Skip to main content

Full text of "HP Journal 1987-05"

See other formats


© Copr. 1949-1998 Hewlett-Packard Co. 


May 1987 Volume 38 • Number 5 


4State-of-the-Art CAD Workstations for Mechanical Design, by Wolfgang Kurz, Dieter 
Sommer, Karl-Heinz Werner, Dieter Deyke, and Heinz P. Arndt Easing drafting tasks for 
designers, this system also provides a shared data base and can transmit commands directly 
to NC machinery. 

12 Example Macro 

14 ME Series 10 Link to HP-FE 

15 The ME Series 10 NC Links 

IO ME CAD Geometry Construction, Dimensioning, Hatching, and Part Structuring, by 
J Karl-Heinz Werner, Stephen Yie, Friedhelm M. Ottliczky, Harold B. Prince, and Heinz 
Diebel Construction lines and circles aid layout, and dimensioning and hatching adapt automat- 
ically to part design changes. 

Alpha Site Evaluation of ME Series 5/10, by Paul Harmon The best way to evaluate 
a CAD system is to design a real product with it. 

Research Report 

Intrabuilding Data Transmission Using Power-Line Wiring, by Robert A Piety 

Widely varing attenuation, noise, and crosstalk are some of the obstacles to overcome. 


3 In this Issue 
3 What's Ahead 
34 Authors 

Editor Ricnard P Doian • Associate Editor Business Manager. Kenneth A Shaw • Assistant Editor. Nancy R Teater • Art Director. Photographer Arvid A Danieison 
Support Supervisor. Susan E Wright • Administrative Services. Typography Anne S LoPresti • European Production Supervisor. Michael Zandwi|*en 

2 HEWLETT-PACKARD JOURNAL MAY 1987 © Hewlett-Packard Company 1987 Printed in Holland 

© Copr. 1949-1998 Hewlett-Packard Co. 

In this Issue 

HP DesignCenter is the name given to a collection ot design automation 
software packages that run on the HP 9000 family of technical computers. 
These products are tools that help electronic, mechanical, and software 
engineers be more effective in their jobs. What they do is often spoken of 
in acronyms — CAE (computer-aided engineering). CAD (computer-aided de- 
sign). CAM (computer-aided manufacturing). On pages 4 to 33 of this issue, 
designers of the HP DesignCenter Mechanical Engineering Series 5 and 
Series 1 0 systems, our cover subjects, describe the structure and operation 
of these advanced two-dimensional design and drafting packages. Automat- 
ing the process of designing mechanical parts presented numerous engineering challenges that 
required contributions in user interface design, memory management and data structure design, 
and algorithm design. For ME Series 10, an additional design challenge was the requirement to 
communicate with systems for finite element structural analysis and systems that translate part 
data into instructions for the numerically controlled (NC) machine tools that make the parts. The 
paper on page 4 describes the ME Series 5/10 products and their user interface. The system is 
designed to be both easy to learn for the novice and powerful enough not to limit or frustrate the 
expert. The command language and its user-definable macro facility are noteworthy. Data struc- 
tures and algorithms for geometry construction, dimensioning, hatching, and parts definition are 
discussed in the paper on page 16. On page 30 is an unusual (for us) report from an HP Division 
that acted as an alpha test site for the ME Series 5/10 and is now using these systems to design 
printer parts. 

Because it's already there, electric wiring has long been used for many things besides carrying 
electric power. In homes, for example, some cordless telephones, intercoms, music systems, 
appliance controllers, and burglar alarms use the power lines, saving the expense of separate 
wiring. Not surprisingly, people are now connecting computers over local area networks (LANs) 
that use the power lines for data communications. On page 35. we have a report on an HP 
Laboratories research project on power-line LANs. The research revealed significant obstacles 
to reliable data communications over power lines, including high noise levels and high signal 
attenuation that vary with time and from site to site. To deal with these conditions, a wideband 
scheme was tested and found to permit usable data rates up to 100,000 bits per second, much 
higher than previously reported. 

-R. P. Dolan 

What's Ahead 

The June issue will cover a variety of topics, including HP's Human Interface Link (HP-HIL) for 
personal computer and workstation input devices, the design of a low-cost HP-HIL graphics tablet, 
interprocess communication mechanisms for UNIX systems, and software verification using branch 
analysis. Completing the issue is a Viewpoint article on the direction of CMOS technology by 
Yoshio Nishi. 

Ttie HP Journal encourages lechrvcal d<scuss'On ol the lopca presented «*i loceril alleles and wM irutMti letters expected 10 be ol interest lo oui readers Letter* must be briet and are subteel 
to ed trng i ettori should be addressed to Editor Hewlett-Packard Journal 3200 Hilli/iew Avenue Palo Alto CA EM3W. US A 


© Copr. 1949-1998 Hewlett-Packard Co. 

State-of-the-Art CAD Workstations for 
Mechanical Design 

Part of HP's DesignCenter, the ME Series 5/ 10 workstations 
simplify the creation of part drawings and the design of 
mechanical assemblies. A shared data base improves 
communication among designers on a project and the 
results can be formatted automatically for use by NC 
manufacturing machinery. 

by Wolfgang Kurz, Dieter Sommer, Karl-Heinz Werner, Dieter Deyke, and Heinz P. Arndt 

takes more than putting a number of wonderful fea- 
tures together. The majority of today's CAD (com- 
puter-aided design) systems for mechanical engineering 
can more or less do the job, but many of the systems demand 
an extensive amount of training before the user can become 
productive. Another concern is that as a user's proficiency 
grows, requests for modification, customization, and new 
features arise. 

To resolve some of these difficulties and to provide de- 
signers with tools to aid them in their designs. HP has 
developed a family of workstations, the HP DesignCenter, 
for use in various technical fields. Two examples are the 
Mechanical Engineering Series 5 and Series 10 (Fig. 1) 

discussed here. These workstations represent one of a 
group of HP projects to develop software for various CAD 
systems and more general interactive graphic systems. Soft- 
ware reusability is a high priority in these projects. This 
not only saves substantial development effort, but also ben- 
efits the user by providing a uniform set (or subset) of 
functions across a range of products. 

Another HP DesignCenter goal was building a CAD user 
interface that enables productive use of a system after only 
a few days, even for a novice without formal training. At 
the same time, the system does not disappoint an expert 
user. The ME Series 5/10 provides over 250 commands and 
functions with different options for many of them. A very 
efficient method of combining these primitives to form 

Fig. 1 . The Mechanical Engineer- 
ing Series 5 and Series 10 Work- 
stations of HP's DesignCenter are 
versatile systems for two-dimen- 
sional drafting (Series 5 and 
Series 10) and mechanical design 
(Series 10) applications. Both fea- 
ture a friendly and easy-to-learn 
user interface designed for techni- 
cal illustrators, drafters, and en- 
gineers. The Series 5 provides 
many tools for the preparation of 
drawings, flowcharts, schematics, 
layouts, and documentation and 
can be easily upgraded to the 
Series 10. which provides many 
facilities tor mechanical engineer- 
ing design such as parametric de- 
sign evaluations, customizability, 
and interfaces to other CAD sys- 
tems (e.g., for finite-element analy- 
sis) and to numerically controlled 
(NC) machinery 


© Copr. 1949-1998 Hewlett-Packard Co. 

5 -.5'DR 




23-' REP iq. 


Fig. 2. A typical ME Series 5/10 drawing. 

more complex command structures provides the experi- 
enced user with a powerful tool for daily design tasks. 

Feature Highlights 

The HP DesignCenter Mechanical Engineering Series 5 
and 10 are advanced two-dimensional design and drafting 
systems for mechanical engineering applications. Fig. 2 
shows a typical ME Series 5/10 mechanical drawing. 

The ME Series 5 and ME Series 10 have the same software 
architecture, basic user interface, and model data struc- 
tures. The ME Series 5 offers a subset of the ME Series 10 
capabilities at an economical price. It is upward compatible 
with and upgradable to the ME Series 10. 

ME Series 10 features include: 

■ Comprehensive construction and annotation capability 

■ Choice of user interfaces — tablet/screen or screen only 

■ Powerful command set to perform fast design modifica- 

■ Sophisticated macro language for parametric design and 
design evaluations 

■ Customizability for specific applications 

■ Interfaces to other CAD systems (IGES), NC machines, 
and finite element analysis systems. 

The ME Series 5 offers the same features with the follow- 
ing limitations: 

■ Limited drawing size 

■ Limited macro capabilities 

■ No editing for hatch parameters, attributes, or fillets 

■ No interfaces to NC, FE, or IGES systems. 

ME Series 5/10 construction geometry simplifies the 
work previously done using a pencil, compass, and ruler 
on a drafting board. Construction geometry consists of basic 
figures like circles and endless lines. There is a mode to 
overdraw the construction geometry, like inking in the pen- 
cil lines. The user can also create real (inked) geometry 
directly. The set of geometry elements includes third-order 
splines (with damping) which can be converted into a con- 
tour of tangent lines and arcs if the geometry is to be trans- 
ferred to an NC system (numerically controlled automatic 
machining equipment). Besides the easy handling of texts 
and symbols, other annotations such as hatching and di- 
mensioning not only comply with different standards like 
ANSI. ISO. DIN. or (IS. but are sophisticated enough to be 
updated automatically when the associated geometry is 
altered. The ability to create assemblies and hierarchically 
structured parts simplifies the handling of large amounts 
of data. Since the data base can be shared with other design- 
ers at other ME Series 5/10 workstations, each designer can 
always have ready access to other parts affecting the design- 
er's particular design. (See the article on page 16 for details 
about ME Series 5/10 geometry, dimensioning, hatching, 
and parts.) 

The ME Series 5/10 products are tailored for the HP 9000 
Series 300 Computers.' but are also supported on the HP 
9000 Series 200 Computers. They take full advantage of 
the different hardware configurations and the broad range 


© Copr. 1949-1998 Hewlett-Packard Co. 

Action Routines 
User Interface. 
Macros, etc 

User Interlace 
Data Base 

of supported peripherals. The operating systems the prod- 
ucts run under are the HP-UX operating system and the 
Pascal workstation systems (Series 200). An interface to 
each operating system is provided to take advantage of all 
of the features of the operating system. 

The hardware configuration can be as simple as a single 
stand-alone 16-bit workstation using a medium-resolution 
monochrome display and a mouse as the input device. For 
more sophisticated applications, it can be extended to a 
network of 32-bit, floating-point workstations with high- 
resolution color displays. B/A3-size graphics tablets, print- 
ers, a choice of disc drives with storage capacities from 
270 kbytes to 571 Mbytes, and E/AO-size plotters with auto- 
matic paper feed to provide multiple frame plots. 

Customers with special applications need to customize 
their systems. The ability to run other programs on-line, 
together with a macro programming capability, makes the 
ME Series 5/10 useful to customers other than those in the 
mechanical and electromechanical design fields. Various 
macros, symbols, text fonts, and standard parts are avail- 

Customizing also means being able to use the system in 
the user"s native language. Localized versions for France. 
Italy, and Germany are available. A Japanese version with 
approximately 4000 Kanji characters is available; it sup- 
ports 16-bit Kanji characters in the editor, macros, menus, 
help subsystem, and text commands. 

ME Series 5/10 is designed to restrict the user as little 
as possible. For instance, the number of elements in the 
model or drawing is bound only by the memory (or virtual 
memory) available. The number of macros, screen tablet 
menu fields, simultaneous text fonts, or symbols is limited 
only by the underlying operating system. 

The ME Series 5/10 is compatible with earlier HP CAD 
products like HP Draft. HP EGS. and HP Design. The valu- 
able data created by these systems can be transferred to 
the ME Series 5/10. An open system architecture not only 
means having interfaces to other HP products, but also an 
IGES (Initial Graphics Exchange Specification) interface to 
systems made by other manufacturers. 

Fig. 3. Software architecture ot 
the ME Series 5/10. 

Software Architecture 

The design goal for the software architecture was to de- 
velop a uniform user interface and operating system inter- 
face for all commands and functions, and mechanisms for 
additional commands, new applications, and different data 
bases. The U-shaped area in Fig. 3 contains all the software 
dealing with the user interface, customizing, command 
parsing. I/O device support, and general operating system 
support. Action routines define all user interaction for a 
specific command or function. The command's syntax is 
only defined inside the action routine. Action doers are 
more general modules containing mathematical algorithms 
and various calculations. 

User Interface 

The large functionality built into ME Series 5/10 is ac- 
cessed by means of a command language with a defined 
syntax. However, for more efficient work and easier learn- 
ing it was necessary to create a friendly environment for 
all commands and functions. There are combined graphics 
tablet/screen and screen-only user interfaces. All tablet and 
screen menus have been defined with ME Series 5/10 com- 
mands using the ME Series 5/10 macro language and can 
be easily customized by the user. 

Catching a Point. The ME Series 5/10 catch mechanism 
allows the user to catch (select) a point in various ways. 
The point caught is the closest point of a preselected type 
within the circular area around the cursor center. It may 
be the closest vertex (end point), intersection, grid, or per- 
pendicular projection point of the cursor center onto an 
element, or nothing. The catch mode can be changed at 
any time. Hence, while creating a polygon, it is possible 
to change the catch mode from vertex to intersection to get 
the third polygon point correctly. The window can simi- 
larly be changed to get a point outside of the current win- 
dow. Catching is confined to the cursor area boundaries. 
These boundaries can be set by the user, who can always 
see the cursor on the screen and has complete control of 
the catch results. Hence, ifs not possible to catch an inter- 
section point in the upper right corner of the viewport by 


© Copr. 1949-1998 Hewlett-Packard Co. 








■ A ■ 



i > 









■ mi 

















u.. , 









;> HM 


kiCI D 






i 1 




1 1 


































■ d 

1 1 

1 1 








; ;4, 











— =r| 








if 1 . . 

Fig. 4. M£ Series 5.'/0 fab/ef over/ay 

indicating a point in its lower left corner when the cursor 
size is five pixels. 

Tablet/Screen User Interface. The ME Series 5/10 tablet 
area can be used for two different purposes: to pick the 
geometry points and to choose commands or functions. An 
overlay (Fig. 4) indicates the softkey locations on the tablet 

Following the desktop metaphor, we put all commands 
and functions that should always be available on the 
graphics tablet, while the items that are mainly used in 
certain design phases are found in screen submenus. This 
defines a clear and easy-to-Iearn structure for the user in- 
terface and makes its use much more effective. For example, 
a user can pick window functions or construction geometry 
commands from the tablet at any time without needing to 
change the CREATE screen submenu for that purpose. 

The tablet overlay consists of: 

■ A column with often-used commands: DELETE, CONFIRM, 

■ Functions that can interrupt the active command at any 
lime and functions for display options. Examples are 

■ Selection items that are normally used together with 
certain commands. 

■ A numeric keypad, 

■ Two columns wilh screen submenu calls. 

On the tablet overlay, commands have a dark gray back- 
ground color, functions and selection items are placed on 
a light gray background, and screen submenu calls are 
printed on a white background. 

All commands and functions that are mainly used in 
special design phases have been collected in screen sub- 
menus with one or two screens each. The screen submenus 

■ CREATE (two screens). General drawing commands for 
real geometry. Hence, a user can create real geometry 
and construction geometry at the same time since the 
construction geometry block is placed on Ihe tablet over- 

■ DIMENSION (two screens). Automatic or manual dimen- 
sioning and unit conversion. 

■ HATCH. Automatic or manual hatch, and simple hatches 
and hatch patterns, 

■ TEXT (two screens). Editing of texts and leader lines and 
I mi! selection. 

■ FILE (two screens). Loading, storing, directory catalogs 
with sorting items, and general file operations. 

■ PLOT. Local plot and spool plot. 

■ MODIFY. Moving, copying, rotating, mirroring, scaling, 
stretching, and isometric, modification of real geometry. 

■ PARTS (two screens). Commands for a single-level or mul- 


© Copr. 1949-1998 Hewlett-Packard Co. 


