BP Equalization via QR Decomposition for Massive MIMO Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The BP equalization algorithm faces challenges with high overhead and limited application scenarios due to its complexity and suitability issues in multi-user detection, particularly in Massive MIMO scenarios.
Innovation Solution
The method involves splitting received signals and channel estimations into real and imaginary parts, performing QR decomposition on the channel estimation matrix to reduce dimensionality, and iterating based on the resulting equivalent signals and noise power to obtain symbol position probabilities, thereby reducing computation overhead and expanding application scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If BP equalization algorithm is implemented using traditional FG-GAI or CHEMP methods, then symbol estimation accuracy is improved, but computational overhead increases and application scenarios are limited
Solution Approach 1:
The patent segments the complex-valued equalization problem into two independent real-valued problems by separating real and imaginary parts of the channel matrix H into Hr and Hi. This segmentation allows each part to be processed independently through QR decomposition, reducing the overall computational complexity while maintaining estimation accuracy.
Solution Approach 2:
The patent transforms the problem from complex domain to real domain by changing the parameter representation. Instead of performing operations on complex matrices directly, the method uses real and imaginary parts separately, which changes the mathematical parameters and reduces computational overhead while preserving the essential signal characteristics.
2Reliability
If BP equalization algorithm is implemented using traditional methods, then performance is improved, but implementation complexity increases due to large matrix operations
Solution Approach 1:
The patent divides the large complex channel matrix H into smaller real matrices Hr and Hi through segmentation. This allows QR decomposition to be performed on smaller real-valued matrices rather than one large complex matrix, reducing memory requirements and computational steps while maintaining the reliability of the equalization process.
Solution Approach 2:
The patent replaces complex arithmetic operations with real arithmetic operations. By substituting complex matrix operations with equivalent real matrix operations, the implementation becomes simpler and more efficient, as real arithmetic is computationally less intensive and easier to implement in practical systems.
3Loss of information
If QR decomposition is performed on full channel matrix H, then complete channel information is preserved, but computational cost increases
Solution Approach 1:
The patent segments the channel matrix H into real part Hr and imaginary part Hi, allowing QR decomposition to be performed separately on each segment. This segmentation preserves all channel information while distributing the computational load across two smaller operations rather than one large operation, thereby reducing overall computational cost.
Solution Approach 2:
The patent performs QR decomposition on the real and imaginary parts separately rather than on the complete complex matrix at once. This partial action approach processes the channel information in manageable segments, reducing the immediate computational burden while still capturing all necessary channel characteristics through the combined processing of Hr and Hi.
Data Source
AI summary
A Belief Propagation (BP) equalization method and apparatus, a communication device and a storage medium are disclosed. The method may include: splitting a received signal Yc, a channel estimation Hc and a symbol estimation Xc into real parts and imaginary parts to obtain a received signal matrix Y, a channel estimation matrix H and a symbol estimation matrix X (S101); performing orthogonal triangular (QR) decomposition on the channel estimation matrix H to obtain an equivalent received signal Ybp, an equivalent channel R and a noise power σ2 (S102); and performing iteration based on the equivalent received signal Ybp, the equivalent channel R and the noise power σ2 to obtain a position probability of per stream symbol (S103).