Oral-History:David G. Messerschmitt

About David G. Messerschmitt

David G. Messerschmitt is a pioneer in the field of communications. His contributions include research on VLSI architecture for signal-processing problems, in particular its modeling and simulation in software, and the development of advanced software techniques, including Blosim and Ptolemy. Dr. Messerschmitt is currently the Roger A. Strauch Professor of Electrical Engineering and Computer Sciences at the University of California at Berkeley, and he has written or co-authored several pioneering textbooks, most recently Software Ecosystem: Understanding an Indispensable Technology and Industry with Clemens Szyperski (2003). He holds twelve patents, has published more than eighty journal articles, and at least 110 conference papers. Dr. Messerschmitt is a Fellow of the IEEE and a member of the National Academy of Engineering. His papers and other contributions have won many IEEE awards, including the IEEE Communications Society's award for the best paper in 1981 and 1987, an award from the IEEE Acoustics, Speech and Signal Processing Society in 1988, and the 1999 IEEE Alexander Graham Bell Medal in 1999.

This interview details Dr. Messerschmitt’s prolific career, beginning with his education and his choice of electrical engineering as a profession. He discusses his work at Bell Labs, his early interest in computers and work on digital transmission, and communications theory and its applications. He also describes his decision to move to the University of California at Berkeley, his work with VLSI technologies, and the development of Blosim and Ptolemy. The interview concludes with a discussion of his most recent projects, his work with the National Science Foundation’s Blue Ribbon Panel on Cyberinfrastructure, and an assessment of his own career.

About the Interview

DAVID G. MESSERSCHMITT: An Interview Conducted by John Vardalas, IEEE History Center, 14 February 2003

Interview #425 for the IEEE History Center, The Institute of Electrical and Electronics Engineers, 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:
David G. Messerschmitt, an oral history conducted in 2003 by John Vardalas, IEEE History Center, Rutgers University, New Brunswick, NJ, USA.

Interview

Interview: David G. Messerschmitt
Interviewer: John Vardalas
Date: 14 February 2003
Place: Messerschmitt's office at the University of California at Berkeley

Childhood, family, and educational background

Vardalas:

Let me start this with a little preamble. The received wisdom has it that the years of our early youth play a particularly significant role in setting the fault lines along which we develop in our adult lives. So before going into the details of your professional life and its achievements let me start the interview with your life from boyhood up to the end of university. At our age I feel probably it is reasonable to push the boundaries of youth to include graduate studies. Having said that, when and where were you born?

Messerschmitt:

I was born in Denver, Colorado, right at the end of World War II. I'm actually not a Baby Boomer, but I come very close. I was born in 1945 and grew up first in East Denver and later in the suburbs west of Denver. I had a math teacher in high school that had taught at the university level previously and whose husband was a university professor. She influenced me very much and inspired me to think in terms of going beyond simply being a professional to become more of a scholar and academic.

Vardalas:

You've jumped the gun on me. I do have questions specific to that issue and I'm glad you raised it, but first I would like to know for the record, did you have brothers and sisters?

Messerschmitt:

No, I'm an only child.

Vardalas:

What were your parents' occupations? Were they scientists or engineers or professionals?

Messerschmitt:

My parents were both college-educated, which was rather unusual for their generation. They were teachers. My mother stayed home with me, but before I was born and after I was about fifteen years old she taught junior high school. My father had done a number of professions, among them teaching. When I knew him he was working for the Colorado State Department of Employment, going around to various high schools giving aptitude tests and advising students on what kind of career opportunities they had open to them.

Vardalas:

Did he try that on you?

Messerschmitt:

No, he felt that was a conflict of interest, so he specifically did not do that for me, but he had one of his colleagues do it.

Vardalas:

And what was the outcome of that?

Messerschmitt:

It was kind of vague like, "Well, I think you can do anything you want to do, so whatever you want to do is fine."

Vardalas:

That's glowing praise. What was your boyhood like? In particular, did you exhibit an early talent and interest in the sciences or math?

Messerschmitt:

I was actually a rather poor student in elementary school. Part of the reason for that was because I had poor eyesight and it wasn't until later that I got glasses to cure that problem. But I found school neither very motivating nor very interesting until about 8th or 9th grade. And I absolutely hated arithmetic. That was my worst and least favorite subject.

Vardalas:

Really?

Messerschmitt:

I did not enjoy anything that was rote memory or rote learning. I also did not find learning languages very enjoyable. However once I got into junior high school in 8th and 9th grade when the mathematics moved into higher mathematics and we got into more in-depth scientific subjects and so forth, then I really found it interesting and did very well. Thereafter I got good grades.

Vardalas:

Who was that high school teacher that motivated and influenced you? Would you elaborate more on that?

Messerschmitt:

Her name was Mrs. Thomas. Her husband was a physics professor at the University of Colorado. She took a personal interest in me and did things like put me in the back of the class and said, "You don't have to listen to class. It's too elementary for you. Here's a book. Study this more advanced material." She took a very personal interest in me and tried to motivate me to think in terms of more scholarly activity as opposed to limiting my horizons to the practice of a profession.

Vardalas:

I see. Had you been studying on your own getting ahead of everyone? How did it become so elementary for you?

Messerschmitt:

No, she just felt that I could move faster than the class and therefore wanted to encourage me to delve out ahead and study more advanced mathematics. At that time calculus was not typically taught in high school though it is today.

Vardalas:

Yes.

Messerschmitt:

We were simply studying trigonometry, and her feeling was that I ought to be able to speed up and get further. She gave me book that were more advanced and set me in the back of the class and said, "You don't have to pay any attention. Just study yourself."

Vardalas:

When thinking back to your days in high school, did you have any career expectations? Did you think of anything you wanted to become?

Messerschmitt:

In junior high school social studies class we went through and exercise where each student was asked to choose a profession and then we went and investigated that profession with a government occupational handbook and some documents and wrote a report. I believe that was in 8th grade. I chose electrical engineer at that time.

Vardalas:

Really?

Messerschmitt:

Mainly because I had been interested in building crystal radio sets and had purchased science kits through the mail. I did science experiments at home, and I really liked the electronics things. So at that point I chose electrical engineering as a profession and did a report on it, and that became the default assumption about what I was going to become. Then when I got into high school and discovered a great interest in mathematics, Mrs. Thomas encouraged me to think about going into mathematics. However what I discovered with a little bit of investigation was that I could actually do both. Electrical engineering was and is quite mathematical, so I saw that I could combine an interest in electronics with an interest in mathematics through studying electrical engineering. Since I had expressed an interest in that, Mrs. Thomas, through her husband, arranged for me to visit an electrical engineering professor at the University of Colorado named Frank Barnes. I went up with my dad one Saturday and met with him. Later on when I went to the University of Colorado as an undergraduate in electrical engineering I sort of latched onto him as a mentor and advisor. I did some undergraduate research with him and have maintained a relationship with him ever since. I have known him for about forty years, and he is still a professor there. He was a fairly new, young professor at the time. He was quite an influence in terms of getting me involved in research while I was an undergraduate.

Vardalas:

When you think back on that report you wrote about the electrical engineering profession, can you recall how realistic your views were? Did it turn out to be what you expected?

Messerschmitt:

Pretty much, although of course I could not have anticipated what happened since, like the electronics revolution, semiconductors and so on.

Vardalas:

In terms of the kind of work electrical engineers do and how they do it, did you have a clear picture of what that would be like?

Messerschmitt:

No, I don't think students in either junior high school or high school who are coming to university really have a clear picture of what engineers do and what the profession is all about. And of course at that time it was changing fairly rapidly. Before World War II engineering was very much of an apprenticeship situation, a kind of hands-on seat-of-the-pants profession. After World War II it became much more a science-based and mathematics-based profession. It probably would not have been possible to know or understand the changes that were happening in the profession at that time. I think that there is a difficulty for the engineering profession in that high school students really don't have a concept of what engineers do, and their teachers generally have not been exposed to the profession either.

Vardalas:

I gather you went into engineering like a lot of engineers, with hands-on experience and an interest in building and working with things as opposed to the purely theoretical. Is that a safe assumption?

Messerschmitt:

I think that is where the interest came from in the first place. One doesn't necessarily want to limit oneself to that, but after playing around with the electronics devices as I did as a teenager, then conclude, "This is really fascinating stuff. Look what you can do with it. You can send signals through the air." This generates the core interest. Then at school one learns that there is a lot more understanding, mathematics and theory behind it. And that's all good, especially if one is interested in mathematics. But something other than theory and mathematics has to stimulate that interest in the first place.

I would say that that is a problem we have in electrical engineering today. There are not nearly the opportunities for students to get hands-on experience with electronics nowadays. For example, it used to be that one could build various electronic things using kits. There was a company called Heathkit that sold kits and young people could build something that actually worked.

Vardalas:

I did that.

Messerschmitt:

That company has long since gone out of business, at least in terms of producing kits. All of the interesting things go on inside little integrated circuits that are encapsulated in ceramic or plastic, so there is no way to really understand or appreciate at that level all the interesting things going on inside there and all of the complexity of what goes in there.

