Telecommunications & Signal Processing Laboratory

Thesis Abstracts, 1987-1989

Vasu Iyengar

A Low Delay 16 kbit/sec Coder for Speech Signals

M.Eng. Thesis, August 1987

Supervisor: P. Kabal

This thesis studies a low delay speech coder operating at 16 kbits/sec. The coding algorithm is a delayed decision tree-coding scheme using the multipath (M,L) tree search algorithm. Two different adaptive synthesis filter configurations are used for mapping the innovations code (excitation sequences) to the output or reconstruction code. The first configuration uses a short-term or formant synthesis filter which reconstructs the speech spectral envelope. The fine structure of the speech spectrum is contained in the innovations sequence in this case. The second configuration consists of a cascade of a long-term or pitch synthesis filter and a formant synthesis filter. The pitch filter first reconstructs the spectral fine structure, and the formant synthesis filter inserts the spectral envelope. Backward adaptation of both the pitch and formant synthesis filters results in no side information requirements for the transmission of adaptation information. Noise feedback encoder configurations are employed to allow for the use of a frequency weighted error measure. The innovations tree is populated using random numbers with a Laplacian distribution, and includes the effect of a backward adaptive gain.

Results of both objective and formal subjective testing of the encoding algorithm are presented. At an encoding rate of 16 kbits/sec, and an encoding delay of 1 ms, the algorithm yields a subjective quality equivalent to 7 bits/sample log-PCM. The encoding algorithm is suitable for use in digital links having low encoding delay constraints, such as the switch telephone network. Recommendations for future studies are given.

John Frixou Michaelides

Nonlinear Adaptive Filtering for Echo Cancellation and Decision Feedback Equalization

M.Eng. Thesis, July 1987

Supervisor: P. Kabal

This thesis studies the problem of nonlinear adaptive filtering for Echo Cancellation (EC) and Decision Feedback Equalization (DFE). The high speed requirements in digital subscriber loops and voiceband data modems have put more constraints on the design of adaptive filters used for EC and DFE due to the significance of nonlinearities present in the echo and communication channels. Different configurations for nonlinear adaptive filters are developed for both echo cancellation and for combined echo cancellation and decision feedback equalization. Most of the configurations are base on the "Table Look-up" structure although the use of nonlinear filters based on the Volterra series and nonlinear compensators (involving separate adaptive linear and nonlinear parts) are also considered. Computer simulations are performed for all the configurations and most of the results are supported by theoretical derivations.

Robert W. Tansony

A Variable Rate Adaptive Transform Coder for the Digital Storage of Audio Signals

M.Eng. Thesis, April 1987

Supervisor: P. Kabal

This thesis describes the development of a transform coding algorithm for the digital archiving of audio signals. The storage of both speech and music is considered. Silence deletion and signal bandwidth estimation are employed to provide a continuously variable bitrate, adaptively matched to the characteristics of the input signal. The economic and operational feasibility of replacing a traditional analog archive with a coder-based digital system is examined. A floating point Fortran simulation shows that the proposed archiving algorithm offers average storage savings of 76% over 16 bit linear-PCM systems.

Thesis titles.