Oral-History:Lars Zetterberg

About Lars Zetterberg

Lars H. Zetterberg was born in Uppsala, Sweden on January 6, 1925. His father built power lines and was involved in the electrical power administration, so Zetterberg’s early interest in technology and engineering is unsurprising. Zetterberg attended the Royal Institute of Technology (KTH) as an electrical engineering major, later specializing in teletechnics. His first job after KTH was in antenna design at FOA (Research Institute for National Defense) for a year, during which time he became interested in circuit theory, statistical applications and information theory.

Zetterberg received his licentiate degree in 1954 and a scholarship to the University of Chicago where he worked on statistics, after which he went to Bell Labs for three months. Upon returning to Sweden, Zetterberg went back to FOA and worked on developing army communications on the battlefield, also pursuing his PhD and developing Swedish air force defense in air traffic control. After becoming tired of military work, Zetterberg worked a brief stint at SAAB, was at the University of Southern California for four months working with Irving Reed, and eventually went back to KTH as a professor from 1965 until his still active retirement in 1990.

Zetterberg’s varied career has included work with delta modulation, radar, detection theory, the Gaussian channel, EEG signals in cooperation with neurophysiologists at Karolinska Hospital in Stockholm, digitalization of speech, image coding, communication theory, and signal processing.

In this ingterview, Zetterberg discusses his involvement in the early stages of signal processing, and explains the differences between signal processing and signal theory. He also talks about the influence others had upon him, the theories of Claude Shannon and David Slepian just two examples. The importance of education is apparent throughout the article as Zetterberg explains the structure of the KTH engineering department, his involvement in creating a graduate program in his area, and his philosophy that those who worked in his group know how their work had a meaning outside of the university. At the end of the interview, Zetterberg also discusses the importance of the development of electronics, particularly the computer, to signal processing.

About the Interview

LARS H. ZETTERBERG: An Interview Conducted By Frederik Nebeker, IEEE History Center, 16 July 1996

Interview #301 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 Staff Director of IEEE History Center.

Request for permission to quote for publication should be addressed to the IEEE History Center Oral History Program, IEEE History Center at Stevens Institute of Technology, Castle Point on Hudson, Hoboken, NJ 07030 USA, or email ieee-history@ieee.org. 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:

Lars H. Zetterberg, Electrical Engineer, an oral history conducted in 1996 by Frederik Nebeker, IEEE History Center, Hoboken, NJ, USA.

Interview

Interview: Lars H. Zetterberg

Interviewer: Frederik Nebeker

Place: Västerås, Sweden

Date: 16 July 1996

Early Life and Education

Nebeker:

So, you were born in Uppsala, 6th of January 1925?

Zetterberg:

That's correct.

Nebeker:

Were you interested in science and technology as a youngster?

Zetterberg:

Yes. My father was working in electrical engineering, so that may have contributed to my interest in technology. But it was clear in my schooling at the time that I was much more interested in natural science than the humanities.

Nebeker:

What did you father do?

Zetterberg:

He built power lines in northern Sweden. He belonged to the electrical power administration.

Nebeker:

Did you go to gymnasium (academic high school)?

Zetterberg:

Yes, in Uppsala.

Nebeker:

And then you went to the Royal Institute of Technology?

Zetterberg:

That's correct.

Nebeker:

What was your field there?

Zetterberg:

To begin with, one had to choose among the seven electives, and I chose electrical engineering. But later on, after two years, it got more specialized. We had only two alternatives: e-technics and power engineering. It was clear that I would take teletechnics. At that time, it had a more physics side also, with Hannes Alfvén, working on plasma physics.

Nebeker:

Did you study with him?

Zetterberg:

No. I considered doing my master's thesis with him, but changed to a subject in circuit theory.

Nebeker:

You must have studied quite a bit of mathematics.

Zetterberg:

Yes, I was interested. We had an old but very good teacher, so I was quite attracted to that.

Nebeker:

Who were the most influential teachers you had?

Zetterberg:

At the university? Professor Erik Hallén. He was a teacher of electromagnetic theory. He was an absolutely outstanding teacher. Then, my predecessor, Torben Laurent, he did electrical circuits. He was also responsible for telecommunication systems, exchanges and so forth, but his main interest was in filter theory. Due to my attraction to electromagnetic theory, my first job choice was antenna technique. I started at FOA, the Swedish defense research establishment. Perhaps you have heard of it? I was working on the design of antennas for airplanes. That was the first project that I worked on.

Nebeker:

What is the name in Swedish?

Zetterberg:

FOA, which is the Research Institute for National Defense. During that year, I came into contact with three subjects, or really three people who had a great influence on my development. One was H. Nyquist, one of the founders of modern circuit theory. I read some more related texts at that time. All of a sudden I had a completely new idea of what circuit theory could be, compared to what I had been learning. Here was a circuit theory that was developed on mathematical grounds. Based on function theory, functions of a complex variable, representing circuits with poles and zeros, and a whole machinery for testing the realizeability and stability. So it was opening a new world. Since my teacher was trained in Germany, he had developed his own ideas of how to design filters. He was in a sense at an end. Design was all based on the theory for guided lines, and it was a completely different approach to design.

Nebeker:

Was Nyquist in Stockholm, or was it his book?

