Adaptive Multi-Channel Least Mean Square and Newton Algorithms for Blind Channel Identification

The problem of identifying a single-input multiple-output FIR system without a training signal, the so-called blind system identification, is addressed and two multi-channel adaptive approaches, least mean square (LMS) and Newton algorithms, are proposed. In contrast to the existing batch blind channel identification schemes, the proposed algorithms construct an error signal based on the cross relations between different channels in a novel, systematic way. 

The corresponding cost (error) function is easy to manipulate and facilitates the use of adaptive filtering methods for an efficient blind channel identification scheme. It is theoretically shown and practically demonstrated by numerical studies that the proposed algorithms converge in the mean to the desired channel impulse responses for an identifiable system.