that the ther- 
mal converter can be isolated from the RF input when it 
requires zeroing. Finally, the signal reaches the thermal 
converter, where the dc value equivalent to the true rms 
voltage of the input signal is generated and converted to a 
12-bit binary number. This number forms the address of a 
lookup table for the power level in dB seen at the thermal 
converter input. A simple calculation is made by the pro- 
cessor using the loss in the step attenuator, gain in the 
amplifier chain, and power at the thermal converter to give 
Ihe power at the IF input in dBm. 

PIN-Diode Switch Isolation 

The C/N switch (Fig. 2) selects either the carrier path or 
Ihe noise path for power measurement. If noise leaks 
through the switch when the carrier is selected, then inac- 
curate readings will result. This is also true when noise is 
selected, but the problem is less severe. 

The required isolation can be estimated by considering 
the maximum noise power at the switch input, -9 dBm. 

D1 02 To Second 



: C3 -< ui 

Fig. 2. ON pin diode switch (one of two channels) 

and the minimum carrier signal expected. -45 dBm. The 
power difference is 36 dB. For the carrier power to be 
accurate to 0.01 dB. it can be shown that the carrier must 
be greater than the noise by another 26 dB. So. an isolation 
of 62 dB is needed. Because of the broadband nature of the 
noise and modulated carrier signals, amplitude response 
with frequency and return loss at the inputs are also of 
importance. The input attenuator (Rl . R2. and R3) ensures 
a good match and allows the dc switching current for the 
diodes to be injected. A positive control voltage switches 
Dl and D2 into conduction and reverse biases D3. D4. and 
D5. This allows power to reach the output. If a negative 
control voltage is applied. Dl and D2 block the signal path 
while D4 and D5 shunt any leakage through Dl to ground, 
D3 is also forward biased and allows R4 to terminate the 
attenuator in its characteristic impedance. A second iden- 
tical section is used for the other input channel, with all 
the diodes reversed so that the same switching voltage can 
be used. The low ground impedances and the level of screen- 
ing required made it necessary to house the C/N switch in 
a machined enclosure with some components mounted in 
individual cavities. 

Software-Enhanced Accuracy 

The overall power meter linearity specification of 0.1 dB 
does not allow the individual attenuator sections to be 
more than 0.01 dB different from the nominal values. It 
seemed an impossible task to manufacture attenuators in 


Power Range 
dBm 1 

Range 1 


10 dB between 

5 dB between 

Range 2 

1 dB 


(12 dB) 

Fig. 3. Power meter ranges With W-dB differences between 
ranges (left), a signal greater than 1 dB peak-to-peak may 
require a range change. With 5-dB differences between 
ranges, as in the HP 3708A (right), a signal must vary in level 
by at least 7 dB to effect a change in range. 

© Copr. 1949-1998 Hewlett-Packard Co. 


Fig. 4. Photograph of program- 
mable input attenuator 

quantity to this standard. The solution adopted uses the 
computing power available to enhance the accuracy. The 
attenuators are measured in the power meter circuit at final 
calibration and the measured value is used by the computer 
in any calculations. These values and others, such as the 
measured gain of the power meter amplifiers, are stored in 
electrically erasable read-only memory (EEPROM) on the 
processor printed circuit board. The idea was so successful 
it was extended. Thirty-two calibration values are now 
stored in this way, including noise bandwidths of filters 
and noise step attenuator values. Consequently there is 
only one gain adjustment in the entire power meter. 

The programmable attenuator used in power meter au- 
toranging is shown in Fig. 4. Each compartment contains 
a printed circuit board loaded with a relay (TO-5 case size) 
and a 7511 attenuator section. The amplitude response from 
10 MHz to 200 MHz shows an almost constant droop at all 
attenuator settings. The droop is caused by the RF loss of 
the TO-5 case header and is compensated by the following 

Power meter calibration is accomplished by connecting 
a 0-dBm level to the power meter input and pressing a key. 
The processor notes the power reading obtained and sub- 
tracts this value from the input power, forcing the meter 
to read 0.00 dBm. The 0-dBm calibration source provides 
the common digital radio IF frequencies of 70 MHz and 
140 MHz. 

Variable-time-constant averaging is another operator con- 
venience made possible by the microprocessor at very low 
cost. Short-term variations in a signal can be averaged to 
give a more stable display when desired. At power-on, the 
power meter time constant defaults to around 150 ms, but 
can be reduced to around 15 ms or increased to 3 seconds. 
Settling time is about 3 to 4 time constants. 

True RMS Thermal Converter 

The HP 3708A's power meter must measure sinusoidal 
unmodulated carriers, carriers with digital modulation, 
and band-limited (pink) noise. It is important that a true 

rms value is measured and not an average value as would 
result from a diode detector. The true rms voltage is the 
dc value lhat gives the same heating effect as the incoming 
waveform, Using a thermal converter with two resistor- 
diode sensors ensures that this happens. The circuit dia- 
gram is shown in Fig. 5. The RF signal is applied to 
Rl, which is in close thermal contact with Dl. The voltage 
across Dl has a temperature coefficient of -2.2 mWC. 
The combination of Rl and Dl heats up by about one degree 
Celsius per milliwatt of RF supplied to Rl. D2 and R2 are 
another resistor-diode pair matched to Dl and Rl, but ther- 
mally isolated from them. The difference between VI and 
V2 is amplified, integrated, and fed back to resistor R2. 
The output of the integrator completes the negative feed- 
back loop and ensures that the power in R2 is the same as 
in Rl. Therefore, the voltage at the output of the integrator 
is proportional to the rms voltage across Rl. 