Tan 2 Arcs 


2 Pis 

Tan Pt Arc 
Ang S Len 

Cen & Pt ~ 




Cen & Ends 

Pt R Angs 

Elem Box 

Divide Len 




Cen R Angs 


Cen Ang R 




Cent Pis 




End Spline 

















File Name 
PhysName [FileDesc 

File Type | File Size 


Oelete Old I Printer 



















Fig. 5. Screen submenus of the tablet/screen user interface. 

tilevel parts structure and parts lists. The PARTS block 
on the tablet overlay contains a small subset of the screen 
submenu functionality and offers only single-level parts 
commands. This enables the novice user to work with 
a flat tree structure (all parts on the same level) whereas 
the experienced user can use the more sophisticated 
parts commands of the screen submenu to create more 
complicated parts structures. 
» SET UP (two screens). Commands for changing the work- 
ing environment, such as setting units, plotting defaults, 

Fig. 7. Example: Catalog function. 

dimensioning options, and dimensioning defaults. 
I SYMBOLS (two screens). The first menu contains a set of 
ISO symbols in different sizes and the appropriate sym- 
bol box elements. Since many users want to use their 
own symbols, the second screen is left empty to offer an 
easy-to-use way to add custom symbols. 
° INFO. Commands for associated texts. 
Screen-Only User Interface. For users who prefer working 
with a mouse or do not want to spend money for a tablet, 
we implemented a screen-only user interface. Our main 














Tan 2 Arcs 




Cen R Anqs 



2 Pis 



Tan Pt Arc 
Ang & Len 

R & 2 Elem 


Cen& Ends 
Pt R Anqs 
































































































































Fig. 6. Screen submenus of the screen-only user interface. 


© Copr. 1949-1998 Hewlett-Packard Co. 

goal was to design both user interfaces to be as similar as 
possible by using the same structure for tablet overlay com- 
mand blocks and their screen-only representation, and the 
same internal structure of the screen submenus like CREATE 
or DIMENSION. The scheme for the screen-only user interface 
consists of five parts: 

■ Calls for changing screen submenus. For example. 
CREATE and DIMENSION as you would find them on the 
tablet overlay. 

■ Calls for changing screen submenus that simulate the 
tablet overlay function blocks like WINDOW. SHOW. 

■ An area in which the changing screen submenus will 

■ A block with the main commands: DELETE. CANCEL. 

■ A numeric keypad. 

With the exception of the changing submenus, all other 
menu parts remain on the screen all of the time. Each menu 
part has a different background color. One additional func- 
tion has been implemented in the screen-only interface. 
The BACK function leads the user directly to the previous 
command submenu after working in a function submenu 

Screen Submenus. Both user interfaces use several screen 
submenus (Fig. 5 and Fig. 6) with the same internal struc- 
ture of commands. This makes documentation much easier 
since one manual describes both user interfaces. 

Some of the menu items are enhanced by a background 
color selected according to a simple scheme. Normally, a 
command (white background) offers several options (black 
or blue background). If a user picks the command name 
itself, the user gets the default option that is always 
positioned to the right of the command name. To select an 
option, the user can pick it directly without picking the 
command name first. However, some commands or func- 
tions can be executed with more than one option at the 
same time. For example, let's take a closer look at the 
catalog function (see Fig. 7). 

First, the user decides whether to see the current direc- 
tory (pick CATALOG or Current) as the default option or 
another directory (pick Name and enter a directory name). 
Second, the user can select one or more options with a 
blue background to select certain catalog items (e.g., File 
Type), or sort the output list according to an item like file 
name, file size, create date, and so on. Finally, the user has 
to decide the destination for the catalog list. For this pur- 
pose, a blue background menu field named OPTIONS at the 
bottom of the screen submenu offers Screen and Printer as 
devices or Delete Old and Append for the destination file 

The different background colors in the screen submenus 
are used to simplify interpretation of the listings presented. 
A white background is used for command and function 
names. A black background shows necessary options that 
the user has to select with the first pick. Blue background 
options can be selected after the first pick. 
User Interface Files. The user interface is completely writ- 
ten using the ME Series 5/10 macro language. It is divided 
into nine files which can be easily modified by the user: 
hp_macro: All macros used in both user interfaces. 

hp_macro_t: Macros and screen menus used only by the tab- 
let'screen interface. 

hp_macro_s: Macros and screen menus used by the screen- 
only interface. 

hp_menu_t: Definition of the menu slot layout for the tablet 
screen interface. 

hp_menu_s: Definition of the menu slot layout for the screen- 
only interface. 

hp_tmenu: The tablet overlay menus for three HP graphics 

OVL9111. OVL46087, and OVL46088: ME Series 5 10 drawings 
of the tablet overlays for the three HP tablet types. 

ME Series 5 10 Language 

In addition to screen and tablet menus, special function 
keys are available. This highly customizable user interface 
system is based on an interactive command language cho- 
sen to be independent of input devices and input tech- 
niques and to provide an easily extensible system. Besides 
the basic commands and functions for invoking the system 
features, there are extensions for expression evaluation and 
macro definition. 

Basic Constituents. Processing ME Series 5/10 input can 
be split into two parts: character processing and token pro- 
cessing. All logical input devices, such as the keyboard or 
a screen/tablet menu, send characters to the scanner. The 
scanner understands the syntax of all token types. It decides 
by looking at the stream of characters if the token is a point, 
a string, a number, et cetera. For example. 4.5 is a point. 
"ENGINEER" is a string, and 3.14 is a number. 

A token is an entity of information (represented by a 
modified Pascal record) containing the token type (e.g.. a 
number), its value (e.g., .3.14), and a link to the next token 
to be processed. The value of a token is in binary format, 
the value of a command token is an internal command 
number, and the value of a macro token is a pointer to the 
macro definition. This format was chosen because it is a 
compact representation and. since tokens are normally pro- 
cessed more than once, processing time is saved because 
most of the work is done when the token is built. The 
possible token types are listed in Table I, 

Table I 

Token Types and Examples 

Token Type 










Math function 





Center Jines 


"Identify circle" 





3D point 




Commands and functions are the primary keywords that 


© Copr. 1949-1998 Hewlett-Packard Co. 


Action 1 

Action K 











Fig. 8. (a) Typical command syntax (b) Function syntax dif- 
fers from the command syntax in that it lacks the typical loop 

invoke a specific system action. We have chosen different 
terms to reflect the two uses. A command (e.g., LINE) is 
normally active until another command is entered. This 
means that the user inputs lines until another command 
is chosen. The typical command syntax is shown in Fig. 8a. 

A function (e.g.. WINDOW) is normally only active until 
the data required for this specific function has been entered. 
Then the function terminates (see Fig. 8b). It does not have 
the typical loop syntax of the commands. Functions are 
those actions in the system that do not change the data 
model, but may change the values of variables defining the 
environment, such as a catch mode. 

A special property of a function is that it can interrupt 
a command. That means that at any stage of the command, 
instead of entering the required data for that command, 
the user can select any function. After completing the func- 
tion the system will ask for data for that command again. 
Even inside a function, another function can be invoked, 
which again interrupts the previous function. 

Both commands and functions are implemented in action 
routines (see Fig. 3). 

Expression Evaluation and Macro Processing. Between the 
scanner, which performs syntactic analysis and generates 
tokens, and the action routines, which consume tokens for 
semantic analysis to carry out the requested actions, is a 
filter to evaluate expressions and expand macros. Expres- 
sion evaluation transforms several tokens into a single 
token. Macro expansion takes a single token and replaces 
it with one or more tokens. For example, the seven tokens 
3.14 x 10 x 10 are replaced by the token 314 and the macro 
token Dac is replaced by the tokens DELETE ALL CONFIRM. 

Examples of expression operators are + . which adds two 
numbers or vectors or concatenates two strings, ROT, which 
rotates a vector, and POS, which returns the position of a 
substring within a string. One special operator is INQ. It 
can be used to read the current environment (e.g., window 

Sub Action 

Sub Action 

Fig. 9. General program structure ol ME Series 5/70 

setting, color, catch mode, and available memory), the type 
and parameters of drawing elements (e.g.. coordinates, 
radius, and color), and miscellaneous items (e.g.. last mea- 
sured distance). 

Macros can be used for the following reasons: 

■ Abbreviations such as Dac for DELETE ALL CONFIRM 

■ Speed (executing a macro is much faster than executing 
the same commands given in character representation) 

■ Localization. For example, using German KREIS for En- 
glish CIRCLE 

■ Variables such as point coordinates, loop counters, etc. 

■ User-defined functions to extend the built-in functional 

■ User-defined action routines to extend the overall system 
capabilities, such as an ellipse routine 

■ Parameterized library parts such as bolts and nuts. 
The macro features include nesting (macros can use other 

macros), local macros, and parameters. To avoid conflicts 
with a global macro with the same name, a macro should 
be declared local if it is only used inside another macro. 
Parameters are the same as local macros, except that an 
initial value must be provided when calling the macro. 
There are also several control structures to be used inside 
tion is provided by the READ function, which includes type 
checking, rubberbanding. catching, and default values. 
Macros can be not only saved in ASCII representation, but 
also stored in a binary format and secured as appropriate. 
The TRACE function can be used to monitor macro expan- 
sion and expression evaluation. It lists all tokens processed 
by the macro/expression filler either to an output device 
or to a file. 

A short example in the box on page 12 explains the ME 
Series 5/10 macro capabilities in more detail. 
Action Routines. Action routines are the lowest-level user 
interface in the system. They cannot be changed by the 
user, and within the same product and revision, they are 
the same for all countries. 

The general program structure of ME Series 5/10 can be 
visualized by the diagram in Fig. 9. An action routine inter- 
faces the user to the sub action level. For example, to create 
a line, the user enters LINE into the system (this can be 
done in various ways). LINE is recognized by the system as 



, I . ♦ ♦ ♦ t ♦ ♦ 


Sub Action Fi 9- 10 - Expansion ol action rou- 

Level tine structure in Fig. 9 


© Copr. 1949-1998 Hewlett-Packard Co. 

a command and the line action routine is called by MAIN 
This routine now handles all interactions with the user to 
create the line the user wants. In general, an action routine 
reads a specific set of tokens and changes the state of the 
system accordingly. The state of the system includes the 
values of global variables, local variables, data structures, 
the file system, et cetera. 

Fig. 10 expands the action routine structure and shows 
some of the ME Series 5/10 action routines, utility denotes 
a set of action routines used by other action routines. Bas- 
ically these routines deal with selection. Other action 
routines shown in Fig. 10 are: 

■ DIM, which handles dimensional data 

■ CREATE, which creates geometric information 

■ MODIFY, which deals with DELETE. SPLIT, MERGE, and 

■ STRUCTURE, which deals with parts and layers 

■ ANALYSIS, which calculates properties of the created 
geometry such as center of gravity 

» DATA_M. which manages data structures (LOAD. STORE, 
and Ml) 

■ SYSTEM_M. which manages system parameters, macros, 
et cetera. 

Store and Load 

Storage Formats. ME Series 10 can store drawings, sub- 
parts, or library parts in two different formats: binary or 
MI (Model Interchange). ME Series 5 offers only the binary 
format. The binary file format corresponds to the internal 
representation of drawings, subparts, or library parts in ME 
Series 5 '10. Very little conversion is necessary when storing 
to another binary format. This results in compact data stor- 
age and provides for fast storage and retrieval of ME Series 
5 10 model data. The MI format is an HP data exchange 
standard similar to ICES that requires transformation of 
the internal data structure to an ASCII format for long-term 
storage and interfacing to external programs. 

Some techniques used to store data are applicable to both 
formats. In particular, the same mechanism is used to store 
lists and object references. Before explaining the way ME 
Series 5/10 stores model data, it is necessary to describe 
some aspects of the internal memory management and the 
data structure of ME Series 5/10. 

Memory Management and Model Data Structure. All ob- 
jects that are created in a dynamic fashion or vary in size 
are stored in the heap storage area. To manage the heap 
area. ME Series 5/10 uses its own heap manager to obtain 
sufficient performance for creating and deleting dynamic 
objects. This helps achieve a satisfactory interactive re- 

Global Dala 

Fig. 11. Objects are linked with relerences (black arrows) lo one another 


© Copr. 1949-1998 Hewlett-Packard Co. 

Example Macro 

DEFINE Center, lines 


(Declare macros used locally) 


(Terminate anv active command! 


(Inquire current color and 

LET C (INO 2011 

linetvoe and save values 

IFT 1 (INO 302) 

in local macros! 



(Restore color) 


(Restore line! vdp) 



READ PNT 'Identify circle' P 

(Ask user to identify a circle) 


(Get element parameters) 


(Exit loop if element=CiRCLE) 


(Else BEEP and continue asking) 


LET P (INO 1 01| 

(Inquire circle center point ) 

LETR (1 .1 INQ3| 

(Inquire radius, scale by 1.1) 



(Switch color to YELLOW) 


(Switch linetype to 


(Draw horizontal center line) 


(Draw vertical center line) 


(Ask for next circle) 


The macro listed above will continuously ask the user to identity 

a circle to add center lines to. The center lines will be yellow, in 

a doLcenter linetype. and will be 10 

& longer than the circle radius. 

The input of any command will terminate the macro and execute 

the command. 

sponse time in ME Series 5/10. 

The storing of objects organized in a heap area takes 
longer than storing an array of bytes containing all objects 
compacted with no free space between the objects. Such 
compact arrays of bytes can be copied from memory to disc 
in a very fast manner. But. in ME Series 5/10, all important 
data structures are organized as linear lists, binary trees, 
quad trees, part trees, etc., with many references between 
the objects. Using a heap storage area offers much more 
flexibility than reserving a fixed amount of storage or- 
ganized as an array. The size of the model data is only 
limited by the installed memory. 

To get the required functionality and to represent the 
dependencies of the objects that ME Series 5/10 creates 
and uses, most objects are linked with references to each 
other (see Fig. 11). If there were no restrictions on the 
references between the objects stored in the heap, then the 
net of objects could be described with a general graph. 

ME Series 5/10 organizes the model data as a hierarchy 
of objects with references only in one direction from the 
higher to the lower levels, with the result that the model 
data can be seen as a directed cycle-free graph. The hierar- 
chy of model data from highest to lowest levels is: 

Highest level ~*Parts 

Faces, dimensions 


Lowest level ~» Model points and global data 

Because of this data organization, the storing of compo- 

nents and other objects can be simplified and done more 

All objects can be divided into two groups. There are 
objects global for all parts (shared by all parts) and objects 
local to a particular part, to which they belong. Examples 

Global objects: 

Dimension line data 

Dimension character data 



Associated text 
Local objects: 

Model points 



Shared object references are used to link local objects to 
global object data and to link local objects in a part to each 
other as listed below: 

Simple components 
All components 



♦model points 
♦associated text 

♦ simple components 


• simple and composite components 
(all components belong to one part) 

♦part (shared and unshared subparts) 

•associated text 

» dimension line information 

♦ dimension character information 

• model points 
'simple components 

For many temporary processes, each sharable object has 
a slot called temp_attr (temporary attribute). The information 
stored in temp_attr is only valid within a single command 
or function. The temporary attribute is used to: 

■ Number objects in a list to transform references, mark 
objects, and assign temporary information. 

■ Produce a listing of the assembly structure or of the 
number of occurrences of all subparts. 

■ Temporarily store pointers to the image of modified shar- 
able components to avoid duplicate processing during 
the MODIFY command. 

■ Find all faces and contours with all their referenced 
components included in a given list of selected compo- 



Fig. 12. Lists of model points and components with refer- 
ences from the latter into the former 


© Copr. 1949-1998 Hewlett-Packard Co. 


• Detect cycles during automatic hatching. 

■ Temporarily store topological information during the 
STRETCH command. 

■ Temporarily mark selected elements during selection 
and support logical operations in the selection mechanism. 

Example: Storing Lists of Objects. As an example, we de- 
scribe the mechanism to store two lists of objects with 
references from the second list into the first list. Let us 
assume that the first list contains model points (MPx) and 
the second list contains components (CPy) as shown in 
Fig. 12. To store this structure the following actions are 

1 . Initialize the temp_attr of all objects in list MP to zero. 

2. Mark all model points referenced from components in 
list CP. 

3. Order all marked model points by assigning to temp_attr 
a sequence number according to the position in list MP 
and remember the total number of model points. 

4. Store the total number of model points followed by the 
model points in list order. The model point count is used 
to detect the end of the list during load and to calculate 
the number of bytes needed to store object references into 
this list. 

