Subspace-based SIMO blind channel identification: Asymptotic performance comparison
Kareem Bonna, Predrag Spasojević, Emmanuel G. Kanterakis · 2016
Blind identification of multiple channels using only second order statistics has been a subject of research since the early 1990s. Two methods of estimating multiple channels, the Cross-Relations (CR) and Noise Subspace (SS) methods, have received attention in previous works for their simple implementation and performance in situations where the channel identifiability conditions are easily met. However, in practice, this is not always the case, particularly since the conditions imply that the channel order is known. Channel order determination may also not be practical depending on the channels observed. Asymptotic results have been given in numerous works for both methods, with different prior distributions assumed for the signal, and under a quadratic (unit energy) constraint on the channels. Previous work has suggested that a linear constraint, as well as l-1 regularization to promote sparsity, may help improve performance when the channels are unidentifiable in practice due to unknown channel order. In this work, the CR and SS methods are reviewed, and the asymptotic performance of both methods is examined though theory and simulation under quadratic and linear constraints, as well as with regularization, when the transmitted signal is a real sequence of IID symbols.