Oral-History:Gottfried Ungerboeck

Revision as of 19:32, 10 October 2008 by EMW (talk | contribs)

About Gottfried Ungerboeck

Technological development has been a global phenomenon. Engineering pioneer Gottfried Underboeck was born in Austria, 15 March 1940, about fifty kilometers south of Vienna 15 March 1940, in a country, like most of Europe would be, devastated by Hitler’s Nazi regime. Underboeck recalls that his parents co-owned a flour mill, but his mother’s untimely death forced his father to abandon it and search for new work. After the war, he remembers, they had to find a safe place to hide from the Russians, “it was not a good time.”

Underboeck studied electrical engineering at the Technical University of Vienna, with an emphasis on communication, where he completed his diploma in 1964. Upon graduation he worked briefly for Siemens in Austria, and then joined IBM, performing tasks that would fall into the current assistant engineer position. While working at IBM, Underboeck continually tried to work toward completion of his PhD. However, he left the university without a doctorate, because he married and needed to save money. Once again, bad luck struck. Underboeck ended up in position that required him to discuss system configurations with customers, a job he was not particularly interested in doing. It was time for a change, so he began to interview for positions with different German countries. When IBM caught wind of this they offered him a new role as a research lab programmer. He stayed, but soon realized that he needed his PhD.

Before he stepped foot into his doctorate program, Underboeck had his dissertation project already completed. He turned his dissertation in his first year, but was forced to wait until 1970 before he officially received his doctorate. Underboeck avoided academia because he was more concerned with application and building rather that paper writing and thinking. The practical component of technology drove him. He was less occupied with theories, and more so with how they would be applied.

IBM, Underboeck believed, should force themselves to the forefront of the modem. He tried to push them in that direction. While at IBM, Underboeck began to work on an industry-changing voiceband modem. Afraid of senior researchers and executives killing his vision before he could fully develop it, he kept is top secret for four years. The technological industry and IBM was a battleground, where some stole ideas from others, and companies stole from each other. In 1985 IBM made him a Fellow and he spent a lot of time fighting to get IBM in the communication chip business. Underboeck worked for IBM until 1998. He thought the company should invest in communication chips, they disagreed.

Underboeck is now live in Switzerland, but works for California-based Broadcom, a company where application is much more central than at IBM. Underboeck is very critical of IBM as a technological company, and believes major structural changes are necessary. He also thinks the marrying of coding and modulation was inevitable, and the practicality should drive technology, not vice versa. That said, he sees today as a struggle for coded modulation technology because it is so complicated.

About the Interview

GOTTFRIED UNGERBOECK: An Interview Conducted by Frederik Nebeker, IEEE History Center, 6 July 2004

Interview # 445 for the IEEE History Center, The Institute of Electrical and Electronics Engineering, Inc., and Rutgers, The State University of New Jersey

Copyright Statement

This manuscript is being made available for research purposes only. All literary rights in the manuscript, including the right to publish, are reserved to the IEEE History Center. No part of the manuscript may be quoted for publication without the written permission of the Director of IEEE History Center.

Request for permission to quote for publication should be addressed to the IEEE History Center Oral History Program, Rutgers - the State University, 39 Union Street, New Brunswick, NJ 08901-8538 USA. It should include identification of the specific passages to be quoted, anticipated use of the passages, and identification of the user.

It is recommended that this oral history be cited as follows:
Gottfried Ungerboeck, an oral history conducted in 2004 by Frederik Nebeker, IEEE History Center, Rutgers University, New Brunswick, NJ, USA.


Interview: Gottfried Ungerboeck
Nebeker: Frederik Nebeker
Date: 6 July 2004
Place: The Ungerboeck home in Switzerland

