The conventional signature-matched RAKE processor for multipath direct-sequ
ence code division multiple access channels is viewed as a regular linear t
ap-weight filter of length equal to the sum of the system processing gain a
nd the user channel memory. In this paper, performance improvements are sou
ght in the contest of adaptive filtering under maximum signal-to-interferen
ce-plus-noise-ratio criteria. The minimum-variance-distortionless-response
RAKE (RAKE-MVDR) filter and the lower complexity scalar optimized auxiliary
-vector RAKE (RAKE-AUX) filter are developed. Bit error rate (BER) comparis
ons with the conventional RAKE signature-matched filter are carried out for
training sets of reasonably small size, perfectly known, and mismatched/es
timated channel coefficients, and extreme near-far system configurations.