Vardalas:

Does that have ramifications later for how engineers approach their work if they have not been able to get the sense of dealing with concrete objects like this early on, or is it just a question of motivation?

Messerschmitt:

I think it is more a question of motivation. A lot more students these days are showing interest in software. The reason is fairly obvious, because even as a high school student they can get hands-on experience with software. They can do some things with it, have some successes and make things happen, whereas it is becoming harder and harder to do that with hardware.

Vardalas:

That's an interesting point. At the University of Colorado electrical engineering department where you went, was it traditionally power engineering? What was your interest? I would imagine that electrical engineering was either radio or power engineering.

Messerschmitt:

It was in a transition stage at that time. In fact, Frank Barnes was instrumental in that transition. He had come from Stanford University. By the time I became a student at CU he was the department chair. He brought in very excellent people from MIT, Stanford and other schools and revolutionized the department in terms of taking on a much more modern approach based on semiconductors. At that time we were just making the transition from vacuum tubes to transistors and fields like microwave engineering – and even optics and lasers – were just coming to the fore. He was instrumental in modernizing the faculty and department, so it was in the process of going from a teaching institution and department into becoming a strong research department. As I was there that transition was occurring.

Vardalas:

Did you develop a sense of where you wanted to go in electrical engineering at this point or did it take your whole undergraduate career before you zeroed in on something?

Messerschmitt:

It occurred gradually. At that time we took power engineering and had a power lab, but what really stimulated my interest were the more mathematical subjects. In particular I took a couple of graduate courses in communications and probability and random processes. I really enjoyed those courses. A professor who was teaching those courses there and had gotten his Ph.D. at the University of Michigan was Dr. Roberts. He encouraged me in that direction and I ended up going to the University of Michigan for graduate work.

Graduate studies and Bell Labs employment

Vardalas:

Why you chose to go to the University of Michigan was my next question. This is the connection.

Messerschmitt:

Yes. When I applied to graduate schools I applied to MIT, Michigan, Stanford and Berkeley, if I recall correctly. The reason I went to Michigan was because of Dr. Roberts. I had gotten interested in the particular field I wanted to study through his course. He prearranged with his former advisor, Ted Birdsall, that Ted would be my advisor if I wanted. Everything was sort of prearranged so that if I went to Michigan I would already have an advisor in the area in which I was interested in working.

Vardalas:

Once you were there, how did you settle on your Ph.D. thesis topic? Was it something you just fell upon, or was it given to you?

Messerschmitt:

Initially it was for a master's degree with an intention of going on for a Ph.D., because I had a National Science Foundation Fellowship that would have supported me nominally through a Ph.D. However this was during the Vietnam War, and it was just at the time at which student deferments were being eliminated. The government was speaking with two distinct voices: one side was saying, "I will give you a scholarship to go to the university," and the other was saying, "but if you pursue this line you will probably get drafted." Therefore after getting my master's I went to work at Bell Laboratories. I had worked at Bells Laboratories during the summer after my bachelor's, so I knew people there. They were quite willing to hire me after getting my master's, so I went to work at Bell Labs after getting my master's. While I was at Bell Labs I had an occupational deferment, and then later on I was able to go back for a Ph.D. During that period in which I was at Bell Labs I was taking some courses at Brooklyn Poly in New York City with the intention of pursuing a Ph.D. on a part-time basis.

Vardalas:

Can I assume that your thesis work or Ph.D. evolved from or was inspired by things you were doing at Bell Labs?

Messerschmitt:

Yes. What I did then was to apply for a Doctoral Support Program at Bell Labs. That program allowed me to go back to school full time at Bell Labs' expense. They paid me some substantial proportion of my regular salary and provided me with a summer job if I wanted it with a tacit understanding that I would come back to Bell Labs after finishing my Ph.D., although there was no strict obligation to do so. Then I went back to the University of Michigan, because that was the known entity from where I had started in this direction. Ted Birdsall continued as my advisor, but I worked on problems that were motivated from my Bell Labs experience – particularly because I was being financially supported by Bell Labs.

Vardalas:

I am going to come back for more detail on your Bell Labs experience, but first, you mentioned that Professor Barnes was your mentor when you went to the University of Colorado. You met him once, and when you got there as a freshman you kind of latched onto him. Did that develop into a personal relationship, did it develop into something, or was it just something you tried initially? And did he have an influence on you beyond teaching?

Messerschmitt:

He was always somewhat of a role model to me because he was in the process of modernizing the department. He was a very accomplished researcher, and he showed a lot of confidence in me that I would be able to do research essentially at the graduate student level while I was a third and fourth year undergraduate. He gave me opportunities to do the kind of research similar to what I would be doing if I were a graduate student. He gave me a laboratory and equipment and research to do research.

Vardalas:

Do any of the kinds of things you played around with at that level stand out in your mind?

Messerschmitt:

There were a couple of projects. One was work on tunnel diodes, which was a microwave engineering topic. Then later he had some ideas about how one might be able to use metastable states of mercury as a memory by exciting the states of mercury optically into a metastable or long-lasting state and then to probe with a light beam to extract that information from the mercury. I was playing around with that for quite a while. I can't say I had a lot of success. I headed later on in my career, partly as a result of that, in a more theoretical direction. I never considered myself to be particularly strong as an experimentalist.

Vardalas:

Was that your first introduction to digital things, this idea of memory and on/off? Did you have a series of courses on digital techniques when you were at Colorado?

Messerschmitt:

No. There was not a lot of visibility of digital techniques at that time. Everything was pretty much analog. Digital computers were in their infancy. We had an IBM mainframe and programming courses and programming assignments where we used the old card decks and submitted card decks to a centralized computer center. We got back a printout 24 hours later. They also had an old Benix G-15 computer that had in the 1950s been the College of Engineering's only computer. And at this point it was just left in a room for students to play with, and so I was one of the students that played with it.

Vardalas:

Did you develop an interest in computers at that stage?

Messerschmitt:

That is when I got some hands-on experience with a computer, but at that time there was no such thing as computer science or computer science departments. My interest in computers was developing at that time, but I would say the real genesis of my interest in digital things was at Bell Labs where they had developed digital transmission systems. That was before I was at Bell Labs, but when I was there during my summer after the bachelor's degree and when I was there after my master's and before getting a Ph.D. I was in a part of Bell Labs that was working on digital transmission technology. Everything I did was relative to digital transmission.

Vardalas:

I see.

Messerschmitt:

That is closely related to representing music and other kinds of signals digitally. In fact, one of the things I worked on at Bell Labs was the digital representation of music, so-called program channels. That probably makes me one of the first people to actually look at some of the issues involved in the conversion of music to digital form.

Vardalas:

I would like to come back to that later. That sounds interesting. Before we leave your educational career, I wanted to ask you, if this is a fair question, about university undergraduate and graduate work. You are now in academia. You teach students and you see what they go through. How would you compare your educational experiences with those of students today in terms of becoming engineers? Would you say it is pretty much the same or different? Have you formed any opinions about that?

Messerschmitt:

Yes. I would say that the students today are extremely impressive relative to my day. I think they are considerably more mature at a given level, particularly in graduate school. And they are particularly stronger in communication skills. One of the experiences I had at Colorado as an undergraduate was that I participated in the IEEE student paper contest. This was based on the research I was doing with Dr. Barnes, and he encouraged me to participate in the contest, so I did. I did not win, but I went at least as far as the regional contest. I remember that at that point public speaking was a completely foreign and very scary proposition. I was totally nervous and scared about getting up before an audience and presenting something technical. Today the students do this kind of thing regularly, are much more experienced at it, and are much better at it than we were at that time.

In that dimension they are very strong. With respect to the technical – that is, whether or not they are good engineers – obviously the electrical engineering profession has changed somewhat. It is much more digitally based and much more of a programming task nowadays. Programming skills are extremely important. Designing hardware today is more similar to software than different and requires a very similar set of skills. There is still this area of analog design, which is very much like what we used to do exclusively in my day of schooling, but I think that in general the students today are much more focused on a combination of theory, digital and programming as opposed to my day when there was much more of a focus on analog, building and measuring.

Post-doctoral employment at Bell Labs

Exploratory development department

Vardalas:

In my next set of questions I would like to get to Bell Labs, because that's a really important part of your life as well as an important part of the history of technology in general.

Messerschmitt:

Yes.

Vardalas:

As a little preamble here, as you probably are well aware, Bell Labs is one of the earliest industrial R&D labs in the United States. GE is the other. We all know how Bell Labs has evolved a remarkably successful organizational machine for generating technological breakthroughs. Therefore it is of great interest to the historians of technology and business to know what went on in Bell Labs. It is very useful when we get the insights of someone who worked there. Whether these breakthroughs were used commercially to benefit the AT&T empire is another issue, but the technology that came out of that place is legendary. And you worked there for ten years.

Messerschmitt:

Yes.

Vardalas:

What do you recall? You already had some summer experience there when you went there. You didn't go there fresh from study to job but already knew what it was like there.

Messerschmitt:

Right.

