Vincent Poor
In April of this year, David Middleton, a pioneer in the field of statistical communication
theory, and a charter member of the Information Theory Society, celebrated his 75th
birthday. Early next year, David's classic book, Introduction to Statistical Communication
Theory, will be published by the IEEE Press in its Classic Re-issue Series, under the co-sponsorship of the Information Theory Society and the IEEE Communications Society.
These two milestones highlight a distinguished research career in our field that has spanned
five decades, beginning at Harvard during World War II and continuing unabated today.
It seems appropriate to take advantage of these occasions to ask David about some of the
perspectives gained from these five decades of work.
Born in New York City on April 19, 1920, David was educated at Harvard, receiving his
Ph.D. in 1947 under the aegis of J. H. Van Vleck (who later received the Nobel Prize in
Physics). During World War II, he was a Special Research Associate at Harvard's Radio
Research Laboratory working (with Fred Terman, among others) on various studies and
applications of electronic countermeasures. After a postdoctoral year with Leon Brillouin at Harvard, he joined the
Applied Physics Faculty, where his investigations included analyses of AM and FM
systems, random noise theory and applications to system design and optimization.
Since 1954, David has pursued his work as a consulting physicist with universities,
industry, and the U. S. Departments of Energy, Defense, and the Navy. He also has
served as Adjunct Professor in physics, mathematical sciences, and electrical engineering at
various academic institutions, among them Columbia, RPI, Johns Hopkins, Texas, Rice
and the University of Rhode Island, where he has supervised a number of doctoral students.
He is the author of over 160 papers, two books, and a third volume, Threshold Signal
Processing, which is forthcoming (probably late in 1996) from the American Institute of Physics Press.
He is a Fellow of the IEEE, the American Physical Society, the Acoustical Society of
America, and the American Association for the Advancement of Science, and he is also the
recipient of a number of prizes and awards.
Somewhat unusually for researchers in our field, David's research has emphasized the
close relationships between the underlying physics and the intended engineering
applications. Among his many contributions to statistical communication theory and
statistical physics may be cited his work with Van Vleck
and the matched filter concept
(1944 - ),
the theory of signals and noise through nonlinear devices (1943 - ), the formulation and development (1954 - ) of statistical communication theory in Bayesian terms (with Van Meter 1954 - 1955),
the
development of statistical-physical models of non-Gaussian noise and interference (1970 - ),
the Bayesian theory (with Esposito) of joint detection
and estimation of signals under
prior uncertainty (1968, 1970),
new approaches to scattering in random media (1970 - ), and the
canonical theory of threshold detection and estimation in generalized noise (1965 - ).
In addition to a good deal of traveling throughout the world, David currently divides his time between homes/offices in New York City and on Cape Cod. The time of this writing found him in New York, where he graciously consented to comment on his life's work to date, and on the perspectives gained therefrom.
VP: You have been involved with the Information Theory Society since the early 1950's. How
have things changed since then?
DM: I am struck by a number of conspicuous changes from the early days. Perhaps the most obvious is the great increase in interest in our field, as evidenced by the much greater number of attendees at Symposia and the number of papers in the Transactions. Also, the much greater diversity in student and other attendees' and authors' origins and professional backgrounds is convincing evidence of the field's vitality and prospects for the future. I am also conscious (for obvious reasons) of the comparative youthfulness of most of our members, always a hopeful sign for the future.
VP: Having been around at the beginning of such fundamental concepts as matched filtering, it must be very gratifying to see such ideas still very much in the center of things.
DM: The ``matched filter'' concept was developed essentially simultaneously and independently by D. O. North (RCA) and by Van Vleck and myself (Harvard Radio Research Laboratory) circa 1943, 1944, as part of engineering efforts during WWII, in both our instances primarily to determine the effects of electromagnetic countermeasures or jamming on radar performance, as well as to help guard against such jamming. The concept of the matched filter is so much a part of our professional tradition that we take its use for granted: it is now ``standard.'' Also, in the early 1950's we have observed the development of a statistical communication theory which has systematically adapted and applied such important concepts as hypothesis testing (e.g., detection) and estimation theory (parameter estimation) to practical communication problems of all kinds and for all types of physical channels. Information theory, and its principal component coding theory, have at the same time also developed by leaps and bounds and is very much alive and productive, of course. (See ``Shannon Theory: Present and Future,'' IEEE IT Society Newsletter, Vol. 44, No. 4, Dec. 1994, for example.)