+ V„ Bias 

^ Balance Adjustment 

Fig. 5. Thermal converter circuit. 


©Copr. 1949-1998 Hewlett-Packard Co. 

Thermal Effects 

Several causes of inaccuracy and nonlinearity in this 
kind of power meter are thermal in origin. They include 
the thermal EMF generated at the copper /Kovar" lead junc- 
tions, temperature drift of the offset voltage of the sensor 
diodes and associated amplifier, and excessive heating 
caused by an overload in the thermal converter. 

The magnitude of the thermoelectric voltages generated 
at the copper/Kovar junctions can be as much as 35 jtV/°C. 
The temperature differences can be caused by the proximity 
of heat-producing components or by the forced air cooling 
system in the instrument. In the HP 3708A's thermal con- 
verter, these effects are reduced by housing the sensor 
diodes and low-level amplifier in a metal housing which 
provides shielding from drafts and reduces any tempera- 
ture gradients in the vicinity of the sensor diodes to neglig- 
ible proportions. 

The effects of offset voltage drifts can be estimated as 
follows. If the ambient temperature seen by the sensor 
diodes and low-level amplifier changes by one degree Cel- 
sius, then the diode balance will have altered typically by 
3 ftV. The effect of one milliwatt of power dissipated in 
the input resistor heats the diode by about 1°C. This leads 
to a reduction in the diode voltage of about 2.2 mV. The 
3-/zV offset adds to or subtracts from the 2.2-raV signal. 
This results in an error of: 

(2.2 mV + 3 mV)/2.2 mV = 1.00137 or 0.012 dB/°C 

Kovar is a trademark ol Westinghouse Electric Corporation 

For an input signal of 10 mW. the effect is reduced and 
the error is only =0.0012 dB/"C. 

For a 30°C change in ambient temperature the total error 
at 0 dBm would be about ±0.36 dB. which is unacceptable 
in this design. Clearly some form of automatic zeroing was 
called for to reduce the size of this potential error source. 

Autozero Circuit 

Errors caused by thermal drift are controlled by the au- 
tozero circuit (Fig. 6). Autozeroing is performed under pro- 
cessor control at power-on and at other convenient times 
such as on a change of instrument mode. Thereafter the 
autozero runs when either of two temperature sensors de- 
tects a 3°C change in temperature. One of the sensors is 
mounted in the housing alongside the thermal converter 
IC. The other sensor is mounted next to the HP 3708A's 
noise source. If the possibility of calibration occurring dur- 
ing a measurement procedure is unacceptable, then this 
facility can be disabled. 

The autozero sequence begins with the RF input signal 
to the thermal converter being removed by switching off 
the final amplifier in the power meter gain block and insert- 
ing full attenuation. The dc feedback loop is broken by 
switching Q4 on. pinching off Q3. The diodes now exhibit 
only the offset voltage to be nulled. The sign of this voltage 
is sensed at the output of U3 by U8. which controls the 
direction of counting in U6, an 8-bit binary up/down count- 
er. The processor releases the inhibit line, allowing the 
counter to run. If U3's output is less than 0V, then the 
counter increments and the digital-to-analog converter 

+V 0 

Adjust I 
Balance « Y A 

DC Output 
to ADC 


Thermal Converter Head 

Autozero Inhibit 
From Processor 





Mil I 

Counter 1 

« - 


Up Down 


Open Loop 


Fig. 6. Thermal converter and autozero circuit used in the HP 3708A's internal power meter 

© Copr. 1949-1998 Hewlett-Packard Co. 


(DAC) U7 gives a greater output, bringing U3 closer to the 
desired OV. This continues until the output of U3 becomes 
greater than OV. The output of UB changes value, forcing 
the counter to decrement and the autozero loop to limit 
cycle: a step up, then a step down. The processor inhibits 
the counter at the end of the cycle after 500 ms. The last 
counter state remains on the DAC input until the next 
autozero operation. This nulls the diode and amplifier 
offset voltages at that temperature. Finally, the loop is 
closed by switching Q4 off, and the power meter amplifier 
is reactivated, restoring status quo. 

Log Conversion 

The dc output from the rms converter is digitized and 
used as the address for a lookup table, stored in ROM, 
which converts the 12-bit analog-to-digital output to dB 
form. This process results in much greater accuracy and 

stability than an analog log converter, yet saves the proces- 
sor considerable time in calculating the logarithm each 
lime. Input power is calculated by summing the individu- 
ally measured values of amplifier gain and attenuator sec- 
tion values algebraically with the rms converter output. 


Thanks are due Geoff Waters, project manager, who de- 
signed the thick-film hybrid amplifiers (see box, page 33) 
used in the gain block and to Ian Matthews who completed 
the rest of the RF design. Dave Stockton contributed the 
C/N switch and Harry Elder was responsible for the product 
design. Brian Woodroffe wrote the operating software. 


