We investigate the optimal signal detection problem in large-scale multiple-input multiple-output (MIMO) system with the generalized spatial modulation (GSM) scheme, which can be formulated as a closest lattice point search (CLPS). To identify invalid signals, an efficient pruning strategy is needed while searching on the GSM decision tree. However, the existing algorithms have exponential complexity, whereas they are infeasible in large-scale GSM-MIMO systems. In order to tackle this problem, we propose a memory-efficient pruning strategy by leveraging the combinatorial nature of the GSM signal structure. Thus, the required memory size is squared to the number of transmit antennas. We further propose an efficient memory-bounded maximum likelihood (ML) search (EM-MLS) algorithm by jointly employing the proposed pruning strategy and the memory-bounded best-first algorithm. Theoretical and simulation results show that our proposed algorithm can achieve the optimal bit error rate (BER) performance, while its memory size can be bounded. Moreover, the expected time complexity decreases exponentially with increasing the signal-to-noise ratio (SNR) as well as the system’s excess degree of freedom, and it often converges to squared time under practical scenarios.

中文翻译：

---

GSM-MIMO 系统的高效内存边界最优检测

我们使用广义空间调制 (GSM) 方案研究大规模多输入多输出 (MIMO) 系统中的最优信号检测问题，该方案可以表述为最近格点搜索 (CLPS)。为了识别无效信号，在搜索 GSM 决策树时需要一种有效的剪枝策略。然而，现有的算法具有指数级的复杂性，而在大规模的GSM-MIMO系统中是不可行的。为了解决这个问题，我们通过利用 GS​​M 信号结构的组合特性提出了一种内存有效的剪枝策略。因此，所需的内存大小是发射天线数量的平方。我们通过联合采用所提出的剪枝策略和内存边界最佳优先算法，进一步提出了一种有效的内存边界最大似然 (ML) 搜索 (EM-MLS) 算法。理论和仿真结果表明，我们提出的算法可以达到最佳的误码率（BER）性能，而其内存大小可以有界。此外，期望时间复杂度随着信噪比（SNR）的增加以及系统的过度自由度而呈指数下降，并且在实际场景下通常收敛到平方时间。