Nebeker: I see that you born on the 15th of March in 1940 in Vienna—the Ides of March. Did you grow up in Vienna?
Ungerboeck: I grew up near Vienna, fifty kilometers south of Vienna.
Nebeker: Would you tell me a little about the family you came from?
Ungerboeck: My family was in the flour mill business and were co-owners of a flour mill of maybe by today’s standards, a very modest size, in a village called Lichtenberg. I grew up there during the war. After the war it was a little more complicated.
Nebeker: You don’t remember anything of the war, do you?
Ungerboeck: Oh, yes. When the war ended, I was five years old, and we had to find a safe place to escape the Russians. With five years, I remember when we came back, and I remember the Russian occupation, and it was not such a good time.
Nebeker: Was it difficult in those years after the war?
Ungerboeck: Yes, but not for us.
Nebeker: Did your father still have the flour business?
Ungerboeck: Well, that had to be restarted, and was not guaranteed this would work again, but eventually it did work again. Then I grew up there. Unfortunately my mother died very early, and she was actually the co-owner of this mill—we had only a small percentage, and after her death, my father had to do something else. So, at this time, we then left and I went to Vienna. That was then already to see me to the end of my studies at the Technical University of Vienna, where I studied electrical engineering with emphasis on communications.
Nebeker: Why were you interested in communications?
Ungerboeck: I was interested not initially when I started to study, the idea was Electrotechnik, electrotechnology, and I picked that because I was fascinated by the fact that there were things going on which you cannot see and I wanted the understand them and thought this would be very interesting. It was strangely such that when I started to study, I did not really have the notion of being actually an engineer.
Nebeker: It was the technology itself?
Ungerboeck: I wanted to study something, and I wanted to study something that was substantial.
Nebeker: You got your engineering diploma in 1964, and then went to work for the Wiener Schwachstromwerke. What was that?
Ungerboeck: Well, this was a branch of Siemens. It was called Wiener Schwachstromwerke because Siemens was a German company and after the war, things had to separated out. At the time when I joined there, it was already becoming again Siemens. But I was there only for a very brief period of time, just six months I was there. Then they got me into the military because I had postponed my military service.
Nebeker: That was compulsory?
Ungerboeck: That was compulsory. So after having completed my study, I still owed them the service, and they got me.
Nebeker: Did they make use of your education?
Ungerboeck: No, not really, not really. It was a bit frustrating because they didn’t really have the means at this time for a real military, and so you had to there, you had to do your service, but it was not at this time, a serious army in my opinion. I was together with a number of colleagues who had studied various things, not all in electro technology. The second half of my service we were collected the high school for technology of the military…
Nebeker: As instructors?
Ungerboeck: And there was a lot of complaining among us that if we are there, then we would like to do something useful, and we didn’t think that we did something very useful, which eventually ended up in the letter to the minister. In any case, I don’t think that was well spent time, but it was necessary for me to do it, and in the end, they made me even an Oberleutnant of the reserve.
Nebeker: Did you have to then serve in the reserves afterwards?
Ungerboeck: After that, I thought maybe I would join IBM.
Nebeker: Was it a one year period you were in the military?
Ungerboeck: I was for nine months in the military. So with some regret, I said goodbye to Siemens and I joined IBM in Austria, actually as Assistant Engineer, a semi-technical person, I would say from today’s perspective.
Nebeker: What systems were you dealing with?
Ungerboeck: The systems engineer with IBM was the one who cooperated with the salesman and with the customers, putting together computer systems, hardware and software configurations. At this time, I had left the university without a Ph.D. degree. I was at this time in need to make money.
Nebeker: Were you married already?
Ungerboeck: At this time we married and I had to earn money. So I went to IBM, and that went quite well. I stayed there only, however, one and half years, because I was thinking, why did I study electrotechnology and now I’m ending up here talking with customers about computer configurations. I am not really contributing to the technology of this.
Nebeker: Was there training in computer systems in your university education?
Ungerboeck: It was not very good, actually. It was not yet very developed at this time. By today’s standards, the computer systems were fairly rudimentary. IBM had essentially the best, some others existed, and it was rather primitive compared to what we have today.
In any case, so I ended up in this semi-sales job, and thought why did I study electrotechnology. So I said, all right, I like to do work for IBM, but I am not going on with this for my life. I want really to taste of a real electrical engineer, working as a real electrical engineer. I said to IBM it is good and fine here, but I have made up my mind that I will go somewhere else, and I went already on my own interviewing with a number of companies in Germany.
Nebeker: You were interested in some kind of research engineer position?
Ungerboeck: At this time I did not really have this research drive; I just wanted to work in real engineering. I mean this Ph.D., academic work, research, that was not really in tradition anywhere in my family. But I thought that I would do something like this, maybe becoming a little better electrical engineer working on the more sophisticated things perhaps. So I interviewed with companies in Germany and I also then said openly to IBM, look, this is not the job for my life. IBM said, “Well, wait a minute. We in IBM, have lots of possibilities for you. They sent me on a tour to England and Arsjubul [correct word?], Switzerland to the research lab here to consider the possibility to become an electrical engineer in IBM. So then I opted for the research lab, thinking something about maybe one year experience approach of working there. I was actually hired as a programmer here.
Then, once being there, I saw what actually research is, and realized as a programmer in the research lab, I don’t really have the best possible future. I need to start some research of my own, I need to get a Ph.D. Unfortunately, I had already these inclinations in Austria also during the time of my military service where I was trying to find something by myself. No professor, nothing, I find my subject by myself. I went to the Technical University Library in Vienna, and I actually picked up some material which looked interesting to me. There was also a reason that IBM in Vienna to look a bit into such things, because one of my jobs became to solve mathematical operations research type of problems for some customers. I gave also programming courses.
Nebeker: So you learned operations research, the simplex method and that sort of stuff?
Ungerboeck: Yes. Some of this I knew already and some of this I picked up during my military service. So at IBM, I had this background, some methods from operations research and working now in communications—what is a good subject? While I was working on the speech processing project, I more or less conversed into a state of transmission, and the problem of detecting binary or discreet signals transmitted over a noisy and distorting channels attracted my attention. I also saw that some of these optimization techniques, about which I had learned by accident by myself, could be applied there.