Vardalas:

As you experienced it, what was the work environment like at Bell Labs when you worked there formally? What strikes you or sticks out in your mind about the people that were there? Was it an environment that encouraged interdisciplinary work? What can you recall of that whole ethos that was Bell Labs?

Messerschmitt:

I think that for a professional career the environment is extremely important. There is an old saying that when people go to good places they end up being good themselves and that is partly a result of the environment as well as their personal efforts. I think that is very true. Bell Labs was unique in that it was partly funded by what was effectively a tax on telephone bills called the license revenues, so portion of everyone's telephone bills was bled off to support Bell Labs. Therefore there was less of an ongoing need to justify one's effort in terms of in a budgetary sense. And there wasn't a lot of effort or time – at least at my level, which was a member of the technical staff and later on as a supervisor – spent in trying to justify what we were doing in a budgetary sense. More time and effort went into coming up with new ideas and trying to explore those ideas. I happened to be within a development organization – not within Bell Labs Research – in a department that was doing what was called exploratory development. What that means is that we were not developing the next product but were looking beyond the next product and trying to stay in close touch with what was going on in research. A lot of what we were doing could also be classified as research, but there was a culture that we wanted to produce something that was practical and would ultimately result in products. There was a culture of prototyping our ideas, demonstrating them and trying them out. I was in kind of a unique position because that was the first place I really got exposure to building hardware and prototyping things and making things really work. That had a lot of influence on my subsequent career at Berkeley. But it was also an environment where there was a lot of freedom to explore whatever directions I wanted as long as it was somewhat relevant to digital transmission. That was the area of the company in which I was working. And I think that everything went back to the boss.

Vardalas:

Who was the boss?

Messerschmitt:

In my case I was in a department led by Bob Aaron – after I got my Ph.D., not before.

Vardalas:

And to whom did he report? Who was next in the chain of command?

Messerschmitt:

Two up was Warren Danielson and one up I have forgotten his name. It's been too many years.

Vardalas:

Okay.

Research and management

Messerschmitt:

Bob Aaron was a unique personality who had worked on the very first digital communication system developments. Although he did not have a Ph.D., he was very oriented to new ideas and research. He may as well have had a Ph.D. He had a lot of credibility within the company and had worked with some of the top executives of the company early in their careers. As a result he was given a lot of freedom. He maintained a department doing exploratory development and was given a lot of freedom to explore directions he felt were exciting. Since he had a track record of producing things that were useful, there was not a lot of near-term pressure.

Vardalas:

Was his management style the same with his staff? Did he give you a lot of freedom too?

Messerschmitt:

Yes, very much so. If we came to him with some idea, he would critique it and say whether or not he thought it made sense and if he thought it was a potentially a good idea, but he gave us a lot of freedom – particularly if his critique was that the idea was promising. In that case, he would give us a lot of freedom to explore it and even get us resources to do that.

He was a very enlightened manager in terms of giving us a fair amount of freedom to explore on our own. As a result he attracted very good people in the first place. A key part of the experience of being at Bell Labs was first of all what area of company one was in – research or development or systems engineering; secondly, the charter of that department; and thirdly, who was leading this department and how much credibility s/he had in the company. How much credibility the boss has with the company determined how much freedom s/he was given. All of these things meant one could go to Bell Labs and have very different experiences depending on the different places within the company in which one worked. I was fortunate to have had a very good experience.

Vardalas:

The cliché is research and development, but you were more in the development side, and then there was the pure research side. Did you personally experience much interaction or symmetry between the people doing the pure research and your group? How did the flow of ideas work between the two?

Messerschmitt:

We had a group of people that, at least the best of them, were pretty much equivalent to the people in the research area in terms of their abilities to be innovative, scholarly, mathematical and whatever. Therefore there was a lot of mutual respect between the groups. We freely collaborated with them. We would reveal our ideas to them and vice versa. However I would characterize my particular department, which we called exploratory development at that time, to be much more characteristic of what would be called research in an industrial environment today.

Vardalas:

Okay.

Messerschmitt:

The industrial environment has really changed for research. There is much less support from top managers, particularly in competitive companies, for doing really far-out and free-ranging kinds of research. There is much more pressure to produce practical things that are useful to the company that meet the needs of business units and to actually prototype and try out things. That was exactly what we were doing in those days. What we were doing under the guise of exploratory development would today be fairly typical of a research environment in a corporation.

Commercialization and Bell Labs research

Vardalas:

I see. If that was that the mandate, how much flowed through to commercialization in the company? Was the work that was done during the time you were there used properly?

Messerschmitt:

That was a frustration for a young person. As you mentioned earlier, there were some problems with Bell Labs and the Bell System being a monopoly in terms of its ability to take the best technological ideas and apply them to the real needs of telephony. There were some amazing things that were accomplished in that era like picture phone and waveguide. Waveguide was a transmission system intended to support picture phone. On the other hand it was not successful commercially. But I would say that was the exception rather than the rule. One must remember that Bell Labs was really supported by what was in effect taxation on telephone bills and Bell Labs, due to the Consent Decree of 1954, was required to share its patents and technology with other companies. Bell Labs made a tremendous impact. For example it can be argued that the entire semiconductor industry was an outgrowth of Bell Labs' research and development from that era of the late forties through sixties. Companies like Fairchild, Texas Instruments and Sony started out with Bell Labs technology. In my opinion Bell Labs can be gauged as a wonderful success in terms of stimulating technology within the country and the world, but it was not nearly so successful in terms of revolutionizing the telephony industry.

There were a number of wonderful successes, of course. I mentioned digital transmission. Digital switching, electronically stored program and switching and direct distance dialing were tremendous successes as well, but there was also a tremendous amount of technology developed that never saw the light of day through the Bell System, though it did in other venues. This was a bit of a frustration for a young person in Bell Labs because one would develop a hot new idea that one thought was quite meritorious, but it was very difficult to navigate that through the multiple layers of management that was necessary to actually get the technology out into the field.

Vardalas:

There is some resemblance to the RCA Labs too.

Messerschmitt:

Yes. So this is probably one of the reasons I left.

Digital transmission and communication theory

Vardalas:

You've been talking about generalities about this frustration without any specifics to yourself. Let me see if I can bring it back to you now. In the Bell Award citation, IEEE stated that your contributions made digital service possible over the existing telephone. Without getting too technical, tell me the details about what kinds of things you did during this period that had to do with digital service for Bell and what made it through the system? And how much was frustration and how much was joy for you?

Messerschmitt:

It was a lot more joy than frustration. I am the type of person that jumps in and likes to work on a lot of things, so I worked on quite a few things during that period. One of those things was the communication theory relevant to transmitting digitally over band-limited channels. And that would have a couple of applications. One application would be to the voice-band data modems that now are becoming passé, but back then were in the process of moving from 300 to 2400 bits per second and so forth. All of those advances in terms of speeds of voice-band data modems were based on theoretical developments in communication theory and what's called intersymbol interference. Intersymbol interference is where digital pulses are transmitted down a line and they disperse and interfere with one another, and overcoming that in the presence of noise. Then later on there were the coding techniques that were an outgrowth of Claude Shannon's original work at Bell Labs back in the late 1940s.

The other application of that was getting higher and higher speeds on coaxial cable and metallic media in the telephone network.

Vardalas:

If you can bridge this for me, what were the theoretical issues that had to be addressed for these applications and how was this bridge made? Because obviously it was a theoretical thing you tackled.

Messerschmitt:

I think the communication theory is uniquely influential in practice. Looking at the history of communications, before World War II, Marconi and others were developing sort of seat-of-the-pants experimental approaches to come up with communication techniques. However, after World War II nearly all of the significant advances in communications were preceded by theoretical understanding. They were based on theoretical understanding and theoretical advances. This is not true of most fields of technology.

Vardalas:

Right, right.

Messerschmitt:

In most fields of technology the theory comes after the experimentation to some extent

Vardalas:

Right.

Messerschmitt:

Oftentimes a theory is used to explain what was seen before trying to carry it further. However communications is a field where theory has usually preceded invention. This is because the mathematical models have the characteristic of being both accurate and tractable, and so the theoreticians can make faster progress than experimentalists.

Vardalas:

Just for the record, what was the nature of the theoretical issues you were tackling specifically?

Messerschmitt:

We were studying techniques for countering intersymbol interference – that is, better and better equalization techniques and better and better detection techniques where these pulses are interfering with each other in the presence of noise. A lot of the early advances in digital communication were based upon better and better techniques for dealing with that problem, which in turn were based upon statistical and theoretical analyses.

Vardalas:

I see.

Messerschmitt:

Later on a lot of the advances were based on coding theory, following up on Shannon's information theory work. Coding theory is an area in which I have not worked very much, but I would say most of the advances after 1980-85 have been based on coding theory rather than on underlying detection and equalization.

Vardalas:

Were there any mathematicians in your group? Did you interact with mathematicians at all in this area?

Messerschmitt:

There were mathematicians working in this general area in the research area at Bell Labs, but we were engineers who had a strong math background.

