COMBINED SUBSPACE TRACKING, PREWHITENING, AND CONTRAST OPTIMIZATION FOR NOISY BLIND SIGNAL SEPARATION

Scott C. Douglas

In many practical blind signal separation (BSS) applica­ tions, the measured mixtures contain additive noise that limits the performances of most existing BSS algorithms. In this paper, we present three new methods for blindly extracting independent sources from noisy linear mixtures. All of the methods combine approximate least­squares sub­ space tracking with contrast­based BSS in an elegant way. One of the BSS algorithms is designed to perform mini­ mum mean­square­error (MMSE) or Wiener estimation of the unknown sources, and novel least­squares prewhiten­ ing and orthogonal contrast optimization techniques are in­ troduced. Simulations verify the robust and accurate be­ haviors of the proposed methods for extracting unknown sources from noisy mixtures.