Nebeker: These optimization techniques were not already applied in the area?
Ungerboeck: Not to my knowledge, applied in the communication theory area.
Nebeker: But you had encountered them in your operations research studies?
Ungerboeck: I encountered such optimization techniques in my readings of operations research and control theory. I had also certain inclination to mathematics, like differential equations and all this stuff.
Nebeker: You liked mathematics.
Ungerboeck: The dynamic evolution that goes on in dynamical systems, I had a certain love for that. So all of a sudden I had my subject for Ph.D. dissertation, which I really, I kept it in my thrall and worked on it at home, etc. During the day I programmed and I worked on the speech processing in the lab. All for certain, I had the whole thing. So now, how do I get the Ph.D.?
Nebeker: You started with the dissertation in a sense, and then went to the Ph.D. program.
Ungerboeck: I had my dissertation actually finished, at least in my head and in written notes the substance was there. Then IBM was instrumental to get me into contact the right professor. And then it turned out that they didn’t trust my education, a university degree of Vienna, so I had to do two semesters with study works which I thought was ridiculous because I was better trained in Austria than the famous ETH. I knew more about mathematics and certain things than I thought that these guys know.
Nebeker: So IBM allowed you to continue work?
Ungerboeck: I shared my time; it was not totally here. IBM allowed quite an amount of freedom, for which I’m very grateful. So after having done my work to be admitted as a Ph.D. and a student, while working at IBM—IBM paid me—and I did also something in speech processing during all this time. I then did my Ph.D. dissertation and presented to the professor.
Nebeker: Who was your professor by the way?
Ungerboeck: This was a Professor Bauman. He was not actually in the area, but somehow I got linked to him. He was a very kind person. He was more of an applied physicist. However, he patiently listened and somehow got convinced that I have something. So very quickly within a few months, I said this is my dissertation. The ETH was screaming, “How can you have a dissertation? He has to be here with years with us!” But I handed in my dissertation—here it is, okay.
Now, what was it actually now? It was the Viterbi algorithm, in some weird form. Actually, nowadays you’d perhaps say it was a variant of the algorithm in this area. Nowadays it’s called the a posteriori [?] probability algorithm which plays a big role in the entropy decoding, which is now the standard decoding. So it was a version of that.
Then of course, I also wanted to publish, write my first paper in the IEEE Transactions of Communications and I submitted this paper. And what did I get? I got a letter from David Forney. And it says, “You almost invented the Viterbi algorithm, except that you didn’t note that here is paper, etc. by Viterbi, etc.” He was at this time, also working this area, doing the proper Viterbi algorithm. So I found this very encouraging. This was my first own contact that I had established into the scientific community and the first recognition, and paper was easily accepted, and that was my first paper.
Nebeker: ETH didn’t object to granting the Ph.D. so rapidly?
Ungerboeck: Well, they let me wait until the fourth year since my matriculation as a Ph.D. student as a student has elapsed. So I had my Ph.D. in 1970. In 1967 I came from Austria and started worked then. By the end of 1968 I had my own sort of little, under the table research project, etc., then I did my study works, etc. Then somewhere in early 1970, I had actually my Ph.D. degree from ETH.
Nebeker: I am curious about that speech processing work you did for IBM. Why were they interested in speech processing?
Ungerboeck: At this time, IBM had been in the second generation of a voice synthesis product. The first one was something analogue and the second one was then digital. This research project was intended to provide further guidance to the development of further technology in this area.
Nebeker: Did you continue to work on that project?
Ungerboeck: Well, I continue to work on this. I was hired as a programmer and I wrote Panelist:/1 simulation programs for that, and I believe that they soon realized that I can work fairly independently. Initially they considered me as a broker, and then I was a research staff member, and then stubbornly I said I want actually to work on the issue of transmission theory.
Nebeker: Were there other people doing that at IBM in Zurich?
Ungerboeck: Not really. Not at the level at which I wanted to pursue. Later then, I did digital transmission and digital magnetic recording. We’re branches pursuit of the research lab, I would say, based on my initial work in this area.
Nebeker: So you said you wanted to work in that area. And they said, okay?
Ungerboeck: At IBM it was to convince people that I work on this. At some time, then I left the speech processing project and I reported directly to the Director as a one-man project, Mr. Carsten Trougheit [sp?]. I am very grateful that he gave me this possibility. He said okay, let him work in this digital transmission area, let’s see what is coming out, and whether his work finds some recognition somewhere, which then for us is a sign that this makes sense. Then some of this recognition then came, and so the area then was established.
The application in IBM was also pretty clear. This was the early time of voiceband modems, and IBM developed voiceband modems in its La Guade development facility at the Cote d’Azur—nice area. In fact at some time I then spent what they call an assignment, six months, working there, living in Ondeep [?] right at the beach. It was a very busy time. I saw that my work could contribute to what they are up to. On the other hand, the beach, etc., was very busy. I had to get up very early, have my first swim in the ocean, and then drive through this damn traffic, somehow try to get to the lab, worked the full day. In the evening there were lots of attractions in the area.
Nebeker: Were your children small?
Ungerboeck: They were very small. It was a good time.
Nebeker: What year was that?
Ungerboeck: That was 1973. I remember well the first stock crash. I thought I had gained some money through the employee purchase plan, and then everything collapsed. So then I came back.
Nebeker: What was the reason for that period at Cote d’Azur? Were there no people there you wanted to work with?
Ungerboeck: There were people to work with, and since I had said I wanted to work on digital transmission, it was clear to say, okay, who in IBM is developing products in digital transmission, and that at this time meant modems in La Guade.
Nebeker: That was where the modem work was being done?
Ungerboeck: That was where the modem work was done. That was very motivating. Already before that I had worked on adaptive equalization. Lockheed and Vedro [?] had been working on this area, which of course interested me very much because that was the key to higher speed transmission over telephone channels. So I engaged in an analysis of the dynamic behavior of the convergence of such adaptive equalizers. I was the first one who had it right, who did the right analysis.
Nebeker: Although you weren’t trained as a theoretician in particular, you just took to that.
Ungerboeck: I think I had very good training in mathematics, physics, and every aspect of electrotechnology from Vienna. As I said already, my impression was that I knew much more about linear algebra and partial differential equations, etc. than the people without the ETH. So my foundations in mathematics were, I believe, pretty good, thanks to this education. They didn’t have much money at the university but, the training by certain professors was excellent.
