A Configurable FIR Filter Scheme based on an Adaptive Multilayer Network Structure

In this paper, we present a design technique of configurable FIR filter architecture based on neural network like (multilayer network) structure. This architecture is a generalization of the configurable adaptive FIR filters and can be implemented on an FPGA. The design is based on a combination of pipelining and folding schemes at the multiplication-addition component level, network cell level, and network layer level. The proposed configurable multilayer network technique can reduce latency and hardware requirements while increasing the throughput of the filter. The weighted input connections, the network cells, and the number of network layers are configurable to fit filter design requirements. The configurable pipelining/folding scheme and the parameter setting characterize the proposed FIR filter architecture, FPGA space, and operation timing requirements. The configurable architecture is compared with several traditional FIR filter structures regarding hardware and time complexity. The potential applications of the proposed architecture are also discussed with respect to traditional adaptive FIR filter performance.