124 IEEE TRANSACTIONS ON INFORMATION THEORY, JANUARY 1916

Dissertation Abstracts

Ming Un Chang, “Quantization-permutation encoding,” MS., Sch. Elec. Eng., Cornell Univ., Ithaca, N.Y., June 1975. Adviser: ‘I. Berger.

Two analog source digitization techniques are discussed-simple quantization and permutation encoding. The difficulties involved in the implementation of these two techniques prompt the consideration of the quantization-permutation (QP) encoding. It incorporates the simplicity of the quantizer in digitizing the analog data and the block coding nature of the permutation codes to provide a fully instrumentable digitization scheme. Since it employs neither variable-length coding of the quantizer output nor permutation coding of the analog source, some degradation in the R versus D performance is suffered in gaining the instrumentability. Analytical formulas are developed to evaluate the mean-squared error performance of the QP algorithm. A simple guard bin technique is introduced to improve this mean-squared error performance, and it easily outperforms the Lloyd-Max uncoded quantizer. As a result, QP encoding is proven deserving of serious consideration as a practical source digitization scheme.

Joseph Ka-Yin Lau, “On binary sliding block codes,” MS., Sch. Elec. Eng., Cornell Univ., Ithaca, N.Y., August 1975. Adviser: T. Berger.

Sliding block codes are an intriguing new alternative to the block codes used in the development of classical information theory. The fundamental analytical problem associated with the use of a sliding block code (SBC) for source encoding with respect to a fidelity criterion is that of determining the entropy of the coder output. In this disserta- tion, several methods of calculating and bounding the output entropy of an SBC are presented. The local and global behaviors of a well- designed SBC are also discussed. The so-called “101-coder,” which eliminates all the isolated zeros from a binary input, plays a central role. It not only provides a specific example for the application of the techniques developed for calculating and bounding output entropy but also serves as a medium for obtaining indirect insight into the problem of characterizing a good SBC.

An easily implementable SBC subclass is introduced in which the outputs can be calculated by simple logic circuitry. The cardinality of this subclass is investigated as a function of the basic parameters of an SBC.

Gurpartap S. Takhar, “On signals over multipath channels for aero- nautical communication,” Ph.D., Dep. Elec. Eng., Southern Methodist Univ., Dallas, Tex., Oct. 1974. Adviser: S. C. Gupta.

The general problem of communication over aeronautical fading channels is discussed in detail. For signals observed as discrete-time sequences, a demodulator structure based on the extended Kalman filter algorithms is developed for continuous-time angle modulated signals. As an example, a frequency modulation (FM) system is considered, and simulation results are presented for a message process with a Butterworth power spectrum.

The demodulation algorithms developed for aeronautical communica- tion are simplified for communication between a synchronous satellite and a ground station. The associated propagation delay is of the order of 0.25 s. The simulation results show that the algorithms perform well in the presence of such large delays.

For digital signals over satellite communication channels, a non- recursive MAP detector structure is developed. An upper bound on the probability of error is derived and numerically evaluated for several amplitude modulation-phase modulation (AM-PM) signal sets. The results are compared with those for additive channels.

Sui Yin Tung, “Buffer-instrumented variable-length encoding,” M.S., Sch. Elec. Eng., Cornell Univ., Ithaca, N.Y., June 1975. Adviser: T. Berger.

A way to use variable-length coding of fixed-rate sources for fixed- rate “real-time” transmission is described. The problem of buffer overflow and underflow is investigated. Computer simulation of the performance of the buffer-instrumented variable-length (B-VL) en- coder for a quantized Gaussian source is studied and compared to that of the quantization-permutation (QP) encoder in the sense of rate versus distortion.

The investigation reveals that the average distortion due to overflow is negligibly small compared to that due to quantization for a buffer length equal to the block length used in the corresponding QP encoder. The B-VL encoder suffers rate loss both because the optimum variable- length code constructed using the quantizer output usually has an entropy rate greater than that of the quantizer and because buffer underflows necessitate bit stuffing. However, procedures are devised that completely eliminate these rate losses without significant increase in distortion. Therefore, B-VL encoding of the quantizer output can outperform QP codes particularly for high rate sources. B-VL coding definitely is more desirable in applications where the concomitant coding delay of QP coding becomes intolerable.

Ali Zolfaghari, “Computer simulation for parameter estimation and target identification of three-dimensional Poisson processes,” Ph.D., Dep. Elec. Eng., Catholic Univ. America, Washington, D.C., 1975. Adviser: T. I. Smits.

Many important non-Gaussian random processes, such as volume reverberation (clutter), can be modeled as Poisson processes but are analytically untractable for parameter estimation. This research involves the adaptive estimation of the target density and target re- flectivity distribution of isotropic three-dimensional Poisson processes from certain experimental statistical data via a sequential computer simulation of the reverberation process itself.

First the three-dimensional Poisson process root-mean square (rms), average number of envelope peaks, amplitude probability density function (pdf), and peak amplitude pdf are studied more thoroughly than previously, using simulated data. It is concluded that the average number of envelope peaks and rms allow a good estimate of the target density, independently of the reflectivity distribution. General empirical curves relating target density with the average number of envelope peaks and with rms are found; also the “target peak loss function,” a potential quantitative measure of echo overlapping rate is studied. Extensive new reverberation envelope amplitude pdf and peak ampli- tude pdf, plus some related quantities, for different simulated target reflectivity distributions and target densities, are presented. As ex- pected, the envelope peak amplitude pdf is found to be the more useful of the two pdf for the estimation of the target reflectivity distribution.

Next, table lookup and adaptive methods for target density estima- tion, at a known target reflectivity distribution, are compared using several intuitvely designed adaptive algorithms.

Then, adaptive methods for target reflectivity distribution estima- tion, when the target density is known, based on the reverberation envelope peak amplitude pdf are developed, and the best of six different intuitively designed algorithms is found. Using error measure “learning curves,” the limitations of the selected adaptive alogrithms at different target reflectivity distributions and target densities are found, and an experimentally motivated stopping rule is introduced.

Finally, a general adaptive estimation method for the simultaneous estimation of both target density and the target reflectivity distribution from the measured reverberation envelope peak amplitude pdf is developed, and its learning curves studied. Areas needing further research are delineated.