Necessity is the Mother of …

by Dave Schinke

A first IVR customer was successful, and the Conversant Model 80 IVR product was being marketed to different customers.  A time came when there were several “big” customers, and the number of manufactured machines needed to increase to about 30 to satisfy these sales. Soon, the Western Electric factory engineers were finding that when a cabinet was constructed, it was taking 3 or 4 days to “qualify” a cabinet:  test it, clear any discovered faults, and make it ready to be shipped.

“We are getting killed on test time.  Is there anything Bell Labs could do to help us with the problem?”   This phrase and question began to circulate in the Dept. and Bob Perdue, my current supervisor, shared the query with a number of us. I’m sure the “hardware” development people were engaged, and what could the “software” folks possibly do to affect the problem? 

Let me pause the narrative and describe in more detail some of the physical features of the Model 80 equipment cabinet. 

• The circuit boards were multi-layer “printed” circuit boards, maybe 10” wide by 14” long. They generally had factory level electrical tests that could assure that the individual board was good when manufactured.  Also, the firmware on the board usually had internal self-diagnostics that would operate on each power-up and determine the board was operational, or set a software error condition for replacement.
• The circuit boards slid into vertical slots on a shelf with guide rails, front retaining clips, and a rear connector would engage with a double row of plated steel pins that were embedded in a “backplane” board.  Each pin went through the board and also protruded out the rear of the “backplane.”  Imagine looking at rows of shiny “porcupine” quills sticking out from front and rear.
• The interconnection wiring on the rear of the backplane was done with “wire-wrapping” tools that stripped an end of wire, wrapped it on a rear pin, then manually pulled the wire to the desired pin, and stripped, wrapped and cut the other end.  This was a manual, human operation and subject to error.  Further, a “bent” pin on the front side could fail to engage the inserted circuit board, producing another fault.

The Model 80 software could interact with circuit boards, run internal diagnostics, etc, and fairly quickly locate circuit boards that had internal faults.  But this “backplane” needed visual inspection, which is a tedious process dealing with the “porcupine” mass of pins. And the backplane carried the audible voice signal for 48 channels.

There is an old saying “Necessity is the Mother of Invention”.  We all kind of scratched around with the problem, but there was no immediately obvious solution that would reduce the testing time.

“Hey, Roy”, I said as I plunked myself down in a chair in Roy Grubbe’s office.

“I want a machine that will test itself” referring to the Model 80 machine. 

“We can make tones, and we can hear tones.  Is there a way to test the voice paths of the machine using its own abilities?”  was my basic query to a fellow staffer.

Roy wasn’t immediately dismissive of this, as many of the things I proposed had previously been.  Roy thought if we could put a “loop back” cable on the cabinet rear phone connector, we could use a tone generator of one shelf to send tones into the detection circuits of the other shelf. We started examining the wiring diagrams to figure out what voice channels could be covered, if any would be missed, etc.  Each Model 80 had two active shelves of equipment to service 48 phone lines.  Each active shelf had two standard 48 pin telephone connectors. (note: two wires made up a standard phone line.)   By some quirk of the circuit pack arrangement that I have forgotten, we needed two sets of cables to join 2 top connectors to 2 bottom connectors.  One pair of cables joined the connectors in a “parallel” arrangement, and the other joined in a “crossed” arrangement.  We did come out with a tentative wiring diagram for the cables that would provide full coverage of the voice paths inside the machine.

Roy started thinking about the test software that would run tests, locate problems and print out errors to point to the backplane locations with faults.  Roy was a remarkable developer.  He worked as quickly as he thought.  It all just flowed into executable code. I explained to our Western Electric engineer what we were trying to do, but that we didn’t have the resources to make the cables. He immediately said, “Just get me the diagrams, and I will have the wiring shop make them quick like a bunny.”  And he did get them made quickly.

This idea of a self-testing machine had become a physical reality. Roy’s software worked and was debugged in rapid fashion.  I started working on the factory floor for several days with the Western Electric test people to get the tests working in practice, get their feedback, and to be the go-between to Roy for software needs, fixes, etc.   I don’t remember how quickly the test procedure stabilized, but it wasn’t very long. 

The self-testing software reduced the time to qualify a machine from several days to less than one day, and significantly boosted the Model 80 production rate.

Author’s footnote:  This story was constructed without the use of Artificial Intelligence, just my fingers on a keyboard.  All errors and omissions are mine and mine alone.