Digital representation of signals

Vardalas:

Okay. What other areas? You said there were a lot of things.

Messerschmitt:

Then I got into the representation of signals digitally. The original digital transmission systems had focused just on voice and telephone speech. One of the things I worked on, as I mentioned earlier, was music. Program channels were a service the telephone companies provided at that time to carry from studio to transmitter radio programs and the signaling for television.

Vardalas:

Oh, I didn't know that.

Messerschmitt:

And so there was an interest then in using digital transmission for these kinds of signals as well as for voice. I was one of the first people to work on the representation of music digitally. Now it's a big deal with the MP3 and all these things.

Vardalas:

Was the big difference between voice and music the larger bandwidth? Was there a challenge in trying to go to music as opposed to just dealing with the voice issues?

Messerschmitt:

Yes, true, it has a greater bandwidth, but that is dealt with fairly easily just by using higher sampling rates. The big issue was that for voice the number of bits needed to encode the signal is dependent upon perceptual effects. Basically a deliberate noise called quantization error is being introduced when any signal is being digitized. One wants to do that in such a way that it will not be perceptible by the user. There was a pretty good understanding of how to do that for voice based on a lot of experimentation. However no one had really thought about music. For example, "How many bits are needed to represent each sample of the music?" and 'How should it be compressed?" What techniques to use to make that whole process as efficient as possible, representing signals with as few bits as possible while making the quantization noise imperceptible.

Vardalas:

Did you have to look at it again because people perceive music differently than voice over the telephone?

Messerschmitt:

Yes. One of the first things I did was to go back to the old literature, going back to Edison who had represented music on phonographs and such media. One of thing they had to deal with in the older analog world was nonlinear distortion. Nonlinear distortion has some similarity to quantization error although there are significant differences. I was trying to understand from the old experimentation in the 1930s and 1940s what people had learned about so-called harmonic distortion of music and how that might apply to digitization.

Music is very different from voice because of the masking effect. The signal actually perceptually masks the noise. In the case of music it is a much broader bandwidth and there is a lot less energy at high frequency. The quantization noise that is introduced at high frequencies becomes much more perceptible. There are a lot of tricks that can be done in terms of coloring the quantization noise color – differential encoding and so forth. We were playing around with those kinds of things.

Vardalas:

What was the outcome? Were there any applications that came out of that?

Messerschmitt:

The outcome of that was I think mainly some practical recommendations for how these program channels could be provisioned over a digital transmission channel.

Time Assignment Speech Interpolation (TASI) and Adaptive Differential Pulse Code Modulation (ADPCM)

Messerschmitt:

Another thing I worked on was Time Assignment Speech Interpolation (TASI). Even though we were using digital transmission for short haul at that time, transmission between central offices within urban areas and all of the long-haul transmission was still analog.

The perception was that digital would not be cost effective in long-haul transmission, partly because it consumed more bandwidth and bandwidth was very scarce. Most of the long-haul transmission was on microwave relay towers going across the country. Today we think of digital as being more bandwidth-efficient than analog. That is one of the big reasons that satellite companies go to digital. The reason it is more efficient today is because we have much more sophisticated compression technologies than we had back then. Things can be compressed in digital form whereas one cannot compress things much in analog form. Back then it was standard wisdom that digital was more consuming of bandwidth and less efficient, so we were looking for techniques that could improve long-haul digital. We grabbed onto one technique called Adaptive Differential Pulse Code Modulation (ADPCM), a technique for encoding speech with fewer bits.

Vardalas:

Fewer bits than the normal pulse code modulation?

Messerschmitt:

Yes. Later the ADPCM became a CCITT standard and was used fairly widely around the world.

Vardalas:

Was that phrase, ADPCM, developed by your group or was that something proposed elsewhere?

Messerschmitt:

I think the latter. Back in the fifties some people had proposed basic differential and encoding techniques. It comes in various forms: delta modulation, ADPCM and so forth. We were trying to make it better and get the bit rate down at a given level of perceptual quality as much as possible.

Vardalas:

Was that an easy and routine thing to do or did that require some hard work?

Messerschmitt:

It was challenging, because it required understanding the nature of the impairments that were introduced and how they could be introduced. In the technology of the day it was hard to conceive how to do that cost effectively. The only technology we had was Medium Scale Integration (MSI). It was also difficult to do experimentation, because in contrast to today the computers of the time could not deal with audio signals in real time, so you always had to design and build custom hardware to try out any ideas.

Vardalas:

Right.

Messerschmitt:

Another technique we looked into was the TASI (time-assignment speech interpretation) principal, trying to take advantage of the pauses in conversation and transmit only the active speech and not the pauses. Of course that game can only be played in large collections of voice channels. It is a statistical game. If there are say 120 voice channels, perhaps only thirty or forty of them are actually speaking at any given time.

Vardalas:

Right.

Messerschmitt:

But then you have to worry about what happens if a hundred or all of them happen to speak at the same time. We worked on something called Digital Speech Interpolation. TASI was something that had been used in undersea cables since the 1950s in an analog context in order to reap more voice channels out of an undersea cable, which was a very expensive facility. We were looking at similar techniques in the digital world, combining the best compression technologies of the time – which was ADPCM – together with this speech interpolation idea. The last thing I actually did at Bell Labs was as a supervisor leading a group building of a prototype digital speech interpolation speech compression system. We put in the network between Boston and New York and carried actual live traffic over it.

Vardalas:

How did it perform?

Messerschmitt:

It performed as expected and the speech quality was quite high. We were able to show that while there was additional impairment due to the various techniques that we were using, there was also a higher quality because in digital transmission over long distances the noise that inevitably accumulates in analog transmission systems is largely eliminated. It does not accumulate in a digital system because of the regenerative effect—bits can be detected and regenerated in spite of noise. Therefore those two balance each other, and we showed that over any distance greater than fifty miles or so our subjective quality was actually higher and we could gain about a 4:1 or 5:1 compression; that is, we could transmit four or five times as many channels as with the straightforward techniques.

Vardalas:

Okay.

Bell Labs resistance to implementing digital transmission for long distance

Messerschmitt:

We thereby made digital transmission more economical in the long distance at that time. At that time also people were worried about whether they should move to digital switching. It's actually kind of strange. At that time the toll (long-distance) switches were digital – that is, represented the speech digitally – but the local switches were analog. Conversely, the transmission was digital locally but analog in long distance. There would be a big advantage to deciding on digital everywhere, which is the situation today, but it was difficult to prove in economically with the technology of the time.

Vardalas:

That's interesting.

Messerschmitt:

And this was mainly due to the underlying economics of line cards and so forth. What we were trying to do, because we were in the digital transmission area, was to put in both digital local switching and digital long-haul transmission. This effort I just mentioned was trying to prove that digital long-haul transmission could be economical. We convinced Long Lines, the long distance part of AT&T, that this was a good idea, and Long Lines said they wanted to do it because they felt that the network would be obsolete if they continued to spend money on analog transmission in long-haul. But it was frustrating, because they couldn't convince Bell Labs to develop it.

Vardalas:

They couldn't convince Bell Labs, but you were working in Bell Labs.

Messerschmitt:

Right.

Vardalas:

What was the part of Bell Labs that could not be convinced?

Messerschmitt:

All the levels of management above me. There were probably five or six levels of management above me.

Vardalas:

Oh, I see. You were trying to champion this, but from the top down you couldn't get that support.

Messerschmitt:

The problem was that it was not a clear economic gain for the near term. We could not prove that it would save money in the near term. The argument that Long Lines made (with which I agreed) was that it would save a lot of money in the long term to have a more modern all-digital network.

Vardalas:

I thought Bell Labs always thought in the long term. That's what surprises me when you say that they did not go along because of the near term problem.

Messerschmitt:

They thought in the long term in terms of research and in supporting a lot of activities such as my own which were exploratory in nature. But when it came down to actually devoting resources to developing technology to go out into the real world they were fairly conservative.

Vardalas:

That's interesting. It appears like a contradiction to me. Was this because separate people were making these decisions or were the same people are making both sets of decisions?

Messerschmitt:

Partly because separate people were making the decisions and I think partly because in a monopoly like the Bell System the risk is asymmetrical. When expending resources to do something new there is a chance of failure, whereas if it is not done there is no chance that someone else will do it and therefore no danger of looking bad.

Vardalas:

Of course.

Messerschmitt:

You have to remember too, that was during a period after the failure of picture phone and waveguide, which was a well-publicized failure. A lot of conservatism came out of that which then overlooked putting important new technologies out into the network. I'm sure that was a factor. A similar kind of thing happened in local switching. I was involved in a couple of task forces where we were looking at the economics of replacing analog with digital local switching. There again it was decided not to develop local digital switching because we could not prove a near term advantage cost-wise. Bell Northern Research in Canada at that time decided to do transmission and switching digitally.

Vardalas:

Yes, I know.

Messerschmitt:

And because they had digital local switches, a little bit later they ended up grabbing a substantial portion of the U.S. market for switches.

Vardalas:

Nortel has its roots in that, benefits from that?

