Design of hardware efficient modulated filter bank for EEG signals feature extraction

In this paper, we propose a new efficient design algorithm to synthesize low complexity FIR filters, which has been applied to EEG signals feature extraction filter bank. A prudently defined bit-level frequency response sensitivity function is designed as a measure for the proposed coefficients quantization decision making. The coefficients are synthesized to meet the specifications while addressing the complexity reduction by maximizing common subexpressions sharing with the aid of instant checking and updating of common subepxressions statistics. By this new iterative algorithm, the filter coefficients are quantized into optimal patterns which are favorable for the circuit implementation with significantly reduced area cost. The effectiveness of the proposed design algorithm is demonstrated using two design examples where the proposed design solution saves about 82.6% and 48.5% of hardware complexity over the baseline implementation and other competing methods.