基于神经网络的高并行大规模MIMO信号检测算法

随着5G和未来移动无线网络的不断发展，大规模多输入多输出(multiple input multiple output, MIMO)是其中的关键技术之一。随着天线数目的不断增加，给接收机的设计带来更高的挑战，复杂度过高的检测算法在实际中难以应用。本文将一种高并行(high-parallelism, HP)检测算法展开到神经网络中，单层神经网络基于该算法的每次迭代，并将其与可训练的权重参数和非线性神经单元相结合，提出基于网络结构HP-Net的方法。通过训练HP-Net得到最优可训练参数，进而提高检测性能。实验结果表明，所提方法相对传统最小均方误差(minimum mean square error, MMSE)算法复杂度更低，并能够得到更低的误码率；同时相对HP并行检测算法误码率性能更优。

Neural network-based algorithm for high-parallelism massive MIMO signal detection

Massive multiple input multiple output (MIMO) is one of the key technologies, as 5G and future mobile wireless networks continue to evolve. With the increasing number of antennas, the design of receivers has brought a huge challenge, and detection algorithms with too much complexity are difficult to be applied in practice. In this paper, a high-parallelism (HP) detection algorithm is developed into a neural network, the single-layer neural network is based on each iteration of this algorithm, which is combined with trainable weighted parameters and nonlinear neural units, and the network structure HP-Net is proposed. Optimal trainable parameters are obtained by training the HP-Net, which in turn improves the detection performance. The experimental results show that the paper method is less complex and can obtain lower bit error rate (BER) than the traditional minimum mean square error (MMSE) algorithm, and has better BER performance than the HP detection algorithm.