Zetterberg:

Yes, a book. The second one was Norbert Wiener, and it was through a former colleague from school, his older brother. He told me about Wiener. My friend’s brother was working at that time at Bofors. That was also something completely new. The statistical approach to signal analysis. The possibility of formulating the requirements of a filter, not, as I had learned in school, to design for L.M. Ericsson for telephone exchanges and lines, but here for some general purpose.

Nebeker:

So it was more of a general analysis?

Zetterberg:

Yes, it was based on the statistical description of the signals. A third great influence I had was Claude Shannon and his information theory. That was through a former teacher, one of the younger teachers at the Institute, who wrote a report on information theory. These three acquaintances completely changed my life. During that first year, I decided I was interested in shifting from antenna theory and design to more of the systems theory. It fit in with what happened at the Defense Institute. At that time, they set up a new group, specifically interested in the systems approach. The areas were primarily radar and communications. So I approached Nils-Henrik Lundquist, and he was interested in cooperating. I did this before going to do the military service to complete my training as an officer.

Nebeker:

A full year in the military?

Zetterberg:

No, about a half a year. The basic training took about a year. I did that while at the university (technical institute). Coming back to the defense institute I changed group affiliation and technical area.

Nebeker:

And that was a result of yourself having encountered these new techniques?

Zetterberg:

Yes, and I felt that we were acquainted with none of these new ideas at school. We were in a poor situation. The professors were talking about their own ideas with very little reference to what was happening outside. When I became a professor I was determined not to do that. I gave a much broader view of what is going on, rather than teaching just my own specialty.

Nebeker:

So, after this period of military duty, you came back with that systems group at FOA?

Zetterberg:

Yes.

Nebeker:

And what was your specific work there?

Nebeker:

I was doing various types of system investigations. It was in radar and in communications, particularly. I should also add that rather soon after I started, I made a decision that I would like to continue my studies. So I enrolled as a licentiate student. That happened rather soon after I made this switch. The problem of course was that there was no research and courses at that time, and the whole higher education was in a very primitive state in the department I had closest affiliation to. Basically I carried on all my studies by myself, taking reading courses.

Nebeker:

Your plan was while you were still working full-time, you would take courses.

Zetterberg:

Yes. At that time that was a good idea. One had a major and a minor subject. In my case it was “TT”, which stands for telegraphy and telephony. I did mostly reading courses. As a minor subject I had chosen mathematics with Professor Å. Pleijel as my teacher. Also there I took reading courses but I did attend seminar series. I did study more of function theory but I also had a course on partial differential equations. At about the same time I studied statistics at Stockholm University for Professor U. Grenander. He was a student of Professor H. Cramér, also at the Stockholm University. Grenander was very interested in stochastic processes, a subject he had written his Ph.D. thesis about. For many years I attended seminars and courses that Grenander gave, and was even active in the seminar activity. When I picked my own subject for the licenciate, I took it from one of the subjects I was working on with FOA, which was natural. It was an analysis of delta modulation, which was a new idea of digital transmission of information, particularly speech. The characteristic feature is that you represent a signal by observing changes in it. Is it positive or negative? If you take the sampling sufficiently, rapidly, then you can say that it is one step up, or one step down. If it is constant, then you go up and down so on the average you will have a constant line. There are more sophisticated developments but that is the case I worked on. So delta modulation was an idea at that time that was essentially established, and to compare it to pulse code modulation was a way to evaluate its performance. The modulator was so simple to instrument, with the electronics of the time, the late ‘50s and early ‘60s. I took my licentiate degree in ’54.

Nebeker:

You were comparing delta modulation with PCM for particular applications?

Zetterberg:

Yes. For speech. But also I looked at what information rate could be achieved. I published that in Ericsson Technichs. It had a certain influence. It was referenced for a long time afterwards. I had an interesting experience much later, when we discussed completely other subjects, like establishing microwave links. To establish army communications networks rather than lines. Then there was a person from the US Air Force, I'm not sure, but he told me when he got to my name he checked to see if it was the right person. He said, "Your paper had a great influence. It resulted in a decision to use delta modulation at that time in the United States Air Force." All of a sudden I understood that it was having an influence. But that was many years later. Therefore one can say that my licentiate was an example of the representation of signals in statistical terms. Coming back to the general comment of what are my references, the interest in using the theories of Wiener. People like Stephen O. Rice at Bell Labs, he was much more useful. He was working with Gaussian processes which allowed him to study what happened when you include both linear filters and non-linearities. I learned much of the analysis technique from Rice, and also from Bill Bennett, who was a colleague of Rice.

Nebeker:

I noticed a term in a number of your things, the "Gaussian Channel." Does that mean Gaussian noise?

Zetterberg:

Correct. But also for example, when I represented speech as a Gaussian process, the spectral density was the same as what could be observed.

Nebeker:

To allow a statistical description of the speech?

Zetterberg:

Exactly.

University of Chicago

Nebeker:

So you received your licentiate, it says here, in '54 or '55?

Zetterberg:

Yes. Late '54. As I said, I had contact with Prof. Grenander and he helped me get a scholarship at the University of Chicago. He had been a teacher there, a visiting professor. He recommended that I go there too, to develop my skills.

Nebeker:

Who were the people there you worked with?

