Channel Estimation Neural Network for 5G NR

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional channel estimation algorithms in 5G NR wireless communication systems face challenges due to limited DMRSs, hardware complexity constraints, and the inability to jointly exploit time and frequency properties, leading to suboptimal channel estimation.

Innovation Solution

A supervised learning method using a neural network is employed to refine channel estimates. The neural network is trained with channel estimates from conventional algorithms as input features and ideal channels as output labels, enabling frequency and time interpolation to enhance channel estimation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If conventional LMMSE-based linear interpolation is used for channel estimation, then hardware complexity is reduced, but channel estimation accuracy deteriorates due to inability to jointly exploit time and frequency properties

Engineering Contradiction:

Improvehardware complexityVSAvoidchannel estimation accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The patent replaces the conventional LMMSE-based linear interpolation algorithm with a machine learning-based channel estimation algorithm. The ML algorithm processes channel estimates from multiple DMRS ports simultaneously, exploiting both time and frequency properties through learned patterns rather than simple linear interpolation, thereby improving accuracy while maintaining hardware complexity constraints.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent combines channel estimates from multiple DMRS ports using a machine learning model that integrates information from different ports, time instances, and frequency resources. This composite approach allows the system to jointly exploit time and frequency properties across multiple ports, achieving superior estimation accuracy compared to processing each port independently with linear interpolation.

Inventive Principle:

Principle #40Composite materials

2Device complexity

If only limited DMRSs within a PRG are utilized for channel estimation, then hardware complexity constraints are satisfied, but channel estimation accuracy deteriorates due to insufficient training samples

Engineering Contradiction:

Improvehardware complexityVSAvoidchannel estimation accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The patent creates a universal channel estimation model that processes channel estimates from multiple DMRS ports simultaneously. The machine learning algorithm is trained on composite data from all available DMRS ports within the PRG, enabling the system to utilize all training samples rather than being limited to a single port's DMRSs, thereby improving accuracy while respecting hardware constraints.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Device complexity

If linear interpolation is performed independently over frequency and time, then hardware complexity is reduced, but channel estimation accuracy deteriorates due to failure to jointly exploit time and frequency properties

Engineering Contradiction:

Improvehardware complexityVSAvoidchannel estimation accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The patent merges the processing of frequency and time dimensions into a unified machine learning model. Instead of performing independent linear interpolation in frequency and time domains, the ML algorithm jointly processes channel estimates across both dimensions simultaneously, learning the coupled relationships between time and frequency variations to achieve superior estimation accuracy.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The patent transitions from separate 1D interpolation operations in frequency and time domains to a unified 2D processing approach using machine learning. The ML model operates on the joint time-frequency plane, capturing correlations and patterns that span both dimensions simultaneously, thereby exploiting the full two-dimensional structure of channel variations.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

4Adaptability or versatility

If precoding varies per PRG, then adaptability to different channel conditions is improved, but the ability to perform denoising in time domain is disabled

Engineering Contradiction:

Improveprecoding adaptabilityVSAvoiddenoising capability

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent changes the processing parameters by applying the machine learning model separately to each PRG while maintaining the per-PRG precoding variations. The ML algorithm adapts to the specific precoding configuration of each PRG by learning from the composite DMRS data within that PRG, thereby preserving precoding adaptability while enabling denoising through the learned temporal and frequency correlations within each precoding block.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12218781B2Enhancement of channel estimation in wireless communication based on supervised learning

Publication Date: 2025.02.04 SAMSUNG ELECTRONICS CO LTD

AI summary

Disclosed is an electronic device including a processor and memory, the processor being configured to perform frequency interpolation on a channel estimation at all resource elements (REs) located where a demodulation reference signal is transmitted, perform time interpolation on a frequency domain interpolated channel obtained from the frequency interpolation, and calculate an enhanced channel estimation based on channel estimates at REs in a frequency domain and REs in a time domain, the channel estimates being output from the time interpolation.