IBM France La Gaude Laboratory Contributions to Telecommunications: Part 2

Abstract

IBM France's La Gaude Laboratory worked at the intersection of computing and electronic communications for more than 40 years. Its research led to specific technical contributions such as voice encoding, trellis-coded modulation, and many others. Part 2 of this two-part article discusses in detail the contributions, inventions, and concepts that were, at the time, landmarks in the state of the art, from the creation of the laboratory in 1959 through the early 1990s.

Full text

Page 1

IBM France La Gaude
Laboratory Contributions to
Telecommunications: Part 2
Michel Bastian, Marc Boisseau, Robert Cohendet,
Alain Croisier, Claude Galand, Etienne Gorog,
Philippe Hernandez, Michel Humbert, Cuong Ngo Mai,
Pierre Secondo, and Robert F. Steen
IBM France’s La Gaude Laboratory worked at the intersection of
computing and electronic communications for more than 40 years. Its
research led to specific technical contributions such as voice encoding,
trellis-coded modulation, and many others. Part 2 of this two-part
article discusses in detail the contributions, inventions, and concepts
that were, at the time, landmarks in the state of the art, from the
creation of the laboratory in 1959 through the early 1990s.
From its inception in 1959, the IBM La Gaude
Laboratory was thrust into the middle of the
convergence of computing and telecommu-
nications. With the targeted recruitment of
valuable engineers and scientists in France,
and cooperation with other leading IBM lab-
oratories, it soon became one of the world’s
most influential laboratories in the crucial
area of applying telecommunications con-
cepts to computer technology and vice versa.
The first installment of this two-part article
dealt with the La Gaude’s general contribu-
tions as well as its history, mission, projects,
and products of historical significance. In ad-
dition to its general contributions to the tele-
communications field, the laboratory also
contributed specific products and processes
in electronic switching systems, digital signal
processing, and computer communications.
In Part 2, we discuss in detail the innovative
contributions, inventions, and concepts that
were, at the time, landmarks in the state of
the art.
Electronic switching systems
Engineers and scientists worldwide recog-
nized the potential of computer-based tele-
phone exchanges in the early 1960s. The IBM
La Gaude Laboratory was one of the earliest
computer company laboratories to explore
the integration of these two technologies.
Every part of the system posed unique chal-
lenges from the voice switch itself to the
provision of uninterrupted service in spite
of computer failures. On the other hand,
computer control also made new functions
possible, which had to be invented. Even de-
velopment process had to be reinvented to
serve diverging national requirements in a
common framework.
Integrated space-division switching network
Beginning in 1964, in the initial design
phase of IBM’s automatic branch exchanges,
the design team considered all the possible
switching principles for a network design.
The dominant technology in the 1960s was
the electromechanical matrix, or crossbar
switch, which was gradually replacing the pre-
vious Strowger switch. Some new switching
system designs used reed relays, tiny pairs of
contacts enclosed in glass bulbs and oper-
ated by external induction coils. At the same
time, other prototypes of PAX or private auto-
matic branch exchanges (PABX) were built
according to the principle of time-division
switching whereby voice was converted to
pulse-amplitude modulation (PAM). Digital
voice, also known as pulse code modulation
(PCM), had appeared in urban transmission
in 1962, but the digital technology of that
era could not yet support time-division multi-
plexing (TDM) switching of PCM streams.
In this new business venture, IBM believed
that it could not depend on components pro-
duced by competitors. Telephone companies
18
IEEE Annals of the History of Computing
1058-6180/09/$25.00?
c 2009 IEEEPublished by the IEEE Computer Society
Authorized licensed use limited to: Dalian University of Technology. Downloaded on October 4, 2009 at 04:26 from IEEE Xplore. Restrictions apply.