1. P.M. O'Neill, "A Monolithic Thermal Converter," Hewlett- 
Packard journal. Vol. 31. no. 5, May 1980. p. 12. 

An Accurate Wideband Noise Generator 
and a High-Stability Reference Source 

by Dayananda K. Rasaratnam 

SET simulates microwave flat fade conditions by 
injecting noise of defined spectral density into the 
IF section of a radio receiver. It automatically maintains a 
selected carrier-to-noise (C/N) ratio by adjusting the noise 
power level so that reliable measurements can be made to 
evaluate the performance of the radio. This requires an 
accurate wideband noise generator and a high-stability ref- 
erence source in the instrument. 

Noise Generator 

The noise generator is designed to meet the following 

■ Flat broadband (10 MHz to 200 MHz) white noise with 
a large crest factor (>15 dB) 

■ A selection of noise-band-defining filters 

■ High accuracy (absolute level error <±0.5 dB) and high 
resolution (<0.02 dB) 

■ Large dynamic range (6 dBm maximum total power to 
-154 dBrn/Hz*) 

■ Fast level tracking within any 10-dB window in the 
dynamic range 

■ Remote control of level and filter selection. 

The noise generator (Fig. 1) is made up of a cascade of 
gain stages, switchable band-defining filters, and program- 

•Toial power in dBm is filter bandwidth dependent 

Fig. 1 . Block diagram of the internal noise generator for the HP 3708A Noise and Interference 

Test Set. 


©Copr. 1949-1998 Hewlett-Packard Co. 

mable step attenuators To meet the required specifications 
and because of the high gain (~100 dB| in the forward 
path, particular care was given during the design phase to 
such aspects as RF grounding. RF screening, mechanical 
layout, and cooling. The requirement for noise with a large 
crest factor was also an important consideration and the 
gains and losses in the path of the generator were distrib- 
uted in such a way that this requirement would not be 

The amplified noise source consists of a cascade of 
amplifiers terminated at the input with a passive resistor. 
Virtually all the noise in the output of the noise generator 
originates from the terminated input stage of this 42-dR 
gain block A variable-gain stage is incorporated at two 
locations in the generator; each consists of a variable pin 
diode attenuator and 34 dB of gain. The filler assembly 
uses pin diode switches to facilitate the remote selection 
of any one of four internal band-defining filters or a filter 
connected across a pair of front-panel ports. One of these 
ports can also be used for the injection of an interfering 
tone. The output amplifier has two outputs. One of these 
outputs feeds into the input of a cascaded pair of step 
attenuators while the other can be selected by a switch for 
connection to the input of the internal power meter. The 
step attenuators are precision programmable units from the 
HP 33320 Series with resolutions of 1 dB and 10 dB. 

The output from the step attenuators is fed to one of two 
front-panel ports via the injection head, depending on the 
selected mode of operation. In the noise generation mode 
the maximum power level at the noise output port is 6 
dBm. In the C/N mode, however, a loss of 6 dB in the 
injection head results in a maximum noise power level at 
the IF output port of 0 dBm. 

Remote control of the noise power level is achieved by 
means of the pin diode attenuators and step attenuators. 
The internal power meter is used to measure the noise 
power level at the input to Ihe pair of step attenuators. 
This measurement is used to determine the settings of the 
step and pin diode attenuators for a given noise level at 
the front panel. 

Step Attenuator Calibration. The step attenuators are spe- 
cially calibrated when the HP 3708A is set up in produc- 
tion. This calibration is effected by comparison with a pair 
of standard attenuators that have been measured accurately 
by an HP standards laboratory. The calibrated attenuator 
values are entered into electrically erasable programmable 
read-only memory (EFPROM) and the HP 3708A's internal 

\ R1*R2 - R,„ = R ou , = z. 
Fig. 2. Bridged- J circuit variable attenuator 

microcomputer uses these values to adjust the noise output 

PIN Diode Attenuators. The pin diode attenuators in the 
variable-gain stages and the means of controlling them are 
designed so that the high resolution and fast tracking capa- 
bilities of the noise generator can be achieved without com- 
promising the specifications for flatness and accuracy. 

A pin diode appears as an almost pure resistance at RF 
and this resistance is a function of dc bias current. This 
property of the pin diode makes it a very suitable element 
for implementing a continuously variable attenuator. Such 
an attenuator with constant input and output impedances 
can be realized with the bridged-T configuration.' This 
configuration also results in good flatness and lends itself 
to temperature compensation. The attenuation of the 
bridged-T circuit (Fig. 2) can be varied while maintaining 
its characteristic impedance Z, p at a constant value by vary- 
ing Rl and R2 in such a way that; 

Rl xR2 = 7.1 = constant (1) 

The forward RF resistance of a pin diode is given by: 

R, = r/T 


where I is the dr. bias current and r and x are constants for 
a given diode type. Assuming a matched pair of pin diodes 
taking the places of Rl and R2 in Fig. 2. we have: 

Z 2 „ = RlxR2 = r 2 /(I, x I 2 ) x 


It can be seen that the product of I, and I 2 must be held 
constant to satisfy equation 1. 

The forward voltage V r of a pin diode is related to its 
forward current I ( by the relationship: 

I f = I, exp(qV f /kT) 


where q is the charge on an electron, k is Bolt/.mann's 
constant, T is the absolute temperature in Kelvin, and [, is 
the reverse saturation current. Therefore: 



r8tl - 




Fig. 3. Method for biasing bridged- T pin diode attenuator to 
maintain its characteristic impedance with varying tempera- 
ture Vi+V.-Va+V, 

© Copr. 1949-1998 Hewlett-Packard Co. 