5. Count the number of components in list CP. 

6. Store the count of components followed by the compo- 
nents in list order and convert model point references by 
storing the value of temp_attr accessed by dereferencing the 
model point reference. 

Storing the Data Structure. Before any data is stored on 
external files, the model data is inspected to check that all 
references point to legal objects. This is accomplished by 
adding a sequence number to the temp_artr of the particular 
object and verifying that all referenced objects have a se- 
quence number in the appropriate range (between zero and 
the number of elements of the particular object type). 

All quad trees for model points are reorganized to delete 
model points that are not leafs and that are no longer refer- 
enced by other objects (i.e., all use counts are zero). The 
leaf nodes that have no other references have already been 

This mechanism is used for all lists representing the 
model data. The order in which lists are stored is important 
to guarantee that during the reload of these lists, all ele- 

ments of other associated lists are already loaded when 
references to these elements are encountered. Translating 
a reference means to follow the reference and get the 
number in temp_attr assigned during previous numbering. 
These numbers are always relative to the beginning of the 
list the referenced element belongs to. 

To get a compact storage format, the numbers are stored 
as one. two, or four-byte integers when the number of ele- 
ments in a list is less than 2 8 , 2 10 . or maxjnteger , respectively. 

All objects that can be referenced only one time need no 
special actions because the contents of these elements can 
be stored without pointer information. All objects of the 
same type within a part are stored together to eliminate 
storing the type flag for each element. Instead, there is one 
type flag stored at the start of the list of objects of the same 

The structure information of the quad tree is also stored 
with the model points. At load time, the tree can be built 
very quickly without time-consuming reconstruction by 
insertion of each model point in a new quad tree. To store 
the parts structure, there is a small difference in the mech- 
anism. Each reference to an unstored part initiates the stor- 
ing of this part and the fact that it was stored is marked in 
the temporary attribute of the part by setting temp_attr to the 
sequence number of the part. The next references to this 
part are stored by using the sequence number read from 
temp_attr for this part. 

Loading Data. The loading of the model data lists is done 
from the lowest to the highest level in the objects hierarchy. 
A transformation table is constructed for each list of shar- 
able objects in which the address for each member of the 
list is stored on the index that equals the sequence number 
in the list. To resolve shared object references, the address 
of the already created object is retrieved from the appro- 
priate transformation table. To minimize the temporary 
storage requirements, the lifetime of the transformation 
table is reduced to the time required to load one part. 

Help System 

A CAD system should not only be simple enough to be 
operated by a user who works with the system only now 
and then, but also sophisticated enough to allow a very 
experienced user to perform very special operations. In 
both cases there is a need for occasional assistance, for 

STORE command 

- -><STORE)- ->♦ >♦-»..->( all) >♦-->♦ >♦-> | f i le name|---> 


•-(MI)->' '-->|partname|-->' ' - - > (DELOLD) • -> ' 

STORE stores the drawing to the named file. If the named file already exists, 
you must use the 0EL_0LD option. STORE can produce either of two formats. 

MI selects the MODEL INTERFACE STANDARD format. This is a text format intended 
for long-term archiving and for interfacing with other systems. 

The default format is an internal binary format. Files in this format are 
smaller than MI f-les, and can be written and read faster. 

Files in sither format are readable with LOAD . 

ALL means to store the entire drawing, from the top part down. You can also 
name a part, and only that part will be stored. 

Fig. 13. Example Irom the help tile 


© Copr. 1949-1998 Hewlett-Packard Co. 

ME Series 10 Link to HP-FE 

The finite element method allows the engineer to analyze the 
benavior of a design accurately before it is manufactured The 
design process can be considered as a two-step loop. The first 
slep is the creation or modification of a design. The second step 
is the analysis of the design If the analysis gives negative results, 
both steps have to be repeated 

HP-FE is a general-purpose finite element system for linear 
structural and thermal analysis of two-dimensional, symmetric, 
and three-dimensional structures consisting of linear elastic ma- 
terial with homogeneous, isotropic material properties. Its 
analysis capabilities include linear elastic, linear dynamic, and 
linear heat transfer solutions 

The program performs m three steps: input of necessary data, 
calculation, and display of results The input process involves 
two steps: describing the geometry of the structure and defining 
the finite element mesh. The user only needs to enter enough 
points to define the boundary of the structure. Given the boundary 
of the object, a cross partitioning Into various subregions must 
be specified The geometry data consisting of points and lines 
(polygons) for each subregion can either be defined directly in 
the preprocessor or transferred from the ME Series 10 system 

The ME Series 10 data exchange program extracts and refor- 
mats the geometrical data required for the HP-FE program In 
the ME Series 10 environment, the user chooses Ihe drawing. 
For manipulations such as partitioning, the whole ME Series 10 
modification capabilities can be used. Then the geometry can 
be passed in a well-defined manner to the preprocessor of HP-FE 
The data exchange sets the third coordinate equal to zero, thus 
embedding the two-dimensional ME Series 10 data in an X-Y 
plane through the origin. Because of different operating systems. 
Ihe file formal is ASCII. After the transfer Ihe user switches to 
the HP-FE program where it is possible to add finite element 
specific information This consists of: 

■ Definition of the mesh 
" Properties of elements 

■ Boundary constraints 

■ Definition of loads 

Entering this data permits finite element analysis to be performed. 

Guenter Voss 
Developmenl Engineer 
Boblingen Engineering Operation 

example, to give the former user a hint about the order of 
input and to show the latter user a different way to deal 
with problems most efficiently. The assistance should give 
guidance, such as when a user just forgets how to input 
the correct command sequence, and should give hints about 
other commands or options that could be used as a better 
solution for the user's problem. 

A configurable and expandable system that can be mod- 
ified by the user should also allow the user to incorporate 
information about a specific functionality into the help 
system. The information presented should be technically 
correct and comprehensive. The technical user should find 
a notation that allows the user to check the syntax of com- 
mands, functions, and macros at a glance. Therefore, we 
chose the so-called railroad notation, where the user can 

travel along and find all options of a given command and 
the sequence in which they have to be entered. 

The ME Series 10 help system is made up of three files. 
One, called help, is shipped with the system. The other two 
files are generated by the system and are called help.i and 
help.t. One is an index file that allows very fast access to 
the formatted text file — its contents are actually displayed 
on the screen. 

All the information about the help system is in the master 
file called help. This file is human readable and can be 
edited with a regular text editor. To explain in more detail, 
let's take a closer look at a description of a specific com- 
mand. Fig. 13 is an original example from the help file. The 
keywords are preceded by the caret '. The keywords are 
later stored in the index file and used as keys to the corre- 
sponding help paragraph. 

The first keyword is the primary keyword, which nor- 
mally is the same as the command name. The other 
keywords may be applicable to more than one command. 
To get from paragraph to paragraph containing these key- 
words, the help command N (Next) is provided. 

The railroad structure shows all the options and the com- 
mand sequence. Words in capital letters (e.g., ALL) are 
known words of the system and have to be entered literally. 
The railroad structure is followed by a more verbal descrip- 
tion of what the command and the options do, hints about 
performance, boundary conditions, and what happens if 
an error occurs. 

The user can write paragraphs and insert them into the 
help file. The system then generates a new index and text 
file so that the user's paragraph is available from then on. 
Searching through the index file is very fast since the index 
file is loaded in main memory. If an exact match occurs, 
the key is found, and the paragraph is displayed. If no exact 
match is found, a second pass through the index file is 
initiated to find the best match. This means that if a simple 
typing error occurs or the exact syntax is not known, there 
is a good chance of getting the proper help message anyway. 

Once a user has a help paragraph on screen, besides the 
above-mentioned Next command, the user can scroll the 
screen to read previous or following paragraphs. If the user 
is already inside a command or function and doesn't know 
how to proceed, the user simply can enter HELP, which 
automatically displays the help paragraph for the current 
command or function. The help system is always available 
wherever the user is in ME Series 5/10, whether in com- 
mand entry, inside a command, in the editor, or even in 
the help system itself. 

Development Process 

The development took place on the Pascal workstation 
system using an HP SRM (Shared Resource Management) 
network, which is also a product configuration. 160,000 
lines of Pascal source code, a few assembler modules, and 
some C language routines were completed in a relatively 
short time. A number of tools were written to automate 
and speed up day-to-day work. Continuous porting of the 
software to the HP-UX operating system and continuous 
testing during the development phase reduced the chance 
of unwelcome surprises at the end. Early prototypes and 
the feedback from our alpha and beta test sites (see article 


© Copr. 1949-1998 Hewlett-Packard Co. 

The ME Series 10 NC Links 

All NC (numenc control) programming languages are based 
on the same principle First, geometric elements like points, imes. 
circles, and contours are defined Then the tools are moved 
along tnese elements For identification, all elements are labeled 
For example: 

L1= LINE 15. 
CI = CIRCLE50.55.3.20 

Traditionally, an NC programmer is given a set of engineering 
drawings of the part to be produced. The programmer has to 
understand and interpret the drawings accurately before pro- 
ceeding to write Ihe NC program. This is where much time is 
consumed and errors are made. Sometimes the amount of 
geometry definition in the NC source program is considerable 
and many typing errors are possible. Therefore, it is desirable 
to provide links from a CAD system to an NC programming system 
to minimize programming time and get code free of error 
These links support NC programming by generating a list of 
geometric elements translated into the corresponding NC lan- 
guage as shown by the example above. In addition, the CAD 
user can also offer a drawing with the label information that 
corresponds with that used in the NC source code 

However, mosl NC interlaces to CAD systems will not support 
drawing changes That means that any drawing change leads 
to totally different labeling of the elements The result is that all 
machining commands in the NC source program must be rewrit- 
ten Noi so with ME Series 10. In case of drawing changes all 
existing labels are not renamed, they are part of the model 

The following NC links are available 


■ COMPACT II (Manufacturing Data Systems, Inc or MDSI) 

■ NCGL (NC Graphics Language of MDSI) 


User Input APT .OUT 

Store Current Pari 

Copy worit2-ASC lo 

Purge mxIiZASC 
Result Message 

External Translator 
APT Program 

Translation and 

APT Output on 

Fig. 1 . Translation process from ME Series 10 to NC language 

Translation Process 

The translation process is started by an ME Series 10 macro 
The translators are implemented as external Pascal programs 
started by the RUN function ol ME Series 10 This function sus- 
pends ME Series 10 and starts the external program (translator). 
Exiling Ihe program will cause ME Series 10 to resume at the 
same point where il was left The translation is based on Ml data. 
If any labeled components exisl in Ihe drawing, only the labeled 
geometry is translated; otherwise, the entire geometry is trans- 
lated. Fig 1 shows the translation process 

Berthold Hug 

Development Engineer 
Boblingen Engineering Operation 

on page 30) helped us tune the feature set and polish the 
user interface. 

The extensive use of abstract data types made it possible 
to keep interfaces stable while experimenting with al- 
gorithms and data structures. For example, the point stor- 
age method changed from a linear list to a quad Iree whili; 
in beta test. The test customers probably did not notice, 
except for the increase in speed. 


Dan Matheson was the project leader of the breadboard 
that led to the ME Series 5/10. We thank Dan for his work. 
Mustafa Soliman was the first person outside R&D to work 
with ME Series 5/10 as a user, and he developed the first 
versions of the tablet/screen personality. Angela Suthurst 
of our publications group assisted with the editing and 
preparation of this article. 


1. Hewlett-Packard Journal, September 198fi. pp. 4-27. 


© Copr. 1949-1998 Hewlett-Packard Co. 

ME CAD Geometry Construction, 
Dimensioning, Hatching, and Part 

by Karl-Heinz Werner, Stephen Yie, Friedhelm M. Ottliczky, Harold B. Prince, and Heinz Diebel 

necessary part of the design process for mechanical 
engineers and drafting and documentation support 
personnel. Using an HP ME Series 5/10 Workstation makes 
this task easier, simplifies the work required to make revi- 
sions later, and provides a data base that can be used by 
other designers and subsequent manufacturing facilities. 
The tools provided with the ME Series 5/10 allow a usei 
to create and manipulate the following simple geometric 

» Construction lines and circles 

■ Lines 

■ Circles, arcs, and fillets (special arcs) 
I Polygons and splines. 

To represent these elements in the computer, the speed 
of geometric calculations and the amount of memory avail- 
able are important constraints. To minimize the amount of 
storage required, we describe the elements above by vector 
algebra and express all quantities as multiples of real num- 

■ Point: 2 numbers 

■ Construction line: 3 numbers — for instance, one point 
on the construction line plus an angle of direction 

■ Construction circle: 3 numbers — for instance, center and 
a radius 

■ Lines: 4 numbers — for instance, two endpoints 

■ Arcs, fillets: 5 numbers — for instance, center, radius, and 
begin and end angles. 

In the case of a construction line it is also possible to 
use the equation y = ax + b to represent the (x,y) points of 
nonvertical lines. 

Geometry Data Structure 

The ME Series 5/10 data structure for geometry has a 
top-down structure with four levels: 

■ Level 3: faces 

■ Level 2: contours 

■ Level 1: construction lines and circles, lines, arcs and 
fillets, polygons, and splines 

» Level 0: model points. 

The Quad Tree. Repeated modifications like scaling a rect- 
angle (consisting of four boundary lines) may introduce 
numerical errors into the values of the endpoints. It is then 
possible that the rectangle will no longer be closed. The 
best closure is obtained if the endpoints of adjacent ele- 
ments are identical. A less stringent form of closure is to 
make the difference of the endpoints less than a given 
epsilon («). 

To overcome the above-mentioned storage problem and 
the topological problem of closure, ME Series 5/10 uses 
the following mechanism. All points used to describe the 
geometric properties of elements are kept in a common 
storage area called model points. A model point is unique. 
There are points with topological relevance such as end- 
points of lines and there are points with no topological 
meaning, for example, the center of a circle. To distinguish 
between these cases, use counters are attached to the points. 

Entering a new point means comparing this point with 
other already existing points for identity. Two points are 
considered to be equal if both coordinate differences are 
relatively less than a variable called data_eps. The usual 
value of data_eps is about 0.000000000001 (10~ 12 ). Compar- 
ing new points with existing points in this way could de- 
grade the performance of the system, so an adequate storage 
format is necessary. The model points are stored in a quad 
tree. 1 The quad tree is a two-dimensional analog of a binary 
search tree. The basic idea of the quad tree concept is to 
associate four quadrants with a given point. Another point 
can then be classified to be in one of the four quadrants. 
Using the standard Cartesian coordinate quadrants, it is 
easy to classify points (see Fig. 1). 

In the kth level of a quad tree there is room to address 
4 k points. So a full quad tree of seven levels can store 
21,844 points. The search time for a specific point is very 
short if the tree has good balance. Typically the quad tree 











Data Structure 

Fig. 1. Quad tree concept. 


© Copr. 1949-1998 Hewlett-Packard Co. 

of a mechanical engineering object is balanced very well. 
If the model shape is equal to a nearly horizontally or 
vertically extended object, the quad tree degenerates to a 
binary tree. To degenerate the quad tree to a linear list, all 
points must be placed along a single line (a shape that is 
uncommon in mechanical drawings). The shape of the quad 
tree depends on input order. If an element referencing a 
given model point is deleted, then the use counters of the 
model point are decremented. If the use counters of this 
model point are zero and the program leaves the model 
point, the model point is deleted immediately. Otherwise, 
the model point is removed within the STORE command. 

An inspection of some complex mechanical engineering 
drawings (about 5M bytes of data) showed the average tree 
level to be 20. with a few branches up to 50 levels deep. 

With ME Series 5/10 a user can create points in the whole 
plane (limited by the longreal format) and the points may 
be arbitrarily close (again limited by the longreal format 
and by epsilon). 

Level 1 Elements. A construction line is considered as an 
infinite line passing through a given point. The internal 
representation consists of a point and a direction vector. 
Construction circles consist of a center (point) and a radius. 
Lines are represented by two endpoints. Arcs and fillets 
are represented by a center and two endpoints. Circles are 
considered as arcs with identical beginning and ending 
points: hence, they are represented by a center and one 
peripheral point. To speed up the system response, redun- 
dant information like radius and angles are also stored for 
these element types. 