Zetterberg:

Let me first mention the man responsible for the statistics department, Professor A. Wallace, but I didn’t have much contact with him. Much more with Professor L.J. Savage, also in statistics, and Professor D.A. Darling in probability theory. There were other people who influenced me, in particular Professor R.R. Bahadur in advanced statistics. Studies at the UC told me how to formulate problems and establish results in the form of theorems. This formal way of doing applied analysis was rather new to me and differed from most technical papers of that time. I found it quite useful. Statistical inference theory was one such area of interest.

Nebeker:

Were there any people who you came in contact with at Chicago who were doing anything like signal processing?

Zetterberg:

No, but I had some contact with the subject of stochastic processes through a lecture series headed by Professor P. Halmos. I also trained myself in mathematics by taking courses in several subjects.

Nebeker:

So you were a graduate student in mathematics.

Zetterberg:

Yes. I had a post grad fellowship in statistics, but basically I had graduate courses in both statistics and mathematics.

Nebeker:

I see you were also for three months at Bell Labs.

Zetterberg:

I took one academic year, including a summer semester. Eleven months in Chicago and then three months at Bell Labs. And then I worked with Bill Bennett.

Nebeker:

What was Bill Bennett's department?

Zetterberg:

He had a very small group. I think I was the only one he had. So he was really a solitary man.

Nebeker:

What was his specialty?

Zetterberg:

He specialized in technical applications of signal processing. The theory, you know, of signal processing had not been developed much at that time. So it was more "signal theory," we would say, or communication theory. A lot of people at Bell Labs were working on that issue.

Signal Theory

Nebeker:

How does signal theory relate to Shannon's information theory?

Zetterberg:

Not a simple question. Signal theory is a way of describing the processes as such. And all kind of processes. At that time they had an interest in stationary processes whose properties, particularly spectral properties were fixed, invariant over time, while the signal itself changed. Shannon definitely was using the concept of stochastic processes but he was interested in how much information a signal can carry. Let me make a comment, signal theory from the beginning was basically continuous functions. More and more the profession became interested in time discrete processes. Shannon assumed time discrete stochastic processes. He asked the question, "How much information does this carry?" It is a measure of the uncertainty. You observe the sequence and you predict what the next value will be. If you are very uncertain there is much information given by the next observed value. If you are more or less certain, the observed variable carries little information.

Nebeker:

So Shannon's theory was based on discrete elements in the signals?

Zetterberg:

Yes, discrete time series. Signal theory could be used for other calculations, such as to find out what happens to the signal when it passes a filter or a non-linear circuit like a detector.

Nebeker:

So that would be signal theory, what happens to a signal.

Zetterberg:

Yes, under various operations. But it also describes various properties of the process.

Nebeker:

And that was what Bennett was doing? Was that a good three months at Bell Labs?

Zetterberg:

Yes. I then came to know David Slepian, who is one of the founders of modern error correcting codes. He had a strong mathematical background, and he based his development of coding theory on finite fields. Error correcting codes are basically of two types. In one case, you deal with information in blocks of data. In order to counteract errors in storage or on a channel, you have to have more data bits than just the ones that are carrying information. You add redundancy. How to do that systematically was really the work of David Slepian.

Nebeker:

What he brought to it was this finite field theory?

Zetterberg:

Yes, it gave a language and theoretical foundation for coding analysis. Of block codes. The other type was convolutional codes, developed at MIT. They are a good example of signal processing. You filter your data sequence, and in that way you add extra bits to what you transmit. And you retrieve transmitted data at the receiver. These two methods were in competition.

Nebeker:

Are both in use?

Zetterberg:

Yes. In recent years they have come together, or into contact.

Nebeker:

Are these different ways of encoding?

Zetterberg:

Yes, they are different ways of designing the code, of how to introduce the redundant data in a series.

Nebeker:

It's not just a different method of analysis, it's different procedures for error correcting codes.

Zetterberg:

Since I had been studying Shannon, I knew about error correcting codes.

I remind you that when I came back from Bell Labs, that was in ‘56, my duties in that group at FOA increased. The situation had changed. From the beginning, FOA was a place where the very advanced ideas of that time were being tried, developed, and used. That was the best place, at least in my field, where one could do research work at that time. But the situation had changed. My boss was Lundquist, and he and the military authorities liked to see work that was much more guided to applied specific military questions. To support the military administration and to buy equipment, they liked to have specific questions answered. We took part in special studies. For many years I worked on how to develop army communications on the battlefield. Certainly this is an interesting subject. So very early on we were working on communications networks, setting up microwave links. That opened up a completely new situation. We had contact with Ericsson, which at that time had worked for the US Air Force, and developed a new type of exchange. A digital exchange.

Nebeker:

A telephone exchange?

Zetterberg:

Yes. At FOA we saw how that kind of exchange would be useful for the type of network we thought about. So I was brought into contact with Ericsson for a long time. We had really proposed it for battlefield networking. The backbone of that. And we added of course radio connections.

Nebeker:

Was it the case that your doctorate work and some of this other rigorous mathematical work, was that separate from the more practical questions you were dealing with?

Zetterberg:

Right. And I was feeling that it was coming into conflict. First of all, I found it tiring, this work with the military authorities. Not the officers themselves but the administration people. They were far behind the FOA people in concepts and knowledge. I really looked for something different. I did not always spend my time on the army communications. During this period I did my Ph.D., while I was also working at FOA. I also took part in the development of the Swedish air force defense. Particularly the center for control and steering of airplanes.