Page 2

produced most of their switching compo-
nents, which eliminated crossbars or reed
relays from IBM’s available choices. Thus,
La Gaude built its first design and laboratory
prototype, a 30-extension-line PABX, with
analog time-division switching. However,
it soon became clear that the analog time-
division approach (1) could not be scaled
up to large switching networks and (2) could
not match the stringent signal attenuation
specifications imposed on the PABX by the
common carrier administrations (these speci-
fications closely matched the capabilities of
metallic contact switches). Because it was a
newcomer and apparent competitor, IBM
could not expect the specifications to soften
in its favor, so by necessity, it had to develop
its own switching technology.
After investigating various electromechan-
ical and electronic designs, the La Gaude Lab-
oratory team arrived at an original solution
based on a matrix of semiconductor cross-
point switches. By the end of 1964 this solu-
tion became the team’s leading option. With
the help of the IBM Components Division,
a physical design could be derived from
the IBM transistor process with minimal
changes.1Each cross point comprised a thyr-
istor as a self-latching switch, plus a diode
and a resistor integrated on a single piece of
silicon.2These first commercially produced
integrated circuit (IC) devices from IBM
were manufactured on the new System/360
computers’ regular production lines of logic
transistors and diodes. Each integrated cross
point measured 1 mm2(see Figure 1). The
cross-point switch matrix was first used in a
field-test prototype in 1966 and then com-
mercially in 1969 with the IBM 2750 switch-
ing system.
The cross points were arranged in matrices
connected in a three-stage folded network.3
The connections were established by matrix
selection and held by constant current sour-
ces fed through the network (see Figure 2).
Voice transmission was bidirectional, but un-
balanced. That is, a single signal path was
established over a common ground plane.
This audacious design choice was made possi-
ble by the packaging technology La Gaude
borrowed from IBM computer hardware. At-
tenuation was still a concern because each
thyristor featured a residual impedance of
about 6 ohms. This was compensated for by
a negative-resistance circuit connected in se-
ries at the central junction point.
Along with the switching network devel-
opment, IBM needed to develop a mass of
auxiliary circuits, such as extension line and
Figure 2. Simplified schematic diagram of the IBM 2750 unbalanced bidirectional switching network.
b
a
Figure 1. IBM 2750 PABX. (a) Circuit diagram and
(b) microphotograph of its integrated cross point
(die size approximately 1 mm2).
April–June 2009
19
Authorized licensed use limited to: Dalian University of Technology. Downloaded on October 4, 2009 at 04:26 from IEEE Xplore. Restrictions apply.

Page 3

trunk line adapters, internal junctors, status
scanners, controller interfaces, and so on.
Many design adaptations were also needed
because of new regulatory requirements and
the product’s geographically wide coverage.
The new and powerful IBM space division
technology from the La Gaude Laboratory
remained competitive for many years, until
the late 1970s when the progress in circuit in-
tegration allowed digital switching to take
over and rapidly supersede all other options.
Control switch-over without service interruption
A telephone exchange or PABX provides a
vital 24-hour, 7-days-a-week function, and
service interruption should be a rare event.
Thus, a PABX with centralized control must
have redundant control components. In
1964 and 1965, the industry standard for re-
dundancy was two identical controllers that
operated synchronously. In the event of a
controller failure, the other would assume
the control tasks. This architecture had a seri-
ous shortcoming because it offered no protec-
tion against software failures, which could
cause both controllers to fail under the same
fault condition in the same operation cycle,
resulting in a short service interruption
and loss of all established connections. In re-
sponse, La Gaude developed a different archi-
tecture, one with a design objective of a
50-year mean time between failures. This
architecture had two controllers, preloaded
with the same call processing and connection
software, but only one of which was in charge
of operations. The second controller remained
in active stand-by, continuously checking
itself. Both controllers sent ‘‘OK’’ signals peri-
odically to a common switch-over control cir-
cuit, a device with a simple function, built of
redundant—majority decision—logic circuits,
and using an independent power source.
The confirmed absence of an OK signal
from either processor would trigger the ap-
propriate recovery action, sending an alert
to the maintenance subsystem. If the prob-
lem originated in the active control unit,
operations would be transferred to the
stand-by unit by activating the physical con-
nections to the switching network and tele-
phone line attachments.
The switching network was self-sufficient
and did not change its configuration during
switchover, so that all established connections
were maintained. The system would only lose
the calls in the process of being established
at the time of switchover. Callers would
simply redial. The newly active controller
gathered information from the network to
reconstruct its configuration information in
memory, thus avoiding the use of possibly
corrupted data from the failing unit.
Two additional features controlled this
mechanism’s proper operation. First, the con-
trollers acknowledged and verified each OK
signal the switch-over logic received. Also,
every night during a low call period, a forced
switchover was scheduled to verify the integ-
rity of all control devices in each unit.
Advanced functions
One
program-controlled telephone switching was
that new useful functions (beyond establish-
ing connections between extensions, trunks,
and tie-lines from A to B) were easily added
and modified. Functions such as call waiting,
call conferencing, and call holding were
virtually impossible using current technolo-
gies. Market studies and focus groups with
customers indicated that customers might
accept various new functions. Program-
controlled switching made so many options
possible that the main challenge became
one of reining in product planners and pro-
grammers who were eager to design and de-
velop every new function imaginable. One
major limiting constraint was that the new
functions had to be accessible through a sim-
ple dial telephone or, at best, a touch-tone
telephone handset. For cost and regulatory
reasons, more sophisticated telephone instru-
ments as we know them today could not be
considered. Remarkably, many of the new
functions introduced, such as double-call
processing, conferencing, call-waiting signal-
ing, and permanent personal phone num-
bers, have since become standard features in
private and central exchanges. In addition,
the IBM PABX offered specialized functions
for corporate facilities. These included con-
tact sense and activation for manufacturing
control, access and attendance control, cus-
tomer updating of telephone system configu-
rations, and remote troubleshooting and
maintenance.
More than a million lines of control code
were developed at La Gaude between 1965
and 1970. The programming activity ex-
tended from code compilers to applications,
including remote maintenance and the real-
time control program—the system’s heart. A
major achievement was the first provision
of direct inward extension dialing—that is,
calling a specific extension without going
through an operator, using individual public
majoradditionaladvantage of
IBM France La Gaude Laboratory Contributions to Telecommunications: Part 2
20
IEEE Annals of the History of Computing
Authorized licensed use limited to: Dalian University of Technology. Downloaded on October 4, 2009 at 04:26 from IEEE Xplore. Restrictions apply.

