A novel and efficient solution to block-based joint-detection using approximate Cholesky factorization

Despite the fact that the zero-forcing block-linear equalizer (ZF-BLE) represents an entry-level solution to the joint-detection problem as applied to the TD-CDMA air-interface proposal for UMTS, it is nevertheless accompanied by a prohibitive computational complexity when supporting long data sequences, multiple users and channels with long impulse responses. A primary contributor to this computational complexity is the need to perform a Cholesky factorization of a sparse yet large correlation matrix. This paper presents a novel technique which exploits the pseudo block-Toeplitz nature of the Cholesky factor to derive an approximate triangular factorization, thereby allowing significant reductions in computational complexity and finite-precision effects at the expense of little or no degradation in performance. The technique is equally applicable to other block-based linear or decision-feedback joint-detection schemes whose operation relies on the Cholesky factorization.