Nebeker:

Air traffic control?

Zetterberg:

Yes, the military air traffic control. And particularly my interest was in radar.

Nebeker:

You have one paper here on detection of moving radar target in clutter.

Zetterberg:

Yes I guess that's the paper. Wasn't that in Information and Control?

Nebeker:

Yes.

Zetterberg:

I did some serious work to really show what could be done

on radar systems. I also did investigations on side looking radar, which was an absolutely new concept. That kind of question was interesting. A new concept that came in, and I would study that with the idea to explain to other people around me what this scheme could be used for.

Nebeker:

Were the same mathematical techniques applicable?

Zetterberg:

Yes. I was investigating the process more than anything. The theory for signal processing was in a very early stage, and it wasn't seen as a separate topic until later. It was still under the heading of signal theory.

Nebeker:

Can you make that distinction for me?

Zetterberg:

Yes. Signal theory is where you have analysis without particular emphasis on implementation. You may study a receiver that contains a filter. You may study that. But signal processing is really the algorithmic aspects of it. How to implement it. If you have an idea about your detector, signal processing is how to implement it. It's not a trivial question to say "What is a digital filter going to look like?"

Nebeker:

So the one is analyzing a given situation. The other is trying to optimize a process.

Zetterberg:

And implement it. Signal processing has much to do with implementation. For example, in signal theory you know what spectral density your process has. But in signal processing you're asking the question, "What is the spectral density of this process here?" Say of speech, or of a radar signal after its return. You'd like to have a procedure for that. One example is the famous algorithm by Cooley and Tukey on the Fast Fourier Transform, FFT. That was one early technique, and it certainly helped to establish signal processing as a separate field.

Nebeker:

So signal processing in the abstract is these algorithms for performing what happens to the signal.

Zetterberg:

Performing analysis. Performing processing. Before there were only the very early ideas had come up on signal processing. I don't think at that time they were viewed as separate entities. That is an interesting question. When was information processing identified as a knowledge field in itself?

Nebeker:

But you felt that some of the work you had to do at FOA was taking you away from what you really wanted to do?

Zetterberg:

Yes, it was too much of the military interaction. Tiring work. So I had to look for something different. I switched from FOA to SAAB, the Swedish Airplane Company. This may seem curious, but it certainly had to do with my engagement in radar investigations as I already mentioned. Another reason was that SAAB at that time had a very qualified group in electronics and system analysis. It was headed by L.E. Zachrisson, a friend of mine from FOA where he worked earlier. So he moved to Linköping and became head of the systems group within the military side of SAAB. For developing airplanes, mostly, and I guess also robots. He was the director of this. And that kind of work I was interested in. So I don't remember if he invited me or if I was attracted myself, but I was fed up with FOA. I shifted over. But I wasn't so sure myself that it was correct. They had an office in Stockholm, and I stayed there and worked for the Linköping group. While there, I did report some of my research results from FOA at a conference in Brussels. Gradually I found that SAAB was not the right place for me. I came to a situation where one of the higher bosses came up to me and told me I needed to move. At that time I had already decided that this was not correct for me.

I had an early contact with TFR (Technical Research Council), I believe that was the name at that time. It was a research foundation. I supplied an application for a research position to be placed at KTH, my old university.

Nebeker:

So they have researchers working for them?

Zetterberg:

Yes. At that time they did have money for a research position. And I was lucky to get that. I only stayed nine months at SAAB. So around January '63 I went to KTH. Again, I had already had some contact with it through a graduate course on detection theory that I gave.

Nebeker:

Can I ask you what detection theory is?

Zetterberg:

The situation is that you are able to transmit information from A to B. On a channel disturbed by noise, Gaussian noise, in the simplest case, it may be positive or negative pulses. You know the rate, and you are asked to decide at a particular moment which one is being sent.

Nebeker:

So you know what sorts of signals are arriving, you have to make a decision.

Zetterberg:

The technique is to optimize that for this specific situation and then introduce a decision unit. You say, it's above zero level or below zero level. In such a case the situation is very simple.

Nebeker:

Is this a branch of signal theory? And is it large enough that there is something called detection theory?

Zetterberg:

Yes. You may think of a much more complicated case like phase modulation, and there it becomes much more sophisticated.

Guest Professor at USC

Nebeker:

You taught a course in that?

Zetterberg:

Right. I then continued my research on coding for the Gaussian channel, which was a major topic of my Ph.D. thesis. So we are now in early '63. Then I had my second visit to the States. I spent part of 1964 at the University of Southern California.

Nebeker:

I know Richard Bellman who's there and at the RAND corporation.

Zetterberg:

I had contact with him. I know him. He was the one who encouraged me to go to USC.

Nebeker:

I didn't know he worked in that area.

Zetterberg:

The one who invited me was Professor S. Golomb. He's really a mathematician interested in a number of things. He worked on number theory for example. That's close enough to coding theory to generate some contributions there. For example, he had a book on binary random sequences. He showed how you can produce these through what you call a shift register that really performs as a kind of digital filter. These then can be used to produce cryptographic systems. There you need random sequences

Nebeker:

Was it specifically to work with him that you went to USC?