Nebeker: You yourself had this inclination to look at the theoretical side of these issues.
Ungerboeck: What else can I do? Here I sit in the research lab. I don’t really want to go on in speech processing forever. I need to create something of my own, I cannot build products. So what can I do? I have to exercise my mind, do something mathematic to explain things and make some progress. And that’s how I became a researcher. But I also am a research engineer. I built everything.
Nebeker: You’ve always been concerned with the application.
Ungerboeck: Yes. And while other people wrote paper, paper, paper, I built, built, built and occasionally I wrote the paper. So after a while, I think I understood where the future is in voiceband modems. At this time we’re at 4,800 bits.
Nebeker: So your approach has been deal with the practical technology, and when it’s useful, do a theoretical analysis rather developing theory for its own sake. Is that right?
Ungerboeck: On one hand, I worked along on problems that are relevant in digital transmission, like equalization, like coding. So at this time it was clear that one should do something with coding to gain coding gains, as we say.
Nebeker: You are back in Zurich at this point?
Ungerboeck: Yes. At first I wasn’t an expert. I was not really familiar with all the people working in coding theory…
Nebeker: Had you read Shannon, for example?
Ungerboeck: Sure I had. But I wasn’t an insider, and so my thoughts were not polluted by what these guys did. They caught themselves up algebraic coding. I mean they had wonderful things here, but they got a little bit narrow. So I said well what is not at the capacity? You are familiar with the notion of capacity in Shannon’s sense of these higher lever channels when used with higher levered moderation schemes. Because the telephone channels are channels which had rather good signal to noise ratio—can never be good enough, but it was good. And so it mandated how the transmission of several information bits per modulation interval per consolation point. You’re familiar with this 16 qualm?
Nebeker: No.
Ungerboeck: Let’s just keep it simple. Amplitude modulation. If you want to transmit one bit, then you just need plus/minus. If you want to do two bits, then you need four levers; eight levers for three bits, etc. Most of transmission theory, you go to complex plane, but let’s forget about this. Just one dimensional. And so now you can ask a question. Here is a channel, it has a certain channel capacity which can be computed from the signal to noise ratio, and speaking, the bandwidth of the channel.
Then, let’s say now I’m using 16 levers. I confine myself to 16 levers. How much information can I transmit, if the signal to noise ratio is very high, I can transmit four bits with 16 choices. If the signal to noise ratio is not so high, then you can transmit fewer than four bits. The question is now, let’s say the number is three bits. How do you encode three bits into a 16 lever signals, sequence thereof, each time three bits.
Nebeker: So you have some redundancy there?
Ungerboeck: Yes. How do you do that? That was the question I asked myself. I started to do capacity calculations in these relations. At this time, this information these information theory guys had not done this before in the complete sense as I then did, and I said, “Good. Now I understand what the capacity is, what I can do with 16 lever amplitude modulation at certain signal to noise ratios.” And I tried to play around. It was clear there is a redundancy, and how to explore this redundancy, and what is the criterion actually. The criterion, I understood well, has to be that if you considered the sequence of, we call it simbers, 16 ??? simbers, a sequence there of, in higher dimensional space, you can consider this a point. So in some higher dimensional space, you have points, and not every point should be used.
Nebeker: So those are the permitted sequences?
Ungerboeck: So what characterizes the permitted sequences? They should have largest possible Euclidean distance from each other.
Nebeker: They should be well separated.
Ungerboeck: Well separated in the Euclidean sense. Not in the Hamming sense, which was the criterion for all these coding guys. It should be in the Euclidean sense, and how do you do that? So I said, well, I guess I need some state machine. I allow a certain number of states, let’s say four. Initially, I did at three, but three is more than two! Certainly more than one. So I played around with some very simple examples, and I saw how the construction, at least what the goals should be and how one could achieve this, and later found the connection between convolutional coding, binary convolutional coding, and mapping the signals in the right way into this higher order of bits. That, then, was trellis coded modulation. But at the same time, I had also done something new in equalization, new at this time, which was called fractional tact [?]. Equalizers which had some bit of proper ???, the so-called T space equalizers, and I had worked in synchronization, and so eventually I understood quite a bit of those things.
Nebeker: Were you working more or less independently?
Ungerboeck: Independently at this time. I was reporting to the Director. A one man project. But then he gave me a person who helped me somewhat. Then I said well, I am not doing this here only for the fun. I have my Ph.D. dissertation. My primary goal is not writing papers, but I want to do something useful that works. I believe the voiceband modems can be improved. So what do you need in order to implement modems, build them? You need a digital signal processor. It was not that clear at this time, so I set out and beat my efficiency for processors.
Nebeker: You mean the hardware, the circuits itself?
Ungerboeck: Well, the architecture first. The computer is primarily governed by its architecture. You have a construction memory, data memory, and you have a CPU which executes instructions and interprets them and does something with the data. So I spent a lot of time coming with my own digital signal processor, which didn’t really lead to much publication, however, a single conference paper on that. After a while, then I got a technician. We actually built the thing, we had the first modem running…
Nebeker: This is because you want to demonstrate that this new…
Ungerboeck: I wanted to build modems. I wanted IBM to be in the modem business and to be successful and have something better than others. And so I built these things and learned how to write my assemblers and all that. Eventually, the first modem ran and had the new concept in equalization and new concept in rapid training, and eventually also then trellis coded modulation was first implemented in it. At this time, then, however, I had already more people and that became a real project.
Nebeker: So IBM had decided that they wanted to produce modems…
Ungerboeck: Well, they did modems La Guade, and I wanted to show the development guys how to do it, how to do something better. So after a while, they realized that my digital signal processor has merit and they developed VLSI, digital signal processor. They screwed it up a little bit. There was a lot of NHI (which was not invented here). They had to screw it up somehow. In Swiss German we say Verschlimmbesserung, which means improve it to the worse. I mean, they absolutely had to get their own things into it, and insisted that my 32 instruction format, they can allow only for 27 bits, and stuff like this—pure nonsense. But eventually it worked. IBM then built its voiceband modem products based on the algorithms which were developed here and also the signal processors.
Nebeker: May I ask, why didn’t it seem enough to you to develop this TCM and these other things, equalization and so on, and just make it known to the people who were building modems?
Ungerboeck: I kept it secret for at least four years. I gave IBM the chance.