VP: Would you describe the work that was going on at Harvard during World War II?
DM: An adequate description is beyond my powers here. I recommend for the interested reader the timely 50th anniversary of WWII article by John T. Bethell: ``Harvard and the Arts of War,'' pp. 32-49, Harvard Magazine, Sept.-Oct., 1995. The parts involving the Radio Research Laboratory (RRL) are most immediately relevant to my own experiences. RRL was engaged in electronic countermeasures against enemy radar. This involved both active and passive jamming (``chaff''). My own theoretical work was principally in noise and signals through nonlinear devices, to predict the effects of jamming on radar systems. For a brief vignette of life in the Theoretical Group of RRL, I refer the reader to my short article ``Some Personal Reminiscinces: Communication Theory and a Nobel Prize,'' IEEE Communications Society Magazine, July 1978, pp. 9-10.
VP: It seems that mathematics, rather than physics, has been the dominant intellectual force in communications and information theory over the past few decades. Do you think that areas such a quantum information theory, and untethered radio and acoustic communications will bring about an increased interest in physical issues?
DM: Definitely. In fact, you have touched upon a pet topic of mine over the years: the critical rôle of the physical model on system design and operation. I suppose my bias stems from my training, but I do believe that much more attention needs to be paid to physical modeling, since we are now forced to deal with quite complex non-Gaussian physical channels. The good old ``black-box-with-additive-gauss-noise'' of yore has largely worn out its usefulness. Already, a great deal of fine work has been done involving quantum channels. Ocean acoustics is another broad area of interest where the physical channel is a dominant factor, as well as in radio communications. The ``scatter channel,'' represented by reverberation in acoustical applications and by ``clutter,'' in radar, for example, is another controlling factor. In all of the above, of course, the common language is mathematics, of the kind needed to describe the physical circumstances. (See also Prof. Sergio Verdú's comments in the above-mentioned IT Society Newsletter).
VP: The current burgeoning interest in wireless communications has motivated a resurgence of attention to the physical properties of radio channels. What rôle do you see for more physically oriented research in this field? And what major problems do you see that need to be attacked by the research community?
DM: My remarks here are closely related to my comments on the preceding questions: the pertinent physics of the channel is a critical factor in radio communications. This is embodied in the electromagnetic noise environment. (I include man-made interference, both ``intelligent'' and non-intelligent, specifically). Here we need adequate analytical models of such noise, which is usually highly non-Gaussian. We must then calibrate such models to the specific environment, by measuring the key parameters of these models. Here both theory and measurement necessarily combine. An important task in this regard is to obtain ``calibrated,'' manageable results. Modeling the environment includes modeling the sources of ``noise,'' of all types. Again, I refer to Prof. Verdú's comments in the above IT Society Newsletter. The propagation characteristics of the channel containing such sources must also, of course, be considered.
VP: One of your major contributions to our field has been the development of analytically tractable and physically meaningful models for non-Gaussian noise. Would you comment on the rôle of such models in our field?
DM: Models of non-Gaussian noise are always a much more complex phenomenon statistically than our conventional Gaussian noise processes. My emphasis has been on the often contradictory requirements of their analytic tractability and physical representability. By achieving a measure of both simultaneously, we can maintain the needed reality. At the same time we can manipulate the indicated mathematics, to obtain results which can then be verified empirically, within measures of acceptable error. This modeling, I feel, is very important in our communication field, because (1) there is so much non-Gaussian noise in the real world, and (2), because systems designed to operate optimally (or near optimally) against Gaussian interference can be significantly degraded or even made ineffective when operating against non-Gaussian noise. This is especially true for the important limiting cases of weak-signal operation. In particular this arises in radar and sonar applications, often where scattered (i.e. signal-excited) noise is the dominant factor, and for telecommunications in general, where the ambient noise environment is controlling. Such generalized noise models, based essentially on physically defined compound Poisson processes of suitable orders, are needed to provide the statistical-physical apparatus for our quantitative analyses and applications.
VP: For most of your career, you have followed the somewhat unusual path (for a researcher) of life as a independent consultant. Has this been an advantage or a disadvantage to you? Would you recommend this career path to young researchers today?
DM: Life as an independent consultant for me (since 1954) has on the whole been quite positive. The obvious advantages are independence with respect to working hours, and the methods, and to some extent the topics of work.