Polygons connect data points by straight line segments. 
Splines connect data points by third-order interpolating 
curves with respect to a set of possible boundary conditions 
and damping associated with each data point. 

On this level the ME Series 5/10 data structure for 
geometry can be viewed geometrically as a set of elementary 
curves with model points as geometric parameters. To- 
pologically it can be viewed as an Euler net. 
Contours and Faces. With ME Series 5/10 it is also possible 
to have composite geometric entities like contours and 
faces. This is an important feature supporting modeling. 
Contours are parameterized by the individual elements 
forming the contour. Faces are parameterized by one outer 
and no. one. or more inner contours. Parameterization 
means association: the changed underlying elements au- 
tomatically influence all higher-dimensional elements. 

Entities on a given level (>0) refer directly to entities on 
the level below. The problem of finding all elements on 
higher levels that refer to a given element is solved by a 
search process using the use count information. Therefore 
it is necessary to have use counters on all levels (<3). For 
instance, a use counter on level 2 indicates how many faces 
share a given contour. 

Parts. It is desirable to collect all these element types in a 
single unit called a part. Then it is necessary to have a part 
element type (see the section on parts). 
Splitting. In a technical drawing, elements such as lines 
and circles are drawn in a common plane. These elements 
can be images of physical edges in different planes. Sup- 
pose now that the drawing is a front view of a simple 
three-dimensional object consisting of two adjacent rect- 

angles. Now add a horizontal line to the lower rectangle 
extending from the left to right boundary (see Fig. 2a). This 
new line may be the image of an edge in the same plane 
as the lines from the lower rectangle or it may belong to 
another plane (see Fig. 2b). 

In the first case, the original rectangle should be split by 
the new horizontal line, in the other case not. How can the 
program know whether to split or not? The splitting func- 
tion in ME Series 5/10 sets a flag. If this flag is true, at the 
insertion of a new simple geometric element like a line the 
system splits all elements passing through the endpoints 
of the new element automatically. If the flag value is false 
nothing happens. In this way the user can decide whether 
to split or not. 

There may be situations where splitting a set of elements 
or splitting at intermediate intersection points is necessary. 
In this case, the SPLIT function can be used. The SPLIT 
function uses the whole selection mechanism, that is. the 
user can select a single element, or several individually 
identified elements, or all elements in a rectangular box. 
and so on. Then, in the case of a single selected element, 
the user is asked where to split. The CATCH mechanism 
can be used to indicate the split point. In the case of several 
selected elements the program calculates the intersection 
points between these elements and splits corresponding 
elements at their intersections (spliUist). 

Let us now describe how the spliUist algorithm works. 
Suppose we are given a list of N elements. To find all 
intersections between these N elements one can consider 
all pairs (element i, element j). where l*i and jsN. and 
test these pairs for intersection. Because of symmetry, there 
are N(N-l)/2 of these pairs. So this algorithm behaves 



Fig. 2. Two-dimensional representations ol three-dimen- 
sional ob/ects mean that the addition ol a line can be the 
image ol an edge in the display plane (a) or it may belong 
to another physical plane (b). 


© Copr. 1949-1998 Hewlett-Packard Co. 

quadratically, which means that splitting large element sets 
will be slow. With N = 1000 we would have to test for 
approximately 1000 2 /2 = 500,000 intersections. Assuming 
500 intersections per second, this would take 1000 seconds. 

In ME Series 5/10 another algorithm is used. As the first 
preprocessing step, the smallest rectangular axis-parallel 
box around the list elements is calculated. This box is 
divided symmetrically into four disjoint subboxes. In the 
second preprocessing step, the list elements are classified 

as either belonging to one of the four subboxes or extending 
across two or more subboxes. Clearly elements belonging 
to different subboxes can have no intersections. The test 
to determine if an element belongs to a subbox uses four 
real comparisons. Hence, the number of preprocessing 
steps is proportional to N, 

Next, those elements that extend over several subboxes 
must be intersected with the corresponding elements. So 
aii element extending over subboxes 1 and 2 may intersect 






! TRX 





Fig. 3. Examples of ME Series 5110 dimensioning. 


© Copr. 1949-1998 Hewlett-Packard Co. 

elements in subbox 1 or subbox 2 or another extending 
element- Usually there are only a few elements extending 
over several boxes in a technical drawing. 

Now we are left with four problems of the same type as 
the original problem. Suppose the distribution of elements 
is uniform in the plane and that there are no extending 
elements. Then each of the four subproblems has tV4 ele- 
ments. The number of intersection calculations is now 
given approximately by 4(N/4)" = N 3 /4. 

Now we can repeat the steps described above recursively. 
With every recursion level, the number of operations is 
decreased by a factor of 4 under the hypothesis given above. 
In the program an empirically selected constant is used to 
stop recursion. If the number of elements is less than this 
constant, the quadratic intersection algorithm is used; 
otherwise, recursion continues. So the actual behavior of 
this algorithm depends on the distribution of elements, the 
number of subbox extending elements, and the cost of pre- 
processing steps (on every recursion level]. 

Coordinates and Mapping 

In vector algebra it is common to represent a planar point 
p in the form p = O + xe, + ye 2 . where (0.e,.e 2 ) is a refer- 
ence system consisting of a reference point O and two 
independent planar vectors e, and e 2 . With respect to 
(O.e, ,e 2 ) the point p has coordinates x.y, that is, p = (x.y). 

Points need to be transformed in two ways. The first is 
to change the reference system and then express a given 
point in the new system. The second is to change a given 
point in a fixed reference system according to some type 
of mapping (move, rotate, etc). The latter transformation 
occurs in the MODIFY command. The former occurs if 
geometric information from one part is compared with that 
of a second part. Expressing both actions mathematically, 
the transformation of a point p to the new point p' is given 

p' = Ap + b 

where A is a 2x2 matrix and b is a translation vector: 

Using homogeneous coordinates, p' = Ap h . where p h = 
(x.y.l) and A is a 2 x 3 matrix: 

To avoid unnecessary arithmetic operations and to op- 
timize certain error handling situations, a control mecha- 
nism was established. In ME Series 5/10 there is a type 
called geo operator, which consists of a matrix, a classifi- 
cation, and a flag. Classification of a matrix is according 
to its geometrical meaning: 

■ Unit (does nothing, usually used to initialize) 

■ Move only (two-dimensional matrix is unit, adds only) 

■ Orthogonal (preserves length) 

■ Conform (preserves shape, includes uniform scaling and 
orthogonal transforms) 

■ Affine (preserves parallelism, includes scaling along an 
axis and all former types) 

■ Perspective. 

The flag indicates that the given matrix preserves orien- 
tation or reverses orientation. 
The usual matrix operations like composition, inverting. 









Fig. 4. Examples ol dimensioning 
commands and functions 


© Copr. 1949-1998 Hewlett-Packard Co. 

and application lo a point use the classification to optimize 
the action. For example, the inverse of an orthogonal matrix 
is the transposed matrix, which is obtained without multi- 

The MODIFY and STRETCH commands in ME Series 5/10 
use the same set of options to define the geometry of a 
modification. The basic options are MOVE, ROTATE, SCALE, 
SIMILAR, and AFFINE. There are suboptions for MOVE, RO- 
TATE, and SCALE. With each modification the options DEL 
OLD or COPY and a repeat factor can be given. 

These commands allow the user to define what should 
be modified, and how it should be modified. The "what" 
part is done by the general SELECTION mechanism for MOD- 
IFY and by a special selection routine for STRETCH. For the 
"how" part the user can define translations, rotations, re- 
flections, and scaling in different ways. SIMILAR allows the 
user to define a transformation built up from a translation, 
a rotation, and a scaling operation by indicating two points 
and the corresponding image points. Affine mappings can 
be defined by three noncollinear points and their image 
points. Affine mappings may destroy the type of an ele- 
ment, for example, a circle will be transformed into an 
ellipse by the affine mapping 

(0,0),(1,0),(0,1) — (0,0), (2, 0),(0,0.5). 

Because ME Series 5/10 in its current version does not 
support ellipse elements, the circle element will be con- 
verted automatically to a spline element in the geometric 
form of the desired ellipse. 

The input to the general transformation algorithm used 
for MODIFY and STRETCH is a list of selected elements and 
a geo operator. For each element type there is a special 
modification routine. Modification of model points simply 
means multiplying the matrix by the coordinates of the 
model point. Modification of simple geometric elements 
like lines or arcs means point by point multiplication. So 
a line is transformed if its points are transformed. Contours 
are transformed if ail constituent elements are transformed. 
Faces are transformed if their contours are transformed. 
Parts are transformed if their underlying elements are trans- 
formed. This can be done for shared parts by composing 
the shared part's matrix with the modification matrix. For 
nonshared parts the part's elements are transformed indi- 
vidually by recursion. (For details of the mapping of faces, 
see the section on hatching.) 

Often two or more simple elements share a model point. 
If the image of such a model point is computed for the first 
time, the address of the image point is stored for further 
use in the temporary attribute associated with each model 
point. To use this information successfully, the temporary 
attributes of all model points that can be addressed by the 
given elements are set to an initial value, say NIL. Then, at 
modification time, an element is read from the list and 
processed depending on its type. The basic modification 
step is to apply the given geo operator to the model points 
associated with an element. If the value of the temporary 
attribute of the model point is NIL, the matrix of the geo 
operator is applied to the coordinates. This gives the coor- 
dinates of the image point. Now we need a new model 
point. So the coordinates of the image point are sent to the 

model point quad tree. If a model point with these coordi- 
nates already exists, the address of this model point is 
returned. Otherwise a new model point with these coordi- 
nates is created. The address of the new model point is 
then stored in the temporary attribute. The image model 
point is passed to the element. This basic modification 
mechanism for model points is the same for modification 
with copy or modification of the original element. In the 
case of copy, a new component record is created containing 
the new geometric (point) information. 

To describe how STRETCH works, suppose the user has 
created a line and wants to stretch the right end of the line 
by ten millimeters. After calling STRETCH, the user indi- 
cates the line endpoint, places the ruler* parallel to the 
line, and indicates two ruler divisions on the ruler's x-axis 
separated by ten millimeters. On the program level, a line 
element is selected, the endpoint not to be modified is 
marked in the temporary attribute with its own address, 
the other endpoint (indicated by the user) is marked with 
NIL, and the translation operator is calculated according to 
the entered points. At modification time, the program 
changes the original 10-mm-translation geo operator to a 
geo operator that fixes model point 1 and shifts model 
point 2 by 10 mm along the line. To do so, the program 
reads the information in the temporary attribute of the 
model points. The new mapping is defined by 1 —» 1, 2 ~» 
2', where point 2' is calculated from point 2 by application 
of the original geo operator. This type of mapping is a 
similarity operation. The new geo operator and the line 
element are passed to the transformation mechanism. So 
on the program level, stretching consists of two preprocess- 
ing steps: topological initialization of the temporary attri- 
butes and the calculation of a special geo operator for each 
(stretch) component from the global input geo operator. 
After these preprocessing steps, the normal modification 
algorithm applies. 

'The ruler facility in ME Series 5/10 closely resembles a T-square II was created to make 
the system resemble the traditional drawing board that most designers and drafting person- 
nel ate familiar with 


Arrow Type 

Text Ratio 

Dimension Unit 

Decimal Place 

Set Frame 

Fig. 5. Dimensioning parameters 


© Copr. 1949-1998 Hewlett-Packard Co. 

Scalar Product Geometry 

The usual geometric calculations used in a two-dimen- 
sional CAD environment are: 

■ Length and angle calculations 

■ Orthogonal and parallel projections 

■ Parallels 

■ Intersections 

■ Tangential problems. 

The basic objects to be dealt with are two-dimensional 
points and vectors considered to be elements of a real two- 
dimensional scalar product space. The usual Euclidean 
distance and the angle between vectors can be expressed 
by a scalar product <,>. 

Two principles are used to perform geometric calcula- 

1. Use a coordinate system related to the special structure 
of the problem. 

2. Use an orthogonal coordinate system. 

As an example, consider the ME Series 5/10 solution of 
the class of tangential problems called Apollonian prob- 
lems. Consider all triples of elements, where an element 
is a point, or a line, or a circle. Given such a triple, deter- 
mine the circle tangential to these elements. 

There are ten special tangent problems, namely: 

(Al) — point point point (point point radius) 
(A2) — line point point (line point radius) 
(A3)— circle point point (circle point radius) 
(A4) — line line point (line line radius) 
(A5) — line circle point (line circle radius) 
(A6) — circle circle point (circle circle radius) 
(A7) — line line line 
(A8)— line line circle 
(A9) — line circle circle 
(A10) — circle circle circle. 

A point can be considered as a circle with radius 0. so 
the class of different tangential problems involving ele- 
ments and points can be reduced to: 

(Cl) — line line line 
(C2) — circle circle circle 
(C3) — line circle circle 
(C4) — line line circle. 

There also must be algorithms for the radius problems: 

(C5)— line line radius 
(C6) — circle circle radius 
(C7) — circle line radius. 

To distinguish between the different problems in the 
user interface, the following options were introduced: 

■ tan3. covering cases A7 through Alt) 

I tan2_pt, covering case A4, A5. and A6 (radius input for 
pt is possible) 

■ tan_pt_pt, covering cases A2, A3, and A4 (radius input for 
pt is possible ) 

o three_pts, covering case Al 

a center, covering the case where the tangential circle is 

given by its center and one tangential element. 

Now. let us consider the solution to problem A 10. Given 
three circles with centers ml. m2. and nut. and radii rl. 
r2. and r3. respectively, we want to calculate the center 
mO and the radius rO of the tangential circle. The tangential 
condition for (ml.rl) is <m0-ml.m0-ml> = (rOrrl) 2 
and likewise for (m2,r2) and (m3,r3). 

In general, there are eight possible solutions to this prob- 
lem. One possible way is to calculate all solutions, compute 
the deviations of the actual tangent points from the approx- 
imate tangent points and return the minimum deviation 
solution. In ME Series 5/10 preselection of the solution is 
performed by analyzing the approximate tangent points 
and temporarily attaching a sign to the radii rl , r2, and r3: 

mO = ml -t-Xd,2 + M x Complement(d 12 ) 

where d 12 is the unit vector from ml to m2 and Comple- 
ment(d 12 ) is d, 2 's orthogonal complement. If ml = m2 = 
m3. then there is no solution except that also rl = r2 = 
r3. so after a possible rearrangement we can assume 
ml #m2. Inserting the unknowns, X, fi, and rO, in the above 
tangent conditions leads to three coupled quadratic equa- 

Decompose the vector d 13 pointing from ml to m3 into 
d 12 and its complement: d 13 = Td, 2 + kx Complement! d 12 ). 
We then get the following system of equations: 

\ 2 + M 2 = (rO + rl) 2 (1) 

(d„-M 2 + M 2 = (r0 + r2) 2 (2) 

(t-\) 2 + (k- M ) 2 = (r0 + r3) 2 (3) 

These equations are solved by standard elimination tech- 
niques. There are two solution branches depending on 
whether k = 0 or k ^ 0. k = 0 means geometrically that 
the three centers ml. m2, and m3 are colinear. In this case 
the radius rO can be found by solving a system of two linear 
equations. If k * 0, then rO is the solution of a quadratic 

Geometry Software Architecture 

The basic software module in ME Series 5/10 for the 

Extension Line 

Dimension Figure 

Fig. 6. Example ol point-to-point dimensioning. 


© Copr. 1949-1998 Hewlett-Packard Co. 

creation of geometry is called ACT_GEO. For each type of 
simple geometric element, a specific action routine in ACT_ 
GEO lets the user create elements of this type in various 
ways. To create a circle tangential to three other circles, 
the ccircle routine in ACT_GEO calls a specific routine in 
CMPNT_CIR. This routine in turn calls a routine in SOLVE 
to calculate the center and the radius of the tangential 
circle. If a solution is obtained, another routine in CMPNT_ 
CIR checks the solution, and if the check is positive, calls 
a routine in MODEL_DS to create a circle element and then 
calls a routine in DRAW to draw the circle onto the graphics 
area. Besides ACT.GEO, the software modules in ME Series 
5/10 are: 

LLGLBLS contains the basic definitions of constants such 
as tt, data types such as points, etc. 

GEO_MTR contains the routines for scalar product geometry. 