Messerschmitt:

Nortel goes way back. It was originally a part of the Bell System.

Vardalas:

Right. As part of Northern Electric.

Messerschmitt:

In terms of their success in the U.S. market, I think a good part of their success during the late 1970s and 1980s was because they had decided to pursue only digital technologies.

Vardalas:

Very interesting. Are there any other things that come to mind? You have covered quite a spectrum of activity. Is there anything else that sticks out in your mind as an important part of that period?

Messerschmitt:

No, that was basically it.

Departure from Bell Labs

Vardalas:

This sounds quite fascinating, all these very stimulating areas of endeavor, and obviously for a person such as you who likes both the theory and the practice it must have been stimulating. Yet you left Bell Labs for academia. Earlier you mentioned frustration. Would you please elaborate on what finally prompted you to leave Bell Labs?

Messerschmitt:

There were three basic reasons. One was a minor amount of frustration. It was a really fun job with a lot of freedom and resources. However in terms of actually getting one's ideas out into the real world, it had proven a little bit frustrating. That was probably not a major reason.

The second reason was a long-term interest in academia as a profession.

I had always been intrigued with the idea of being a professor, so when the opportunity came along to do that I decided to pursue it. I must say that the people in Bell Labs were very supportive of that. They understood that it was very important to have people in academia turning out good students who could come work at Bell Labs, and they were also very supportive of my career personally. Whatever was good for an employee, they tended to put that ahead of the company, and I give them a lot of credit for that. The third reason was just a geographical presence. I had grown up in Colorado and really prefer the weather and culture of the Western U.S.

Comparison of academic and industrial research

Vardalas:

As an aside, your interest in being a professor and in academia, do you attribute that to your high school teacher? She was trying to get you to think that way.

Messerschmitt:

She probably got me started in that direction. The fact that my parents were both teachers probably had some influence. And the fact that when I was in college I looked up to the faculty and thought it must be a pretty neat job. And I would say too that within Bell Labs I had as much freedom I'm sure as I would in any industrial context to pursue my ideas wherever they led. However it was still somewhat constrained relative to being a professor where there is really no constraint. In an industrial concern one is within a particular department with a certain charter and deviating from that charter is looked at askance. In my case, I was in the digital transmission area, so my activities for the most part were supposed to contribute to that domain. There are really no constraints in academia – whatever one wants to get into and whatever one wants to research, as long as you can find support for it. There is no boss in the traditional sense. It is really like being self-employed and one defines one's own agenda.

Vardalas:

Have there been any times in your life in academia until now when you said, "I wish I had some of that industrial environment here"? Have you ever thought that some portion of that kind of environment would have been useful in academia?

Messerschmitt:

The good thing about the industrial environment is that there are a lot of resources for doing something concrete, for example prototyping or developing hardware or something that requires a lot of resources. Things that are bigger and more ambitious are hard to do in academia. As long as one stays on the conceptual side one can do anything and it can be more free ranging, but it comes to doing something in terms of actually developing or building something there are many more possibilities in the industrial world.

Vardalas:

It's an interesting contradiction if you think back to what you said about why you left Bell Labs. The inability to have an impact on the real world was a bit frustrating, and you just said that in the university world it's a similar kind of thing.

Messerschmitt:

One view of that would be that as long as I am not going to have an impact on the real world, I might as well be in an environment where I have more freedom. But I don't actually think that way, because in academia there are ways to make an impact. I think one of the strengths of this department at Berkeley is that we put a lot of premium and weight on impact. That manifests itself in various ways. One tends to make impact more through others. For example through influencing industry, which still can be done in academia. Through consulting and continuing to be involved in industry one can make an impact, but it's a little bit less direct.

Vardalas:

In your CV, you classified the history of your research into eight categories. We have covered the first part and part of the second, which involved Bell Labs. I would like to go through these and have you make some observations about these areas. In doing that, I would like to ask you to consider five questions or so as you reflect on these various areas. Why did you become involved or what was it about in each of those areas? And for each area what would you say were your achievements and failures? You must have had some failure sometime in your life. The third question is, how would you assess the contributions of your work within the great scheme of things for this area? The last two questions are: How does each area represent a natural continuity from a previous area or to a next area? And if an area is an important change, how does it represent a discontinuity in where you went.

Messerschmitt:

Okay. Sure.

VLSI (very large scale integration) architectures research

Vardalas:

From 1977 to '90 you had an interest in "VLSI [very large scale integration] architectures for telecommunications and single processing functions such as ISDN U-transceivers and video signal processing." Will you please make some observations based on those questions I asked?

Messerschmitt:

Yes. As I observed earlier, people are influenced by environment. When I arrived at Berkeley in 1977 it was a pretty unique time. There just happened to be here a very strong group of people who were looking at the new VLSI technologies and trying to apply them to systems problems.

Vardalas:

Oh really?

Messerschmitt:

And I came from Bell Labs where I had a good understanding of one very important system, which was the telephone system and telephony, and they were looking for opportunities to apply VLSI technology in new ways. That is, they were looking for a confluence of new circuit techniques that met unique system needs. This was a unique thing. Berkeley was the only place where this kind of thing was happening. It was not even happening very much in Bell Labs. By all measure it should have been going on in Bell Labs, but like all large companies Bell Labs was organized in a kind of stovepipe fashion. There was one vice presidential area that for electronics technologies, another vice presidential area for systems (in fact several of the latter) and it was almost impossible to get interaction going on across those boundaries. One basically had to go through a vice presidential level, so it was really too compartmentalized. What was being done at Berkeley was a unique opportunity for me. The techniques that were developed between 1977 and 1982 had a major impact on the telephone industry. Some techniques were demonstrated – like switch capacitor filters and charge redistribution codecs – which made it much more economical to do things digitally within the network. As I mentioned, Bell Northern Research had gone into digital switching and Bell Labs had not. One of the reasons that happened was because Bell Northern Research was paying much closer attention to what was going on here at Berkeley.

Vardalas:

Really? I didn't know that connection.

Messerschmitt:

And they pulled those technologies – switch capacitor filters and charge redistribution codecs – out of Berkeley and applied them. These had a dramatic effect in terms of reducing the line costs of the analog-to-digital conversion function, which is a major part of the cost of the local switch. I was not very involved in that, but at the same time we were looking at digital transmission on the subscriber loop; that is, not only making the switch digital but also going digital all the way out to the customer premises.

Vardalas:

Right.

Messerschmitt:

Some of my colleagues here at Berkeley and I were looking at that as a combination of communication theory and VLSI technologies and so on. We also made a major impact in that area in defining some techniques based on echo cancellation that allowed full duplex digital transmission on the subscriber loop. There again, echo cancellation was a known technique before this full duplex transmission. It had been proposed in the theory groups at Bell Labs some years before, but was generally considered to be impractical because of the high accuracies required in the implementation. Techniques were developed here at Berkeley to make the algorithms adapt to the imperfections of the technology to get around that accuracy problem. More importantly, we were oriented toward actually prototyping our ideas. That is something that the electronics people were naturally apt to do but which I was also quite accustomed to doing having come from the part of Bell Labs in which I had been. I think it was our building an actual chip that worked, in all these things I mentioned, that was really very important in terms of convincing people that this was not just an academic exercise but that it was real and could work in the real world. So I think it was partly as a result of that that the echo cancellation techniques were adopted by the International Standards for ISDN U-transceivers. ISDN transceivers are now used in the digital subscriber loop as well.

Vardalas:

During this time was there any active cooperation from Bell itself?

Messerschmitt:

No. Actually our closest association was with Bell Northern Research. They were much more actively interested in these things that were being pursued. There was also a company on the peninsula called TRW Vidar, which was in the telephone equipment manufacturing business. Jack McDonald, who was their vice president, was in close touch with us.

Vardalas:

Was this radio signal processing also related to functions like video signal processing?

Messerschmitt:

Yes. The history has been that the VLSI technologies improved with Moore's law. What is hard at one point in time becomes very easy later on. While it was very challenging to do things at voice band speeds in 1977, it was not many years before that became a piece of cake because the technology had advanced so much. We were always looking for the problems that stressed the technology, that make it hard, and video was the next obvious frontier.

Vardalas:

I see.

Messerschmitt:

Then that got me into VLSI architectures.

Application of VLSI architectures to signal processing

Vardalas:

Let me read the next bullet. It says, "Concurrent VLSI architectures for high-performance signal processing, emphasizing recursive and memory-bandwidth problems." Would you please explain and elaborate on that and its significance?

Messerschmitt:

We were interested in signal processing contexts where the sampling rates or the speeds or the functions that needed to be realized were so high-performance that they were difficult to realize in the technology of the time. Of course the technology advances, but at any given time there are challenging problems.

Vardalas:

What was considered high performance signal processing at that time?

Messerschmitt:

At that point something like video at broadcast resolutions, and later on at high-definition TV resolutions. One of the strengths of the VLSI technologies was that they allow lots and lots of devices. They really emphasize concurrency; that is, they emphasize doing a lot of things at the same time, in parallel.

Vardalas:

Okay. I see.