I, x I 2 = (I s _ x y exp[q(V, + V 2 )/kT] (5) 

and I, xl 2 will be a constant if V, + V 2 is a constant, but 
only at a fixed temperature. 

Temperature compensation is necessary if the charac- 
teristic impedance of the pin diode attenuator is to be main- 
tained in a changing environment. Fig. 3 illustrates a means 
of biasing the attenuator and maintaining its characteristic 
impedance with varying temperature. 2 The voltage de- 
veloped across pin diodes 1)3 and D4 by the reference cur- 
rent I r is applied across pin diodes Dl and D2 in the at- 
tenuator. I r is adjusted to give the desired characteristic 
impedance at the attenuator ports. 

The reverse saturation current I s can be written as: 

I s = Mf(T) 

where M is a constant for a given diode and f(T) is a function 
of absolute temperature. Then equation 5 can be written as: 

I, xl 2 = (M, x M 2 )f 2 (T) exp (q(V, + V 2 )/kT] 

Since the same current flows through both D3 and D4, 
from equation 4 we have: 

I? = (M n xM 4 )f 2 (T)exp |q(V 3 + V 4 )/kT|. 

If V-, + V 4 is maintained equal to Vi + V 2 and l r is held 
constant as in Fig. 3, then IjXl 2 = I 2 (M, x M 2 )/(M 3 x M 4 ) 
will be constant, independent of temperature, and so will 
the characteristic impedance. 

The choice between using voltage or current drive to 
control the attenuator, although influenced by such factors 
as resolution, was primarily determined by the requirement 
for temperature stability of the attenuation setting. The 
attenuation A as a function of control current I, is given by: 

The variations of r and x with temperature and the con- 
sequent effects on the attenuation are small for the particu- 
lar diodes used. The temperature variation of the charac- 
teristic impedance Z„ has also been compensated for. Thus, 
a significant reduction in the temperature dependence of 
the attenuation setting is achieved with the use of a temper- 
ature-stable current, rather than a voltage, to control the 

Further provision for maintaining the accuracy of the 
noise generator under varying environmental conditions 
is made with automatic adjustments to the drive currents 
of the pin diode attenuator in the second variable-gain 
stage. These firmware-driven adjustments are made when 
a characterization process (see below) is carried out: they 
supplement the hardware temperature compensation tech- 


The high resolution and fast tracking capabilities of the 
noise generator are achieved by using the pin diode at- 
tenuator in the second variable-gain stage — the fine atten- 

uator. While continuous monitoring of the noise level with 
the internal power meter ensures accuracy and stability in 
the noise generation mode of the HP 3708A, this technique 
cannot be adopted in C/N mode, when the power meter is 
preoccupied with continuously monitoring the incoming 
carrier level. In this latter case the attenuator is driven from 
a table in RAM. This table is generated by a routine that 
uses the internal power meter to make measurements of 
the noise level at chosen settings of the fine attenuator and 
then interpolates linearly between these cardinal points. 

The time taken to characterize the attenuator is largely 
determined by the number of measurements, i.e.. the 
number of cardinal points. Hence, it is desirable to keep 
the number of these points to a minimum. The attenuation 
characteristic of equation 6 can be rewritten as: 

A = 20 log |a|(l c + Vl 2 + b)/2r' + 1 ) dB (7) 

where a = 1.563. b = 1.31. and d = 0.8 are typical values 
with Z„ = 75fl. A plot (Fig. 4) with these typical values 
shows that this characteristic is nonlinear. The cardinal 
points must therefore be appropriately spaced if a minimal 
set with a given resolution is to be obtained. In the HP 
3708A, a single fixed set of cardinal points is used. This 
set is derived by numerical methods from the second de- 
rivative of the theoretical characteristic. 

A special hardware adjustment procedure ensures that 
deviations of practical characteristics from the theoretical 
curve are small enough to be accommodated within speci- 
fication. The required resolution is obtained by using a 
12-bit digital-to-analog converter (DAC). The table of at- 
tenuation versus 12-bit word settings is generated at power- 
on and whenever the temperature of the noise source has 
changed by more than a few degrees Celsius. 

Reference Source 

The reference source in the HP 3708A Test Set is de- 
signed to provide a highly stable 0-dBm sinusoidal tone at 
the front panel. Either of two frequencies (70 MHz and 140 
MHz are standard) can be selected by operating a pushbut- 
ton. Primarily intended as a calibration source for the resi- 
dent power meter, this output can also be used as a general- 
purpose reference and as a source of interference in the C71 
(carrier/interferer) mode of the instrument. 

-8 -7 -6 -5 -4 -3-2-1 0 1 2 

Control Current l e (mA) 

Fig. 4. Plot of attenuation versus control current lor equation 
7 with a = 1.563. b = 1.31, and d = 0.8. 


©Copr. 1949-1998 Hewlett-Packard Co. 

