Reduction of quantization effects in adaptive filters

This paper analyzes and discusses several methods for reducing the quantization effects in adaptive filters. Described methods are "offset truncation rule", "error feed back quantizer", "memory size expansion method" and "leakage algorithm". In this work, the quantization effects in the tap Weight control loop of an adaptive equalizer are analyzed assuming additive input noise. Then convergence speed and the tap weight variance in each method are examined by computer simulation and theoretical analysis. The results are as follows: the offset truncation rule is better than the rounding rule in terms of convergence speed and behavior for a high S/N input signal, the error feedback quantizer and memory size expansion method have good performance almost the same as a non quantized structure, and the leakage algorithm is very effective when the input auto-correlation matrix has small eigen values.