MU-MIMO Beam Selection and User Pairing Using Deep Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The challenge in multi-user MIMO systems is the computational complexity and inefficiency of existing methods for selecting optimal beam combinations and user pairings, which are essential for maximizing the multi-user priority metric (MU-PM), making real-time applications infeasible.

Innovation Solution

The use of deep neural networks (DNNs) to select multi-user MIMO candidate beams and paired users, trained to maximize the MU-PM, allowing for efficient and computationally inexpensive beam selection and user pairing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional heuristic and combinatorial-search schemes are used for beam selection and user pairing, then the implementation is simple, but the computational complexity is high and real-time applications become infeasible

Engineering Contradiction:
Improvebeam selection and user pairing efficiencyVSAvoidcomputational complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent replaces conventional mechanical search algorithms (heuristic and combinatorial search) with a deep neural network-based system. The DNN model, trained offline to maximize multi-user priority metric, performs beam selection and user pairing through neural network inference, substituting complex computational search processes with efficient neural network forward propagation, thereby achieving real-time performance while reducing computational burden

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The deep neural network model is pre-trained offline using extensive channel data and heuristic/combinatorial search results as training labels. This preliminary training phase captures optimal beam selection and user pairing patterns, allowing the trained model to make rapid predictions during real-time operation without requiring complex computations at execution time

Inventive Principle:
Principle #10Preliminary action

2Productivity

If deep neural networks are used for beam selection and user pairing, then the computational complexity is reduced and real-time applications become feasible, but the system requires training data and model deployment infrastructure

Engineering Contradiction:
Improvereal-time beam selection capabilityVSAvoidsystem implementation complexity
Core Design Contradiction:
ProductivityVSEase of manufacture

Solution Approach 1:

The patent creates a computational model (deep neural network) that copies and generalizes the decision-making patterns of optimal beam selection and user pairing from training data. Instead of implementing complex search algorithms directly, the system uses a trained neural network model that has learned optimal patterns from offline training, enabling real-time inference without requiring complex implementation logic

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The deep neural network acts as an intermediary between the raw channel state information and the final beam selection/user pairing decisions. The model processes input channel data through multiple layers of neural network computations, transforming raw observations into optimized scheduling decisions, thereby simplifying the overall system architecture while maintaining high performance

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3815255B1Methods and apparatuses of multi-user multiple-input multiple-output beam selection and user pairing using deep learning
Publication Date: 2025.12.17 NOKIA TECHNOLOGIES OY
  • EP3815255B1 patent drawingFigure 1
  • EP3815255B1 patent drawingFigure 2
  • EP3815255B1 patent drawingFigure 3

AI summary

Systems, methods, apparatuses, and computer program products for multi-user (MU) multiple-input multiple-output (MIMO) user pairing selection are provided. One method may include selecting multi-user multiple input multiple output (MU MIMO) candidate beams using deep neural network(s) (DNNs), and selecting paired users based on the selected beams. The deep neural network(s) (DNNs) are trained to maximize multi-user priority metric (MU-PM) or a heuristic of the multi-user priority metric (MU-PM).