General-Purpose Wideband Thick-Film Hybrid Amplifier 

During Ihe development of the HP 3708A's iniernai noise 
source ii Became obvious that a general-purpose amplifier could 
Be designed for use in many sections of the instrument The 
objective was to produce a 10-dB amplifier with a frequency 
response as flat as possible from 1 0 to 200 MHz, two independent 
outputs linear up to levels of + 10 dBm. high gain stability, and 
high reverse isolation The result is a hybrid amplifier that is used 
in 14 positions m the instrument In addition, a second high-level 
hybrid and a discrete noise power amplifier use the same circuit 

Choice of Circuit Configuration 

The use of high-f T transistors precluded any transformers in 
the feedback loop. In addition, experience has shown that at 
ihese frequencies it is inadvisable to include more than two 
stages within a feedback loop because of a tendency towards 
high-frequency instability at anything other than small loop gains. 
However, m general, for a given overall gain and a given number 
of stages, Ihe sensitivity of a multistage feedback amplifier to 
changes in the active elements is reduced as the number of 
stages within each feedback loop is increased. Thus, a cascade 
of feedback pairs has a lower sensitivity than a cascade of single- 
stage feedback amplifiers. 

In 1962. Cherry' showed that the sensitivity of a cascade of 
alternate series and shunt feedback stages is very nearly as 
small as that of a cascade of feedback pairs. 

The alternate cascade approach is based on the introduction 
of a gross impedance mismatch between adjacent stages so 
that there is negligible interaction Also each stage operates 
under the nearly ideal conditions tor which its stable transfer 
function is defined, and the overall gain is approximately the 
multiplied individual stage transmittances. There is high reverse 
isolation, thus Ihe use of cascaded single stages introduces a 
considerable amount of flexibility into the design It was therefore 
decided to design a two-stage amplifier consisting of a series 
feedback stage followed by a shunt feedback stage The high 

♦ 15V 

isolation between stages allows the input and output amplifiers 
to be independently optimized either manually or by CAD tech- 
niques, and the low ouiput resistance of the shunt feedback 
stage is a convenient point to connect two isolated matched 

Circuit Design 

The design uses general-purpose transistors biased at a col- 
lector current corresponding to their maximum f, of typically 5 
GHz A simplified circuit schematic is shown in Fig 1 

The design was refined by CAD techniques to determine first- 
order dependence of circuit parameters on element values As 
a starting point for optimization, some approximate low-frequency 
design equations were used 

The gam is 20 log (0.5R/R o n) dB, where n is the interstage 
coupling efficiency (the input resistance of the shunt feedback 
stage is not zero and therefore a small portion of the first stage 
output current flows into the next stage's collector supply resis- 
tor) In practice, n is approximately 90%. 

Signal current flowing in R, reduces current available to the 
load and the value of R, is a compromise between output loading, 
the required stage gain stability, and the output resistance. To 
a first-order approximation, output resistance = R,/(J 

The input data available for Ihe CAD process includes the 
initial element values and the measured s-parameters for both 
transistor types The s-parameters were measured on devices 
embedded in representative bias and collector supply networks. 

Small variations in collector current cause variations in transis- 
tor gain of the order of 0. 1 dB/mA This becomes very significant 
when there are nine amplifiers cascaded, as in the HP 3708A 
noise source. Consequently, a special bias supply circuit is used 
to stabilize the collector currents 

The flexibility of the design approach is used to advantage in 
some positions in Ihe instrument where 26-dB gain is obtained 
from two cascaded hybrids by effectively removing the 75JI ter- 
mination at the input to the second hybrid at all but gigahertz 

; Output 1 

Fig. 1. Schematic ol wideband 
thick-film amplitier. 

© Copr. 1949-1998 Hewlett-Packard Co. 


Fig. 2. Connection of two hybrid circuit amplifiers in HP 3 708 A Test Set to obtain 26-dB gain 


10. 009dB O. lOOdS MAG <U0F> 9. 907dB 

O. ODOdB 10. OOOdB MARKER 70 325 000. OOOHi 


10. OOOdB 10. OOOdB MAC «UOF> -70. 47Bd8 
O. OOOdB 10. OOOdB MARKER 70 323 OOO. OOOHi 
MAG CUDF> — 3S. 05BdB 
I 1 1 1 1 1 1 1 1 1 |OdB 

AMPTD 1 5. OdBn. 

Fig. 3. Amplitude response and input return loss versus fre- 
quency tor wideband amplifier, 

frequencies as shown in Fig. 2 In this application the margin for 
stability is increased by terminating the unused second output 
of the first hybrid with a 75fi load. 

Measured Performance 

The amplifier forward gain changes by 0.04 dB at 200 MHz 
for an ambient temperature variation of 40°C. Results at 10 MHz 
are at least three times better. 

The amplitude-versus-frequency response and the input return 
loss of the thick-film hybrid amplifier are shown in Fig. 3 at 0.1 
dB and 10 dB per division, respectively, measured over a fre- 
quency range from 10 to 200 MHz. Fig. 4 shows the output return 
loss and reverse transmission at 10 dB per division over the 

CENTER 10S 000 000. OOOHi 
AMPTD 15. OdB« 

SPAN 1 BO 000 000. OOOHi 

Fig. 4. Output return loss and reverse transmission versus 
frequency for wideband amplilier 

same frequency range. 

