Blind identification and equalization of MIMO FIR channels based on second-order statistics and blind source separation

Abstract

Abstract:In the single-user scenario of data communications, the identification and equalization of the channel can be accomplished blindly (i.e., without training sequences) using second-order statistics (SOS) if suitable diversity in the received signal is exploited. Diversity can be temporal, spatial, or both. In this paper one well known SOS-based method for blind identification and equalization (BIE) of communication channels is extended to the multi-input scenario. It is shown that the application of this SOS-BIE procedure reduces the system to a problem of blind source separation (BSS) of instantaneous linear mixtures. The users' signals can then be recovered in a second stage through appropriate BSS techniques, typically requiring higher-order statistics.

Exploiting independence for co-channel interference cancellation and symbol detection in multiuser digital communications

“…For ISI cancellation, we use the extended version [4] of the blind channel identification method of [2]. This method ('TON94') is then followed by a CCI-cancellation stage based on ILSF [1] or ICA (Sec.…”

“…case. It is proved in [4] that in the multiuser case the estimated channel matrix is of the formĤ = H(IM+N ⊗ Q H ), with Q ∈ C K×K an unknown unitary matrix. In the noiseless case, this estimated channel matrix results in the ISI-free spatial mixture…”

“…Typically, the multipath propagation effects are well approximated by finite impulse response (FIR) filters, and then the MIMO model is characterized by a channel matrix composed of the polyphase coefficients. Hence, one can first identify the channel matrix and then employ a conventional detector, such as zero-forcing or minimum mean square error (MMSE), to obtain the source symbols from the channel estimate [2,3,4]. The main drawback of this approach is the requirement of estimating a 'nuisance' parameter such as the channel matrix, when the actual parameters of interest are, in most cases, the users' symbols.…”

Tensor Decompositions With Several Block-Hankel Factors and Application in Blind System Identification

“…We mention that subspace-based methods for convolutive ICA have already been proposed [46], [25]. However, the methods in [46], [25] essentially ignore the tensor structure of the problem.…”

“…We mention that subspace-based methods for convolutive ICA have already been proposed [46], [25]. However, the methods in [46], [25] essentially ignore the tensor structure of the problem. For this reason, these methods can only handle the overdetermined case ("more outputs than inputs") while the methods here can also deal with the underdetermined case ("fewer outputs than inputs") and are guaranteed to find the decomposition in the exact case.…”