Using 2nd-order information to reduce average coefficient length in IIR filters

Developments in ASIC and FPGA technology have increased interest in area- and power-efficient implementations of digital filters. We propose a method of obtaining reduced space and power IIR implementations using variable length coefficients. Our method uses the Fisher information matrix to include 2^nd-order effects in the determination of coefficient wordlengths. For Hwang´s filter, our algorithm yields designs with more than 16% improvement in the average coefficient length, compared to uniform length coefficient design for the same performance. The resulting coefficients can be implemented using different implementation approaches, including multiplierless designs and lookup table methods.