Date of Award
1989
Degree Type
Dissertation
Degree Name
Doctor of Philosophy in Electrical Engineering
Department
Electrical, Computer, and Biomedical Engineering
First Advisor
R. Kumaresan
Abstract
Conventional information processing methods suggested for the parallel implementation of DSP algorithms in VLSI hardware use weighted binary numbers. Weighted number systems suffer from high circuit complexity and low noise immunity. In this study we explore methods for representing signal and algorithm parameters using an unary representation in which all the digits have the same weight. We introduce a deterministic counterpart to stochastic computing, which enables us to control the error of computation without sacrificing the simplicity of the computing elements and the fault tolerant properties of stochastic computing. We study the properties of sequences derived from linear feedback shift-register generated PN sequences and their application to scalar and vector multiplication.
We first demonstrate how binary weighted sequences derived from PN sequences can be used to represent numbers exactly. A necessary and sufficient condition for the exact multiplication of two n/2 bit numbers to n bit accuracy represented by an nth order PN sequence is proved. We also indicate how a simple multiplier based on bit-serial architecture can be implemented based on these principles.
Next we investigate the properties of comparator sequences, which are binary sequences derived by comparing the state of the shift-register with the number to be represented. We study the interaction of binary weighted and comparator sequences and show that two n bit numbers can be multiplied to n - 1 bit accuracy using an nth order PN sequence. This technique is then extended to compute the inner product of an unknown data vector and a known coefficient vector. The proposed vector processor offers flexibility in trading off accuracy with silicon area and time of processing. Applications to digital filtering and neural networks are also discussed.
Recommended Citation
Gupta, Prabhat Kumar, "STOCHASTIC AND DETERMINISTIC COMPUTING USING PSEUDO-NOISE SEQUENCES APPLIED TO DIGITAL SIGNAL PROCESSING" (1989). Open Access Dissertations. Paper 1698.
https://digitalcommons.uri.edu/oa_diss/1698