Analog and Digital State-Space Adaptive IIR Filters

Adaptive recursive filters are often implemented using direct-form realizations. In this l ;. thesis, adaptive algorithms are presented for state-space systems so that the performance of various filter structures may be investigated. Through the use of simulations, it was found that much faster adaptation rates and much improved round-off noise performance may be obtained using structures other than direct-form when final pole locations can be estimated. Since the resulting algorithms are gradient-based, where the gradient signals are obtained as the output of additional fibers, both digital and anuZog adaptive recursive filters can be realized. A new or?honomZ Zu&fer jiZter structure is presented which has some properties making it attractive for analog adaptive filtering. Specifically, the structure is derived from a singly-terminated LC ladder and has the properties that it is always scaled for optimum dynamic range and its integrator outputs are orthogonal when white noise is applied to the system input. To demonstrate the practicality of analog adaptive recursive filters using the methods in this thesis, experimental results from a discrete prototype are given. As well, the design details and experimental results for a monolithic realization of a continuous-time programmable filter is presented, thus showing the feasibility of practical fully integrated analog adaptive filters. Finally, the effect of DC offsets present in analog implementations is investigated and formulae derived so that the these effects can be estimated and reduced.