Performance improvements for fine-tuned adaptive recursive filters

Adaptive recursive filters are often implemented using direct-form realizations. However, direct-form realizations are known to have very poor sensitivity and roundoff noise properties, especially in the case of oversampled filters. Recently, a gradient-based technique was proposed for adapting a single column or row of a state-space filter where gradient signals are derived using only one extra filter. Results are presented showing that these structures are useful in oversampled filtering where an estimate of the final pole locations is known. It is shown that these adaptive structures can have much faster adaptation rates than direct-form realizations. The adapted filters can also have significantly better roundoff noise performance