Modified Split Learning Autoencoder for 5G CSI Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 5G communication systems face challenges in efficiently compressing and decompressing channel state information (CSI) for effective feedback, particularly in ultra-high frequency bands, which affects data transfer rates and transmission distances.
Innovation Solution
An electronic device employs an autoencoder based on modified split learning, where a neural network performs data compression and decompression, enabling efficient CSI feedback by transmitting intermediate outputs between the device and the base station, and updating models using gradient values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional CSI compression methods are used, then feedback overhead is reduced, but compression efficiency and accuracy are insufficient for ultra-high frequency bands
Solution Approach 1:
The patent divides the CSI feedback process into two segments: the electronic device performs compression using an encoder model and transmits intermediate outputs, while the base station completes decompression using a decoder model. This segmentation allows optimized compression at the device side and efficient decompression at the base station side, simultaneously reducing feedback overhead and maintaining high data transfer rates in ultra-high frequency bands
Solution Approach 2:
The patent introduces intermediate outputs as a mediator between the electronic device and base station. Instead of transmitting full CSI data, the device compresses CSI to generate intermediate outputs that contain essential channel information. These intermediates serve as a compact representation that reduces feedback overhead while enabling the base station to reconstruct accurate channel state information for high-rate data transmission
2Measurement precision
If full CSI data is transmitted, then channel state accuracy is maintained, but feedback overhead and transmission resources are excessive
Solution Approach 1:
The patent extracts essential channel information from full CSI data through the encoder model, generating compact intermediate outputs that contain only the most critical channel state characteristics. This extraction process maintains measurement precision by preserving key channel properties while dramatically reducing the quantity of feedback data transmitted between device and base station
Solution Approach 2:
The patent transforms CSI data from its original high-dimensional form into a compressed parameter representation through the autoencoder model. By changing the parameter representation from full channel matrices to compressed intermediate outputs, the system maintains channel state accuracy for beamforming and resource allocation while reducing feedback data volume suitable for ultra-high frequency communication
3Loss of information
If complex compression algorithms are used, then compression ratio is improved, but computational complexity and training difficulty increase
Solution Approach 1:
The patent segments the complex compression algorithm into two simpler models: an encoder model at the electronic device for compression and a decoder model at the base station for decompression. This segmentation reduces the computational complexity and training difficulty at each individual device while achieving high compression ratios, making the system more practical for deployment in 5G ultra-high frequency bands
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of operating a base station, includes: transmitting a reference signal to an electronic device comprising a first model and a third model; receiving a first intermediate output from the electronic device; obtaining a second intermediate output by inputting the first intermediate output to a partial model excluding an output layer from a second model comprising a second neural network for data decompression; transmitting the second intermediate output to the electronic device; receiving a first gradient value from the electronic device and updating weight parameters of the partial model; and generating a second gradient value different from the first gradient value and transmitting the second gradient value to the electronic device.