Nebeker: You didn’t think they would--
Ungerboeck: No, I wanted IBM to be successful.
Nebeker: No, what I mean is, within IBM. Why wasn’t it enough for you to tell the people working on the actual modems, “Here is a better way of doing it. You do it,” rather than you yourself building the signal processor?
Ungerboeck: Because you have to show them a prototype that works better than what they know. So I came with my machine, and then I carried it into their base lab, and they did everything possible to it to prove that it is not good.
Nebeker: And they wouldn’t have listened to you if you just told them you had these good ideas for a new type of modulation?
Ungerboeck: Some listen, and some did not listen. There was also the habit that a manager does have to be technically very knowledgeable. In fact, in the development lab in La Guade it was something like an unspoken rule that the manager should not compete with his people on technical work; he should administer and see that he stays in power. There is a lot of this. This also in the big other companies.
Nebeker: So that meant managers withdrew from the research?
Ungerboeck: The managers were not those who were very knowledgeable technically.
Nebeker: So it was sort of a selection process that someone would…?
Ungerboeck: Look, there were people who were just interested in having power, and they played the power struggle; and then there were some people who worked technically. However, it somehow worked, I remember very well with them, because I think actually they got from me what should have enabled them to actually corner the market. They wasted time.
Nebeker: You mean that they wasted time in exploiting this?
Ungerboeck: Yes, I mean I was almost like and enemy to them.
Nebeker: To hear some new ideas coming from…
Ungerboeck: A terrible guy, yes. But on the other hand, they had somehow to stay in the market, and so they had also to take things.
Nebeker: And you said that you kept this secret for those four years…
Ungerboeck: From the outside.
Nebeker: From the outside, in order to give IBM a chance.
Ungerboeck: I played this somehow safe, a little bit safe. I gave a conference paper in 1976 where I described the beginnings in the first few simple good codes on trellis coded modulation. I was from there on—these people were skeptical. Some, now later say they saw immediately the importance of it.
Nebeker: What conference was this?
Ungerboeck: It was the Information Theory Symposium in 1976. In retrospect, they saw the importance. In reality, I saw that they were screwing around with it and I observed. They said I had the channel paper written, and because I was building modems, etc., I was just sitting on this paper. It went through one phase of review. I then did not continue on it, kept it and was waiting, what is going on to put it out there as soon as there was the danger that somebody else would do the same thing.
Nebeker: Now, as a result, so you presented the paper…
Ungerboeck: So I had priorities, some are established, but it was not very willing to make this very public
Nebeker: Right. But of the people at the conference, did any of them pick up on this and write papers in those next years?
Ungerboeck: Surprisingly, no. Although, now they all say they saw the importance of it. Then eventually, in 1982 I published the whole subject. I believe, there was not much to add. And that, however, then lead to an explosion of things. They screwed it up in every possible way, and added also, somewhat. My work was on building signal process, algebraic coding, writing signal processing code, and trying to get IBM in the forefront of this business, which however, was not the case.
Nebeker: But they did, you say, build some modems using your idea?
Ungerboeck: They had, and these modems worked quite well, but then they did not succeed in properly marketing it because the salesman were non-technical and they were not interested in modems; they were interested in selling 360 systems and 390 systems and not a little item. So the sales organization of IBM was not fit to serve this market.
Nebeker: Yes. When was it that IBM actually produced some modems with your coding?
Ungerboeck: That was maybe the first 14.4 modem, lease-line modem, that had all the things in it. That was in ’82.
Nebeker: About the same time you published the whole theory.
Ungerboeck: Yes. Around then.
Nebeker: So might you say that as soon as you published the full account, there was a great deal of interest?
Ungerboeck: As soon as I published, there was a great deal of interest, and every possible variation has been tried.
Nebeker: Yes, I saw a very nice chart somewhere here with all the different implementations. We are very interested in that story, how after the publication it was picked up by different people.
Ungerboeck: Well, what happened then, and this has been heavily discussed at the ITU CCITT at this time. In the modem area, you could not just build a modem based on some good ideas you have, but you have to make it to stand. That was then especially so when the area lease-line modems started to decay and the switched-line modems came along. In the switched network you need to have really standard products talking to each other. And during all this time, beginning in ‘82 or so, there was all this discussion of coding, which then first lead to lease-line modems with initially 14.4 kilobit and straight forwardly extended to 19.2 bit. Then came the first important switch-line modem which was the V-32 modem. Initially 9.6 kilobit with a simple 8 step trellis code.
Nebeker: Were you involved yourself with any of this standardization CCITT?
Ungerboeck: Yes, I was involved there, I went to many of those. But I never was a full timer there. Other companies like Motorola had a team of four or five people just working for the standard. I went there when I had time in between my building signal processors, so I was a little bit handicapped here.
Everyone wanted to somehow have a piece of this invention and they did things like rotationally invariant TCM, multi-dimensional TCM. Most of this was not really very important, I would say. The multi-dimensional TCM, four dimensional, in fact in my paper in 1982 I said of course there exists the possibility of multi-dimensional telescope measure, but I don’t think very highly about it because of such and such. And I didn’t initially present four dimensional telescopes. But really I had the concept of them, that was clear. I had an argument against it. So what was left over to the others who wanted a piece of the action is multi-dimensional, so that when there is a 4 there is 8; 6 there is 32, etc. Only a particular 16-state 4-dimensional codes really make sense. That is all that is left.
Nebeker: The others haven’t proved useful.
Ungerboeck: Right. The 8-dimensional, 6-dimensional, all this, has gone away and today actually, the most important telescope 16-state 4-dimensional code. That is being used in ADSL, VDSL versions which have more recently been derived from ADSL. That is really today where probably you find the most used—voiceband modems have almost disappeared today. They exist only for those people in their hotel room have no other means to dial up. And fax machines, of course, still using voiceband modems, yes. However, the users of trellis coded modulation are now in ADSL, primarily. A very simple trellis code is being used in the US standard for ??? TV transmission. A very simple trellis code, 4 state, 8 amplitude levers is being used.
