ML Performance Achieving Algorithm With the Zero-Forcing Complexity at High SNR Regime

Although zero-forcing (ZF) detection is well-known for its low computational complexity in multiple-input multiple-output (MIMO) communication systems, it suffers from significantly poor performance. The sphere decoder (SD) method, on the other hand, achieves the maximum likelihood (ML) performance yet imposes a high computational complexity. We propose a low-complexity detection scheme, concatenated with the SD method, which verifies the reliability of the ZF equalized observations via some predefined regions and thresholds obtained by the channel realization. We design the threshold analytically, such that the method achieves the ML performance. With the designed threshold, we prove that the method achieves the ML performance and the ZF computational complexity at the same time with probability one, at high signal-to-noise ratio (SNR). The theoretical analysis is corroborated with numerical simulations. The simulation results also show that the proposed method achieves the ML performance very rapidly as the SNR increases.