A recursive least-squares algorithm with multiple inputs and outputs, and a cylindrical systolic implementation

A recursive least-squares (RLS) algorithm with multiple inputs as well as multiple outputs is presented. The corresponding processing architecture is a cylindrical systolic array that uses a basic two-input-two-output decorrelation processing element (PE) as its primitive building block. Because of the time recursive property, the implementation only involves scalar computations, so no vector or matrix operations are required in each PE. An efficient algorithm is derived on the basis of various architectural considerations combined with the use of an algebraic interpretation of a geometrical approach used previously. The algorithm presented generates as many output channels as input channels, which is a requirement for many signal processing tasks. One application is to adaptive pulse Doppler processing for radar systems in which a bank of filters is used to recover the entire Doppler band and each Doppler subband is processed in such a way that it is totally decorrelated with the inputs of the remaining subbands.< >