Block processing for one-dimensional adaptive channel estimation algorithms for OFDM systems

Adaptive pilot-aided channel estimation in frequency domain for orthogonal frequency-division multiplexing (OFDM) transmission is considered in this paper. More specifically, adaptive block and non-block processing algorithms for the first one-dimensional filtering stage of a two-stage channel estimator are compared. The estimator consists of two one-dimensional filtering stages (2 times 1D) as a common and efficient approximation of the optimal one-stage but two-dimensional (2D) estimator. The optimal block algorithms for the first stage, namely the normalized block least-mean-square (NBLMS) and block recursive least-squares (BRLS) algorithm, are deduced from the standard non-block normalized least-mean-square (NLMS) and recursive least-squares (RLS) algorithms, respectively. They are compared by simulations with complexity reduced, simplified versions of the optimal block algorithms (sNBLMS, sBRLS). Most of the block processing gain in terms of improved excess error and increased convergence speed can be achieved by the simplified approaches, whereas the non-block processing algorithms are always outperformed by their block processing counterparts. Furthermore, if the amount of input data for block processing is reduced, i.e., block size is decreased, nearly the whole block processing gain is preserved by the optimal and the simplified block algorithms. Additionally, the convergence of the NBLMS and sNBLMS can be accelerated by increasing the according step size to values, for which the NLMS algorithm would already suffer from misconvergence.