Messerschmitt:

The question is, "How can we utilize that dimension of the technology doing a lot of things in parallel to speed things up?" There are certain inherent limitations in the algorithms and functions that one is trying to realize in their ability to achieve concurrency. One limitation was recursive algorithms, where the output of one operation is fed as the input into the next operation. Seemingly that reduces one's opportunity to parallelize.

Vardalas:

It looks more sequential.

Messerschmitt:

It becomes more sequential. And so what we showed for example was that in a large class of algorithms by unwinding those loops, by calculating multiple steps at a time, one could actually create more concurrency at the expense of additional processing overall. We could create tradeoffs of more concurrency for additional processing, and that exactly the kind of tradeoff that makes sense in VLSI.

Vardalas:

What do you mean by memory bandwidth problems in this context?

Messerschmitt:

Yes. A lot of the issues in high-performance digital processing have to do with the fact that when things are stored in memory they have to go off chip, and that is inherently a lot slower. This is because of the much higher capacitance that must be driven and so on. What we were looking at was ways to reduce the need to go to memory and to go to memory in a way where the higher latencies of getting information from memory would have less impact on the throughput of what was being realized.

Vardalas:

Everything was kept on the chip?

Messerschmitt:

A lot of this has the characteristics of trading off latency for throughput. With greater latency – that is, greater delay in getting things done – a higher throughput of getting things done could be achieved. That is a basic trick that is played in a lot of processing, and in networks as well.

Vardalas:

Where did this find its greatest application? Was this in all areas of signal processing or were there some specific areas where you saw this being applied or could be applied?

Messerschmitt:

It is a little bit like optical communications. I mentioned how important theory was to communications, and that's true of radio communications and wire pair communications. It is not really true of optical communications, because optical communications is an area where the physicists stay ahead of the communication theorists. By getting smaller dimensions and faster devices and so on, they can improve throughput. There has not been nearly as much impact of communication theory techniques on optical communications as there has been in other areas of communication. This work in that era has a little bit of a similar characteristic. By coming up with these techniques of expanding the reach of the applications we had a difficult time staying ahead of the people who were just improving technology and making it move a lot faster. I'm not aware of a lot of applications for that particular thing, which would be applications where it was feasible using these techniques but not feasible without these techniques. There is sort of a range of performance where applications using straightforward techniques would work. The straightforward techniques were advancing very quickly in that era due to Moore's law.

Vardalas:

You were doing some catch-up with them. Right, right.

Messerschmitt:

I am not sure that that work ever had a lot of practical impact. It probably was used in some places, but was not highly visible.

Vardalas:

Okay. So it wasn't generic enough that you could always stay up with whatever was going on in advancement? You couldn't just generically port it over to something else.

Messerschmitt:

Yes.

Application of advanced software technology to signal processing and to communications systems simulation

Vardalas:

What about the one that says 1984-present, "Application of advanced software technology to signal processing and communications systems simulation and modeling"? Was that continuous from preceding work or did that represent a change in your work?

Messerschmitt:

It's continuous in the sense that in the course of doing all this algorithm work and turning it into hardware we always had the need to do simulations as well as develop theories. I looked at that as an opportunity to explore the use of advanced software technology in the same way that we were exploiting advanced hardware technology. I took some pains to learn about the best software technologies of the time, which in the early period was C programming and UNIX, which was just moving out of Bell Labs – through Berkeley actually – into widespread use. A little bit later on came object-oriented programming technologies. So while we were doing this we saw an opportunity. We had the need to do simulations and we saw the opportunity to do that in an advanced way that would contribute to the use of the software technologies for these kinds of things. The first thing that I did personally – and this was really my way of learning C programming – was to develop a system called Blosim. It was a block diagram simulation system that tried to put some structure around building reusable modules that could be mixed and matched by connecting them up in new ways. We used Blosim, and I think a number of other people used Blosim, for signal processing simulations and I think probably some commercial digital signal processing simulation systems were influenced by that. And in that era I took on a student, Edward Lee, who really took off in that direction on the software side for his Ph.D. thesis. Then we hired him onto the faculty and he's been working in that area ever since. I collaborated with him a lot, particularly in the early years. So after Blosim he developed another system called Gabriel, which was based on Lisp programming.

Vardalas:

Lisp still has uses then.

Messerschmitt:

It did, yes. Gabriel was never very widely used. It was sort of a stepping-stone to the next stage, which was Ptolemy, named after the astronomer, and again Edward and I got together. Object-oriented programming was a brand new thing at that time. Then languages, the compilers were just coming out – C++ and some other options, such as SmallTalk, and we asked, "How can we use this object-oriented technology?" We defined Ptolemy, a simulation system that supported a wide-ranging set of computational models – and was able to freely mix these models, which was the innovative part – and used the capabilities of object-oriented technology very heavily to achieve that. Ptolemy is still alive today as an active research project. It is the basis of a lot of Edward's research. I moved on to other things and Edward has continued to pursue that.

Vardalas:

Is Ptolemy commercially used? Is it something that is used in industrial settings? Where is it used now?

Messerschmitt:

There is a fairly substantial user community, mostly in universities and to some extent in industry. There is a commercial product out of Agilent, which is in part based on an early version Ptolemy. But generally I think it remains way out in front of most of the commercial options.

Vardalas:

I am fascinated by the use of simulation for this area of work. I usually think of simulation modeling as being an economical alternative to the cost of hardware prototyping. Simulations are first done in order to save that step and make it much more efficient when a prototype is finally built. Is this a similar kind of logic to what drives use of this in this area?

Messerschmitt:

Back in the early era, yes, that was the main motivation. Because with the technologies of the time in order to get reasonable performance one had to think about hardware implementation. Of course with intervening decades and Moore's law, today most things in signal processing are actually done in embedded software within real products. It is less common to build hardware. Even in cell phones, all the algorithms that do the various communication signal processing functions of compressing the voice and communicating it over the radio channel and so on, most of that is done in software in the cell phone.

Vardalas:

A simulation of that is done in software.

Messerschmitt:

The actual implementation is a software implementation, rather than a hardware implementation.

Vardalas:

What is the simulation? I'm trying to understand the simulation.

Messerschmitt:

That means that the simulation is not modeling a hardware realization but is an intermediate step toward an actual product. But during the phase in which the algorithms are being designed and tested, that can be done in simulation. One likes to have a lot more tools to help do that, and one doesn't worry nearly as much about efficiency. However at the end of the process when one is satisfied that the algorithms are working properly, then it desirable to be able to just press a button that will create the software that is actually the implementation.

Vardalas:

I understand. Okay.

Messerschmitt:

That's the old area of Electronic Design Automation (EDA). As I mentioned earlier, today hardware design is probably much more like software design and takes similar kinds of skills.

Vardalas:

In your CV you mention that you have worked on application software technology from 1984 to the present. Yet you just said that you and Dr. Lee went on separate paths and that he is still doing this. In what area are you continuing this interest?

Messerschmitt:

I am continuing in the sense of continuing to track the current activities in the software industry and keep up with the latest developments in software and think about how they might be applicable for example within the telecom industry.

Signal processing and fast packet network transport in advanced video services

Vardalas:

From 1990 to '98, which is also of more recent vintage, "Interaction of signal processing and fast packet network transport in the context of advanced video services." What prompted you to get into this? Is this a continuation from everything else?

Messerschmitt:

Yes, it was an outgrowth of the fact that we were doing signal processing with video as one of the main applications – primarily because video was a technology driver. I had also come from another track, which was communications, and then moving up one layer to be interested in networking. That was sort of a natural evolution. Then it became fairly evident to me that there was too great a separation between these communities – that is, the people who did networks were not thinking about the source coding problem and the people that were doing source coding were not thinking about the networking issues. As a result, the solutions they are coming up with are arguably quite sub-optimum. There is too much separation, too much layering. We started thinking about this – and not only us, but a number of other people. It comes in the general category of joint source channel coding. By the way. Shannon had proven back in 1948 that those two problems could be separated, that source coding could be separated from channel coding. However that was under a set of strong assumptions that are not operative in real life, and in the context of a simple communications channel rather than a network.

Vardalas:

Okay.

Messerschmitt:

In real life there is a lot to be gained by thinking about the confluence. As an example, when speech or video is encoded some bits are more important than other bits in the sense that if you change them somewhere in the communication process it has more of a perceptual impact. Perception is all that really matters in terms of how thing are done overall. It makes a lot of sense in the transport part of the problem to think about bits as not all being created equal but some bits are more important than other bits. Therefore if something has to be given up somewhere because the noise or traffic is temporarily very large, it makes sense to prioritize the bits, getting some bits through at the expense of other bits. That's what we mean (as a simple example) in terms of the joint source channel coding. We were trying to look at that a little bit more deeply in this body of work, bringing in the notion of latency. One thing that is very hard for packet networks to do is to get information through with low latency because they tend to trade traffic capacity for greater latency. We were particularly interested in the dimension of some of the information in the source coding say of a video signal is more susceptible to latency than other information. It is important that it get through early, consuming more network resources.

