Neural Network Beam Management for Millimeter Wave Alignment

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In wireless communication systems using millimeter wave bands, the large channel propagation loss and physical limitations of beam width in beam training processes hinder precise beam alignment and increase communication resource consumption, leading to performance degradation.

Innovation Solution

An electronic device equipped with a neural network trained via a deep-learning algorithm estimates the angle of arrival (AoA) distribution for reference signal received power (RSRP) pattern data from multiple candidate beams, enabling efficient beam management and selection of optimal beams for wireless communication.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If the number of candidate beams is increased to improve beam alignment precision, then manufacturing precision (beam alignment precision) is improved, but device complexity (beam training overhead) increases

Engineering Contradiction:
Improvebeam alignment precisionVSAvoidbeam training overhead
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The neural network is trained in advance using offline machine learning with simulated RSRP pattern data and ground truth AoA information. This preliminary training enables the network to directly estimate high-resolution AoA distributions from RSRP measurements during runtime, eliminating the need for exhaustive beam training procedures and reducing beam training overhead while maintaining precise beam alignment.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If the beam width is made smaller to improve measurement precision, then measurement precision (beam alignment precision) is improved, but ease of operation (spatial constraints) deteriorates

Engineering Contradiction:
Improvebeam alignment precisionVSAvoidspatial constraints
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The patent replaces the mechanical approach of physically narrowing beam width (which faces spatial constraints) with an information-processing approach using neural networks. The network estimates high-resolution AoA distributions from RSRP pattern data, achieving precise beam alignment through computational methods rather than mechanical beam narrowing, thus avoiding spatial constraint limitations.

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

3Measurement precision

If the number of candidate beams is increased to improve beam alignment precision, then measurement precision (beam alignment precision) is improved, but loss of energy (communication resource consumption) increases

Engineering Contradiction:
Improvebeam alignment precisionVSAvoidcommunication resource consumption
Core Design Contradiction:
Measurement precisionVSLoss of energy

Solution Approach 1:

The neural network performs beam alignment estimation in advance through offline training, enabling rapid runtime operation. During actual communication, the pre-trained network directly estimates AoA distributions from RSRP measurements without requiring extensive beam training procedures, thereby reducing communication resource consumption and energy loss while achieving precise beam alignment.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4503467A1Electronic device for performing beam management based on neural network and operation method thereof
Publication Date: 2025.02.05 SAMSUNG ELECTRONICS CO LTD
  • EP4503467A1 patent drawingFigure 1
  • EP4503467A1 patent drawingFigure 2~3A
  • EP4503467A1 patent drawingFigure 3B

AI summary

An electronic device includes at least one memory, a communication interface including an antenna array configured to form a plurality of candidate beams, and at least one processor operatively connected to the communication interface and the at least one memory. The at least one processor is configured to generate a plurality of pieces of reference signal received power (RSRP) pattern data based on an RSRP measured in each of the plurality of candidate beams with respect to a signal received from an external device; estimate an angle of arrival (AoA) distribution for each of the plurality of pieces of RSRP pattern data, by applying each of the plurality of pieces of RSRP pattern data to a neural network that is trained based on a deep-learning algorithm; and perform a beam management for a wireless communication with the external device, based on the estimated AoA distribution.