Wireless Terminal Neural Network CSI Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large-scale MIMO systems face challenges in efficiently compressing and reconstructing channel state information (CSI) due to high resource overhead, especially with traditional methods like quantization or codebook-based methods, which lead to significant performance loss when dealing with multiple channel models with varying powers, delays, and angles.
Innovation Solution
A terminal and base station system that performs characteristic domain transformation on the channel matrix using fully connected networks to compress and decompress CSI feedback, allowing for more accurate data recovery with reduced overhead, employing neural networks for transformation and compression processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If complete channel state information is fed back using traditional quantization or codebook-based methods, then the channel state information is preserved, but the resource overhead increases significantly
Solution Approach 1:
The patent segments the channel state information processing into two stages: first transforms the channel matrix into characteristic domain using neural networks to extract essential features, then applies compression to the transformed characteristics. This segmentation allows preserving critical channel information while reducing feedback overhead by transmitting only the compressed characteristic features rather than complete channel state information.
Solution Approach 2:
The patent changes the domain parameters by transforming channel matrix data from spatial domain to characteristic domain through neural network-based transformation. This parameter transformation converts the original high-dimensional channel matrix into a lower-dimensional characteristic representation that captures essential channel properties, thereby reducing feedback overhead while maintaining information quality.
2Productivity
If the number of antennas at the base station is increased to improve MIMO performance, then the throughput increases, but the feedback overhead increases with the increase of antennas
Solution Approach 1:
The patent extracts essential channel characteristics from the complete channel matrix through neural network transformation. By taking out only the most important characteristic features that represent the essential channel behavior, the system can support increased antenna counts for higher MIMO throughput while keeping feedback overhead manageable by transmitting only the extracted characteristics rather than full channel information.
Solution Approach 2:
The patent applies dimensionality reduction by transforming the channel matrix from its original high-dimensional form (scaling with antenna count) into a lower-dimensional characteristic domain. This dimensionality change allows the system to accommodate more antennas for improved throughput while the compressed characteristic representation keeps feedback overhead from scaling proportionally with antenna count.
3Quantity of substance
If quantization or codebook-based compression is applied to reduce feedback overhead, then the resource overhead decreases, but channel state information is lost
Solution Approach 1:
The patent introduces neural networks as an intermediary between the channel matrix and the feedback transmission. The neural network transforms the channel matrix into characteristic domain representation that serves as an intermediate form, capturing essential channel information in a compressed format. This intermediary transformation preserves critical channel state information while achieving compression, avoiding the information loss associated with traditional quantization or codebook-based methods.
Data Source
AI summary
The present invention provides a terminal and a base station in a wireless communication system, and methods executed by the terminal and the base station. The terminal comprises: a processing unit, configured to perform characteristic domain transformation on a channel matrix to obtain a transformed channel characteristic and compress the transformed channel characteristic to obtain a compressed channel characteristic; and a transmitting unit, configured to transmit the compressed channel characteristic, as feedback information, to the base station.