Remembering Glen

Mike McCammon

I just finished reading Ron Stoughton's mail; I'm really impressed with his memory. I was at the New York end of the conference in 1965 or 1966 when the communications interface decided to only receive. At that time the interface was a box the size of a refrigerator (the icebox), and communicated over 2000 baud synchronous telephone lines. One of Glen's earlier inventions was the method of encoding graphics as 4bit vector codes. The equipment came from BBN, who in those days supplied equipment for the "teleputer." The "voice input system," with Glen announcing his key pushes and me relaying them over the telephone to Curt Lemon, worked fine, as Ron described so well. It was a good thing Glen didn't type any faster.

When I started as a Math Graduate student in 1964, I expected to get a Masters and a teaching credential and become a high school teacher. Then I took an applied math course from Glen. At the start, the on-line system was running in Canoga Park, connected by the "icebox" as described above. When the machine crashed at night, Glen would call the guard at Ramo Wooldridge and get him to go down the hall and push the buttons to start the machine running again. By the end of the course, the RW 400 had arrived from Canoga Park, along with Ray Bjorkman and Dennis, and I got a summer job working on the vector math operations for that machine. By the end of the summer, Glen had convinced me I could get a PhD in math (he was wrong) and gave me a job as a research assistant on his grants. If this story sounds a lot like many others, its because Glen affected so many of our lives in such fundamental ways.

But many people have talked about the years when Glen was on campus, setting up UCSB. Let me say a little about what went on after he left.

The first story really begins before he started Culler-Harrison. In 1968 I was married and "almost" through with my PhD (not in math though, happily Glen had moved to EE, and I went with him). He talked to a college in Washington, and I got an invitation to interview for a job on the Information Sciences faculty at Washington State University. I got the job, and moved to Pullman, Washington that fall. Then I found that Glen had been talking to them about coming there himself, as head of the department. He came to visit that fall, gave a series of lectures and saw the town. It looked like the deal was set. The faculty in the department was excited by the prospect. Then the winter came. It fell to Ð50 degrees one day over Christmas, and stayed so cold and snowbound that you couldn't get in or out of town for a week at the end of January. Shortly thereafter, Glen came to his senses (actually, I think it was more concern by his doctor about his health), and stayed in Santa Barbara. In April, Culler= Harrison was started.

I finally finished my PhD under Glen in 1970, and stuck around at WSU until 1972, then came back to work for Glen. I spent the next twenty years working for him, as the company name changed, and new companies came and went. For most of that time, it was a small organization, rarely less than ten or more than twenty people.

By the time I returned, Glen had developed into a system architect, no longer limited by the machines that other people designed. The first "Culler Architecture," called Serial Number 0 as far as I can remember, had been completed by then to provide the computing power Glen needed for his speech analysis system. The machine was very fast for the early 70s, ran at 8 MHz, and used a unique "multiple operation per instruction" coding to let the smart programmer take maximum advantage of the precious computing resources. Shortly thereafter, a second generation machine, SN/1, which had an attached processor, the original AP 120 array processor, to greatly expand its computing powers. This machine, and a second, almost identical one (SN/2) were the vehicles for a number of innovative computing applications, all related to signal processing, and often speech, but never leading to a voice recognition system.

Glen had gone back to ARPA funding to support the company around 1972, so the justification for SN/1 was an "acoustic research" facility at Moffet Field. Initially the machine was used mostly as an FFT box, for looking for specific frequencies in signals gathered from underwater sensors. But Glen looked at the real processing they wanted to do, which was to be able to track the source of the sound. This required correlating signals from many listening posts, using two dimensional transforms and other processing. The only machine this algorithm ran on then was the Illiac IV, a massively parallel machine which used a single instruction stream to control 64 to 256 processing units. He figured out how to organize the data handling, using disk for intermediate storage, to be able to perform the same task on our machine with one fourth the performance of the Illiac IV.

Culler-Harrison got back into networking about this time, with a very distant host connection to the UCSB imp. Dale Taylor developed the code, which ran on SN/0; this machine was in turn connected to SN/2 with a parallel host to host connection. This system was used primarily to experiment with sending speech over the ARPANET. The speech had to be encoded down to low bit rates, around 2K to 4K bits/second, in order to have any hope of traversing the ARPANET as a continuous stream. The speech compression was based on work that John Markel and A.H. Grey, both from UCSB Engineering, had done on using Linear Predictive Coding. Glen was never satisfied with the methods they published, so we spent many hours trying variations on the algorithms, particularly for determining the pitch parameter for the speech. The work was very successful, and the first network communication using these algorithms was between Culler-Harrison and Lincoln Labs in Massachusetts in November of 1994. Later extensions of this work led to Network Voice Conference tests involving many participants, with copies of each packet sent to all the listeners and a chairman program to decide who could transmit.

This speech work evolved into the design of processors designed for cheap implementation of the coding, which required about a million multiplies and adds per second. Glen developed a 12 bit machine design the LPC array processor, to do just that. It was built on four 7 inch square boards, and fit in a very fancy metal case, with nicely streamlined curved covers. It even did the computing using 12 bit arithmetic. Later, this machine led to the CHI 5, a 16 bit array processor which could be programmed for FFTs as well as LPC speech algorithms. This design was reduced to a single ASIC though a collaboration with Ed Greenwood at Motorola around 1980.

I've left out the Floating Point Systems story. In the days when Glen was still in the Engineering Department on campus, Tektronix developed the first storage display unit with a large (10") screen. The man who brought this unit to Glen to try was Norm Winningstad, then working at Tek. Years later, Winningstad had started a company called Floating Point Systems, which built add in floating point processor cards for DEC and other mini computers of the time. He needed a fast array processor for a customer who built seismic data processing units for the oil exploration industry. He came to Glen, and he agreed to license the AP 120 design to FPS, in a few months, FPS introduced the AP 120B array processor, which became a very successful commercial product in the days when mini computers and even mainframes were very poor at numerical computations. Unfortunately, all that Glen ever got from this arrangement was enough money to keep the company going during the constant search for interesting work.

There are many more things that could be told, including the whole saga of Culler Scientific Systems, which grew quickly with the aid of much venture capital funds to build a mini Super computer, then died even faster when sales didn't match the venture capitalist's expectations. That will have to wait for now, since this saga is already late.

Back to Glen Culler Honorary Lecture Home Page

Back to College of Engineering