SOLVE contains the equation solver routines for all the; 
geometric problems ME Series 5/10 can handle. 

MINLDIST contains routines that calculate the distance and 
the projections from points to elements. These 
routines are used by CMPNT.ANY and MODEL_ACC to 
identify elements by a pick in the graphics area. 

MAP.GEO contains the calculus to deal with geo operators. 
Routines from this module are used by part routines, 
modification routines, and routines associated with 
the user coordinate system. 

MODEL_DS contains insert, get, change, and delete routines 
for elements realized as abstract data types in a stan- 
dard format. 

CMPNT^ANY contains routines that perform element type 
independent calculations such as: 
Routines to get other geometric representations of 
elements (e.g., circle data from an arc). 
Distance from point to element. 

Tangent vector at a point to an element. 
Intersection point(s) between two elements. 
Approximation of an arbitrary element by a point 

MODEL_ACC contains all the search routines, using MODEL 

CATCH uses the MODEL_ACC routines to catch according to 

the catch mode. 
MODEL_CHG contains split, merge, delete, etc. subroutines. 
CMPNTJJN. CMPNT_CIR. CMPNT_FIL, etc. contain the high- 
level routines to create elements. 
There are two levels in this software structure with re- 
spect to data structure. On the low level, objects such as 
constants, numbers, points, and vectors are defined and 
can be exchanged between routines. On the high level, 
beginning with MODEL_DS, one deals basically with abstract 
objects such as faces, contours, lines, and model points. 
For example, to catch to the intersection point of two lines 
on the screen, the user indicates this point approximately 
by a pick with the graphic cursor. Then CATCH calls routines 
in MODEI — ACC that identify the part, part matrix, and line 
elements of the intersection lines. The intersection routine 
in CMPNT_ANY gets all this information and decomposes 
the line elements into beginning and ending points, which 
are transformed by the part matrix to a suitable common 
coordinate system. This point information is then passed 
to a routine in SOLVE that calculates the intersection point 
relative to the coordinate system. After transformation of 
the intersection point back to the world coordinate system, 
the MODEL_ACC routine checks to see if the intersection 
point is in the range of the graphic cursor. If this is not the 
case, the CATCH routine passes a NOT FOUND to the calling 
action routine. If there are more than two elements passing 
through the cursor range, then all possible intersections 


© Copr. 1949-1998 Hewlett-Packard Co. 

are calculated and the closest to the indicated point is 


Dimensioning a given mechanical part means to give a 
description of the physical size, shape, and position of the 
geometry of that part in a human readable form. Hence, 
dimensioning is one of the most important operations in 
creating a detailed technical drawing (Fig. 3). There are 
national and international standards for dimensioning: 
ANSI. ISO. DIN. and JIS. for example. 
CAD User Expectations. CAD systems should help the user 
concentrate on the design process. Dimensioning a part in 
a technical drawing should be easy and fast. The system 
should provide a complete set of dimensioning functions. 
It should be possible to dimension according to a given 
standard. Often it happens that the geometry is changed 
after dimensioning. In this case dimensioning should be 
associative, that is, follow the modified geometry automat- 
ically. This system behavior saves a lot of work and time 
for the designer. Changing of dimensioning should be easy. 
Sometimes it is important that a system be able to perform 
isometric dimensioning. For factories using both metric 
and nonmetric dimensions it may be important to have a 
dual dimensioning capability. 

There are many routine tasks: conversion of units, plac- 
ing of dimension and extension lines, spacing checks, 
changing the height of a dimension text for microfilm pur- 
poses, etc. It is expected that a CAD system will perform 
these tasks with as little as possible user interaction. 
Dimensioning in ME Series 5/10. One of the ME Series 5/1 0 
design principles is fidelity of the created data, that is, the 
user creates geometry that corresponds to the physical size 

Fig. 8. Isometric dimensioning can be given to shared parts 

of the part that will be manufactured. Hence, dimensioning 
a part in ME Series 5/10 means to make the inherent geo- 
metrical properties of the part visible in human readable 

The following dimensioning commands and functions 
are implemented (see Fig. 4): 

■ Line dimensioning. The distance between two points or 
the length of an edge can be represented by either datum 
or chain dimensioning. Horizontal, vertical, or parallel 
dimensioning is optional. Coordinate dimensioning is 
possible. To create a dimensioning line, geometrically 
defined points such as the endpoints of elements or the 
centers of circles are needed. 

■ Radius or diameter dimensioning. 

■ Arc dimensioning. 

■ Angle dimensioning. Suppose a user wants to dimension 
the angle between two lines. By digitizing the text posi- 
tion in one of the four possible quadrants defined by the 
lines, it is possible to choose the angle of interest. The 
dimension figures can be expressed in decimal as well 
as in degree-minute-second notation. 

The shape of a dimensioning can be optimized by the 
user by changing one or more of the following parameters 
(see Fig. 5): 
* Text standard used 

■ Height and width of dimension figures 

■ Direction of dimension figures 

■ Gaps between edge and extension lines 

■ Length of extension lines over the dimension lines 

■ Number format. For example, the number 0.100 can be 
expressed as 0.1 00. or 0,100, or .100, or 0,1 , and so on. 

■ Position of dimension figures with respect to the dimen- 
sion line 

■ Units (mm, inch, fractional dimensioning in foot and 
inch, dual dimensioning). 

Data Structure. There are three design principles for thr; 
dimensioning data structure: flexibility, associativity, and 
compactness, Flexibility means including a complete set 
of dimension parameters in the data structure so that a user 
is able to generate dimensions according to a particular 
standard. Associativity means using references to geometry 


Fig. 9. Finding the closed outer contour (solid line) The al- 
gorithm backs out of "dead-end streets" like abed (dashed 


© Copr. 1949-1998 Hewlett-Packard Co. 

to let the dimensioning automatically follow any modifica- 
tion of the geometry- Compactness means to store only the 
most basic pieces of an actual dimensioning so that it is 
possible to reconstruct the expected display image from 
the stored data. 

As an example, we consider a point-to-poinl dimension- 
ing (see Fig. 6). The picture on the display consists of the 
geometry, the dimension line, the extension line, and the 
dimension figure. Stored in the data structure are references 
to the geometry, the values of parameters, and the dimen- 
sion text location in relative coordinates with respect to 
the geometry. Drawing this dimensioning means adding 
the nonstored dimension data by using an algorithm so 
that the complete display picture appears. 

What do relative coordinates mean? In the case of our 
example, the user must indicate two vertex points and the 
text position. The text position point is transformed into 
the coordinate system where the origin is defined by the 
first-indicated point, the x-axis by the line connecting the 
geometry points, and the y-axis by the line parallel to the 

extension lines passing through the origin. The length unit 
on the x-axis is defined by the distance of the vertex points; 
the length unit on the y-axis is defined by the internal units 
(mm). The information on this user-defined text location 
is kept even if the text figure does not fit between the 
extension lines. In such a case, the dimension figure will 
be drawn outside the extension lines. If, after a modifica- 
tion, the text figure will fit, the system remembers the text 
position and draws the text at the user-defined position. 

To make the data structure more compact, a set of param- 
eters that usually have equal values across a drawing is 
kept in a special buffer as a single item. As long as these 
parameters are used, any newly created dimensioning 
points to this item in the buffer. If one of the parameters 
is changed, the system searches in the buffer for an identical 
item. If such an item is found it will be used; otherwise, a 
corresponding item will be generated. The buffer is or- 
ganized as a linear list. 

Dimensioning is attached to a part. A part is an element 
that contains other elements of any type, including parts. 



0 1 135 GREEN SOLID 





0 33333333333333 1 0 GREEN SOLID 





0 0 70710678118655 45 GREEN DOTTED 


0 070710678118655 135 GREEN DOTTED 


Fig. 10. Examples ol hatching. 
Hatch defines the location, orien- 
tation, and size ol the pattern on 
the face. 


© Copr. 1949-1998 Hewlett-Packard Co. 

The dimension data is stored in this part. 

Associativity of dimension data is an important feature 
in ME Series 5'10. Therefore, the dimensioning algorithms 
follow a high-level description of the algorithms invoked 
by the MODIFY command. The user selects geometry ele- 
ments to be modified. These elements are kept in a separate 
list. The system searches the dimensioning data pointing 
to geometry in this list and collects the data for the selected 
elements in a separate dimensioning list. The geometry is 
then modified sequentially and the affected dimensioning 
is automatically adjusted to the values of the new geometry. 

The text location of a dimension is defined with respect 
to a mixed coordinate system. One axis of the coordinate 
system (the x axis) is changed automatically by the new 
geometry values, but the other axis value must be calculated 
from the transformation type and the old value. In addition, 
for stretching, the transformation can change from element 
to element. So the dimensioning must be modified in paral- 
lel with the geometry. 

Parts in ME Series 5/10 have their own transformation 
(see section on parts data structures). The purpose ot this 
transformation is to express the part content in the parent 
coordinate system. This transformation is classified by its 
geometric meaning. In addition to the mathematical mean- 
ing (length, angle, parallelism preserving), there is informa- 
tion on the use of the part data. In the case of an isometric 
view, the part data does not represent physical data, but 
instead a view of the physical data. Hence, in ME Series 
5/10 an additional part identifier is included to cover the 
cases of modeltype, viewtype, and detailtype parts. 

By default, the system assigns a modeltype identifier to 
a new part. Using the ISOMETRIC command with shared 
parts, the system automatically assigns the viewtype iden- 
tifier. On the other hand, the result of the DETAIL command 
is a shared detailtype part (see Fig. 7). 

The part identifier is used to draw the part dimensioning. 
The display appearance of a dimension is generated by a 
mapping mechanism, converting the stored data to the dis- 
play data. In the case of a modeltype identifier, the geome- 
try information is transformed by the part transformation, 
and then the normal display mechanism applies to this 
data. For a detailtype identifier, the system behaves as in 
the modeltype case except that the dimension figure is 
calculated from the original value. In the case of a view- 
type identifier, the geometry information remains un- 
changed while the display mechanism is adjusted to in- 


LINE . . . 

ARC . . . 

LINE . . . 
ARC . . . 


Fig. 12. List of commands that could have been used to 
create the drawing shown in Fig. 11. 

elude the part transformation. In the current state of the 
program, an isometric dimensioning can be given to shared 
parts only (Fig. 8). 


Besides the geometric information, the ME Series 5/10 
data structure also contains topological information. The 
data structure can be viewed as a topological net. A topolog- 
ical net is a finite set of objects from two abstract classes 
called knots and edges obeying the following rules: an edge 
connects two knots (equal or unequal) and at least one edge 
extends from each knot, 

The elements of the topological model are: 

■ Model points: the knots of the topological net. 

■ Primitive components: edges starting and ending at 
model points (e.g.. lines, circles, circular arcs, splines, 
and polygons). 

■ Contours: an ordered set of primitive connected compo- 
nents where every component occurs only once in the 

■ Faces: defined by one outer and none. one. or more inner 

Primitive components are connected if they share at least 

Fig. 11. Simple ME Series 10 part 
drawing (a) Geometry (b) Parts 


© Copr. 1949-1998 Hewlett-Packard Co. 

one model point. The components of a contour are ordered 
so that every component except the first and the last is 
connected with both of its neighboring components. A 
property of the contour is that the components are con- 
nected with no other components in the contour except 
their neighboring components. If the first and the last com- 
ponent of a contour are connected, the contour is called 

Recognition of Faces in Topological Nets 

ME Series 5/10 includes a mechanism that recognizes 
and creates faces identified by the user. There are various 
ways in which a user can identify faces to the system. The 
easiest is to indicate a point on the face. Because faces are 
defined by their boundaries, there are two tasks: finding 
the closed outer contour and finding any closed inner con- 

Finding the Closed Outer Contour. An infinite ray starting 
at the point indicated intersects at least once with every 
contour that includes the point. Starting with the compo- 
nent the ray intersects first, the algorithm tries to find a 
closed contour including the point. If no such contour can 
be found, the next component intersected by the ray is 
used for the search. 

The beginning and ending points of the start component 
for the contour search are defined so that the component 
is run through counterclockwise relative to the point that 
has been indicated. Starting at the endpoint we search for 
connected components until the beginning point of the 
start component is reached. If a point is reached where we 
have the choice between several components to go along, 
we use the one leading to the left. Since we run around 
the point in a counterclockwise sense, going left ensures 
thai we find the innermost contour of all the contours that 
include the point. 

If a point is reached where no connected component can 
be found, we change direction and go back the same way 
until we come to the last point where we had the choice 
between several components. There we take the component 
leading to the left. This is the one we would have taken if 
the dead-end street we just took was not there. 

Only two rules are used to determine the connected com- 
ponents. The first tells what to do when reaching a knot 
that has several branches, and the second tells what to do 
when reaching an endpoint of the net. 

Keeping track of the route results in a list of components 
where preceding components are connected by a common 
point. This list contains all the dead-end streets we tried, 


Right Hole 

Fig. 13. Desired modification of Fig. 12 parts structure (see 

so they are to be sorted out. Every dead-end street forces 
a return. In the list a return is documented by a component 
followed by itself. If we go back along a dead-end street, 
we pass the same components we passed on our way in 
but in opposite order. Hence, every dead-end street has the 
following appearance in the component list: ...a-b-c-d-d-c- 
b-a.... (see Fig. 9). 

Removing the return component d followed by itself 
yields: ...a-b-c-c-b-a... Going on to remove each component 
followed by itself results in a contour with no dead-end 
streets. It is obvious that the algorithm not only works with 
simple dead-end streets but also with complex connections 
of dead-end streets. 

Removing all dead-end streets may give a closed contour 
as we have defined it above, but it need not. It also may 
result in a few closed contours connected by bridges con- 
sisting of one or more components. Bridges exist if we still 
have components in our list that occur at least twice but 
are separated by other components. If we split our list at 
these bridges we get several closed contours. One of them 
is the outer contour we are searching for, the others are 
inner contours connected with the outer contour. To select 
the outer contour we search for the contour inside which 
the indicated point lies. 

This test depends on the fact that a point lies in a contour 
when the number of intersections of any ray starting at the 
point with the contour is odd. Since the direction of the 
ray is optional, it is a good idea to choose a vertical or 
horizontal ray to make computation easy and quick. 
Finding the Closed Inner Contours. First, the number of 
components that are used for the search of inner contours 
is reduced by a box test using the box of the outer contour. 
Then starting with an arbitrary component, we try to find 
contours by searching for connected components. If a point 
is reached where more than two components are connected, 
all components reachable from that point are recursively 
searched. Thus, a subset of components is gained. This 
subset forms a net. The outline contour or contours of the 
net are to be found. This can be done using the same al- 
gorithm described above for finding the face's outer con- 
tour. The only difference is that we now use a point outside 
the contour we hope to find. 

The start component for the outline search is the compo- 
nent reached first when an arbitrary ray from a point out- 
side is shot through the net. The beginning and ending 
points of the component are ordered the same way as above. 
The turn-left rule used for finding the outer contour of the 
face led to the innermost contour around the point. Since 

H ► Lisl ol Lines 

H ► List of Arcs 

H ► List ot Circles 

| ► Point Tree for these Elements 

Fig. 14. Graphical representation ot a part data structure in 
ME Series 10. 


© Copr. 1949-1998 Hewlett-Packard Co. 

the point now lies outside the net. the turn-left rule leads 
to the outermost contour of the net. If only points occur 
connecting two components, the contour is found when 
the beginning point of the first component is reached again . 

When all contours in the box of the outer contour are 
found, we have to sort out those that do not lie in the outer 
contour and those that lie in other inner contours. The 
following rule is used: A contour / lies in a contour O if I 
does not intersect O and one point of I lies inside O. 

Hatch Data Structure 

In mechanical engineering, the exposed cut surfaces of 
sectional views are indicated by hatching. Symbolic hatch- 
ing is used to distinguish various materials. In ME Series 
5/10, symbols for materials can be created by superimpos- 
ing sets of parallel lines. The symbols are called patterns. 

The data structure for the hatch in ME Series 5/10 consists 
of three elements: 

■ Simple pattern — all the data needed to define a set of 
parallel lines. It contains display information like color 
or linetype and geometric information defining the dis- 
tance between the lines and the orientation and location 
of the line set in the pattern. 

