New Common Subexpression Elimination Method for FIR Filter Design

A common subexpression elimination (CSE) method based on identical weight criteria is presented to reduce silicon area and power of finite impulse response (FIR) filters. The novelty of the method is twofold. Firstly, it only requires a few small size adders by selecting subexpressions composed of identical weight coefficient bits and then eliminating common subexpressions of different weight. Secondly, it utilizes the folded direct-form structure instead of the widely used transposed structure, thus approximately 50% registers can be reduced when compared with conventional methods. In order to find the optimum common subexpressions, a matrix search process with low complexity is introduced. Implementation examples show that our method offers an average reduction of 46% cell area and 69% power consumption over the existing H-CSE method, and an average reduction of 45% cell area and 68% power consumption over the V-CSE method.