The input and output return losses are better than 30 dB in a 
7511 system The intrinsic high reverse isolation is approximately 
90 dB at 10 MHz and is still greater than 60 dB at 200 MHz 


1 E M Cherry, "An Engineering Approach to the Design of Transistor Feedback 
Amplifiers," Journal or British Institute ot flado Engineers, February 1963 

Geoffrey Waters 
Section Manager 
Queensferry Telecommunications Division 

The circuit (Fig. 5) is made up of two crystal-locked 
transistor oscillators, either of which can be switched into 
a level-stabilizing feedback loop. Hence, only one oscillator 
is active at any given time. The oscillators are based on a 
well-known common-base configuration. 3 However, they 
also incorporate a special pin diode negative feedback con- 
trol circuit. A pin diode is placed in effect between the 
collector and base of each oscillator transistor and the dc- 

controllable RF resistance of these diodes is used to control 
the gain of the active oscillator. 

Leveling Loop. The leveling loop operates in the following 
manner. The output of the active oscillator is connected 
to the input of a broadband (10 MHz to 200 MHz) RF 
amplifier by a pin diode switch. One output of this dual 
output amplifier is fed via an attenuator pad to the front 
panel while the other output is fed via a linear, passive 


©Copr. 1949-1998 Hewlett-Packard Co. 

Oscillator i 



To Oscillator 2 




70 140-MHz Level- 

RLC level-equalizing network to a Schottky-diode detector. 
Level equalization is necessary so that the output levels of 
the two oscillators at 70 MHz and 140 MHz are the same 
at the front panel. The dc voltage from the diode detector 
is amplified to a suitable level and compared with a refer- 
ence voltage from a precision dc source. The resulting error 
voltage is amplified, integrated, and used to drive the pin 
diodes in both oscillators to close the loop and thus 
stabilize the output level of the reference source. 

To enhance the level stability of this output , the detection 
circuitry is driven with the highest signal level that will 
not compromise the harmonic performance of the RF am- 
plifier. The circuitry also includes temperature compen- 
sation and is isolated from the effects of small powersupply 

An oscillator will only come to life if its loop gain is 
greater than unity. When the amplitude of the oscillation 
has built up to the required level, however, this loop gain 
must be reduced to unity. It is also necessary to restrict 
the crystal drive voltage so that the transistor is not driven 
into cut-off and the crystal power dissipation limit is not 
exceeded. The high accuracy required of the output level 
of the reference source precludes any significant harmonic 
content in this output, and the usual method of determining 
the level of an oscillator by amplitude limiting would re- 
quire filtering. These problems are avoided in this circuit 
by adopting automatic gain control (i.e.. the leveling loop). 

At power-on, the feedback around the oscillator is au- 
tomatically adjusted from a low value which ensures start- 

Fig. 5. Circuit diagram of HP 
3 708 A s internal reference source. 

up to the precise value at which the loop gain is unity and 
the output is at the desired level. The spread and drift in 
device parameters are also automatically corrected. The 
crystal drive voltage is maintained at a known fixed value 
by the control loop to ensure that this voltage is not exces- 
sive. Since the oscillator is restricted to operating in its 
linear region, the harmonic content of the output is low 
and additional filtering is unnecessary. 
Frequency Switching and Tuning. Ganged solid-state 
switching ensures that only one oscillator is active at any 
one time and that the oscillator outputs are isolated from 
each other. Either oscillator is switched off by forcing its 
transistor's base-emitter junction into a high impedance 
state while at the same time isolating its output by means 
of a pin diode switch. 

The frequency of each oscillator is adjusted at the factory 
by monitoring a dc tuning voltage at the output of the 
integrator in the control loop (Fig. 5). This voltage is pro- 
portional to the dc current in the pin diode across the 
collector and base of the active oscillator transistor. The 
frequency of the oscillator is tuned by the variable inductor 
in its tank circuit until this voltage is at a minimum. At 
this frequency the negative feedback through the pin diode, 
and hence the positive feedback via the crystal, will be at 
a maximum. This implies that the impedance of the crystal 
is at a minimum and that it is operating at or very near its 
series resonance frequency. This tuning method chooses 
the optimum crystal frequency and minimizes the possibil- 
ity of losing crystal lock because of environmental vari- 

© Copr. 1949-1998 Hewlett-Packard Co. 


alions and aging. Therefore, the tolerance on the frequency 
of the oscillator is essentially determined by the tolerance 
on the frequency specification of the crystal. 


The author wishes to acknowledge the contributions of 
several members of the design team. The thick-film hybrid 
amplifier (see box on page 33) used extensively in the noise 
generator and in the reference source was developed by 
Geoff Waters who also designed the output amplifier. David 

Stockton completed the design of the filter assembly. Harry 
Elder was responsible for the mechanical design. Brian 
Woodroffe wrote the operating software. 


1. N, Kadar, "This voltage-controlled rf attenuator..." Electronic 
Design 15. July 22. 1971. 

2. R.S. Viles, "Need a PIN-diode attenuator?," Electronic Design 
7. March 29, 1977. 

3. Quortz Crystal Circuits Design Handbook, Magnavox Company. 
For! Wayne, Indiana, 1965. 

Automated Radio Testing Shortens Test 
Time and Enhances Accuracy 

by John A. Duff 

WHEN TESTING the flat fade performance of a 
digital radio, a series of repetitive measurements 
have to be made to produce a complete curve 
(Fig. 1). If a typical characteristic plot consists of ten mea- 
sured points (each averaged over three readings), the over- 
all test would take about one hour. By using the HP 3708A 
Noise and Interference Test Set to vary the IF C/N level 
directly, this test time can be reduced to around 20 minutes 
while increasing the measurement accuracy and repeatabil- 
ity. If the HP 3708A is controlled remotely by computer 
via the HP-IB (IEEE 488/IEC 625), and the same is done 
with the bit error rate (BER) test sets to produce an inte- 
grated system, this time can be cut to around five minutes. 
When automatic graph plotting and results storage func- 
tions are provided, the savings in engineer time become 
even more substantial. 

