A modular pipelined implementation of a delayed LMS transversal adaptive filter

The problem of efficiently realizing a delayed least-mean-squares (DLMS) transversal adaptive filter is investigated. A time-shifted version of the DLMS algorithm is derived. Due to its order recursive nature, the restructured algorithm is well suited to parallel implementation. In addition, a pipelined systolic-type architecture which implements the algorithm is presented. The performance of the pipelined system is analyzed, and equations for speedup are derived. The pipelined system is capable of much greater throughput than existing conventional least-mean-square (LMS) implementations, making it a good candidate for real-time applications where high sampling rates are required. Also, due to its high modular nature, the system is easily expandable