■ Pattern — a simple pattern or a connection of simple pat- 

■ Hatch — defines the location, orientation, and size of the 
pattern on the face (see Fig. 10). 

A pattern can be shared by hatches, and a hatch can be 
shared by faces. 

In many CAD systems, hatch is a separate element not 
coupled with geometry. In ME Series 5/1 0 another approach 
has been chosen. Hatch is considered to be an attribute of 
faces, Hence, hatch is coupled with geometry through faces. 

Modification of Faces and Contours 

The composite components, contours and faces, are up- 
dated automatically when the underlying simple subcom- 
ponents are modified. However, in the case of a MODIFY 
copy operation, hatches are copied automatically. The user 
cannot select faces and contours directly. Selection works 
on the simple component level. So the generic input for 
the MODIFY command consists of a list of simple compo- 
nents. This means that faces and contours pointing to these 
simple components have to be recognized. If all the simple 
components of a contour are in the list, the simple compo- 
nents are replaced by the contour. If all the contours of a 
face are in the list, the contours are replaced by the face. 
The result of this preprocessing is a list containing parts, 
faces, contours, and independent simple components. 

In ME Series 5/10, composite components can share their 

subcomponents. So we must take care that a component 
belonging to several composite components in the list is 
not modified twice or more. Therefore, the first step of the 
modification algorithm is to initialize the temporary attri- 
bute associated with each component. Only if the value of 
the temporary attribute of a component has its initial value 
at modification time xvill this component be modified and 
the address of the image component stored in the temporary 
attribute. The modify information stored in the temporary 
attributes is also used by an UNDO of the current MODIFY 
command where the inverse transformation is applied to 
all components with noninitial temporary attributes. This 
is especially important for the STRETCH command, since 
the original topological information about how to stretch 
is condensed to the temporary attributes. 


Up to this point we have discussed a number of elements: 
lines, circles, arcs, and so on. Now we turn to a new ele- 
ment: parts. A part is an element that contains other ele- 
ments of any type, including parts. A part usually corre- 
sponds to some physical object, like a screw, or to an assem- 
bly of physical objects, like a pump. But parts can be simply 
groups of elements that the user finds convenient to manip- 
ulate as a unit, and may have no direct physical meaning. 

Parts in a drawing are like directories in a file system. 
A part is an element, just as a directory is a file, and a part 
contains elements, just as a directory contains files. A part 
can contain a part, just as a directory can contain another 
directory. Parts impose a hierarchy on the elements of a 
drawing, just as directories do with a file system. 

It is common to represent this hierarchy graphically as 
a tree, much like a directory tree. Fig. 11 shows a simple 
ME Series 5/10 drawing representing two symmetrical 
holes, and the parts structure for this drawing. Elements 
are represented as circles. The elements on the bottom line 
are lines (L) and arcs (A); the other elements are parts. Each 
hole is a pari consisting of a line and an arc. The two holes 
are named Left Hole (l,H) and Right Hole (RH). The holes 
belong to a part called Top. which is present in every ME 
Series 5/10 drawing. 

The two holes could also be drawn without parts, in 
which case the only elements of the drawing would be the 
two lines and the two arcs, plus the everpresent Top part. 
So why should a user bother to create the part structure in 
Fig. 11? There are several reasons: 

1. A part can be treated as a single unit when modifying, 
deleting, storing, or loading. Assume the user wants to 
move the two holes closer together, for example. The user 

Right Hole 

Line ST Arc 

Righl Hole 

Fig. 15. Data structure representation of part shown in Fig. 


Fig. 16. Different data structure lor Fig 1 1 if LH and RH were 


© Copr. 1949-1998 Hewlett-Packard Co. 

can identify the part to be moved by picking any element 
belonging to it. With complex parts, this can be very con- 

2. Parts can be reused. The user might keep a library of 
commonly used screws, for example, and load them as 
parts when needed. If a drawing is made completely from 
library parts, the user needn't actually draw anything: the 
user only needs to say where the parts go. 

3. Parts can be shared. When two parts in a drawing are 
shared, any change to one is reflected immediately in the 
other. The two holes in Fig. 1 1 could have been created as 
shared parts, for example, although for simplicity they were 

One goal of ME Series 5/10 is to treat parts like other 
elements as much as possible. We have tried to avoid creat- 
ing two sets of commands, one for use with simple elements 
and the other for use with parts. Instead, we try to use the 
same commands for all elements. For example, there is no 
DELETEPART command; there is only DELETE, which applies 
to parts as well as other elements. 

Nevertheless, there are a few special parts commands. 
Two commands are used to create parts. With CREATESUB- 
PART. the user specifies which elements will belong to the 
new part with the requirement that the elements must al- 
ready exist. INIT_SUBPART creates a new part with no ele- 
ments; the user can then add the elements that will be in 
the part. Whether CREATE.SUBPART or INIT_SUBPART is used 
is the user's choice. The drawing and parts structure in 
Fig. 11 then could have been made with the commands 
listed in Fig. 12. 

EDIT_PART chooses the active part, which is very much 
like the current directory of a file system. Only one part 
can be active at a time. The active part is the one to which 
all new elements belong. In Fig. 12. INIT_SUBPART both 
created a new part RH and changed the active part to be 
RH. LINE and ARC then created elements in RH. 

The active part is also important for identification with 
the tablet or mouse. Suppose the user wants to delete RH. 
The user picks DELETE from the menu, and then picks, say. 
the arc of RH. How does the system know whether to delete 
just the arc. or the entire part? It doesn't, without some 
convention. The convention we have chosen is that the 
element to be deleted (or modified, or identified for some 
other purpose) is always in the active part. Ia our scenario, 
if the active part is Top, the system deletes RH. At that 
level, RH is an indivisible unit. But if the active part is RH, 
the system deletes the arc. What if the active part is LH? 
In this case, the system has no way to tell what the user 
wants, so it just beeps. 

The active part is so important that we have done two 
things to make plain to the user where the part is. First, 
the status line contains its name. Second, elements in the 
active part are drawn with the color and linetype chosen 
by the user (default values are white and solid), while all 
other elements are drawn in magenta and dotted, making 
them less prominent on the screen. 

Two commands are used to rearrange the parts structure: 
part, but leaves its member elements behind in the same 
place. GATHER pulls elements into the active part. As an 
example of the use of these two commands, suppose we 

want to modify the parts structure of Fig. 1 1 so that it looks 
like Fig. 13. That is, all four simple elements are to be 
members of the same part. One possible way would be to 
edit the Top part and then enter SMASH_SUBPART "LH". Now 
the drawing has the structure shown in Fig. 13, and the 
job is half done. To finish, we edit RH and then enter 
GATHER and identify the line and arc that used to belong 
to LH. This produces the result we want. 

VIEW is a command that causes a part to be viewed in a 
viewport without its surrounding context. Normally, a port 
has a view of the Top part, but it is sometimes convenient 
to view some other part while you are working on it. Nor- 
mally it suffices just to change the window, but if two parts 
overlap, only a view can show one without the other. It is 
even possible to have several viewports, with an overview 
of Top in one port and views of other parts in other ports. 
Parts Data Structures. Internally, each element is rep- 
resented by a block of memory containing the element's 
color, linetype, and other properties common to all ele- 
ments. In the following discussion, we call this block the 
element's head, and represent it as a circle. At the end of 
the head is data specific to the type of element represented. 
A line, for example, has pointers to its endpoints. A con- 
struction circle has a radius and a pointer to its center 
point. A part has a pointer to a block we call the part's 
body. Graphically, we might represent a part as shown in 
Fig. 14. The part body contains lists of its member elements, 
one list per element type. Having one list per type makes 
the system faster in several situations than it would be if 
all elements were in a single list. These situations arise 
when the system knows that only elements of certain types 
are interesting for the command. CONVERT_SPLINE. for 
example, converts splines to lines and arcs, and needn't 
consider any elements other than splines. Its also possible 
for the user to qualify a command with an element type, 
as in DELETE CIRCLES ALL, which deletes circles. 

Each part has its own point quad tree, containing the 
model points for all its member elements, except those that 
themselves are parts. This arrangement is much more suit- 
able than having a single point tree. It is very simple to 
delete a part with the current scheme, for example. There 
is no need to restructure a large point tree; we just throw 
away the tree of the deleted part. This scheme is also con- 
venient for point catching, since we can easily search for 
points in visible parts without having to look at points in 




Fig. 17. Simple version of a draw routine The matrix repre- 
sents a mirroring transformation. 


© Copr. 1949-1998 Hewlett-Packard Co. 

parts that are not being viewed. 

The fact that a part's head is separate from its body is 
the key to shared parts. Fig. 15 shows a data struct ure 
representation of the drawing in Fig. 11. Notice that LH 
and KH contain different elements. One part can be changed 
independently of the other. 

If LH and RH were shared, the display would look exactly 
the same, but the data structure would look like Fig. 16. 
The structure is much the same as that of Fig. 15. but with 
one essential difference: the two parts LH and RH share 
the same body, and thus the same elements. In this case, 
a change in one part is a change in the other, since internally 
there is only one representation of the two parts' contents. 

How can a data structure with one line and one arc in 
Fig. 16 look the same as one with two lines and two arcs 
in Fig. 15? How does ME Series 5/10 know that the shared 
line and arc are to be drawn twice? How does it know 
where to draw them? The answer is that each part head 
contains a matrix telling how to transform the part contents 
for the part instance represented by the head. 

The matrix can represent several types of transforma- 
tions: none at all. a translation, a rotation, a scaling, or a 
mirroring. In Fig. 17. the matrix represents a mirroring. 
The matrix can also represent combinations of transforma- 
tions as well as parallel perspectives. The matrix is created 
by the MODIFY command when it notices that the part has 
been declared shared by the command SHARE_PART. If the 
part is not shared, the MODIFY command must transform 
and possibly copy every element in the part. This can be 
slow for large parts. Hence, another advantage of shared 
parts is that the execution of MODIFY commands is much 

Since shared parts can be nested, there may be several 
matrices that apply to a given element. To draw an element, 
the system needs to transform it with each matrix along 
the path from that element to the Top part. To draw every 
element (when the user changes the window, for example), 
the draw routine must walk over the entire parts tree, main- 
taining at each point the product of matrices from the Top 
part to that point. A (very) simplified version of a draw 
routine is shown in Fig. 17 in a mixture of Pascal and 
English. The routine shows the sort of recursive approach 
that pervades all of the parts software. 

Selection Mechanism 

In many ME Series 5/10 commands, the first action is to 
select elements. Examples are DELETE. MODIFY. SPLIT, 
CHANGE.COLOR. CHANGE_FILLET. etc. For these commands. 
ME Series 5/10 uses a common mechanism called selection. 
In other commands, such as GATHER and STRETCH, more 
information lias to be selected, so these commands use a 
different selection mechanism. 

Selection works either locally in the current part or. if 
the user has entered GLOBAL, in all parts. 
Implicit Selection. Suppose the user wants to delete only 
one line. Probably the easiest way to do this is to call 
DELETE and to pick this element on the screen. On tin; 
action routine level, the delete act ion routine passes control 
to the selection action routine. This routine finds a [joint 
token from the user pick on the screen. This piece of infor- 
mation causes a search of the data structure to find tin; 

closest element to this point within the search range, hi 
this case the search is successful and the address of the 
line element is passed to the delete action routine. This is 
what is called implicit selection. The user poses an easy-to- 
solve problem and the system need not have more informa- 
tion to solve the problem. 

Other features of the implicit selection are given by the 
automatic vertex and box option. In picking a vertex, the 
system selects all elements that share this vertex. Now 
suppose the user entered a point and no elements passing 
through the search circle around this point were found. 
Then the system interprets this point as a box corner and 
asks for the other (diagonal) box corner. Then the elements 
completely inside this box are searched and passed to the 
calling action routine. 

Explicit Selection. To get explicit selection the user enters 
the SELECT command. For example, the user wants to delete 
three lines scattered around the screen. Probably the easiest 
way to do this is to use DELETE SELECT pick! pick2 pick3 CON- 
FIRM. SELECT triggers the selection action routine to expect 
input until the loop is finished with CONFIRM. As described 
above, a point token lets the selection routine search for a 
closest element VERTEX and BOX can now be entered as 
explicit options. 

With explicit selection the user can select elements with 
more options. There are the logical operators AND. OR, EXOR. 
NOT. ADD. and SUBTRACT. The element types POINTS. LINES, 
CLES. PARTS. GEO, and C_GEO can be used. Colors and 
linetypes can also be used as search patterns. For example, 
it is possible to select all blue arcs included in box 1 and 
not included in box 2. The user can also select all elements 
with the associated information string "remove me" that 
are not of long-dotted linetype. In connection with informa- 
tion literals, wild cards can be used. SELECT INFO "•er" 
selects all elements whose information strings end with 


1 II Sonet, "The Quad tree and Related Ejlerschical Data Struc- 
tures. Computing Surveys. Vul. 16. no. 2, June 1984. 


© Copr. 1949-1998 Hewlett-Packard Co. 

Alpha Site Evaluation of ME Series 5/10 

by Paul Harmon 

WHEN WE FIRST HEARD al HP's Vancouver. 
Washington Division of a new software package 
being developed in one of HP's facilities in Ger- 
many for two-dimensional computer-aided drafting (2D 
CAD), there was skepticism about the value of developing 
such a product within HP. Several software packages of 
that general description already existed and some were 
very good. Within HP. HP Draft had just been released on 
HP 9000 Series 200 Computers, and ECS-200 was getting 
a major overhaul that greatly increased its performance as 
a mechanical engineer's tool. Outside HP, several other 
programs developed by other manufacturers were avail- 
able. Beyond that, many felt that recently announced sys- 
tems that incorporated three-dimensional CAD capability 
were much more powerful than any possible 2D system 
and. though many times more expensive, represented a 
better investment. 

At HP's Vancouver Division (VCD), we had two projects 
that were making use of CAD to design next-generation 
personal workstation printers. The first project used HP 
Draft on our HP 9000 Series 200 Computers and was an 
alpha test site for the second generation of EGS-200. The 
second project used another company's three-dimensional 
solids software on dedicated hardware. Another group, still 
working with pencil and paper, was considering making 
the switch to CAD. 

Consensus as to which system the whole lab at VCD 
should adopt was difficult to arrive at, since each option 
had advantages as well as disadvantages. 

The Ideal CAD Workstation 

As we worked with our CAD tools, we began to see what 
an ideal system should include: 

■ The hardware should be inexpensive so that each en- 
gineer can be provided with a personal station, eliminat- 
ing timesharing (which we found to be deadly to produc- 

■ The system should be powerful enough to execute the 
CAD program quickly while being flexible enough to use 
for other engineering tasks such as instrument control, 
custom analysis, document preparation, and electronic 

■ The individual workstations should be networked so 
that data can be shared instantaneously within the de- 
velopment organization, and eventually between organi- 

■ The operating system should be easy to comprehend and 
allow multitasking for shared tasks such as instrument 
control, long batch mode analysis, etc. 

■ The program must be accurate enough to cover the range 
of sizes we would have in a typical layout. 

■ The software should incorporate a three-dimensional 
solid modeler as well as an easily understood two-di- 
mensional graphics editor with a full feature set. 

■ It should be possible to customize the user interface and 
write special subroutines, or macros. 

■ There should be links to both finite element ( FE) analysis 
programs and numerically controlled (NC) machinery 
for CAD/CAM integration. 

■ Links should exist to outside vendors, possibly through 
the Initial Graphics Exchange Standard (IGES), 

■ There should be easy access to spreadsheet, word pro- 
cessing, and electronic mail to provide an integrated 
workstation for maximum productivity. 

■ Above all. each part of the system should be reliable. It 
doesn't take long to eliminate all productivity gains if 
the system fails often. 

