Colored-Noise MIMO Detector Using ZF Filtering and DNNs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current MIMO detectors face suboptimal performance due to channel variations and complexity in matrix inversion, especially as the number of antennas and users increases, and deep-learning based detectors struggle to adapt to real-time channel variations.
Innovation Solution
A colored-noise learning-based MIMO detector using zero forcing (ZF) filtering and deep neural networks (DNNs) is employed to decompose colored-noise components, allowing for joint detection of data streams and optimizing performance across varying channels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional MIMO detectors use matrix inversion methods, then detection can be performed, but performance becomes suboptimal and complexity increases with the number of antennas and users
Solution Approach 1:
The patent replaces traditional mechanical/mathematical matrix inversion methods with a deep learning-based neural network system. The neural network is trained offline to learn the inverse channel characteristics, and during operation, it directly processes received signals without performing real-time matrix inversion, thereby reducing computational complexity while maintaining or improving detection performance
Solution Approach 2:
The patent performs channel inversion operations in advance during the offline training phase of the neural network. The network learns the inverse channel characteristics beforehand, so that during actual data detection, the system only needs to perform simple forward propagation through the trained network rather than complex real-time matrix inversion
2Measurement precision
If deep-learning based detectors are used, then detection accuracy can be improved, but adaptability to real-time channel variations becomes difficult
Solution Approach 1:
The patent segments the detection process into two distinct phases: offline training phase where the neural network is trained on channel characteristics, and online detection phase where the trained network processes incoming signals. This segmentation allows the system to achieve both high detection accuracy through offline learning and reasonable adaptability through the structured two-phase approach
Solution Approach 2:
The patent employs colored-noise learning that adapts to different channel conditions by learning noise characteristics specific to each channel scenario. The system can adjust its detection parameters based on the learned channel and noise characteristics, improving adaptability to real-time variations while maintaining detection accuracy
3Productivity
If the number of antennas and users increases, then system capacity improves, but detection performance deteriorates due to increased complexity
Solution Approach 1:
The patent replaces traditional matrix inversion detectors with a deep learning-based detector that does not suffer from the same complexity-performance tradeoff. The neural network can handle high-dimensional input data from multiple antennas and users simultaneously without the exponential complexity increase that plagues traditional methods, thereby maintaining detection performance as system capacity scales
Data Source
AI summary
A method for receiving data by a terminal in a wireless communication system according to the present document comprises: receiving a channel signal and a reference signal (RS) from a base station; and generating a sequence by filtering the RS, and decoding the channel signal on the basis of the generated sequence, wherein the filtering is zero forcing (ZF) filtering, and the decoding of the channel signal is a selection of one parameter from among parameter sets generated in accordance with a colored-noise machine learning process.