Blind adaptive interference suppression for near-par resistant CDMA

Multiuser detection techniques can potentially solve the near-far problem in code-division multiple-access (CDMA) systems. However, these techniques often assume knowledge of system parameters, suck as signature waveforms for all users, and associated timing, which may not be available, or may be inconvenient to obtain in practice. Recently proposed adaptive minimum mean squared error (MMSE) detectors do not require knowledge of signature waveforms; however, these techniques require a training sequence for adaptation. Here we propose a blind multiuser linear detector which requires only knowledge of the desired user's signature sequence (and associated timing). In particular, it does not require the use of a training sequence. Our approach is to decompose the filter impulse response into the sum of two orthogonal components: the matched filter corresponding to the desired user plus an adaptive filter. We show that if the adaptive filter is chosen to minimize the energy (i.e., variance) of output samples at each symbol interval, then a scaled version of the MMSE detector is obtained. Based on this observation, a simple adaptive gradient algorithm is derived, and numerical examples are presented which illustrate its performance in a synchronous CDMA system.