2023/02/17 by Hufei Zhu, Zhu, Hufei, Yanze Zhu +8
#eess.SP
paper · pdf · doi:10.48550/arxiv.2302.08660
This paper investigates low-complexity and memoryefficient detection for vertical Bell Laboratories layered spacetime architecture (V-BLAST), where recursive and square-root implementations exhibit different tradeoffs in computational complexity. Two new improvements are first developed for the recursive implementation. The first adopts a more efficient partitioned-matrix inverse relation, reducing the dominant complexity of the inverse-construction step by a factor of 1.67. The second reformulates symbol estimation and interference cancellation using quantities available from the detection-error covariance matrix, thereby avoiding the need to retain the inverse covariance matrix during recursion. The resulting recursive algorithm reduces the dominant complexity of the existing speedoriented implementation by a factor of 1.3, while achieving a speedup of approximately 1.86 and reducing the principal matrix-storage requirement by about one half relative to the existing memory-saving implementation. Since the proposed recursive algorithm offers lower worst-case complexity, whereas the inverse-Cholesky square-root implementation provides lower best-case and average complexities, a hybrid scheme is further developed to switch adaptively between them. The hybrid scheme preserves the worst-case dominant-complexity bound of the recursive implementation, attains the best-case complexity of the square-root implementation, and slightly reduces the average processing cost. Analytical operation counts and numerical results verify these gains and demonstrate robustness to moderate variations in a switching threshold.