AI-Assisted Rate Adaptation for Resource-Limited Wireless Nodes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Resource-limited nodes in wireless networks cannot perform AI-based rate adaptation due to the high computing power and energy consumption required for neural network training.
Innovation Solution
An information exchange method that allows resource-limited nodes to obtain pre-trained neural network information from more powerful nodes, enabling AI-assisted rate adaptation or joint channel access without the need for local training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If resource-limited nodes perform neural network training locally, then AI-based rate adaptation performance is improved, but energy consumption and computing power requirements increase beyond available resources
Solution Approach 1:
The patent extracts the neural network training function from resource-limited nodes and relocates it to access points with sufficient computing resources. The trained neural network models are then distributed to stations for inference execution, separating the training process from the inference process and enabling resource-limited devices to benefit from AI without bearing the training overhead.
Solution Approach 2:
The patent introduces access points as intermediary entities that perform neural network training and then distribute the trained models to stations. This intermediary role allows resource-limited stations to access AI-based rate adaptation capabilities without directly performing training, thus resolving the contradiction between performance improvement and energy consumption.
2Reliability
If resource-limited nodes perform neural network training locally, then AI-based rate adaptation performance is improved, but computing power requirements exceed available resources
Solution Approach 1:
The patent extracts the computationally intensive neural network training function from resource-limited nodes and relocates it to access points with sufficient computing resources. The trained neural network models are then distributed to stations for inference execution, separating the training process from the inference process and enabling resource-limited devices to benefit from AI without bearing the training overhead.
Solution Approach 2:
The patent creates copies of the trained neural network models from access points and distributes them to multiple stations. Instead of each station independently training their own models, stations receive pre-trained model copies that can be executed with minimal computational resources, thus resolving the computing power contradiction.
3Device complexity
If conventional rate adaptation algorithms with manual parameter selection are used, then device complexity is reduced, but transmission performance and adaptability deteriorate
Solution Approach 1:
The patent applies preliminary action by pre-training neural network models at access points before distributing them to stations. This pre-processing of the AI model enables stations to execute sophisticated rate adaptation algorithms with minimal local computation, achieving high transmission performance without requiring complex local training capabilities.
Solution Approach 2:
The patent creates copies of the trained neural network models from access points and distributes them to multiple stations. Instead of each station independently training their own models, stations receive pre-trained model copies that can be executed with minimal computational resources, thus resolving the computing power contradiction.
Data Source
AI summary
This application relates to an information exchange method and a related apparatus. The method includes: A resource-limited node (for example, a STA) requests to enable an AI-assisted rate adaptation function, and after agreeing to the request, a node with a more powerful function (for example, an AP) sends, to the STA, related information of a neural network (such as a structure, a parameter, an input, and an output of the neural network) that has been trained, so that the STA determines the neural network based on the related information, and performs inference and decision-making based on data observed by the STA, to obtain a rate adaptation decision result. According to embodiments of this application, a basis can be provided for the resource-limited node to implement AI-assisted rate adaptation, and further, performance can be improved.