Zetterberg:

No, I was going to be a guest professor. The person I cooperated with was actually Irving Reed.

[End of tape 1, side b]

Nebeker:

So you were' at USC in 1964.

Zetterberg:

Yes, they invited me to stay a year, but I felt that I could not do that, since I had been appointed a professor at Lund Technical University. I would have liked to stay, but I came back to Sweden. I worked for four months at USC. I continued to work on the coding of phase modulated signals, and worked on a paper together with Irving Reed, who had positions both at USC and RAND. I also met Bellman and he proposed that I should put part of my Ph.D. thesis in his book series. But I didn't undertake that work. But it was a very interesting time for me. I had achieved three published papers. These became the basis for research at KTH when I became professor there.

That time at USC was much too short, but I had an appointment as professor at Lund. I was now in the communications theory department, and I had to establish a number of courses. I only stayed for one semester since I got the same professorship in Stockholm. I felt there was a need to establish graduate education in my area. I could see there was a great need for that kind of competence. This need was coming up in industry. It was also in academia. My main interest was in graduate education. I am not uninterested in undergraduate education, but my main interest was in graduate education. I could see that it would take several years before I could graduate anyone in Lund. The other reason was the climate and the landscape. It is flat, and I have a background in the countryside and so I liked to be able to be walking in woods. So for these reasons I moved to Stockholm. Actually, before I came to KTH I had a graduate student there. I had established my first Ph.D. student. I had already given graduate courses while being docent at KTH. I continued on the coding for the Gaussian channel. Now we come to the Stockholm position which lasted from early '65 until my retirement in '90. It actually went further since I still have a position as retired professor. I am a sort of consultant. A few of my students had not finished when I retired.

Nebeker:

So it's not simply an emeritus position, it's some active position.

Zetterberg:

Yes, I'm being paid a small fee for what I'm doing. I hope I'm doing something useful, and I wanted to see that my students finish.

Nebeker:

You wrote in this article about TTT in Stockholm that this chair was authorized in '63.

Zetterberg:

That's right. I did not hold the position permanently at first, and I shared it with another professor, named Gunnar Fant. But that was for a very short time, and since then I moved to USC and Lund.

Nebeker:

So even before you were in the States you had a position.

Zetterberg:

Yes. Even before I was in Los Angeles at USC. I gave graduate courses, that was my contribution. Already I was docent, as we call it.

Nebeker:

So in '63, the government authorized three new professorships, replacing the two in radiotechnique and telegraphy/telephony. And those were in applied electronics, communication theory, and communication systems. What is the academic structure? You have an EE academic department, or section?

Zetterberg:

We had a school of electrical engineering.

Nebeker:

Do you remember the branches, or divisions of that?

Zetterberg:

There were two main divisions. One was teletechnique, and the other one was power engineering.

Nebeker:

That's the old "strong current" and "weak current" division.

Zetterberg:

In addition we also had two common departments, namely, Electromagnetic Theory and Electrical Measurement Technique. The weak current division contained departments in Applied Electronics, Communication Theory and Communication Systems.

Nebeker:

So when you came there, the weak current branch consisted of these three professorships, right?

Zetterberg:

Yes, but soon after a Speech Laboratory was added as a separate department headed by G. Fant. On the heavy current side we had Electrical Machines, Electrical Power systems and Plasma Physics.

Coding Theory and Signal processing

Nebeker:

So you were named to this chair in communication theory. I'm afraid we don't have a lot of time, so maybe we can summarize the twenty-some years you were at Stockholm? What would have been your main areas of work?

Zetterberg:

Should I talk about the group or my personal interests?

Nebeker:

If we could get your personal main lines of work in this period first.

Zetterberg:

The first period was that coding theory. Very soon, I came to ask myself, "I have 25 years here. Is it reasonable that I will go on educating people in coding theory for all these 25 years? Can I defend this regarding the needs of Swedish society?" And I said " I can't do that." You should remember that I had an interest in communication theory more generally. I had seen the subject of signal processing coming up, and I had a great interest in that. At that time I was called up by Dr. Lennart Widén at the Karolinska Hospital in Stockholm who said, "I am a neurophysiologist, and we have EEG signals to analyze. Could you help us do that?" I had been reading Norbert Wiener, and he discussed the EEG, the spectral properties and so forth. So I said, "Sure, we can do that. We know what a spectrum is and so forth." But of course the situation was far more complicated than that. More and more we were involved, and I was more and more fascinated by the subject. That really brought me into contact with modern signal processing. I did take part in some modest way of developing this area, formulating a model of a certain process, and estimating the parameters in that model.

Nebeker:

The model is of how the brain generates these signals.

Zetterberg:

We were only interested in the spectral properties of what is called the background activity of the EEG, which is fairly stationary, at least during observations of reasonable length, a stationary process. So we were given the question, "Could you estimate this spectrum?" It had to do with the interests of the neurophysiologists. They have to identify what they call waves, peaks in the spectrum. The alpha peaks at around 10 Hz, the beta at about twice that, and the lower delta rhythm at quite low frequencies. So from my point of view, there was a way of making a precise description of the spectrum. We looked at it from a Fourier technique point of view, and I found that more interesting. Gradually that continued to grow. We worked together with a Ph.D. student Arne Wennberg at the Karolinska Hospital. I had people in my laboratory developing equipment for calculating the correlation function that was used in the computer then, to calculate the model parameters of the spectra. It is not only the background activity that is of interest. The change in the background activity, what is called epileptic spikes, is not complete epileptic seizures, they are just isolated events, but it is an indication of the reason for a grand mal or some other more heavy epileptic activity. Dr. Wennberg did sleep recordings and identified that kind of activity. There are also changes in sleep stages, though we were not particularly engaged in that. To identify the changes in the spectrum, that brought us into finding what they call event detection. You detect a change in the process properties. And that is a general topic, it's a subtopic in signal processing, and people are writing books on the subject. But we made some modest contributions in that area.

