I join Nicolet
I had a new BS in physics in 1966. I answered an ad on a fine spring day and I became a technician at Nicolet. Things did not go well and I was fired a month later. I got my job back by claiming I knew something about NMR. That was a lie. Things continued to go badly and I got fired again. This time I got my job back by claiming, truthfully, that I knew something about minicomputers and the PDP-8 in particular.
How Nicolet got into the computer business
I told Nicolet that computers were the future. I convinced Nicolet to buy a PDP-8 and I designed an interface between it and the Nicolet 1070. It was an awkward marriage because the 1070 had 18 bit words and the PDP-8 had 12 bit words. Moreover, computation was done in the PDP-8 floating point format which assigned 24 bits to the mantissa and 12 to the exponent. I wrote simple programs to do integration, FIR filtering and the like. These were not "killer applications."
The Nicolet 1070/PDP-8
There was a killer application for the 1070/PDP-8 combination. It was Fourier analysis via the FFT. The first person to recognize this was Jack Krauss. Jack was a former TMC employee. He showed up in my office one day wanting to know if I could implement an FFT on the PDP-8. He had a paper by a French mathematician named Alain Connes which explained how to do it. Connes is one of the people who immediately understood the significance of the FFT. He also had a gift for explaining complex things, in this case, the Cooley-Tukey algorithm. Connes is a pretty interesting guy and you can learn about him here. His paper must have been clear because I got an FFT running on the 1070/PDP-8 in 2 weeks.
The FFT, in its most natural form, operates on complex data and it produces a complex result. However, many applications produce only real data, the imaginary data being set to zero. The result of an FFT in that case is an array with complex conjugate symmetry. That means that half of the array can be easily reconstructed from the other half. This fact can be exploited to reduce memory requirements and computation times by half. Jack Krauss explained this to me and asked if I could write the code to do it. The Connes paper was silent on the matter. This was a much harder task and it took four weeks to do it.
It took the PDP-8 almost 10 minutes to do a 1k real Fourier transform and almost an hour to do a 4k. The poor PDP-8 had to evaluate a Taylor series approximation whenever a sine or a cosine was needed. There was no room in memory for a lookup table.
Now the Nicolet 1070/PDP-8 had a killer application.
The connection between TMC and Nicolet.
Bob Schumann met Jack Krauss at a trade show. It was a chance encounter. Schumann
recruited Krauss to be Shumann's boss. Because of the recent self-destruction of TMC,
Krauss was available. Eventually Krauss brought other ex TMC employees with him. These
are Richard Cushing and Peter Langner.
I spent a month in Europe with Peter Langner in 1970. We even attended the Leipzig Trade Fair in East Germany. There were lots of Russians there, but no Americans. I saw a Russian computer up close. The cold war was definitely on at the time. Leipzig was Peter's hometown. He was a little cagey about how he escaped to the West.