This is particularly true when talking about a service like video teleconferencing where the overall end-to-end delay makes a big difference in the perceptual effect. It's not very important in something like video on demand and playing back a movie. If one sees it say two seconds later, it really doesn't make any difference, but in a two-way conversation two seconds of delay is disastrous.

Vardalas:

Right, right.

Messerschmitt:

Therefore we really focused on thinking about the source-coding problem in a more generalized way that would take into account what was harder for the network to do and what it was easier for the network to do.

Vardalas:

And what were your findings? How would you assess what came out of this?

Messerschmitt:

We came up with a technique. We went over to the optometry department where they were accustomed to doing subjective tests with visual things and set up subjective tests and showed that it worked pretty well. However I cannot say that it has yet been picked up in the sense of people actually building these systems. A number of different approaches that have been pursued, including different coding techniques, and then there's a school of thought that says this whole area is addressing efficiency, but efficiency doesn't really matter because bandwidth is free.

Vardalas:

Is it free?

Messerschmitt:

The last time I looked at the rates that the telephone companies charge, prices are still fairly substantial, but there is a school of thought that says, "Don't bother about being efficient. Worry more about the functionality of what you are trying to achieve. And do new things because the technology is advancing so rapidly both in communications and in processing." It is a little bit like the problem I was mentioning earlier with the recursion. One is always chasing the advancing technology. It may be that problems being addressed currently will go away in two or three years just through advances in technology.

Vardalas:

I see.

Messerschmitt:

This is probably one of those areas. The technology, the capabilities of the access links to the network such as DSL and the capabilities of the backbone network are advancing fast enough that people are satisfied with what they are getting using very simple and easy techniques.

Influence of business on technological research

Vardalas:

I was just going to make a sarcastic remark that Gordon Moore made that from his perspective the telecom industry is still not moving fast enough. They are dragging their feet on making bandwidth grow as fast as Moore's law did with semiconductors.

Messerschmitt:

That gets into some of the more recent things that I've been thinking about in terms of looking at the overlap of business and economic considerations with technology. Where that arose is in part because having been around so many years and worked on so many different things, I feel that my added value today comes mostly from a broad perspective. And having worked on signal processing, networking, communications, software, etcetera, I have a pretty fair understanding of all those technologies, so that is where I feel I can make more of a contribution today. Secondly, the advance in technology. We have seen a geometric advance in technology, and this includes all three areas of information technology – communications, storage and processing. Over a large number of years a geometric advance has a big impact. The question is, "What is that impact?" Well, if I look back over my involvement in these technologies going back to the late sixties, what has changed – and it is a radical change – is that back in the sixties and seventies we were having a really hard time doing things with adequate performance at low enough cost to make it practical in the real world. Therefore we addressed relatively few applications – telephony, television and radio were basically just three or four applications – and all of the effort went into advancing those applications and making them more cost-effective and making them available to a larger number of people. In that era the charter of the Bell System, the whole reason for existence of the Bell System was to extend to universal service; that is, to get telephone service out to everyone and to get the cost down so it would be used more and have a greater impact on the citizenry. That was one of the reasons that there was less interest in new technologies and in doing new things. There was a very clear charter, which was to do the things we were doing, only better and cheaper. In contrast, today I would characterize it that we have a set of amazing technologies that can do almost anything that we can imagine at a reasonable cost and with a reasonable performance. I would argue that the nature of the problem is very different today. The nature of the problem today is, "What are we going to do with these technologies?"

Vardalas:

Right.

Messerschmitt:

And there is a much greater diversity of applications. It's not just two or three applications but dozens or hundreds. There is an opportunity to have a lot more applications or to do the applications we are doing better. It is likely to be cost-effective if we just conceptualize what to do. In today's world, while there are certain notable exceptions to this observation, it is generally much less important to worry about doing things efficiently and at low cost and a lot more important to pay attention to what we are doing functionally. Of course there are notable exceptions. For example it is still not economical to bring high-definition TV and high-speed Internet access to everyone's home. There are certain bottlenecks like the one you mentioned that Gordon Moore mentioned. The more important question becomes "what do we do," not "how do we do it." One of the aspects to the "what" question is, "What do the users want? What does the world want? What will the world buy?" That is what they call in business school marketing. The second question is, "How does the marketplace and technology work? How does technology actually get out into the marketplace? What is practical to do and what is not?" based on various issues. For instance what the economists call network effects means that certain types of technology have the characteristic that as more people adopt it, it becomes more valuable to everyone. Or lock-in, the way that I create value as a software supplier. Because the cost of the actual good that I am selling is practically zero in terms of its replication and distribution costs, most of my economic value is in locking in the customer – getting my customer so dependent upon my software that they are willing to continue to pay me high prices for upgrades and maintenance of the software. I want to make it very hard for them to move to an alternative solution. Issues like that become very important in terms of the success of technology. In my view, issues like, "Are we going to standardize what we are doing? Are we going to try to do something proprietary? Are we going to try to do something proprietary and then make it a de facto standard? How do we lock in our customers to our technology? How do we bypass any obstacles that might be introduced by network effects?" are the most critical design issues. They are engineering design issues, not just business and management issues.

Vardalas:

You are saying they are both.

Messerschmitt:

They are really both, yes.

Education in policy and management

Vardalas:

Would you say there is also a regulatory political issue involved in all this?

Messerschmitt:

There are all kinds of regulatory issues, and societal issues like security and privacy, etcetera. These are all design issues that are part of the engineering design process. And yet we don’t really teach this aspect to our engineering students. In my view, by not doing so we are disadvantaging them later on in their careers. Therefore what I decided to do the last six or seven years, after serving as department chair and then moving back into teaching and research, was to focus on that aspect of business and economics. I have been really trying to formulate better what I mean by these kinds of non-technical issues being part of the design process and what should be taught to engineers about this. This also coincides with an interest I have in business, because like many people in academia today I have had the experience of starting a company and getting involved in all of the business issues involved in how to make it successful. I have an intellectual interest in that and also have a fairly broad perspective across different technologies. I have been working more on the curriculum side than the research side.

Vardalas:

Besides this issue with curriculum, do you see any policy issues relevant to this? There are also issues for management, ideas for companies to use, and obviously important issues about how companies develop and manage technology in respect to their customers. Do you see that as a part of what you want to do, or are you strictly focusing on curriculum now?

Messerschmitt:

In the curriculum area, I have been teaching courses that mix business and engineering students, specifically getting them together so that they interact with each other. We have a program on campus that encourages that called Management and Technology. More recently we have a School of Information Management Systems, which brings another group of students very interested in these issues.

Vardalas:

Excellent.

Messerschmitt:

Those three groups of students participate in the management of technology classes. With Hal Varian, who is an economist and the dean of the School of Information Management Systems, I developed a course on called Strategic Communication and Computing Technology. I also started a service course in network applications for students across the campus.

Economics and industrial properties of software

Vardalas:

The last thing you mentioned working on to the present, which seems to be a continuation, economics and industrial properties of software. Was that part of this whole current endeavor? Or is that something distinct in itself?

Messerschmitt:

It is part of it, but curriculum cannot really be separated from research. The two are intertwined. What I found for example when looking for literature on which to base the course on the strategic technology was that there was really a dearth of literature in the software area, its non-technical aspects. There is obviously a huge amount of technical literature, but there is real lack of literature on the non-technical aspects of software.

So I developed a lot of material on software as part of this course.

Vardalas:

You are about to publish a book. Does that material find itself in this book?

Messerschmitt:

Yes. I started looking around for other people who had an interest in the non-technical aspects of software and found Clemens Szyperski at Microsoft Research. He had written a book on component software, and it was one of the only books on software I had ever seen that talked about the economic and business aspects of it and relating it to the technology. It was clear that he had similar interests, so I got together with him and we decided to write a book called Software Ecosystem. It is going to be published by MIT Press later this year. What we are trying to capture in that book is the aspects of software that of interest to other professions like lawyers, economists and business people and explaining software technology and the processes that surround software. These processes include provisioning, operations and usage. We explain those processes to these people filtering the issues we cover by what is important to them and explaining it in a way they will understand. Basically the goal is to encourage more scholarly research particularly in the area of software in its non-technical aspects.

Vardalas:

Interesting.

Messerschmitt:

One of the themes of the book is that software as an economic good is different from any other economic good.

We explain how it differs and the various ways it differs. It shares properties with almost every other kind of economic good, including information and material goods and even services, but it mixes these properties in very unique ways.

Vardalas:

Interesting.

Messerschmitt:

So in this book we try to capture that thought and explain it.

Vardalas:

I look forward to reading this book.

Messerschmitt:

That is one thing that mixes curriculum and research. In a sense it is partly a research activity, but more in the sense of trying to identify what the issues are than it is in actually answering all the questions.

Patents

Vardalas:

That could be important, because the questions are more important sometimes than the actual answers. The right questions set people thinking along certain directions, which is very crucial. Let me take you back for a second. I have some very exploratory questions. I believe you have twelve patents, or somewhere in that ballpark. Of which are you most proud? Were any of them a particularly satisfying experience to produce? I would like to know the fate of these patents as well. Have any of them been put to commercialize use for instance?