It is this automation of the fade measurement sequence 
that the HP 3708S Noise and Interference Measurement 
System addresses. This system consists of an HP 3708A, 
any HP bit-error-rate test set with HP-IB capability, and 
software written in HP Pascal and running on HP 9000 
Series 200 and Series 300 Computers (see Fig. 2). The sys- 
tem offers comprehensive measurement support facilities 
in addition to performing the measurement sequence. 

Interface Ergonomics 

Simplicity of use was a prime requirement for the system; 
nonexpert production personnel should be able to perform 
measurements with the minimum of complication. At the 
same time the system should provide a comprehensive set 
of functions for an experienced design or test engineer. 

To achieve this dual interface functionality, the system 
has a strict hierarchy of commands that allow access to the 
detailed features progressively, each command requiring 
only a single keystroke. To avoid complication, there are 

Example Carrier/noise plot 


B IB 12 14 1G 

carrier / noise dB 

Fig. 1 . Flat lade performance curve lor a typical digital radio 
obtained by automated testing. 


©Copr. 1949-1998 Hewlett-Packard Co. 

never more than seven options at any command node in 
the system, and common functions are represented by the 
same key each time. Of course this interface method is 
neither novel nor unique, but in the HP 3708S it is com- 
plemented by displaying a tree of the command structure 
with the user's current position highlighted. This allows 
users to visualize their path through the system, avoiding 
the common problem of becoming lost in a maze of special 
function key definitions. Furthermore, the key required for 
a particular option corresponds to that option's position 
(as read from the left) in the following level on the tree. 
This can be seen in Fig. 3. where the command to enter 
the curve manager is 3. 

System Localization 

Readability and clarity were obviously important fea- 
tures for the system interface. To cope with the large 
amounts of information and data to be viewed and changed 
by a user, a general-purpose screen forms package (see box. 
page 38) was developed which is capable of reading pre- 

defined full-screen pages of text from files, displaying 
them, and reading and writing data to and from windows 
on them. Not only does this provide an easy-to-use user 
interface, it greatly simplifies the task of programming the 
various pages of text required and exchanging information 
with the user. 

A secondary effect of this package, brought about by the 
ease of text manipulation within the system, led to an over- 
all strategy of allowing full localization of the system text. 
That is, all the system screens, prompts, error messages, 
and even keystrokes can be changed without altering the 
program source code. This means text can be converted to 
other languages for support abroad, or customer-preferred 
formats can be accommodated. 

Three separate files contain the localizable text: one for 
the screen display forms, one for help information (also 
stored as complete forms), and one for the user prompts, 
error messages, and key definitions. Each file is easily al- 
tered using the standard system editor and following sim- 
ple design rules. Thus, no special-purpose builders or com- 

© Copr. 1949-1998 Hewlett-Packard Co. 


A Reusable Screen Forms Package 

A common requirement for all software systems is a method 
for displaying full pages of text quickly and simply on the screen, 
and being able to put extra information and messages on the 
form or read data entered by a user from it The Forms 200 
package for the HP 3708S System enables pages of text to be 
read from files and displayed selectively. Both text and numbers 
can be written to or read from predefined windows (fields) in the 

Two types of form are available, a basic one for text display 
only (useful for giving standard information, such as in help 
facilities), and another more complex form with interactive data 
fields and display enhancements. The two sets of forms, named 
help and screen, respectively (see Fig 1 ), are held in separate 
ASCII text files for simple and easy update or relocalization 

The help forms file is opened each time a form is required, 

and searched until the required form is found by name. This 
reduces the amount of memory required for forms which are, by 
their nature, used infrequently. The SCREEN forms Me. however, 
is opened once only during activation, and the forms set up in 
an internal data structure. This means that the more common 
forms, which have data windows and display enhancements, are 
set up in advance, ready for rapid display. 

To make the forms as simple to change as possible, a what- 
you-see-is-what-you-get (WYSIWYG) approach is adopted Each 
form contains the 24 lines of text as they are to be displayed, 
with the additional data fields of screen forms being specified 
by putting a field marker on the form at the start position required, 
followed by a row of an identifying character. These serve to 
give the field's length, but are not shown when the form is dis- 
played The type of field marker used determines which of the 

She! p. for m 

Mils la the first line of form teal, and appears at the tup of the screen. 

The HELP ferns file is opened and read each time a form is required, 
and this loin found by name tie "help_f orm" ). 

supper ted. 

HELP forma are designed for simple display of ta«t information, 
such as pages of help informal ion lite this one. 

the form is displayed until a Ley is pressed by the user. 

f» piannun of «4 lines are alluued. 

§screei._f or* .OVERLAY . INHB.UNQT 

This is the first line of form text, and appears at the top of the screen. 