Nebeker:

Can you explain how this work brought you to signal processing?

Zetterberg:

As I said, here they showed us a signal, and we had to extract the information from it. And you should remember, this is the mid-60s, '67 or '68. Computer technology was very primitive, you know. We had a simple HP desk calculator, and it was extremely simple. How are you going to convert the signal into data? At FOA the people had been sitting there doing it manually while I was employed there.

Nebeker:

Manual digitization.

Zetterberg:

Yes. We had in our department at KTH an electronics group, and they had to build everything from scratch. The recording, for example. Of course we used a Revox recorder, but still, we built a lot of equipment. More and more, deeper and deeper we were involved, and in parallel I was developing theories about how these signals were generated. Gradually that brought more students to me.

Nebeker:

What has been the net result of all this work on EEG?

Zetterberg:

First of all, we got equipment established at the Karolinska Hospital. It was being used to some extent in regular treatment, but more so for special studies. But they did use it for special cases, such as to see how different chemical agents would affect the EEG. Dr. Wennberg worked with some other people doing research and using our equipment.

Nebeker:

Well, you have quite a few publications, I see.

Zetterberg:

We were really established. We were really established in the area of processing of EEG. We started that field in the EEG area, bringing completely new means and new concepts into it. I had my students go to conferences and we published.

Nebeker:

Would this be an example of a case where electrical engineering made a big difference to medical practice?

Zetterberg:

Yes, but of course they were not mature enough to really absorb the technique. We really had to educate Dr. Wennberg. We gave him special lectures and books. He had to take some of our basic courses so we could communication with him. That's really the basic problem when you go to work in an area like medicine. At the same time, we engineers studied basic neurophysiology.

Nebeker:

Did this particular application bring important ideas to signal processing?

Zetterberg:

Yes, model formulation and system identification turned out to be useful in completely different areas. After many years, it came to a stage where we stopped EEG analysis for two reasons. One was that the technical research council had changed name and policy with more emphasis on industry related projects. The result was that they no longer liked to give us money. They still had a program in medical techniques, but they liked to give money to projects that they thought would give returns sooner than they thought our project would. The other reason was that students came up to me and said that they had come to our department to study communication theory, not to become neurophysiologists. We decided we had to stop. But for many years, we used the EEG as an example in laboratory work for students. Event detection was one of my students Ph.D. subjects and he used the EEG. And in cooperation with a colleague at the University of Uppsala, he used seismic signals. So these are examples of work on signal processing, rather than a subject in itself. I personally continued for a longer time by investigating the mechanisms for the generation of EEG and the appearance of epileptic activity. I used to have this contact with a group in Utrecht, headed by Fernando Lopes da Silva. I spent a month with their group. We had several publications on the mechanism of the production of alpha activity. It was fascinating.

Nebeker:

Is that working with a neurophysiologist?

Zetterberg:

Yes. It was really fascinating.

Nebeker:

Was that work fruitful?

Zetterberg:

Yes, we wrote a paper and a report together. I know it has been continued by others. I'd like to point out that the education we gave our students on signal processing was useful also for technical work. Two people at the Ericsson company, much later said, it's exactly engineers with the education you gave on signal processing we need. No one from Ericsson company told me at an early stage that I should engage myself in signal processing. That in a few years they would need people with that education. They did not. It was a completely other motivation for my engagement in signal processing, as I have explained. But eventually, it turned out to be exactly the kind of education Ericsson needed for their engineers developing mobile radio. My students went straight to do development work at Ericsson.

Nebeker:

I think signal processing is a field like control theory with tremendous range of applications, and it seems that certain ideas that are formulated mathematically can be fruitful in very different areas.

Zetterberg:

Take that speech signal. One has to process a signal that is more or less stationary, and you are asked to describe the characteristics of that signal. My interest then was digitalization of speech. But to do that efficiently you should know what the properties of the signal are for a specific instance or time span. We could move the technique, the basic knowledge, from EEG analysis to speech. I never meant to be really successful in speech research, or to try to establish a laboratory in speech. We never went that far. One reason was that already at KTH there was a speech laboratory headed by Gunnar Fant, and I felt it was completely wrong for us to also do that. I tried to establish contact with them. But it never became really good. One reason was that they were so poorly educated in signal processing, that we didn't have the language in common. Much too late they discovered that situation. They have had other interests also. They like to produce artificial speech from written text, and they have been successful in that. But they have been far less successful in going the other way, coming from speech to the written text. That would have required much more knowledge in signal processing I think.

Practical Applications

Nebeker:

But it's clear from what I've learned about your work that you have continually been concerned with the practical applications of all of this. Is there any danger in signal processing of the mathematically inclined people going their own way, developing theorems and so on that are not very important in practice?