Page 4

phone numbers assigned to each exten-
sion phone. In Germany, a similar function
existed in certain areas where the extension
number could be dialed in addition to the
public phone number. The IBM 2750 offered
German direct inward dialing. Full direct inward
dialing within the public numbering system
was first introduced in Europe as a feature of
IBM 3750.
Another achievement was the realization
of corporate and government private net-
works of PABX, self-managed with signaling
and routing facilities independent of the
common carrier services. These private net-
works had advanced security features and
used an integrated numbering system. In
the 1970s, the French government acquired
such a network from IBM for its own critical
communications needs.
The IBM 3750 switching system offered
remote maintenance in 1972. A support cen-
ter received malfunction alarms, initiated di-
agnostic routines, isolated a subset of the
system, and downloaded program updates.
Altogether, this was a substantial innovation
in the telecommunications world. Concur-
rently with the development of the IBM
PABX, La Gaude also built an innovative
computer-controlled test and qualification
system known as TESA (Telephone Evalua-
tion System Automat). TESA simulated the
PABX’s environment (users, trunk lines, and
operators), checked its operation, and mea-
sured its performance under intensive traffic
loads.
Single development process
Prior to IBM’s entry in the field of tele-
phone switching, PABXs were designed to
national specifications and usually developed
under supervision by the PTT laboratories.
The installations including interconnection
of the various component subsystems were
assembled on the customer premises. IBM in-
stead adopted a closed-box design, with a com-
mon base and preinstalled country-specific
features. This approach, revolutionary in the
telephony field, required extensive testing
and demonstration by IBM to obtain national
agreements in Europe.
In 1966, La Gaude operated a prototype
system to prove IBM’s closed-box feasibility
and to obtain PTT’s formal agreement before
connecting the system to the public network
in France. Beginning in 1968, one field test
system was installed in a dozen major coun-
tries. Installations were scheduled according
to local availability of skilled installers,
financial
agreements, and strategic interest. France (a
second installation), Germany, the UK, Italy,
the Netherlands, Denmark, Sweden, Brazil,
Finland, Austria, Belgium, and the US partici-
pated in these system field tests using the IBM
2750 Switching System, which was still in
development.
La Gaude’s planning efforts, negotiations,
field-test installations, PTT agreements, and
operations experience afforded IBM, other
manufacturers, and
trations early awareness of these systems’
requirements. Through such logistic and
technical achievements, La Gaude was in-
strumental in accelerating and optimizing
the standardization process for electronic
PABXs and private networks in Europe and
beyond.
resources, telephonecompany
telephoneadminis-
Signal-processing architectures
and algorithms
In the 1960s and 1970s, the early days of
computer communication, IBM was a verti-
cally integrated manufacturer. From com-
ponent production to customer service,
it produced nearly everything computers
required, including transistors and logic cir-
cuits in large volume on automated facilities
of its own design. These components were
specified, designed, and intended only for
computing applications. Accordingly, they
fit in rigid automated design practices, from
individual circuit conception to final assem-
bly test and maintenance in the field. An in-
dependent internal organization, IBM Product
Assurance, certified any purchased component
not produced by IBM. Products developed
with these components required additional
engineering to meet the IBM manufacturing
and maintenance organizations’ operating
requirements.
To the La Gaude Laboratory engineers
developing telecommunication products,
the use of in-house components was both
an opportunity and a constraint. In those
days, even the design of a simple amplifier
was a challenge to an engineer constrained
to transistors not specified for linear opera-
tion. Of course, as digital technology became
cheaper and better over the next few decades,
virtually all signal-processing designs moved
to the digital domain. The constraints placed
on designs from La Gaude in the 1960s
forced the engineers to think digitally long
before the practice became common, which
helps explain why so much early work
came from La Gaude.
April–June 2009
21
Authorized licensed use limited to: Dalian University of Technology. Downloaded on October 4, 2009 at 04:26 from IEEE Xplore. Restrictions apply.

