Federated Autoencoder Pairs for Client-Specific Channel Feedback

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing wireless communication systems face challenges in efficiently managing channel state information reporting due to varying environmental conditions and device-specific impairments, leading to suboptimal physical layer link performance across different clients.

Innovation Solution

Employing autoencoder pairs trained using federated learning techniques to adapt neural networks for individual clients, allowing for personalized compression and reconstruction of measurement data, thereby facilitating improved channel state feedback and communication performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a single neural network model is used for all clients, then device complexity is reduced, but physical layer link performance deteriorates due to inability to adapt to individual client environmental conditions

Engineering Contradiction:
Improvephysical layer link performanceVSAvoidneural network model complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the single neural network model into multiple client-specific models, each tailored to handle the unique environmental conditions and device-specific impairments of individual clients. This segmentation enables each model to specialize in optimizing performance for its specific client context, thereby improving overall physical layer link performance.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements local quality by training separate neural network models for each client, allowing each model to capture and adapt to the specific characteristics, environmental conditions, and impairment patterns unique to that client. This ensures that each client receives customized processing optimized for their specific requirements rather than a generic approach.

Inventive Principle:
Principle #3Local quality

2Reliability

If client-specific neural network models are trained for each client, then physical layer link performance is improved through adaptation to individual conditions, but device complexity and training requirements increase

Engineering Contradiction:
Improvephysical layer link performanceVSAvoidmodel training and management complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent employs a universal server-based training platform that handles the training of multiple client-specific models centrally. This universal system receives raw data from all clients, processes it through the federated learning framework, and generates customized models for each client without requiring complex local training infrastructure at individual client devices.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The server acts as an intermediary between raw client data and final customized models. It receives data from clients, performs the complex training computations, and distributes the resulting models back to clients. This intermediary approach simplifies the overall system architecture by centralizing computational complexity while maintaining client-specific optimization.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If raw measurement data is transmitted to the server for processing, then channel state feedback accuracy is improved, but network bandwidth consumption increases

Engineering Contradiction:
Improvechannel state feedback accuracyVSAvoidnetwork bandwidth consumption
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts and transmits only the essential raw measurement data required for training the neural network models to the server, rather than transmitting complete processed results. This selective extraction minimizes the quantity of data transmitted over the network while preserving the information needed for accurate channel state feedback generation.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system performs preliminary processing and feature extraction at the client devices before transmission to the server. By pre-processing the data locally to identify and transmit only the most relevant features, the system reduces the amount of data that needs to be transmitted while maintaining the accuracy needed for effective channel state feedback.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4200755B1Federated learning of autoencoder pairs for wireless communication
Publication Date: 2026.01.28 QUALCOMM INC
  • EP4200755B1 patent drawingFigure 1
  • EP4200755B1 patent drawingFigure 2
  • EP4200755B1 patent drawingFigure 3

AI summary

Various aspects of the present disclosure generally relate to wireless communication. In some aspects, a client may determine, using a first client autoencoder, a feature vector associated with one or more features associated with an environment of the client. The client may determine a latent vector using a second client autoencoder and based at least in part on the feature vector. The client may transmit the feature vector and the latent vector. Numerous other aspects are provided.