Messerschmitt:

Within the electronics and software information technology industries I think that patents play much less of role than they do in some other industries like biotechnology. But this issue of intellectual property is an important aspect of the management of technology.

That is an important part of the course that we teach to these engineering and business students. But in our industries I think that copyright plays a much bigger role than patent. Patents are for the most part used by large companies as a defensive mechanism. Generally speaking, with a few notable exceptions, companies do not try to exclude others from using particular technologies because they have a patent – which would be an offensive use of patents. It is more they want to have a patent portfolio so that if other companies sue them for patent infringement they will have something they can use in defense. That is, they can look at what the other company is doing and identify possible infringements of their patents and counter-sue. For that reason, I have never viewed patents as a terribly important element of what I do, in the electronics, communications, and software industries.

Vardalas:

But whether they were licensed or not, did they find application? Were they something that had durability and were used? A lot of patents just sit there, "Oh, that's a nice idea," and then nothing ever happens with it. Right? Did something happen with the patents?

Messerschmitt:

Yes. We came up with several patents on this digital speech interpolation work at Bell Labs. While AT&T never used that technology in terms of developing it, there is at least one other company that did use digital speech interpolation. I wouldn't say that it was widely used. This is probably mainly because AT&T made a conscious decision not to use it and that eliminated the biggest market by far. In the case of the ISDN transceivers that we developed, we patented some of the basic techniques that were used in the echo cancellation. And some of those ideas became part of international standards.

Vardalas:

Okay. That is important. Yes.

Messerschmitt:

The other work that was done here at Berkeley in which I was not as directly involved in charge redistribution codex and switch capacitor filters were obviously patented. That technology was used commercially very extensively. One of the issues however, particularly in the university, is that those of us that are doing research find that patents are actually counterproductive. This is because they tend to get in the way of our relationship with industrial sponsors who want to support our research. The extent to which the university is trying to derive value from the research outcomes through licensing of patents is a deterrent for potential sponsors to actually support research.

Vardalas:

Yes.

Messerschmitt:

Therefore we sort of view patents as something that gets in the way rather than something that we like to encourage.

Vardalas:

I can see that. They create whole intellectual property departments in universities with staffs of lawyers.

Messerschmitt:

Yes, and particularly in this area. It works pretty well in areas like biotechnology where the patents are much more fundamental, but in areas like information technology I would argue that the payoff has never offset the cost.

Vardalas:

Yes. Some historians would argue that overall patents have not been as essential to company success as people think they are. Trade secrets are much more important to a company than that.

NSF Blue Ribbon Panel on Cyberinfrastructure

Vardalas:

Would you tell me very briefly about how you got involved with the NSF's Blue Ribbon Panel on Cyberinfrastructure? What does it mean in the context of everything else? It seems to gel very nicely with what you have said about your current interests.

Messerschmitt:

I first got involved in Washington through the Computer Science and Telecommunications Board of the National Research Council. I was a member of that board for five years. That board looks for policy and government issues that are relevant to computer and information technologies. Through the process of being on that board for five years and meeting four times a year I learned a tremendous amount about the various policy and legal issues relative to the information technologies. Then as I was winding up that engagement I got involved in the Advisory Board for the Computer and Information Sciences portion of NSF. And when they formed this Blue Ribbon Panel they looked for people whose advice they presumably value to appoint to it. Originally this panel was to provide advice to CISE (Computer and Information Science and Engineering) but ended up defining a much larger program that spanned all of NSF. That was appointed basically out of CISE, so my name was suggested.

Vardalas:

I read the first draft by following the link provided on your website. What has come of it? Has a final draft been produced?

Messerschmitt:

Yes. The final report is finished and available on the web.

Vardalas:

What are your conclusions about what impact this report is actually going to have in terms of action plans? What do you think is going to happen?

Messerschmitt:

I think the need to do something in terms of providing a much more solid foundation of information technology for all of science and engineering research is fairly obvious. And that is kind of happening anyway. Hopefully this report will point out that it is happening in a bad way. It is happening, as is traditional in information technology, that a lot of people are doing a lot of things in an uncoordinated fashion. As a result over time information technology will actually become an obstacle to people in different fields collaborating or exchanging data. There will also be tremendous duplications of effort with people solving the same problem multiple ways. The report will hopefully cause NSF to decide that they need a coherent program that is coordinated across all of NSF and across all of scientific and engineering disciplines. Conscious decisions need to be made about what should be common and what should be local; what should be centrally supported and what should be software provided directly to end users and supported locally; and what commercial software foundations to build upon across the whole community so that there is a lot more commonality and coordination. That way over time there will be less duplication of effort and more economies of scale. But at the same time it is important to be very cognizant of the need for flexibility so that the special needs of individual fields are met. This needs to be done in such a way that in the future it will be, because of information technology, easier (not harder) to form new sub-disciplines or collaborations across disciplines or to use data from another discipline.

Vardalas:

Right. Did you stick to the same budget estimate that you had in the first draft in this report? Is it still $650 million?

Messerschmitt:

No, it went up to a billion dollars a year.

Vardalas:

Do you realistically think, given the current status of budget issues that are going on, that this has any chance?

Messerschmitt:

Yes, I think it has a rather good chance. First of all there is a plan in place. It might be a little bit at risk, but there is a plan in place to double the NSF budget. That’s an increase of much more than a billion.

Vardalas:

There is a plan. Okay.

Messerschmitt:

The government, starting with Clinton and as I understand it the Bush Administration concurs with this, that over a period of time they would like to double the budget of NSF. This is a stronger commitment to civilian and basic long-horizon research. Of course to make that happen in reality, NSF needs to define compelling ways to spend the money that are cost-effective and have outcomes that are worth the funds expended. I see this initiative as a very worthy way to spend a billion dollars, a way that will have a big impact on science and engineering research and will also have a number of commercial spin-offs that will result in economic benefits to the country, and the country staying at the forefront.

Vardalas:

Interesting.

Messerschmitt:

The other thing that is happening is that other countries are pulling out ahead of us in this particular area. For example the so-called earth simulator in Japan is the world's fastest supercomputer. Many people in the government look at that and conclude that the U.S. has been under-investing in infrastructure of information technologies supporting scientific research. Some Europeans, particularly the British, are considerably ahead of us in defining what they call E-science, which moves science much more toward investigative tools that are digitally based.

Vardalas:

Okay. I see.

Messerschmitt:

At the same time there is a high recognition that there is a tremendous amount of data being collected in scientific endeavors at very high cost and there is a significant chance that a lot of this data will be lost and not properly curated and kept over a long period of time unless we set up specific programs for making sure that those data are systematically captured and preserved.

Vardalas:

You are optimistic then that given this attack that the Bush Administration is undergoing that there will be budget money available to promote this initiative at NSF?

Messerschmitt:

Well of course you are talking about a lot of larger issues in terms of fiscal and monetary policy that are probably beyond my expertise, but I would say that if the government wants to maintain this position of doubling the NSF budget in ways that are beneficial to the country – and not just for the purposes of spending more money – that this is a very meritorious direction to go. Therefore I think that it is a fairly easy sell, within NSF and across a broad range of directorates – and not only in CISE but also in Congress and Dr. Marberger and the Administration.

Assessment of career contributions

Vardalas:

My final question, since this is for oral history and archival purposes and this transcript will be sitting for the ages somewhere, how would you like future historians of communications technology to assess your career as an engineer? If you were to read about yourself, how would you like your career to be assessed in this area, what you've done and its contributions?

Messerschmitt:

I would be really nervous if they were looking too closely, but I guess I would like to be assessed as someone who kept up with the times. I did not continue to pursue my Ph.D. thesis through a thirty-year career at greater and greater depth but rather was fairly agile in terms of identifying the most modern issues and addressing them. I would also hope to be viewed as someone who kept a broad perspective, was not tied down to any particular narrow area, but was really focused on the big picture and tried to make contributions that were important in light of the big picture.

Vardalas:

Would you give the same answer if I asked you how you want engineers in the future to think of you and your contribution? Would it be the same kind of answer?

Messerschmitt:

It would be a similar answer. A corollary is that I personally hold the opinion that many engineers, partly as a result of engineering education, are too narrow and are too intensive too early in terms of specializing in particular areas. I would hope that engineers, and particularly engineering educators, would view me as a role model in terms of how one can think about broadening perspective and broadening education to keep up with the needs of society. I would argue in part for the reasons I mentioned about advancing technology and in part because of the increasing impact of technology on society, there is a greater need to take into account the uses and impacts of the technology as a part of the engineering design process itself – and by inference as a part of engineering education. That has happened much more in some more mature fields like civil and nuclear engineering, and it needs to happen in electrical engineering as well.

Vardalas:

That's interesting. It has been an absolutely fascinating interview. Thank you very much again for agreeing to be part of this IEEE oral history program.

Messerschmitt:

You are very welcome, it was a pleasure. It is at once fun and sobering to reflect back on one’s career.