Adaptive Length IIR Filters Implemented with Imprecise Computing

This paper explores the use of imprecise computing to implement adaptive length IIR filters in order to meet realtime constraints. An adaptive-length filter can provide a resource manager with alternatives that can be used if resources become scarce. This paper discusses the use of standard IIR filters used in controls applications, such as a Butterworth filter, and how this filter can be implemented in prioritized stages that can be skipped if necessary. The paper also examines the use of filter banks with different size filters. In both cases, transition management methods such as bumpless transfer are investigated to mitigate the transients induced when switching between filters.