AI-Based Compressed Sensing for MIMO Beam Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional beam searching methods in mmWave communications are time-inefficient due to excessive beam training overhead, necessitating improvements in compressed sensing measurement and prediction for enhanced network efficiency.

Innovation Solution

The implementation of AI-based compressed sensing for beam prediction, where user equipment (UE) performs beam sweeping to generate measurement matrices, compresses these matrices to obtain sensing matrices, and conducts weighted beam measurements to predict optimal beams efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If exhaustive beam searching method is used to measure beam quality information for all candidate beams, then measurement precision is improved, but loss of time increases due to excessive beam training overhead

Engineering Contradiction:
Improvebeam quality measurement precisionVSAvoidbeam training overhead
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent extracts only the essential beam quality information needed for optimal beam selection rather than measuring all candidate beams exhaustively. Compressed sensing techniques extract key channel state information from reduced measurements, enabling accurate beam prediction without exhaustive searching through all beams in the codebook.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent performs preliminary beam training to acquire channel state information and train AI models in advance. By pre-training neural networks with channel characteristics during initial beam training phases, the system enables fast beam prediction without requiring exhaustive measurement of all beams during data transmission.

Inventive Principle:
Principle #10Preliminary action

2Loss of time

If hierarchical beam searching method is used to search among wide beams first then narrow beams, then loss of time is reduced, but measurement precision deteriorates as beam selection is based on approximate metrics

Engineering Contradiction:
Improvebeam training overheadVSAvoidbeam quality measurement precision
Core Design Contradiction:
Loss of timeVSMeasurement precision

Solution Approach 1:

The patent implements feedback mechanisms where AI models continuously learn from beam measurement results and channel state information. The system uses measured beam qualities to update neural network predictions, refining beam selection accuracy over time while maintaining reduced training overhead through intelligent prediction rather than exhaustive searching.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent changes the approach from fixed hierarchical searching to adaptive beam selection using AI models. By transforming beam quality metrics into predicted optimal beam selections through neural networks, the system achieves accurate beam prediction without being constrained by hierarchical search limitations or approximate metrics.

Inventive Principle:
Principle #35Parameter changes

3Productivity

If AI models are used for matrix training and reconstruction in compressed sensing, then productivity is improved through faster beam prediction, but device complexity increases

Engineering Contradiction:
Improvebeam prediction efficiencyVSAvoidAI model implementation complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent introduces AI models as intermediaries between raw channel measurements and beam selection decisions. The neural networks serve as mediators that process compressed sensing measurements and channel state information to predict optimal beams, simplifying the overall system architecture while enabling fast prediction without requiring exhaustive beam searching or complex real-time optimization.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4498616A1Beam prediction for MIMO wireless communication systems
Publication Date: 2025.01.29 MEDIATEK SINGAPORE PTE LTD
  • EP4498616A1 patent drawingFigure 1
  • EP4498616A1 patent drawingFigure 2
  • EP4498616A1 patent drawingFigure 3

AI summary

Apparatus and methods are provided for AI-based compressed sensing for beam prediction. In one novel aspect, the UE obtains one or more sensing matrices with matrices training and performs weighted beam measurements. In one embodiment, the UE performs beam sweeping to generate one or more measurement matrices, obtains one or more sensing matrices by compressing the one or more measurement matrices with matrices training, and performs weighted beam measurements to generate one or more weighted beam measurement matrices. In one embodiment, the matrices training is performed by the UE or by the wireless network using an AI model. In one embodiment, the UE performs measurement matrix reconstruction based on the one or more weighted beam measurement matrices and the one or more sensing matrices. In one embodiment, the UE predicts an optimal beam based on the one or more predicted weighted beam measurement matrices.