This paper is concerned with issues and techniques associated with the deve
lopment of both optimal and adaptive (data dependent) reduced-rank signal p
rocessing architectures. Adaptive algorithms for 1D beamforming, 2D space-t
ime adaptive processing (STAP), and 3D STAP for joint hot and cold clutter
mitigation are surveyed. The following concepts are then introduced for the
first time (other than workshop and conference records) and evaluated in a
signal-dependent versus signal independent context: 1) the adaptive proces
sing "region-of-convergence" as a function of sample support and rank, 2) a
new variant of the cross-spectral metric (CSM) that retains dominant mode
estimation in the direct-form processor (DFP) structure, and 3) the robustn
ess of the proposed methods to the subspace "leakage" problem arising in ma
ny real-world applications. A comprehensive performance comparison is condu
cted both analytically and via Monte Carlo simulation which clearly demonst
rates the superior theoretical compression performance of signal-dependent
rank-reduction, its broader region-of-convergence, and its inherent robustn
ess to subspace leakage.