Low Complexity Optimized AOR Method for Massive MIMO Signal Detection

Massive multiple-input multiple-output (MIMO) efficiently resolves the contradiction between the rapid growth in capacity demand and the limited spectrum resource by deploying a large number of antennas. However, it involves certain obstacles such as designing low-complexity signal detectors. Linear minimum mean-square error (MMSE) achieves excellent performance for such large wireless systems but suffers from high computational complexity as the system dimensions increase. To overcome this problem, several iterative detectors with improved variants have been developed. However, these detectors suffer from performance loss and slow convergence when it comes to large system configurations.

To address these challenges, we introduce an improved variant of traditional accelerated over-relaxation (AOR) called an optimized AOR (OAOR) signal detector. AOR performance is highly dependent on its acceleration and relaxation parameters. To determine these parameters systematically, we incorporate an innovative approach using the Nelder-Mead Simplex optimization algorithm. This heuristic optimization technique allows us to find the optimal acceleration and relaxation parameters, thereby achieving superior detection accuracy with faster convergence for challenging system configurations.

We have also proposed a novel, low-complexity adaptive regularized initialization method for our detector. Initialization is crucial for signal detectors as it helps to enhance performance and achieve better results. This approach provides a better starting point that facilitates faster convergence and more accurate detection. These significant modifications amplify the efficiency and facilitate the faster convergence of the OAOR method.

Our proposed method improves signal detection accuracy while decreasing computational complexity from O(K³) to O(K²). This makes it suitable for signal detection in large-scale antenna systems within modern wireless communication. Simulation results of the proposed method indicate superior performance compared to existing techniques in detection accuracy, error rates, faster convergence, and reduced computational complexity.