I did a presentation once about trellis coding and other things, so I said trellis coded modulation is widely used, not only in voiceband modems—and I forget about lease-line modem—in V32, V17, V34, V90, V92. These are the voiceband modems. Then you have it used in digital subscriber links, primarily nowadays this means ADSL, but there is also SHDSL standard which uses another family of trellis codes of very high complexity. Potentially code is programmable and they allow up to 220 states, the people programmed. But in fact they have implemented up to 512 states. Cable modems, use it for downstream and upstream submission. Ethernet, one gigabyte uses a particular four dimensional code out of my code tables. Wireless LAN, of course, ??? ??? 11 A & G, they use trellis codes. We are actually 20 years later. A lot of things have happened. Tool book [?] coding, the channel area is now concatenate codes with etherdiftic [?] coding. That will eventually replace trellis coded modulation. The standard so far, has always been in almost all the links you have nowadays, your trellis coded modulation. On both of this, you have re??? coding. Okay. Now, we are the time where this would be replaced by etherdiftic [?] coding.
Nebeker: In this period in the ‘80s and ‘90s when TCM was getting applied in many areas, were you involved in actually implementing it in different systems, like the satellite communications?
Ungerboeck: Well, I went on with IBM until 1998. In 1985 they made me an IBM Fellow, and I was spending a lot of time fighting to get IBM in the communication chip business. I also worked on other projects. After the modem area, came the high-speed transmission cable and transmission, etc. So I did these things, including chip development. I was a manager since about the early ‘80s, and I had also sometimes quite a few people to worry about.
Nebeker: Did you continue to have a research group after you became an IBM Fellow?
Ungerboeck: Yes. I had all these project responsibilities and I was the one who went against the IBM rule to work technically as a manager.
Nebeker: Did that give solace?
Ungerboeck: Yes, it gave lots of promise. In the end, I was actually writing a white paper on why IBM should be in the communication chip business, etc. In talking to IBM management, I saw that IBM wants the bottom line and not risk in this highly technical feat of communication chips.
So in 1998 I said goodbye to IBM, and for a short period of time I thought that I would work as a consultant. In fact, I was already in a team of two other people in Germany. However, it did not last very long and I got a job offers mostly from the US. So my freedom lasted only for about six months.
Nebeker: So you intended to work as a consultant, purely.
Ungerboeck: At least for the while, to see how things were going. That did not last very long. Initially Broadcom in Irvine convinced me to join them as a technical director. A technical director actually was something like IBM Fellow. That is a person who gets quite a bit of independence to pursue what he thinks is right or to pursue things which have a certain level of difficulty. So I’m doing this in step time, and I’ve worked through all the areas which are relevant. Initially I worked on cable modems, coding questions, mainly on TSL. Right now I’m working on 10 gigabit Ethernet.
Nebeker: How big is Broadcom?
Ungerboeck: This is 2,700 people, with revenue of well over $2.7 billion by the end of this year, being a fables communication chip manufacturer, where communication chip is rather broad, that includes lots of things for encryption that includes m-backing [?] coders, decoders. Recently that includes also network storage systems, SNA (storage network architecture), network processors. These are very fast processors, somewhat specialized, which are used in switching systems.
Nebeker: How has you work for Broadcom been? Do you feel that your ideas have actually gotten into many products?
Ungerboeck: Yes, I think I have some influence there. People are listening to me. One of the big differences is that in Broadcom the technically good people are always the managers. But with my boss, who is the Chief Technology Officer, I can talk about these things. Never could at IBM. IBM would always have task forces—what should we do? We don’t have task forces in Broadcom; we know what we are doing. Our problem is to be successful and to develop things such that they have an advantage over the products of the others. Not what we do. What we do is obvious.
Nebeker: It is interesting to me that it sounds you have always been very applications-oriented. I can imagine it must very satisfying to come up with good ideas and just get more and more good ideas, but you’ve always wanted to see those ideas in products. How has it actually worked with Broadcom?
Ungerboeck: When they approached me as well as others, there were actually in the end three irresistible offers from the West Coast, and I had a hard time to make up my mind. But I said, “Look, I live in Switzerland and I don’t want to move to the United States. I have my family here, my children, everything, my sailboat and my house. I don’t think that I want to disrupt all this for a move.” They said, okay. Good, all right, so you are telecommuter. So I traveled to the United States, let’s say on the average every two months.
Nebeker: Otherwise, you have a computer connection?
Ungerboeck: Well, of course, that is nowadays easy. How do you communicate? Email with attachments. And what do we attach? We attach presentations, PowerPoint presentations or reports and we discuss them by phone. Once you know the people, you don’t have to see them. You have to have a document hich you can discuss by phone.
Nebeker: You use a telephone a lot?
Ungerboeck: Yes, I use the phone, well, a lot. Let’s say maybe one telephone conference per week. That is enough. Meanwhile, you have to prepare something. Every telephone conference is based on a power-point presentation.
Nebeker: I see. So that is the principle way of communicating ideas is you prepare a presentation and then talk about it?
Ungerboeck: They send me a presentation and we talk about it, yes. Then, now already six months, I am in work with this 10 gigabit thing, and that requires these meetings every two months somewhere; and it requires each time a presentation or two by Broadcom. And we hope we are a pretty significant driving force.
Nebeker: It sounds like you’re happy with the choice you made there.
Ungerboeck: I think this was the right thing. In fact I am a little bit proud that I was 58 at this time. I said, “IBM, I don’t think that you’ll make it.” I saw that they would not make it. Because the way management worked and how sales worked, I didn’t think that they would be successful in the communication chip business. In fact, I was right because they have disappeared. IBM is not in the service business, and has selected technology areas. So I am proud that I had the guts to say no. It was nice to be an IBM Fellow, around a lot of money, etc., but I thought that they were actually wasting what I did.
Don’t write all that. I am grateful to IBM. At the time where it was important for me to become a researcher, they gave me the freedom and it was in particular Carson Target [?] who supported me, lifted me.
Nebeker: I wanted to be sure we covered these big questions that the Marconi Foundation asked us. What were the most significant inventions or breakthroughs? Have we named the things?
Ungerboeck: I would say the most significant ones are I believe I was the first one to understand convergence of adaptive equalizers, which however is not what all people would concede. The second one, I came up with a version of maximum likelihood sequence decoding, which in a way is the compliment to the version of David Forney. Mine is working on the outputs of a matched filter, and in David Forney’s formulation, it is called the widened matched filter. There are advantages to use this or that. The version which I came up with was an advantageous to use in the GSM phones. The GSM phones have a particular way of sending packets with a so-called mid-amber [?]. The training sequence is in the middle, and with this training sequence, [inaudible] my version of the Viterbi algorithm, is more advantageous. All the GSM phones are using that, or initially at least. Nowadays they are written into dual [?] coding and so on. Then was trellis coded modulation, and then I would say architectural digital signal processors, for which I got little credit outside of IBM, but it was years. I see now, in Broadcom how our digital signal processors work—well, they work pretty much the way I conceived this back in the early 1980s.
