Performance and VLSI implementation of novel adaptive filtering algorithms

Thesis/Dissertation · Sat Dec 31 23:00:00 EST 1988

OSTI ID:6596513

Evans, J B

Implementation and performance of several adaptive filtering algorithms is investigated. Algorithms for simplifying and improving the performance of stochastic gradient adaptive algorithms are presented and analyzed. VLSI implementations are created, and area/speed comparisons are developed. A stochastic gradient adaptive filtering algorithm using a power-of two quantizer on the coefficient update is analyzed first. The power-of-two quantizer algorithm allows the coefficient update of the adaptive algorithm to be reduced to simple shifting operations. Previous analyses have studied the power-of-two quantizer algorithm under fairly restrictive conditions. This analysis relaxes these restrictions. Under the assumption that the inputs are Gaussian, both convergence and steady state error for a finite bit power-of-two quantizer adaptive algorithm are analyzed. Stochastic gradient adaptive filtering algorithms using variable step sizes for the coefficient update are then analyzed. The variable step size algorithm improves the convergence rate, while sacrificing little in steady-state error. Expressions describing the convergence of the first and second moments of the coefficients are used to study the mean square error evolution. In addition to the initial convergence rate and the steady-state error, the algorithm performance is investigated when a power-of-two quantizer algorithm is used. Analytical results are verified with simulations encompassing various applications. Implementations of a variable step size, power-of-two quantizer algorithm in 2.5 micron Complementary Metal-Oxide-Semiconductor (CMOS) process are carried out. A 8 MHz integrated circuit with 4 taps, using dedicated hardware, and a 32 kHz chip with 63 taps, using multiplexed processing hardware, are presented. The CMOS cell designs are used to develop area/speed comparisons of various stochastic gradient adaptive filtering implementations.

• Research Organization: Princeton Univ., NJ (USA)
• OSTI ID: 6596513
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990200* -- Mathematics & Computers
990300 -- Information Handling
ALGORITHMS
ARRAY PROCESSORS
COMPUTER ARCHITECTURE
COMPUTERIZED SIMULATION
CONVERGENCE
DESIGN
DIGITAL FILTERS
ELECTRONIC CIRCUITS
GAUSSIAN PROCESSES
IMPLEMENTATION
INFORMATION SYSTEMS
MATHEMATICAL LOGIC
PERFORMANCE TESTING
SEMICONDUCTOR DEVICES
SIMULATION
STOCHASTIC PROCESSES
TESTING