The CAD systems we were familiar with provided less 
than half these capabilities. When we asked about ME 
Series 10. we found that it was being designed to run under 
the HP-UX operating system (HP's version of AT&T's UNIX'" 
operating system] on HP desktop computers, provide links 
to FE and NC packages that we were already successfully 
using with HP Draft, and provide a macro language which 
would allow a high degree of flexibility. Althuugh lacking 
three-dimensional solids modeling capability. ME Series 
10 was always intended to be the user interface for the new 
three-dimensional modeling system (ME Series 30) intro- 
duced by HP in November 1986. Since the description 
sounded so close to our concept of the ideal CAD system, 
we offered to be an alpha test site. Although HP's Boblingen 
Engineering Operation in Germany (the authors of ME 
Series 10) initially thought Vancouver. Washington was 
too remote to be an effective test site, effective lobbying 
from Trent Christensen of Lake Stevens Division (the U.S. 
support engineer for ME Series 10) and our CAD experience 
with other systems convinced them it was worth their time 
to add us to their list. 

Testing ME Series 10 

A team from HP's Lake Stevens Division and Boblingen 
installed our first revision of ME Series 10 on our local 
shared network and we immediately began using it to de- 
sign some of the parts for a new printer. As we went through 
the design process, we discovered some very impressive 
new capabilities, some very irritating limitations, and our 
quota of bugs. These observations were collected over the 
network and sent to Boblingen once a month via HP Desk. 
HP's electronic mail system. The ME Series 10 authors 
would then respond as they deemed appropriate. 

From the outset it was apparent that we were working 
with a winner, although one that needed a little smoothing 
out. Engineers familiar with HP Draft were up to speed on 
ME Series 10 (creating drawings, doing isometrics, etc.) in 
a couple of weeks from a cold start. (The isometric shown 
in Fig. 1 was created by an engineer less than a week after 
he transferred to ME Series 10 from HP Draft.) 

Yet these same users have not found the program to be 


© Copr. 1949-1998 Hewlett-Packard Co. 

a toy that is restrictive or lacking in power. As the experi- 
ence of our group using ME Series 10 progresses, we find 
the program a powerful tool that aids us in the development 

One of the disadvantages of being an alpha site is the 
lack of manuals. It is a credit to the program that our users 
were able to transfer their knowledge from HP Draft to ME 
Series 10 so quickly using only the on-line help provided 
within the system augmented by startup help horn the local 
ME Series 10 guru. Still, some of our engineers found it 
difficult to use the compact flowchart style information 
presented therein and would have preferred an illustrated, 
written manual. 

The biggest disadvantage of being a test site for new 
software is the potential for losing data to software bugs. 
We (and several other sites) came up with several bugs, 
including one in the beginning that occasionally destroyed 
not only the drawing currently in memory, bul its stored 
copy as well! However, if ME Series 10 had been too fraught 
with bugs we would have stopped using it. Because the 
bugs were few and quickly fixed by the authors, we used 
ME Series 10 throughout our development project. 

Missing Capabilities 

A powerful macro language is included with ME Series 
10 that lets you virtually redefine the product if you desire. 
Still, with experience we found some language functions 
missing thai we could not easily work around. The ME 
Series 10 development team was very responsive to our 
requests, and as each successive version came out. we were 
pleased to find one complaint after another disappear. The 
more we use the ME Series 10 macro language, the more 
we wish to do with it. However, it is unlikely that the 
authors will be able or willing to satisfy all of our requests. 
We noticed in our evaluation that alpha test sites never 
ask for fewer features, always more! 

One of the limitations we found in the early versions of 
ME Series 10 was a lack of any layering capability. Al- 
though it provided a method of viewing parts individually 
(using the parts tree), no method to view groups of parts 
from different assemblies was provided. Layering was 
added and the user can now have a part on several layers 
at once, a nice advance over previous CAD systems. 

Dimensioning was quite irritating. In trying to provide 
a completely automatic method for the engineer. ME Series 


Fig. 1 . Isometric drawing created 
by engineer with one week's ex- 
perience on ME Series 10. 


©Copr. 1949-1998 Hewlett-Packard Co. 

10 would place dimensions in accord with standards set 
up by ANSI. While admirable, not all situations are foreseen 
by this august body and occasionally (but too often) we 
found the program putting dimension text where it decided 
the text should be and not where we wanted it. The ME 
Series 10 development team saw nothing wrong with this 
state of affairs and would have released the product with 
completely automatic dimensioning intact. While we are 
all lor ANSI standards, we did not like I he program enforc- 
ing the letter of the law and convinced them that most 
engineers would not like it either. ME Series 10 now offers 
greater freedom for dimension placement. 

In the early versions of ME Series 10. plotting was dif- 
ficult. Although we usually prefer flexibility, when it 
comes to Dlotting we want no surprises. ME Series 10 has 
the capability lo plot anywhere on the page and in the early 
versions the user was required to tell it where to plot on 
a page. The authors have added macros to make normal 
plotting (i.e.. putting a D-si/.e drawing on paper) straightfor- 
ward. Users simply tell the program to send the contents 
of the screen to whatever paper is in the plotter. Custom 
positioning on a sheet is still possible for the advanced user. 

When we first received the program, there were no plans 
for manual hatching. Automatic hatching is very fast and 
quite powerful but there are circumstances where the user 
wants lo do it manually and the development team had to 
be convinced of this. ME Series 10 now has a manual hatch- 
ing mode. 

Sometimes You Don't Get What You Ask for Immediately... 

There are a few functions in ME Series 10 that we feel 
still need development and as an alpha test site we had 
the opportunity to present these to the program's designers. 
The program's authors had several more pressing issues to 
attend to before the introduction of ME Series 10 and. as 
a consequence, some (in our opinion) needed improve- 
ments did not make it into production code. We have been 
told these will be incorporated in a future revision, how- 

...And You Don't Always Want What You Ask for! 

Sometimes the authors refused to make a requested 
change and we later realized that they were correct in doing 
so. For instance, at the beginning we requested that ME 
Series 10 set up the drawing area with the user-selectable 

"electronic sheet of paper" (A. B. C. etc.) analogy, just as 
HP Draft had done. We were comfortable with the analogy 
and fell a little lost without it. The authors refused and as 
we became more familiar with ME Series 10 we found that 
being locked into standard formats was indeed restrictive. 
We came to appreciate the infinite work surface of ME 
Series 10. 

As another example, we often asked the authors to pro- 
vide more macros with the base system. They politely told 
us to go write them ourselves. After a hesistant start we 
found that writing macros was as easy as Ihey said it would 
be. By encouraging us to write macros, the authors showed 
us the means for achieving great flexibility and control. 
We use our fluency with the macro language to help gel 
maximum performance from the program. We often write 
small procedures to help with particularly repetitive or 
automatable tasks (like "While I'm at lunch, plot this draw- 
ing, load that one. plot it, and then load another one."). 

Some Functions Impressive from the Start 

The macro language of ME Series 10 and the way it is 
implemented are extremely impressive and the key to very 
high productivity. All eight engineers using the program 
have developed individual menus and commands, collec- 
tively referred to as personalities. Macros are easy for the 
interested novice to learn, yet powerful enough for the 
experienced user. We have written macros ranging in com- 
plexity from simple one-line typing aids to a set that pro- 
vides an on-screen, full-function RPN calculator and 
another set that helps an engineer design cantilever beams. 

A simple screen editor is provided within ME Series 10 
to help in macro development. Basic system instructions 
are keywords such as LINE, ARC, FIT. and EDIT_PART, which 
make it easy to remember command names. Because of 
these features, it is productive for us to create small macros 
as we work. If we feel some feature of ME Series 10 is 
awkward or a function is missing, we can usually add the 
function through macros or change the operation of the 
feature on the spot and store il for future use. This provides 
easy adaptability that actually gets used often. 

We are continually impressed with a function designed 
to help construct isometric and general three-dimensional 
views of the part we are designing. Using AFFINE, a user 
can create two-dimensional isometric, dimetric. and trimet- 
ric drawings in a third of the time required to do so without 

Fig. 2. Using a group of files in an 
SRM directory as a development 
team's master layout changes the 
information flow within a project 
from (a) to (b). 


© Copr. 1949-1998 Hewlett-Packard Co. 

it. This functionality relieves some (but not all ) of our desire 
for three-dimensional solids modeling, because AFFINE 
makes it fairly easy for us to provide the shop with general 
views of a part we wish to have created. As an added benefit 
we often find mistakes during the process of creating the 
isometric view, saving us time and embarrassment. With 
complex parts the model maker rapidly gets a good feel for 
what is to be created. Although the value of an isometric 
view is less for a simple part, the process used to create 
the view is then so easy that we almost always make one. 

ME Series 10 supports group interaction during the de- 
sign process well. The parts tree structure allows layouts 
to have logical part groupings. This means subassemblies 
can be moved around and edited easily. In fact, subas- 
semblies can be kept in several different files and pulled 
in only when needed. This allows engineers to work on 
their designs and yet communicate changes in near real- 
time to the entire product team. We have our printer 
geometry divided between several files in a master layout 
subdirectory on our Shared Resource Management (SRM) 
network which the engineers work in and add information 
to. Product development time is conserved by lowering the 
number of build cycles required to find interferences be- 
tween parts. We can do this because each designer can 
view on a local workstation how a particular design re- 
lates to other parts. Hence, our intragroup communications 
have become more effective (see Fig. 2.) 

The two-dimensional analysis function built into the 
code was well received. It allows calculation of area, 
perimeter, and moments of inertia for any shape you can 
sketch. Dimensions can be moved around once created and 
automatically change in value if the part is stretched or 
changed. Dimensions for the whole drawing can be con- 
verted from metric units to inches with a simple toggle. 
Hatching stretches with the part. It is possible to change 
the size of a fillet after putting it in place, a great help. 
Three different types of layout grids are included (dot, line, 
and ruler). Full HP Draft construction line functionality is 
included. In short, ME Series 10 now has all the capability 

expected of a two-dimensional system plus some added 
special features. 

Using a translator, we can move ME Series 10 drawings 
directly into programs for driving our model shop's NC 
milling machine. The existence of this link (which also 
exists for HP Draft) has greatly enhanced the efficiency of 
the shop and has yielded a noticeable drop in part defects 
caused by poor communication compared to the time be- 
fore our link was in place. 

Throughout our testing of ME Series 10 we found the 
development team at Boblingen to be enthusiastic and re- 
sponsive to our suggestions. They were surprised that we 
preferred the option of having manual methods available 
rather than relying on automatic methods completely and 
learned that mechanical engineers like to do things their 
own way. regardless of how it is done down the street. We 
were impressed with the capabilities of ME Series 10 and 
saw an opportunity to define a tool for ourselves, getting 
a few steps closer to that ideal CAD station. In the process 
we learned a lot about the way we use our tools and in so 
doing increased our productivity. I would like to stress 
that our testing was done while we were actively designing 
a new printer as a team. If we had felt we were being slowed 
down by ME Series 10 we would have gone back to HP 
Draft because finishing the design and bringing our own 
product to market was, of course, our highest priority. The 
fact that we never looked back once we got up on the 
program is a testament to its basic functionality and the 
support we received from Boblingen on bug fixes. 

The ME Series 10 authors now have a product that has 
passed through the hands of real-world users. Using us as 
an alpha test site involved extra effort for both groups, but 
each found benefits coming from the work. I hope we can 
do it again. 


Members of the design team at VCD who tested ME Series 
10 included Rick Berriman, Don Bloyer, Dave Gast, Larry 
Jackson, Brian King, Dave Pinkernell, and Steve Rasmus- 

© Copr. 1949-1998 Hewlett-Packard Co. 



May 1987 

4 — CAD WnrkBlalinns 
Karl-Heinz Werner 

Karl-Heinz Werner holds a 
PhD degree in mathe- 
matics trom Saarbrucken 
University and has been at 
HP since 1 983. He has con- 
tributed to the development 
ol HP DesignCenler ME 
Series 1 0 and Series 30 as 
well as other products Karl- 
Heinz was born m Volk- 
lingen and is now a resident ot Deckenplronn He's 
married and has two children. His outside interests 
include amateur movie-making, photography, 
optics, and music 

Dieter Sommer 

Dieter Sommer came to HP 
in 1984 the same year he 
received his degree in 
. computer science from the 
I University ot Stuttgart. He 
I was responsible lor the tab- 
let overlay and screen 
menus tor HP Design- 
Center MESerieslOandis 
now working on product 
enhancements Born in Stuttgart, Dieter lives in 
Magstadtano is married. He sings in a choir, plays 
violin, and likes dancing. 

Dieter Deyke 

■ With HP since 1979, Dieter 
\ Deyke contributed lo the 
development ol HP Design- 
Center ME Series 1 0 and 
ME Series 30 He has 
| worked on storage, a 
nacro language, com- 
mand decoding and low- 
level drivers His Diplom In- 
genieur was awarded by 
the Engineering School at Esslingen He was born 
in Gdppmgen and now lives in Gartringen He's 
married and has three sons and a daughter 

Wolfgang Kurz 

Wolfgang Kurz |omed HP in 
1981 and has developed 
soltware for several CAD 
systems, including HP De- 
signCenter ME Series 10 
He was born in Muhlen- 
rahmede and studied tor 
his Diplom Ingenieur at 
^^^^ Paderborn He continued 

his studies at the Ruhr Uni- 
versity of Bochum and at Purdue University, from 
which he received an MSEE degree Woltgang is 
a resident of Bdblingen and lists flying sailing and 
bicycling as his favorite recreational activities 

Heinz P. Arndt 

A native ot Stuttgart, Heinz 
Arndt studied computer 
science at the University ol 
Stuttgart and received his 
degree in 1983 He pined 
HP the same year and has 
worked on software for vari- 
I ous CAD systems His de- 
sign contributions on HP 
DesignCenter ME Series 1 0 
include the plotter, binary storage, and pans of the 
internal data structure. Heinz is married and enjoys 
travel, sports, and reading, especially science fic- 
tion He's also interested in personal computers and 
artificial intelligence. 

16 — ME CAD Geometry: 

Karl-Heinz Werner 

Author's biography appears elsewhere in this 

Harold B. Prince 


A native ol Atlanta Geor- 
gia, Hal Prince received an 
MA degree in music Irom 
Harvard University m 1976 
and an MS degree in com- 
puter science from Yale 
University m 1979. He de- 
veloped UNIX software for 
a small Los Angeles com- 
pany belore pining HP m 
1984 His contribution for HP DesignCenter ME 
Series 10 centered on the parts and UNIX system 
interfaces. Hal lives m Gartringen. Wurtlernberg 
and enjoys reading, mathematics, and European 

Friedhelm M. Ottliczky 

Fnedhelm Ottliczky re- 
ceived his degree in 
mechanical engineering 
from the Technical Univer- 
sity of Karlsruhe and pined 
HP in 1984 His primary 
contribution at HP has been 
the topological data struc- 
ture for HP DesignCenter 
ME Series 10 Fnedhelm 
was born m Kunzelsau and now lives m Weil. 
Schfinbuch His leisure activities include classic 
cars, motorcycling, and skiing 

Stephen Yie 

Stephen Yie pined HPs 
BOblingen Engineering Op- 
eration in 1 983 and contrib- 
uted to the development ol 
HP DesignCenter ME 
Series 1 0 Recently, he has 
worked on software locali- 
zation and applications 
and has provided technical 
support for ME Senes 10. 

He has a bachelor's degree in mathematics from 
the University of Cologne and a 1 982 doctoral de- 
gree in mechanical engineering Irom the 
Rheinisch-Westfaliscn-Technische Hochschule al 
Aachen Stephen was born in Jakarta, Indonesia, 
and now lives in Bbblmgen He and his wife have 
three children His recreational activities include 
tennis, aikido. playing piano and guitar, and 
Chinese cooking 

Heinz Diebel 

Heinz Diebel is a native ot 
Stuttgart and holds a Dip- 
lom Ingenieur from the Uni- 
versity of Stuttgart. With HP 
since 1 980, he has worked 
on a number ol internal 
R&D projects and has con- 
tributed to the design of HP 
DesignCenter ME Series 1 0 
and ME Series 30 Heinz 
now lives in Boblingen and enpys tennis, swim- 
ming, and flying motorgliders. 

30 — Alpha Site Evaluation 
Paul Harmon 

t * Paul Harmon pined HP's 
McMmnville Division in 
1981 after receiving his 
BSME degree Irom the Uni- 
versity of Washington He 
has worked on mechanical 
engineering designs for a 
number ot R&D projects 
and was involved in prod- 
uct teslmg for HP Design- 
Center ME Series 10. His work on prmlhead inter- 
connections has resulted in a patent application 
In addition fo his HP work, he is studying lor a mas- 
ter's degree in mechanical engineering from Stan- 
lord University Born in Hermiston, Oregon, Paul 
and his wife and new daughter live in Vancouver. 
Washington He enjoys motorcycles, sports cars, 
programming, and church activities. 

