Coefficient dither in fixed-point fir digital filters

This paper presents dynamic coefficient dither as a method to mitigate coefficient quantization error in FIR filters. The uncorrelated, shapeable noise from dithered coefficients may be preferable, in certain contexts, to the frequency response distortion due to static fixed-point implementation. A theoretical framework for the analysis of coefficient dither is developed. Performance, fundamental tradeoffs, and optimization of dithered filter implementations are discussed. Oversampling is shown to improve the SNR. The theoretical results are validated with numerical experiments.