Performance analysis of blind channel estimation without eigendecomposition

Abstract

The analytical performance of the blind estimation of finite-impulse response (FIR) channels without eigendecomposition in CDMA systems is examined. The channel problem arises when CDMA signals are subject to frequency selective fading. Blind channel estimation is desired when training data is unavailable. Blind channel estimation techniques in the literature often rely on subspace decomposition of received signals. Blind channel estimation can also be coupled with a multiuser receiver design in the context of a minimum output energy (MOE) criterion. While subspace decomposition is avoided, the inverse of the data covariance matrix of the received signals is needed (Xu, Z. et al., Proc. IEEE Asilomar Conf. Signals, Syst. Comput., p.689-93, 2002). Previously (Ozdemir, O. and Torlak, M., Proc. IEEE Int. Conf. on Commun., vol.5, p.2621-5, 2004), we proposed a blind channel estimation method using simple covariance matrix transformations. We now use analytical and simulation results to show that the channel estimates converge to the eigendecomposition based channel estimates at medium to high signal-to-noise ratio (SNR).