35 = Power-Line LAN Z=! 
Robert A. Piety 

With HP since 1972, Bob 
Piety is a development en- 
gineer at HP Laboratories 
He's currently designing 
software and hardware for 
video graphics systems 
and previously charac- 
terized RF transmission 
properties ot ac power 
nes, the subject of his HP 
Journal paper. Among other accomplishments he 
has developed IC lest systems and has designed 
and built digital and analog controllers He's also 
working toward an MSCS degree Irom California 
State University at Chico. When Bob is not working 
or studying he enioys reading, woodworking, gar- 
dening, skiing, swimming, photography, camping, 
electronics, and computers. Several years ago he 
and his wile built their current home and land- 
scaped the grounds 


© Copr. 1949-1998 Hewlett-Packard Co. 

Intrabuilding Data Transmission Using 
Power-Line Wiring 

An investigation of the transfer and noise characteristics of 
intrabuilding power lines has indicated the feasibility of their 
use for local data communication at data rates greater than 
1 00 kbits/s. Within certain constraints, data rates of 1 Mbits/s 
or greater are possible. This paper discusses typical power- 
line characteristics in the 1 -to-20-MHz region and one 
implementation of a 100-kbitsls spread spectrum data link 
operating in the 3.5-to-1 0.5-MHz range. 

by Robert A. Piety 

Utility companies have used this technique for con- 
trol, telemetering, and communication over high- 
voltage lines since the early 1920s. 1 Most nonutility users, 
however, operate on the low-voltage, 120/240V lines. Uni- 
versity campus radio systems have often used the power 
lines for campus-wide broadcast-band programming. Other 
devices that use carrier-current communications include 
cordless telephones, intercoms, music systems, appliance 
controllers, and burglar alarms. Recently, local area net- 
works (LANs) that communicate over the power lines have 
become commercially available. 

Virtually all existing carrier-current systems operate in 
the 1 0-to-500-kHz range,' with the exception of the campus 
radio systems. Use of such low frequencies usually limits 
the data bandwidth of these systems to less than 10 kHz. 
Until recently, such data rates and bandwidths have been 
sufficient for most applications. However, the current ex- 
plosive growth in the use of distributed computers and 
terminals in offices and factories has created a demand for 
higher-bandwidth LANs. 

Several obstacles must be overcome when dealing with 
the power line's hostile environment, High noise levels 
and high signal attenuation are common. Furthermore. 

0 T 

Fig. 1. Typical attenuation and 
background noise for a 1 00-foot - 
long power-line path (reference = 
0 dBm. resolution bandwidth = 10 


© Copr. 1949-1998 Hewlett-Packard Co. 

these conditions not only vary from site to site, 3,1 but (along 
with impedance) also vary with time. To evaluate these 
difficulties, a variety of measurements were carried out at 
several different commercial buildings. 

Line Characteristics 

First, the power-line impedance was measured to deter- 
mine what type of coupling network would be required. 
The results were site and frequency dependent— imped- 
ance magnitudes in the l-to-10ft range at 100 kHz to the 
20-to-120ll range at 10 MHz were observed. The results 
indicate that 50fi provides a fair match, especially at the 
higher frequencies. Transfer characteristics measured from 
earth ground to the energized (hot) conductor were not 
substantially different from those taken between the com- 
mon and hot lines. Hence, subsequent data was usually 
taken between the hot and earth ground conductors. 

Using a spectrum analyzer and a sweep generator, attenu- 
ation and background noise measurements were made at 
several locations, including different buildings in the Palo 
Alto. California, area. Since ail buildings tested were com- 
mercial structures and wired in accordance with the U.S.A. 
Uniform Building Code, attenuation results did not vary 
widely. Noise characteristics, however, did vary consider- 
ably, being worst in areas containing heavy electrical ma- 
chinery. Background noise at these locations in a 10-kHz 
bandwidth was as high as - 20 dBm at 1 MHz, dropping 
to -65 dBm at 10 MHz. Typically, in most other areas, the 
noise was 20 dB lower. Also observed were random, in- 
frequent, broad-spectrum noise spikes of high (>0 dBm) 

The signal coupling between the three power-line phases 
and between branch circuits is quite high above 100 kHz. 
Typically, a signal injected into one circuit was detected 

at the same levels within 3 to 5 dB in other circuits that 
were routed in the same conduit. It is extremely difficult, 
if not impossible, to isolate a specific signal path because 
of this coupling. Hence, data for attenuation versus distance 
can fluctuate wildly. Therefore, radial distances were used 
to specify the path length for a given situation (rather than 
the actual, often unknown, route). Over 30-meter radial 
distances the typical attenuation was 20 to 30 dB at 1 MHz 
and 40 to 60 dB in the 10-to-20-MHz range as shown in 
Fig. 1. 

Much of the attenuation is caused by equipment plugged 
into the power-line network. For example, the typical office 
and laboratory environment contains numerous pieces of 
electronic equipment such as power supplies, oscillo- 
scopes, CRT terminals, desktop and other computers, disc 
drives, printers, etc. Most of these remain plugged into the 
power line whether or not they are switched on at the time 
of the test. Virtually all have a capacitor and/or inductor 
line filter designed to shunt and/or block radio frequencies 
traveling in either direction. The shunting effect of these 
filters is responsible for much of the attenuation observed. 
Switching the equipment on or off usually makes little 
change in this attenuation. 

A comparison experiment, in which all the equipment 
was unplugged from a 100-foot-long circuit path, produced 
interesting results as shown in Fig. 2. The top curve shows 
the power-line attenuation with the equipment unplugged. 
Because each wire shares the same conduit with other cir- 
cuits and there is excellent coupling between them, the 
effects of all the equipment coupled to the line cannot be 
totally removed — only minimized. When the equipment 
was plugged in. there was a 20-to-30-dB increase in average 
attenuation, as well as spectral differences in the position 
and depth of attenuation notches and an increase in their 

0 T 

Fig. 2. Attenuation losses caused 
by equipment on power line 
Upper curve is for equipment re- 
moved and lower curve is for 
equipment plugged in (resolution 
bandwidth = 70 kHz) 


© Copr. 1949-1998 Hewlett-Packard Co. 

number. These notches are a result of resonances caused 
by the lumped and residual reactances associated with the 
equipment and line filters, as well as standing waves on 
the lines. Because of these effects, there are often significant 
differences in attenuation at any given frequency between 
outlets close to one another. 

Attenuation not only varies with location, it also varies 
with time. This is caused by the ever changing load on the 
power line with equipment frequently being plugged in. 
unplugged, switched, or moved from one location to 
another. Sharp attenuation notches not only continually 
form and disappear, but change frequency and depth. Fig. 
3 shows broad and narrow attenuation changes (as much 
as 40 dB) recorded over a three-week period in one labora- 
tory-and-office environment. It is clear that the use of a 
narrowband communication scheme in such a hostile en- 
vironment is impractical unless extremely high transmitted 
power is used. Although it may be possible to deal with 
these narrowband problems through manual or automatic 
carrier- frequency changes, wideband communication 
methods appear better suited for this environment. Ap in- 
teresting form of wideband communications considered 
for this situation is spread spectrum. 

Spread Spectrum 

Spread spectrum transmission is a relatively new com- 
munications method where the baseband signal is spread 
over a wide frequency band — much wider than the required 
bandwidth for the information itself. Spread spectrum com- 
munication appears very suitable for power-line applica- 
tions because it has high immunity to narrowband interfer- 
ence and varying attenuation notches. This immunity is 
approximately the RF bandwidth divided by the data rate. 
(This ratio is called the processing gain.) Another advantage 

of this type of modulation is that the transmitted energy- 
is spread over such a wide frequency band that there is 
little or no interference with narrowband systems. One 
disadvantage in these systems is that the wide-bandwidth 
front-end amplifier is more likely to be subject to overload 
from strong interfering signals. 

There are several spread spectrum schemes including 
wideband FM. frequency hopping, direct sequence, chirp, 
and various hybrid forms employing combinations of the 
above. (Wideband FM is not usually considered spread 
spectrum since, in this case, the spreading is a function of 
the information signal alone.) A process or signal other 
than the information itself is used for spreading the signal 
bandwidth in all conventional spread spectrum systems. 5 

The basic principle of frequency hopping involves the 
radio carrier's hopping from one discrete frequency to 
another in a pseudorandom fashion. The receiver must 
have knowledge of the frequencies, their specific hopping 
sequence, and their rate, and must be synchronized with 
the transmitter before successful reception of the informa- 
tion is possible. 

Chirp spread spectrum uses a swept signal. The receiver 
must know the frequency range and sweep rate to detect 
the signal successfully. Dispersive delay lines matched to 
the angular rate of frequency change are often used to recon- 
struct the chirp signal. 

Direct-sequence spread spectrum uses a high-rate 
pseudorandom digital sequence to modulate the radio-fre- 
quency carrier directly. Detection requires synchronously 
mixing a copy of the pseudorandom sequence or using a 
SAW (surface-acoustic-wave) device 6 or tapped-delay-line 
correlator that is matched to the specific spreading code. 
This latter type of system was chosen for our power-line 
communications because it is simple to implement; it also 


© Copr. 1949-1998 Hewlett-Packard Co. 

demonstrates the capability of low-power data links to op- 
erate reliably on the power line. 

Similar work with spread spectrum communication on 
power lines was done earlier by Heinz Ochsner 7 in Switzer- 
land. Although the frequencies and data rales reported by 
him are much lower than those used here, the obstacles, 
methods, results, and conclusions are quite similar. 

The Hardware 

To evaluate power-line data transmission feasibility, a 
direct-sequence spread spectrum transmitter and receiver 
were built as shown in Figs. 4 and 5. A 31-bit pseudoran- 
dom spreading code, clocked at 3.5 MHz. is used to biphase 
modulate a 7-MHz carrier. For each data bit, all 31 spread- 
ing code bits (chips) are sent. The data bit can invert the 
chip sequence in antipodal modulation or, for simplicity, 
can amplitude modulate the output amplifier as done in 
this case. Thus, a one is represented by the transmission 
of the 31 chips and a zero by an absence of signal. The 
resulting data rate in the system is 112.9 kbits/s. 

The (sin x/x) 2 output envelope is shown in Fig. 6. Since 
90% of the energy is contained in the main lobe from 3.5 
to 10.5 MHz, the lower and upper lobes can be filtered out 
without significant performance degradation. (In Fig. 6, a 
27-MHz clock was used instead of 28 MHz; thus the fre- 
quencies of spectral features shown are offset slightly.) 

The receiver amplifies the signal and digitizes it using 
a ground-referenced comparator. The resulting digital sig- 
nal is shifted through a 124-stage shift register whose paral- 
lel outputs are analog summed to either of two nodes. Since 
the chip rate is 3.5 MHz, and the carrier is at 7 MHz, there 
are two complete sine waves for each of the 31 chips. Since 
each sine wave has a positive and a negative peak, there 
are a total of 124 such peaks for each data bit. Thus, each 
register output is connected through a resistor to the sum- 
ming node in a manner corresponding to the expected po- 
larity of the signal at that particular instant in time. 

As the pseudorandom signal stream shifts through, the 
summing nodes show only a small background signal 
whose relative amplitude is determined by the specific 
pseudorandom spreading code used. A sequence with a 
low background value was specifically chosen. When all 
124 bits shift into their expected positions, the now-corre- 
lated signal produces a strong peak at the summing nodes, 
as shown by the simulation in Fig. 7. This bipolar signal 
is squared, integrated, and synchronously detected to re- 
cover the original data. 

Since an analog delay line was not practical and the 
receiver's clock is not synchronized with the transmitter, 
two identical shift registers were built, the second being 
clocked at a 90° offset from the first. By using the summed 
and squared outputs of these correlators and summing them 
again, the need for clock synchronization is eliminated. In 
retrospect, a phase-locked loop and a single shift register 
could have been a simpler and better choice. 

The transmitter output power is 5 mW, which corre- 
sponds to a very low power density of 0.75 nW/Hz and is 
equivalent to 7.5 /nW in a 10-kHz bandwidth. Over a 100- 
foot transmission path through the power lines the data 
was recovered with typical bit error rates (BERs) of 10 _B 
to 10 ~ 7 . Increasing power did not substantially improve 
BER, although the working distance was extended. This 
suggests that unusually strong transients, rather than the 
typical background noise, are responsible for most of the 
residual errors. Since these transients are relatively in- 
frequent, retransmission or redundancy-error correction 
can be used to overcome their effects. 

Interference tests indicated that the noise level radiated 
by the transmitter was approximately the same as that 
radiated by a disc drive. Both were found to be - 70 dBm 
at a distance of 3 meters using a 1-meter vertical antenna 
and a calibrated spectrum analyzer. Average background 
noise in the building was - 85 dBm. 

Although performance differences between balanced and 
unbalanced coupling to the power line were slight, balanced 
transmission may be a better choice since the power lines 
then radiate less energy than for unbalanced transmission. 


This work has shown that the usable data bandwidth of 
the power lines is much higher than had been previously 
reported, thus permitting data rates suitable for high-speed 
LANs to be transmitted over a building's existing power 
lines. Because of the dynamic nature of the power-line 
environment, especially pertaining to narrowband interfer- 
ence, spread spectrum techniques are well suited for this 
application. The insurmountable transients that do inevit- 
ably occur are relatively infrequent, so reasonable error 
rates are attainable. By using the full bandwidth available 
and raising the spreading code rate, data rates of 1 Mbit/s 
or greater are possible. Here again, the key to power-line 
communications is to use wide-bandwidth signals to 
minimize the effects of narrowband interference and chang- 
ing attenuation patterns. 


- 4 



AC Line 

Spread Spectrum 

Fig. 4. Power-line spread spec- 
trum transmitter 


© Copr. 1949-1998 Hewlett-Packard Co. 

Reliable, high-speed, power-line data communication is 
well-suited to installations where wiring expenses for such 
communications are excessive or not practical, such as fac- 
tory floors, older buildings, temporary or movable installa- 
tions, and certain computer-peripheral interfacing situations. 


The author wishes to thank Zvonko A. Fazarinc, director 
of the Measurement & Communication Laboratory, Hew- 
lett-Packard Laboratories, for suggesting the problem and 
for providing the opportunity to do this work, and to R. A. 
Baugh. R. D. Crawford, B. ). Elliott, J. P. Freret. and R. L. 
Wheeler for iheir contributions throughout this investigation. 


1. M.P. Perry and M.R. Stambach, "A System Analysis of Power 
Line Carrier Communications with Gas Insulated Conductors and 
Overhead Lines," Canadian Communications &• Power Con/erence, 
1980, pp. 240-243. 

2. "Summary of an IEEE Guide for Power-Line Carrier Applica- 
tions." IEEE Transactions on Power Apparatus and Systems, Vol. 
PAS-99, no. 6. November/December 1980, pp. 2334-2337. 

3. S.N. Talukdarand J.C. Dangelo, "Uncertainty in Distribution PLC 
Attenuation Models," IEEE Transactions on Power Apparatus and 
Systems. Vol. PAS-99, no. 1, )anuary/February 1980. pp. 328-334. 

4. J. Shekel. "Using Electric Power Distribution Lines as a Communi- 
cation Network," IEEE International Conference on Circuits and 
Computers, Vol. 2, 1980, p. 1010. 


Fig. 6. Frequency spectrum ol 
spread spectrum signal 


© Copr. 1949-1998 Hewlett-Packard Co. 


Fig. 7. Correlator output. 

-5 0 5 It 



5. R.C. Dixon. Spread Spectrum Systems, John Wiley & Sons, New 
York. 197fi. 

6. W.R. Shreve. "Radio Data Link," Hewlett-Packard Journal. Vol. 
33, no. 1. January 1982. p. 7. 

7. H. Ochsner, "Data Transmission on Low Voltage Power Distribu- 
tion Lines Using Spread Spectrum Techniques," Canadion Com- 
munications 6- Power Conference, 1980, pp. 236-239. 




© Copr. 1949-1998 Hewlett-Packard Co.