Also, the financial rewards are larger on the whole than those
encountered in ``company'' life, unless, perhaps, one goes the management route.
Obvious disadvantages are the short-term nature of financial support,
which must constantly be proposed for anew, the need for several
``clients'' in parallel to ensure a measure of fiscal continuity, and a broad enough spectrum of expertise to gain the needed support.
In brief, this requires a quasi-continuous ``hustle,'' and an appropriate amount of travel.
Basically, one must have the temperament for independent operation and the ability to handle the uncertainties involved. In all of this it is important to keep in contact with colleagues, including, of course, the people one
works for and with, as well as to keep close to the ``cutting-edge,'' as it were, of the areas under investigation. This involves a rather diverse continuing education, in effect. Finally, one must also cope with the fact practically and psychologically, that one is never fully part of the ``group'' as it were, where the group may be a university faculty, a business, or a government office.
To young researchers I generally do not recommend this career path. One must establish sufficient depth and breadth in one's field of knowledge and experience first, as well as determining whether or not one can be comfortable handling the uncertainties and discontinuities, as well as the partial isolation, of such a professional life. At a later and more experienced age, the path of independent consultant may be attractive to some.
VP: You have been associated with many universities over a period of more than fifty years. Would you comment on the changes in communications education that you have seen, and the issues that you see arising in the future?
DM: I should say at the outset here that my affiliation with various universities over the past half century (with the exception of my time at Harvard) has been limited (by choice) to that of Adjunct Professor, responsible over time to a sequence of doctoral students and their PhD research. Thus, my exposure to both undergraduate and graduate educational programs has been comparatively limited. Nevertheless, I note a few changes from ``my day'': (1) training in and use of computers for numerical results and modeling; (2) more applied mathematics; (3), and more out-of-school practical experience. I feel, however, that not enough attention to the underlying physics of the channel is being given; (you can see my bias arising again), but I believe it has merit. Engineers, at least those whose center of gravity lies on the theoretical side, really need more of the applied physics component. I realize that there are only so many hours in the day, but deficiencies in this area can and will be limiting. Perhaps five-year undergraduate programs (as in many architectural schools, for example) may work here, with a bit more course time also in the graduate (doctoral) programs. My point is not just standard physics (for physicists), but a properly crafted program which integrates engineering applications as well. Finally, let me state the obvious: education is a continuing process which each of us must experience in the regular course of our work.
VP: What do you consider to be the major impact of computers and computer technology on communications research generally?
DM: There is no question that the computer, properly used, can be and normally is, an essential modern tool for solving the problem of quantification here: at the final stage, prediction, performance and comparison require numbers, whatever the system. In fact, the computer most often is the instrument whereby theory is made real in the numerical sense. Having said that, and making my apology for the obviousness of the preceeding remark, I would like to say, to paraphrase an earlier (circa the '40's) statement about the concept of impedance, that a computer is not (yet) a substitute for thought.
VP: You have had extensive contact with Soviet and Russian scientists for a number of years. Could you comment on the changes you have seen there?
DM: My contacts have been of three types: (1) visits to the Soviet Union, (four times in the period 1973-1984), principally as a guest of the Acoustics Institute (Moscow) of the then Soviet Academy of Sciences; (2) encounters with Soviet, and then Russian colleagues at scientific meetings here and elsewhere outside the USSR, and (3), through correspondence. Certainly, with the collapse of the USSR I observe much more openness and ready exchanges. On other hand, because of financial problems, and the general loosening of government structure, visits to and from Russia appear to have dropped off considerably, except for those which have support from US and other scientific and engineering groups. [On a small personal note, I have found that mail sent to Russia, at least to various of my colleagues there, rarely seems to arrive. Or if it does, the answering mail does not get out. Perhaps this is just bad luck, but I do notice a significant difference from USSR days.] In any case, I hope very much that things will improve there and that our Russian colleages will be able to visit us unconstrained by organizational and fiscal problems. I look forward to that happy time and may it come soon!
VP: On behalf of myself and the readers of the Newsletter, I would like to thank you for taking time to share these thoughts with us. Certainly, your work over these past fifty years has left an indelible mark on our field. I am sure I speak for the entire community in saying that your future contributions are eagerly awaited. Congratulations, David, both on your recent 75th birthday, and on the forthcoming re-issue of your book.
DM: Thank you, Vince, for suggesting this interview, and my thanks, as well, to the Editors of this Newsletter for giving me the opportunity to reply to these questions.