This paper presents a novel framework for link-level performance abstraction for multiple input multiple output (MIMO) receivers using a neural network model. The link-level performance abstraction is widely used to predict the receiver performances through a lookup table (LUT). As opposed to the classical LUT-based techniques, in the proposed neural network-based approach, the dataset for different channels and receivers is generated from the link-level simulations in order to train the neural network. The output performance values for MIMO wireless system are defined in terms of various features extracted from the input received codewords, derived primarily from the received post-detection signal to noise ratio (SNR) values. The redundant features are removed before training the neural network. Finally, the neural network model is incorporated into the link-level simulation chain, replacing the receiver. The performance of the proposed framework is then evaluated for different channel conditions. Experimental results provide a good close link-level approximation for different receivers subjected to various modulation and coding schemes. We show that the neural network-based link-level performance abstraction outperforms the classical LUT-based link-level abstraction technique with exponential mapping function under various modulation and coding schemes.

本文提出了一种使用神经网络模型对多输入多输出 (MIMO) 接收器进行链路级性能抽象的新框架。链路级性能抽象广泛用于通过查找表 (LUT) 预测接收器性能。与经典的基于 LUT 的技术相反，在所提出的基于神经网络的方法中，不同通道和接收器的数据集是从链路级模拟生成的，以训练神经网络。MIMO 无线系统的输出性能值是根据从输入接收码字中提取的各种特征来定义的，这些特征主要来自接收到的检测后信噪比 (SNR) 值。在训练神经网络之前去除冗余特征。最后，神经网络模型被纳入链路级仿真链中，取代了接收器。然后针对不同的信道条件评估所提出的框架的性能。实验结果为经受各种调制和编码方案的不同接收器提供了良好的链路级近似。我们表明，在各种调制和编码方案下，基于神经网络的链路级性能抽象优于具有指数映射函数的经典基于 LUT 的链路级抽象技术。