Low Complexity Equalization for Doubly Selective Channels Modeled by a Basis Expansion

We propose a novel equalization method for doubly selective wireless channels, whose taps are represented by an arbitrary Basis Expansion Model (BEM). We view such a channel in the time domain as a sum of product-convolution operators created from the basis functions and the BEM coefficients. Equivalently, a frequency-domain channel can be represented as a sum of convolution-products. The product-convolution representation provides a low-complexity, memory efficient way to apply the channel matrix to a vector. We compute a regularized solution of a linear system involving the channel matrix by means of the GMRES and the LSQR algorithms, which utilize the product-convolution structure without ever explicitly creating the channel matrix. Our method applies to all cyclic-prefix transmissions. In an OFDM transmission with subcarriers, each iteration of GMRES or LSQR requires only O(K log K) flops and O(K) memory. Additionally, we further accelerate convergence of both GMRES and LSQR by using the single-tap equalizer as a preconditioner. We validate our method with numerical simulations of a WiMAX-like system (IEEE 802.16e) in channels with significant delay and Doppler spreads. The proposed equalizer achieves BERs comparable to those of MMSE equalization, and noticeably outperforms low-complexity equalizers using an approximation by a banded matrix in the frequency domain. With preconditioning, the lowest BERs are obtained within 3-16 iterations. Our approach does not use any statistical information about the wireless channel.