Reduced-complexity decision feedback equalization for high-speed communications over sparse channels

Ian J Fevrier, Purdue University

Abstract

Two modified decision feedback equalizers (DFE) structures are developed for efficient equalization of long, sparse, multipath channels with large delay spread such as those encountered in high speed communications. A modified DFE structure called the partial feedback equalizer (PFE) is first developed. The PFE has a simple implementation and partially cancels postcursor inter-symbol interference (ISI) before feedforward filtering. The complete feedback equalizer (CFE) which has a more complex implementation but completely cancels the postcursor ISI prior to feedforward filtering is then developed. Unlike the DFE, these structures allow the channel's sparseness to be exploited for efficient hardware tap allocation before this property is degraded or destroyed by feedforward filtering. One tap allocation scheme is to use a thresholded partial feedback equalizer (TPFE) or a thresholded complete feedback equalizer (TCFE), where taps are allocated according to a thresholding strategy. The TPFE and the TCFE are shown to yield significant reductions in the number of complex-multiply operations that must be implemented in hardware for speed, while maintaining performance comparable to the conventional DFE. Consideration is given to the computation and adaptation of the inherently channel-estimate based PFE and CFE algorithms to facilitate operation over time-varying channels. Fully adaptive algorithms are developed, as well as fast algorithms which directly compute the optimal filter settings from the channel estimate and knowledge of the symbol and noise variance. The performance of the DFE-type structures on long, time varying channels is limited when the channel-estimate is updated based on a finite impulse response (FIR) model, due to the large number of coefficients that must be updated. Therefore consideration is given to channel parameterizations based on the underlying multipath gains and delays which are adaptively estimated, from which the channel estimate can be computed. In the linear parametric (LP) approach, multipath path gains are adaptively estimated while the multipath delays are assumed to be known. The non-linear parametric (NLP) approach extends the (LP) approach to include adaptive delay estimation. Simulations show that LP and NLP-based DFE-type structures have improved tracking performance when compared to their FIR-based counterparts.

Degree

Ph.D.

Advisors

Fitz, Purdue University.

Subject Area

Electrical engineering

COinS