Rank-independent convergence for generic robust adaptive cascaded cancellers via reiterative processing

This paper describes an adaptive radar method that significantly improves the signal to interference-plus-noise ratio (SINR) convergence performance of a block-processed cascaded canceller that uses a generic robust adaptive algorithm in its multiple building blocks. It is shown empirically that implementing a simple reiterative processing technique, whereby the canceller output channels are directed back to its own input channels multiple times, produces a subsequent overall adaptive convergence rate that is approximately independent of the effective rank of the input sample covariance matrix. It is noted that cascaded cancellers lend themselves to practical real-time implementation as systolic processors due to their highly parallel / pipelined signal flow structure. However, it is shown that reiterative processing provides the desired SMI-like convergence independence feature when used in conjunction with generic (i.e., nonGSCC) robust adaptive cascaded cancellers; without reiterative processing this feature is generally lost. Thus, by using reiterative processing, generic cascaded canceller building block algorithms may be made robust to realistic data without concern for placing an unintentional, rank-dependent, convergence limitation on the processor. Results for several disparate robust adaptive algorithms support this conclusion.