Large-scale MIMO downlink precoding method based on deep learning

Abstract

The invention discloses a large-scale MIMO downlink precoding method based on deep learning. According to the method, a base station calculates a precoding vector corresponding to each user terminal for downlink precoding transmission through a universal framework or a low-complexity framework by using instantaneous and statistical channel information of each user terminal according to traversal achievable rates of all users or an approximate utility function maximization criterion of the traversal achievable rates. The frame is based on the structure of the optimal solution of a precoding vector: a Lagrange multiplier is given, and the direction and the power of the precoding vector can be respectively expressed as a maximum generalized feature vector form and a closed form. The universalframework calculates an optimal Lagrange multiplier through a deep neural network, and then calculates a precoding vector through an optimal solution structure; and a low-complexity framework decomposes a precoding problem into an instantaneous sub-problem and a statistical sub-problem, and the instantaneous sub-problem and the statistical sub-problem are respectively calculated and recombined. According to the method, the downlink precoding can reach the nearly optimal reachable sum rate performance, and the calculation complexity is relatively low.

Description

technical field [0001] The present invention relates to wireless communication downlink precoding, in particular to a massive MIMO downlink precoding method using machine learning. Background technique [0002] In recent years, people's demand for wireless data rate has been significantly improved, and the precoding that dynamically controls the power and phase in the base station (BS) to improve efficiency and performance has aroused widespread interest in different forms. [0003] For quasi-static and low-speed situations, the instantaneous channel state information (CSI) is relatively accurate. At this time, regularized zero-forcing (RZF) precoding, signal-to-leakage- and-noise ratio, SLNR) precoding and weighted minimum mean square error (weighted minimum mean square error, WMMSE) precoding can achieve good performance. For high-speed scenarios, the relatively short coherence time will lead to more challenges in CSI acquisition. At this time, channel obsolescence will b...

AI Technical Summary

Problems solved by technology

For high-speed scenarios, the relatively short coherence time will lead to more challenges in CSI acquisition, at this time channel outage will be inevitable, and traditional precoding methods will be seriously deteriorated

Existing methods such as joint spatial division and multiplexing (JSDM) and beam division multiple access (BDMA) use statistical CSI, which works well in mobile environments, but it ignores Instantaneous CSI is lost, so the performance in low-speed scenarios is not ideal

[0004] The posterior channel model proposed by robust precoding utilizes both instantaneous and statistical CSI to maximize ergodicity and rate, but it is difficult to directly obtain the optimal solution

Existing iterative algorithms can achieve near-optimal performance, but their cubic-level computational complexity needs to be further reduced before they can be applied to real-time systems

Images