The SCREEN forms file is opened once only, and all ttit forms read and 
sat up internally. Each form is then used by name tie "screen_f orm" here). 

the data fields are specified as- 
ilaaaa or 


uher e field 'a* is five characters lontj arid uill display any data in 
hall -bright inverse video < 1NHB specif lad in the torn header above!, 
field *b' is ten characters long arid displays data underlined ( UNOl specified!. 

This form ulll be overlaid on top of the previous one I UVERLHY specified), and 
ue can let information < including data fields) from the previous form 
" shou through' this one by specifying a series of blank Lines: 

(til forms display and field control functions are performed by 
high-level calls from the application. 

n maaimum of 24 lines are alloued. 

Fig. 1. HELP (lop) and SCREEN 
(bottom) forms (or HP 3708S soft- 
ware system. 


©Copr. 1949-1998 Hewlett-Packard Co. 

two available display modes the field is to use. each mode map- 
ping to one of ten possible video highlights tor that form Succes- 
sive SCREEN forms can also oe overlaid on previous ones to 
allow a pseudowindowmg feature Finally, since no record * 
made of the contents of a field, any field can be used to read 
or wnte text and real or integer number types (unlike many other 
forms systems) 

Besides its simplicity, the WYSIWYG approach has the advan- 

tage that, because the forms are read directly from text files and 
created by the standard editor, no special-purpose builder or 
compiler utilities are required to generate the forms Once the 
form file has been edited the application program can be simply 
rerun with no other intermediate steps necessary 

Integration mto application programs >s eased by a procedural 
interface to the subsystem enabling all internal forms details 
and data to be hidden from the caller 

tC) l-JBG 

EXECUTE naeaurenent 
— *;< — "■•■ PARAMETERS 
CURVE nanagor 
RESULTS Presentation 

<ersion 94.09 /C 

Setup none: SAMPLE 
Nunber of curves: t 

H ... HELP information 
R ... RETURN to nam RMW 


your selection ... 








Presentat ion 


define display load store define print temp file screen 





Fig. 3. HP 3708S user menu. 
Here the number of the sottkey 
label position from the left is used 
to select the next level ol the com- 
mand tree For example, 3 selects 
the curve manager 

pilers are required to produce the forms or language-depen- 
dent text. 

Measurement Accuracy 

To improve the HP 3708S fade simulation method further 
over the traditional technique other than simply automat- 
ing the process, several features are included to enhance 
the measurement accuracy and repeatability. Each point 
on the fade characteristic is averaged over a user-program- 
mable number of BER readings and the point is plotted as 
the arithmetic mean, with an error bar drawn between the 
upper and lower values measured. 

A conventional BER measurement is made over a fixed- 
length gating period (e.g.. 10 seconds), that is, a fixed-length 
window in which the error detector logs transmission er- 
rors. However, because of the random nature of error oc- 
currences and hence their nonuniform distribution (Pois- 
son) in time, a certain number of errors are normally required 
to assure a correct BER measurement. Put mathematically, 
for a truly random error process that is time invariant, a 
BER result based on N errors has a standard deviation of 
100/VN percent. Hence, if 100 errors are counted, the cal- 
culated BER has a 63% probability of being within 10% of 
the long-term BER. 

Consequently, any timed measurement gating period is 
a compromise between being short enough to produce a 
quick test, yet long enough to allow a minimum number 
uf errors to be received at low error rates. The HP 3708S 
overcomes this by providing an adaptive-length gating 

period. The controller reads the error count from the BER 
detectoi and slops the period when the required number 
of errors have been received. This error limit is user-pro- 

Correct, calibrated operation of the HP 3708A depends 
upon the IF input to the instrument having little or no 
noise component, so that the signal-to-noise ratio at the 
injection point is determined only by the noise added by 
the instrument. However, most radios will have some noise 
that is intrinsic to the circuitry before the HP 3708A inser- 
tion point, and this could produce inaccurate values from 
the HP 3708A. A mathematical function is provided in the 
system to compensate for these errors by adjusting the val- 
ues set up on the HP 3708A, given the intrinsic signal-to- 
noise ratio of the radio sections. 


I would like to thank those engineers around me who 
have contributed ideas and given guidance in the develop- 
ment of this system. 

© Copr. 1949-1998 Hewlett-Packard Co. 


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


July 1987 Volume 38 • Number 7 

Technical Information from the Laboratories of 
Hewlett-Packard Company 

Hewlett-Packard Company. 3200 HilMew Avenue 
Palo Alto. California 94304 U S.A 
Hewlett-Packard Central Mailing Depanmeni 
P O Box 529 Startbaan 16 
1180 AM Amstelveen The Netherlands 
Yokogawa-Hewlett-Packard Ltd . Suginami-Ku Tokyo 168 Japan 
Hewlett-Packard (Canada) Ltd 
eway Drive. Misstssauga. Ontario L4V 1M8 Canada 

Bulk Rate 
U.S. Postage 


I A a. I f\ f~ f \ |— * |i p~ (~~\ a tosubGcnbo change your address, or delete your name from ou' -naiimg list send your request to Hewlett Packard 

UnANut \J I MLJL/ll tOO. Journal 3200 Hlii„ew Avenue Palo Alto CA 94304 U SA Include iW ok3 address label « any. Allow 60 days 


© Copr. 1949-1998 Hewlett-Packard Co.