Zetterberg:

I don't think so. It has been shown many times that fundamental research later often finds applications. I have as a philosophy that everyone who worked in my group should know that what they do has a meaning outside the university. Hopefully they even should have a contact outside the university. We should be free to formulate our own problems. I never liked the idea of my university group being a consultant to Ericsson, or others. I have at times admitted that specific solutions we made have been directly applied and really led to contracts. But that has been because the results have been found to be directly useful outside our group. I found it to be a good idea that people should be aware of the meaning of what they have been doing, not only for the internal science but also for the possible applications.

Nebeker:

So you're saying that there's also a danger of being too closely tied to applications.

Zetterberg:

Right. All this time I have felt that we should not be consultants. I have heard others say the same. The reason is that if you do that you become a poor consultant. You can't operate in an effective way.

Nebeker:

So, the freedom to follow the questions where they lead is very important to be effective.

Zetterberg:

We have had people who have started at Ericsson and have come back to us. I have encouraged that. It's been a good hybridization of their studies. They have a background, they have a problem area they are interested in. It's really fine. Still, even if they are partially employed at Ericsson, they are doing the work at the KTH. A special form of cooperation with industry is to establish the position of an adjunct professor. That I have been running for a number of years, and it is a good way of getting information from industry to the university. I have engaged Jan Uddenfeldt, a former student of mine, who was to be responsible for the development of mobile radio, the research side of it. He of course introduced that kind of problem in our group. In general, it is very interesting to look back and see how we came to work on specific subjects. The decision I made to engage Jan Uddenfeldt I knew would lead us into mobile radio development.

Another example of the influence from outside was when we worked on the coding of images. We left coding of speech because there was a group in Gothenberg that had gone much further on. We used speech like EEG as an example for developing methodology, rather than doing full-fledged contributions in the coding technique as such. In image coding, it was a different story. We were approached by a person from the Tele administration, then working in a development and research group there. He said to me "I would like you to get yourself into the coding of moving images." I said, "Why should I? There is already a group at Linköping University." He said, "I think it would be good. You ought to do that." Because, first of all, they have a special line of development, and it's good to have competition in this area. He engaged me as a consultant in one of the European development groups, the Cost program. This was started much earlier than the European Union sponsored research. The Cost project is driven by the tele administrations. He engaged me in that, and I got a graduate student, and we started off together studying the coding of moving images. We were aiming for a video conference system. That turned out to be quite successful work.

Nebeker:

How much of your earlier work was applicable to that?

Zetterberg:

Quite a lot. Signal processing, but also model formulation, algorithmic approaches and analysis. A lot of it. The difference is that you can't really build a model of an image or an image sequence that is in any sense stationary or has fixed properties. In the EEG and the speech, it is much different. An image was a completely new situation for us. People have tried, and do to some extent, operate with conventional techniques like taking small blocks of picture elements forming spectra by doing FFT, which more or less assumes that the spectrum properties are the essential ones. We did that to some extent, but also started looking at the situation from a much more fundamental point of view. Looking at a picture as having contours and areas with a structure or pattern. A texture, in fact.

So I have had four Ph.D. students working both on conventional spectrum transform technique to compensate for movements. But also the new technique that we call model-based image processing. My interest in image coding has a certain relevance outside science since I have an interest in art. In fact, I'm a water color painter. So I have an interest in working out how you describe images, and how you characterize them.

Nebeker:

Are you actually working on the algorithms for effective coding?

Zetterberg:

Yes. One of my students, after his exam, went to work for a while at Northern Telecom, and then to the Boston area where he joined a company, Picturetel. He and two students from MIT established a video conference system, which is a successful one. His name is Staltan Ericsson. Now he has left Picturetel and has his own company, and his idea is to use a powerful Pentium PC and establish coding on a program and do it with a powerful but still regular PC.

Nebeker:

It seems that there is one area we haven't mentioned, and we're running out of time. Maybe you can just say a word or two about the active RC networks.

Zetterberg:

That has never been my own field. It was my colleague K. Mossberg who has been in charge of the circuit network side. He was teaching circuit theory, and had two Ph.D. students first on the active RC networks and then on switched capacitor networks. Both of them are basically using analog technique. One of his students related this to what you could do with digital circuits, but basically it is an analog technique.

The Computer Revolution

Nebeker:

Maybe if I could, I'd like to pose a couple of big questions. If you have some thoughts that you want to express immediately maybe this can be something that can later on be filled out a little bit. How has the emergence of the electronic computer after World War II influenced the field of signal processing?

Zetterberg:

Let me first take up the question from a personal point of view. The appearance of the computer has been absolutely fundamental for driving its development. It meant a lot that we acquired a PDP-11. First we had the very simple desk calculators made by Hewlett-Packard, but it meant a lot that we acquired a PDP-11. We established a course in signal processing that must have been one of the very earliest university courses in signal processing.

Nebeker:

What was the year?

Zetterberg:

I think that was the early 1970s.

Nebeker:

1973.