Nebeker: You talked about the one in particular, the digital signal processor you developed to implement your TCM.
Ungerboeck: Well, actually I developed two generations of digital signal processors; IBM developed VLSI versions derived from them. One screwed up more than the other, and it went actually into products.
Nebeker: So those were influential?
Ungerboeck: Yes, right, within IBM. Another thing that I may perhaps be proud of is that although that I did do with the team, more is the manager, I came up with what is called a magnetic recorder with PRML. This is, of course, just a blank communication theory to the magnetic recording channel, and this was something where I cannot say that I was the only one. I was the one who said, you stop this, and you stop this, and we do this. This is PRML.
Nebeker: And that was developed at IBM in Zurich?
Ungerboeck: That was developed at IBM in Zurich. It took years until it then became in IBM the mainstream technology, magnetic recording, and gained note invented here. As soon as IBM didn’t have it, it was taken over by the entire industry. And like modems, IBM is out of magnetic recording, and Ma Bell, Agear [?]. they are now having the entire market of the chips. Seagate and others are making the actual products.
Nebeker: So we have another instance of IBM having the good ideas very early and not being successful in the marketplace.
Ungerboeck: Absolutely.
Nebeker: Is it this management structure that you see as the problem?
Ungerboeck: IBM was never, able in my opinion, to sell technology products on the basis of chips, where there is a disk drive or there is a modem. IBM came out of this big computer systems selling them as a bundle initially together with the software and providing all the service. Now IBM has been more of a service company. That is where actually the growth is, and a few selected technology areas have remained.
Nebeker: Yes, I see. So their whole business wasn’t the right one for taking a new technology and bringing it successfully to market?
Ungerboeck: At least not these kinds where I had my experience and my capabilities to contribute. It was too little—a modem cost $2,000 initially, or maybe $5,000 even. Peanuts—nothing. Totally unimportant. I was, I believe to be first one to have developed, based on my signal processors, a V32 modem, fully functional. The others, for example, Kodak at this time, Motorola, they did not have at this time a V32 modem based on the single signal processor, one single processor executing the whole algorithm. What did IBM do? The product should have been developed in La Guade. And what did they do? They ordered me to go to Rockwell, tell them everything about it, and then let Rockwell build it for IBM. So Rockwell got the bootstrapped in this whole thing. The IBM guys responsible for the contracts, however, were not clever enough to make this into a business. In other words, then Rockwell was able to increase the price and IBM made no money. And I know how it went—it was free, a gift from IBM. So gradually some pressure internally in me built up and I said then, okay, I need to leave.
Nebeker: Have we covered the first question?
Ungerboeck: First question, so the most significant things I would say it started with the analysis of the convergence of adaptive equalizers, then was maximum likelihood sequence decoding by a variation on what Forney has done. Both are important and both are used in a lot of applications. Trellis coded modulation, sequence processors, magnetic recording.
Nebeker: I think to some extent we have answered the second question of how these fit into the overall development of communications, but perhaps you have comments on that. It is a huge question.
Ungerboeck: I think this was just in the mainstream of digital transmission and technology and storage technology. I was fortunate to be able to participate in that, contribute my share. I have learned a lot from others here. I have a little action, a little piece of this whole action by myself.
By the way, in trellis coded modulation, you should not make the mistake, this is not modems. It starts with modems, because at this time, it was possible with digital signal processors, which were not terribly fast, to just at these low rates, execute more sophisticated algorithm. Now the voiceband modem is almost over, is essentially over, and all this has now just been going on at higher speeds, and now it is at 10 gigabit.
Nebeker: Yes. But also, all of these areas of communications and recording that many of them have taken up such ideas, so…
Ungerboeck: Yes, it’s in all of them.
Nebeker: Have there been any of your inventions that have not gotten the role that you think they deserve? You have explained how some of these were slower because of the particular IBM structure or whatever, but are there things that you think did not get the attention deserved?
Ungerboeck: Not really. I believe here, because my paper writing activity was only an add-on, not the main thing. In my career I didn’t write as many papers as other people, and I believe I didn’t get enough credit for the first correct analysis of linear adaptive equalizers, and somehow my work on digital signal processors was never really published in the full journal paper, just the single conference paper who wrote on that.
Nebeker: I can imagine part of the explanation is that university professors necessarily are writing a lot more papers, whereas an IBM researcher…
Ungerboeck: Don’t worry about the noise which is causing your signal processor on the back plane to not work, and to spend hours, days, weeks on such topics. I should also say, together with Joachim Hagenau we built, the two of us, we built the first satellite modem, with trellis coded modulation. That was in ’83 I believe. At this time, I took, kind of sabbatical to work in Munich, Germany in what was called, DFVLR, now TLR. It is like the 10 nausa [?], well not quite. The Dutch research organization for DFVLR, that is how it was called initially, the consult institute—nobody could pronounce it! Deutsch Forschungscenter fuer Luft und Raumfahrt [?]. So with Hagenau, we did the first satellite modem with trellis coded modulation under a contract for ??? That was a nice experience because I escaped from the research lab. I said I go on sabbatical there and this is the project.
Nebeker: I read in one of the articles about you, that because of the very efficient TCM coding that the satellite dishes could be much smaller than…
Ungerboeck: Well, yes, but that is the more popular. So let’s be a little more technical. What does it mean? You have a coding game of 3db, okay. The sequence trellis codes begin at 3db; below that, we don’t worry. Okay, so what does it mean? 3db means in power a factor of two. So now how can you explore this? You can use less transmit power, you can cover more distance, you can use an antenna with less gain. That means in the case of a dish antenna, one that is smaller. 3db means actually means make the area of the antenna-- if you want to use the coding gain for the sole purpose of getting away with a small antenna, all other parameters are kept the same, then is only half the service, and people like to say this is not a 30 meter antenna. And now if the radius becomes the square root thereof, it is divided by the square root of two, that is also many million dollars.
Nebeker: It’s a way of explaining it for the layman.