Page 5

The digital echo modulation process pre-
sented the first opportunity for a digital de-
sign in 1969.4This involved the synthesis
of signals from logical circuit outputs and
summing networks of resistors. After generat-
ing a pulse sequence digitally and passing it
through a low-pass filter, modulated signals
with perfectly defined frequency and time
specifications were formed (see Figure 3). Al-
though perhaps obvious today, at the time
this design was a significant innovation in
generating precise amplitude and phase
modulated signals. This technique was later
extended to multiphase and quadrature am-
plitude modulation (QAM) signals.5
As logic circuit performance improved,
digital techniques were also introduced in
receivers. The underlying principles were
known, but their application to voice band
signals wasn’t obvious considering the pro-
cessors’ large computation load and speed.
A first step was the use of hybrid (partly dig-
ital) techniques in automatic equalization,
based on delta modulation and analog trans-
versal filtering.
The transition to purely digital processing
resulted from the digital multiplex filter
(DMF), an innovative signal-processing archi-
tecture developed at the La Gaude Laboratory
in 1970. Other researchers independently
developed a similar concept called distrib-
uted arithmetic.6The DMF comprised a
short transversal filter with four taps, where
four successive signal samples could be multi-
plied with fixed coefficients and summed
(see Figure 4). This circuit became a common
building block in digital filtering. Several
filter blocks were realized on the same
hardware by time multiplexing on preas-
signed time slots.
The DMF was built in bit-serial arithmetic;
it was the only affordable technology avail-
able, but relatively low in performance.
To accelerate processing, the DMF used a
RAM to store the pre-computed sums of prod-
ucts for the 16 possible combinations of signal
bits present at any instant at the shift register
outputs. With this arrangement, the sum of
the four products was executed within the
time normally necessary for a single multipli-
cation. Performance was further enhanced via
a double accumulator where the propagation
of carried-over bits was avoided—an arrange-
ment known as carry saving accumulator.
A dynamic field-effect transistor (FET)
DMF chip clocked at 2.5 MHz was fabricated
in prototype quantities in 1972. At 416,000
partial products per second, its performance
was remarkable for the time. The chips pro-
totyped in an IBM manufacturing facility
worked satisfactorily in laboratory applica-
tions. Unfortunately, the DMF turned out
to be the only dynamic FET component in
development at IBM and would have had to
bear the whole cost of the manufacturing cer-
tification process; the project was therefore
abandoned.
Early stored-program signal processor
Besides its IBM certification problems, the
DMF was difficult to use in the design of
modems because of its synchronous nature:
each filtering operation had to be assigned a
time slot and its output re-inserted in a dif-
ferent slot. This required external storage
components and control that were more com-
plex than the filter itself. To make things
worse, customer requirements for modems
became increasingly complex and included
many different transmission modes in the
same hardware depending on the type of
transmission line, fall-back situations, and na-
tional regulations. In addition, the need for
fast synchronization on multipoint lines led
to a requirement for operating the modem
in a completely different mode during the
signal acquisition phase.
IBM France La Gaude Laboratory Contributions to Telecommunications: Part 2
Figure 3. Pulse sequence and resulting spectrum from early digital echo modulator.
signal
accumulators
shift registers
RAM
Figure 4. Digital multiplex filter (or distributed
arithmetic)
22
IEEE Annals of the History of Computing
Authorized licensed use limited to: Dalian University of Technology. Downloaded on October 4, 2009 at 04:26 from IEEE Xplore. Restrictions apply.