Zetterberg:

I'm sure that was a very early regular course. So that was fundamental for us. What happened was that, to begin with, we did all the EEG work for example on the big computer. We tried several computers at different installations in order to get the best performance. But all of a sudden the university authorities changed the policy and asked for money. Real money, not allocated time. Everyone was saying, "Gee. This can’t go on." So people bought their own computers. You could do experiments with it. You might want to attach equipment to it to do the spectral analysis. Various things could be done. For quite some time we added equipment to the computer until finally we stopped and said, "No more! No more homemade junk!" We had a lot of problems with the computer. We decided to only buy things that were standardized and tested. It was absolutely fundamental. We had our own computer and we could do whatever we liked with it.

I now would like to comment on the question in general. The development of electronics in general and computers in particular has been closely tied to the development in signal processing or more generally in information processing. Development in electronics has made possible implementation of more advanced algorithms but also the other way around. Advances in theory and concepts has been driving in the development of more advanced electronic circuits. A good example is given by the development of the digital mobile radio.

Nebeker:

I can imagine that a computer is also central as a tool of analysis and working out algorithms and analysis. It's also of course crucial in a lot of the communications systems themselves. They have to have that processing power. Some of this could be electronic systems that predate the computer.

Zetterberg:

You can also see that I had students at one time that were interested in building their own microprocessor. I said, "Okay, go ahead." And they really did that. It was used for the processing of speech signals. We had a guy from the University of Loughborough in England staying with us. And there was one from France. Over the years we've had many, and we've had a close connection with ENST in Paris. We would exchange master's thesis students. Also, we had one of their teachers. Both these interactions, with Loughborough and ENST had exactly the same motivation. They were doing signal processing like we were. And as I've said, everything started in a sense with communication theory, Shannon and others. After a while both I and other people also, for example David Forney, had started with coding theory and discovered they needed signal processing in order to design channel equalizers. But now several of us are back into coding theory. I have retired and turned back to coding theory. I didn't say that in that paper I sent you. That's very interesting to see that.

In a sense information theory was not enough to solve the problems. Take speech coding and picture coding. I think information theory has made surprisingly little progress in digitization of speech and particularly of pictures. It was really a powerful use of signal processing that has made most of the progress. For example in mobile radio, a handset is filled with signal processing equipment. You have to do the coding and decoding of speech. You have to have channel equalizers to clean up the channel. There are sophisticated detectors. Since you receive sequences, not individual units or symbols. And it contains equipment for error correction. One can see signal processing as a pre-processor. In most of the equipment, in communication but also in control and in other areas, before you really apply the final information processing. But you need a lot of signal processing before it becomes efficient and effective. I would say that signal process contributed far more to the gain in performance. Information theory in the narrow, Shannon sense, has had its importance as giving us a language and a concept to discuss these things with, and a measure to compare with. In communication they are able to formulate limits. But much of modern communication equipment has been made possible by signal processing.

Nebeker:

Another big question. There are a couple of application areas that were extremely important in pushing the theory. Gyroscopic systems in control theory, for example, and negative feedback control in telephony. You've mentioned a number of areas in your own experience that have contributed to signal processing. Is it possible to name the most important applications?

Zetterberg:

In communications I will first mention satellites and data transmission on the telephone network. Later came mobile radio, driving the development both on speech coding and on data transmission on the radio channel; image processing we have already talked about but it does apply to both still and moving pictures. Process control is another very important area and I like next to make a general comment on the interrelation between communication theory and control theory, with reference to the situation in this country. NUTEK is the name of the consortium that presently gives money for research at the universities. They have had for a long time a program that covers signal processing for communications. Presently, I'm chairman of an advisory group that gives out money for that research, but earlier I was one of the applicants for it. The program has engaged people both from control theory and from communications. It is very essential that these two groups have been brought together. People on the communication side have been aware of much of the technique. To be honest, a lot of the theory of signal processing has been developed in relation to control theory, particularly the field of system identification. K.J. Åström in Lund and L. Ljung from Linköping have made essential contributions in this field. Internationally there are a number of people who have contributed. That technique has been made available so many people have been made aware of it in communications. But on the other hand, communications contain problems that have attracted people having a control theory background. That has also been fruitful for both areas. There are several groups here in Sweden, I'm thinking of a particular one in Uppsala, that has a background in control theory and they are working on communication problems. On channel equalizers. It has really been mutually fruitful. Does that answer your question?

Nebeker:

I was also curious if there are particular problems. You've mentioned this coding of moving images. Are there other particular problems that have driven a lot of the research in signal processing.

Zetterberg:

Yes. Speech coding and processing is definitely an area. And picture coding. There are also other areas such as data transmission. Another areas is seismic signals. They were very early to apply efficient algorithms. Since signal processing has an origin in mathematical statistics, particular the theory of discrete time series. I'd like to particular emphasize seismic technique for discovering oil. Not only the passive seismic registrations but also the active form, when you "sound" it. You have other areas, like the military side, radar and sonar. These are users of signal processing, although I am not sure they have contributed to the development of the basic theory. Definitely seismic. For example the modern wavelet theory had its origin in seismic investigations.

Nebeker:

I'm afraid we're running out of time. Thank you very much.

Post scriptum

1. Concerning your question, Mr. Nebeker, about important applications of signal processing I should have mentioned the use of antenna arrays, e.g. for mobile radio. The subject has recently become a topic of highest importance.

2. The interview came largely to be devoted to signal processing which gives a somewhat misleading impression. Part of our research could just as well have been called communication theory./LHZ