Ungerboeck: Well, okay. But in fact, these coding things are used in various ways. For example, if the antenna exists, the big Intels [?] of 30 meter dishes exist, then what do you do? Well, you transmit faster, or you get away with less power in the satellites. What the engineer is doing, they are going ??? ??? the limit. Somewhere the speed is getting up to a value where the system still works well, but you don’t have an overload of margin, because if you allow too much margin, this is actually waste and then your competitor comes and eats up the margin for some competitive advantage. So therefore, you always getting some how to do it.
Nebeker: That’s right. We’re actually getting into this third big question of how the work has contributed to the general well being.
Ungerboeck: I don’t know. I mean, this is a concept that people have, that the technology contributes to the well being. Well, it is actually true. Communications helps, water purification helps, electricity helps.
Nebeker: Well, I think there are many researchers it is not clear that research breakthroughs have made a difference in everyday life, but your work has gotten into so many…
Ungerboeck: Generally, one can say communication, unlike other technologies, is more on the helpful side to mankind. Also, some harms. For example, all this garbage that is floating around in the Internet.
Nebeker: That’s right, and broadcast television and so on. I think what is special about this area is that, when a person just looks at all these application areas where TCM has been important, it’s perfectly clear to people why that is valuable. Whereas, some types of, you know, microwave technique or something like that, you might have to point to a medical application or say, it requires more explanation why the technical work is important to the average person.
Ungerboeck: Most clear where the value of technology, where technology is most useful and contributes to the well being is, in my opinion, not TCM or communications, but the technology for the house with heating, with light, maybe the application of technology to producing good food. Automobiles are also fairly practical, and the airplanes. And trellis coded modulation is a small improvement in communications.
Nebeker: If you look, though, at how economies have grown in the last few decades, so much of the growth has been in these information and entertainment areas rather than the production of material goods, and it is in those areas that these coding breakthroughs are very important.
Ungerboeck: I would be cautious with using too many superlatives. This is little additional technique. I would not even call it a break-through because if I would not have done this, somebody else would have done this. It was somehow the time was right for going into this direction. Marrying modulation and coding, it was inevitable.
Nebeker: Well maybe inevitable, but it wasn’t what the dominant view at the time was.
Ungerboeck: Just for this brief incident of time, the few years where people tried to use coding for bandwidth constraint channels where you squeeze several bits per hertz through these channels, there were a few years where they did not have the right answer immediately. But it was inevitable, that this would eventually be approached in the correct way, a solution would be found. Then, by the way, in the area of trellis coded modulation is almost over because the new standards will most likely now use concatenate codes with etheridiftic [?] coding. Now, they of course, also exaggerate, because this was also some problems, and they exaggerate the merits of it. Today it is actually a struggle. In the standards, what do we do? Should we continue with the established way of how to reach ??? in the trellis coded modulation? Should we now jump ahead and do one of the new schemes? If you carefully look at it, then you see that the advantage is not that great. It’s for sure more complex. It is exactly a question which we will be wrestling with next week. What do we use? [Inaudible] where so called low density [inaudible].
But on the other hand, one can say at this time I was there at the right time, where a new thinking in modulation coding had started. It would have started anyway. I was just lucky to be there and be not spoiled by the academics in algebraic coding theory.
Nebeker: What about this last question, Did the inventions get used in ways you did not expect?
Ungerboeck: No. No, because as I said, I am an engineer primarily and a little bit of a researcher, and I did things for a certain purpose. So the only thing that I might say surprised me was the speed at which the solid-state technology developed and made possible more complex systems than I actually thought would be feasible. So back, for example, again with trellis coded modulation, I initially thought, okay, maybe gaining 3db with four states, that is relatively simple, and 3db is a sorted factor of two. Some people say the dish can be made smaller until you save millions of dollars. I did the codes up to 512 states, but this was just because my program was able to bring those codes out. But I thought, well okay, we’ll save states, that’s it. Similar signal processors, when I did my first signal processors, I had in mind to run them at 10 mega instructions per section—an instructional rate of 10 million instructions per second. And some of the things actually made only eight. And now, 250 or 500 mega instructions per second, and now what can you do with this? And then also, one word [?], it goes in and out per time, per instruction of the data memory. Now they built vector processors. They grab a vector of 16 bit data words [?] at the same time. And I thought, okay and the one multiplier, 16 x 16 at that rate, that’s already a little wonder. And to make thousands.
Nebeker: Yes. Well, that probably surprised everyone, the way this solid-state technology has advanced.
Ungerboeck: So the scale surprised me, but not what in principle is being done.
Nebeker: Not the achievement itself or the application areas.
Ungerboeck: Right. Let me say, in Kodak, there is one thing which I unfortunately had the time to do, and I wasn’t able actually until recently to follow up with the literature. That is in the DNA coding. There is a coding. There is a redundancy much the same way as in trellis coded modulation. There are symbols. The symbols are the amino acids out there; there are 21.
Nebeker: The amino acids, yes. They are encoded by three of the nucleotides?
Ungerboeck: Three of them encode an amino acid. So there are four. We have four times four, times four, okay, so 64. And in the formation of protein they are mapped into these amino acids, but there are only twenty. And so the code-on [?] which of the three nucleotides, they can express more. So several combinations of a code-on are mapped into the same amino acid. So there are actually more letters in the alphabet to express a certain sequence of amino acids. This is like you have 16 levers in trellis coded modulation to express each time ??? use three bits. And now the others, so there is a redundancy here. And there is redundancy in the sequence of the code-ons, and that’s, for example, something that I wanted to have time to look at, and I never made it to spend more than an hour on that.
Nebeker: In your retirement, you can pursue that!
Ungerboeck: Maybe, but then I believe I discover they have already figured out everything.
Nebeker: Well, maybe nobody has looked at that from that point of view.
Ungerboeck: Well, nowadays people look at everything. So when the Ph.D. students are desperate in looking at the good problem-- I would say this is something I regretted, and I’m still an electrical engineer, no time for code-ons.
Nebeker: Are there things we didn’t cover, or things I didn’t ask about that you would like to comment on?
Ungerboeck: No. No negatives about IBM. I am grateful for them, but they didn’t make it and that is why I left. But that doesn’t need to be said.
Nebeker: Thank you very much for the interview.