Precoder Selection via Trained Model for Massive MIMO
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current precoder optimization methods in wireless communication networks are computationally complex and infeasible for systems with large antenna arrays, such as massive MIMO, due to their high computational complexity, especially when dealing with finite-alphabet signal constellations.
Innovation Solution
A low-complexity alternative is proposed using a trained computational model, such as a machine learning-based model, that learns the mapping between approximated precoders and optimal precoders during an offline training phase, allowing for efficient precoder selection based on new channel observations in the online inference phase.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional precoder optimization methods are used to achieve optimal precoder selection, then communication performance is maximized, but computational complexity becomes infeasible for systems with large antenna arrays
Solution Approach 1:
The patent applies preliminary action by pre-training a neural network model offline using optimal precoder calculations on training data. The trained model is then deployed for online inference, where it rapidly selects precoders without requiring complex real-time calculations. This separates the computationally intensive optimization work (done beforehand) from the real-time operation (done quickly during inference), resolving the contradiction between optimal performance and computational feasibility.
2Measurement precision
If exact optimal precoder calculation methods are used, then precision of precoder selection is maximized, but computational resources required become prohibitive for massive MIMO systems
Solution Approach 1:
The patent uses copying by training the neural network on a dataset containing channel matrices and their corresponding optimal precoders. The network learns to copy the mapping relationship between channel conditions and optimal precoder selections without needing to recalculate optimality conditions during real-time operation. This allows the system to achieve precise precoder selection by copying learned patterns rather than performing expensive real-time optimization calculations.
Data Source
AI summary
Embodiments herein relate to a method performed by a network node for handling communication in a wireless communication network. The network node selects a precoder given a channel based on output of a trained computational model, trained with approximated precoders or channel matrices mapped to preferred precoders, and transmits data over the channel using the selected precoder.