Trained and Robust Parameter Based Path Sampling for Low Complexity MIMO Detection in 5G-NR

In this paper, a low complexity MIMO detection algorithm based on path sampling is proposed to achieve near-optimal performance with limited number of paths. Conventional list sphere decoders (LSD) have good detection performance but at cost of high computation complexity, while the random sampling (RS) based MIMO detectors may suffer from inevitable performance loss and sample repetition. To overcome these issues, we first propose a data-driven method, called the trained parameter based path sampling (TPbPS), to optimize the parameter of sampling distribution, and design an approach to improve the generalization performance. Further, we take advantages of the well-trained sampling distribution and derandomize the RS to develop the robust parameter based path sampling (RPbPS), while all the sampling paths are determined by the trained sampling distribution as well as the fixed and uniformly-distributed numbers. Combining these techniques yields an enhanced MIMO detection design that non-trivially advances the state-of-the-art, and can provide significant performance and complexity gains over the traditional LSD and RS methods.