Wireless AI/ML Model Deployment Through Control and Data Channels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing communication systems lack a standardized method for managing the transmission of AI/ML models due to capability limitations of terminal devices, necessitating a flexible and efficient approach for deploying these models.
Innovation Solution
The method involves sending information associated with AI/ML models through control messages or data channels from network devices to terminal devices, allowing deployment under capability constraints, using control and user planes for flexible model deployment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If AI/ML models are deployed on terminal devices, then system adaptability and intelligence are improved, but terminal device capability limitations are exceeded
Solution Approach 1:
The patent segments the AI/ML model deployment process into multiple components: model information is divided into model identifiers, configuration parameters, and weight data, which are transmitted separately through control messages and data channels. This segmentation allows the system to manage complex model deployment while respecting terminal device capability limitations by transmitting only necessary portions of model data.
Solution Approach 2:
The network device acts as an intermediary between the model source and the terminal device. It manages the storage, processing, and transmission of model information, filtering and preparing data before sending it to terminals. This intermediary role protects terminal devices from directly handling overwhelming model data while still enabling advanced AI functionality.
2Adaptability or versatility
If model information is transmitted through control messages, then deployment flexibility is improved, but transmission overhead increases
Solution Approach 1:
The patent merges model information transmission with existing control message structures in the communication protocol. Model identifiers, configuration parameters, and weight data are integrated into standard control messages and data channels already present in the system, avoiding the need for separate dedicated transmission channels and reducing overall transmission overhead.
Solution Approach 2:
The system transmits only partial model information (model identifiers and configuration parameters) through control messages, while more extensive weight data is transmitted through efficient data channels. This partial action approach balances deployment flexibility with transmission efficiency by sending minimal necessary information through overhead-prone control channels.
3Measurement precision
If terminal devices acquire model information through model training processes, then model accuracy is improved, but deployment complexity and time consumption increase
Solution Approach 1:
The network device performs preliminary processing of model information before transmission to terminal devices. It pre-processes weight data, extracts essential configuration parameters, and prepares model information in advance, eliminating the need for terminal devices to perform complex model training processes and reducing deployment complexity while maintaining model accuracy.
Solution Approach 2:
Instead of requiring terminal devices to perform full model training, the system transmits pre-trained model copies (model identifiers, configuration parameters, and weight data) from the network device to terminals. This copying approach allows terminals to deploy accurate models without undergoing the complex and time-consuming training process themselves.
Data Source
Figure 1~4
Figure 5~8
Figure 9~12
AI summary
Embodiments of the present application provide a wireless communication method, a device, and a storage medium. The method comprises: a core network device or an access network device sends first information to a terminal device, the first information comprising information associated with a first model, and a sending mode of the first information comprising: sending the first information by means of a control message and/or sending the first information by means of a data channel.