Rank-2 MIMO Parameter Selection Using QR-Based MIB Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for selecting transmission parameters in MIMO channels, such as those used in LTE standards, face inefficiencies in estimating mutual information per bit (MIB) and block error rate (BLER), particularly when using maximum likelihood decoding, leading to poor prediction of link performance.
Innovation Solution
A method utilizing machine learning algorithms, specifically artificial neural networks (ANNs), to estimate MIB values based on a four-parameter representation derived from QR decomposition of the effective channel matrix, allowing for efficient calculation of MIB and BLER for rank-2 MIMO channels, and selecting optimal transmission parameters (RI, PMI, CQI) to maximize throughput while maintaining a BLER below a predetermined value.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If EESM method using PP-SNR is used to estimate MIB for MIMO channels, then the estimation process is simple and fast, but the prediction accuracy of BLER is poor especially for MLD
Solution Approach 1:
The patent transforms the channel matrix H into a four-parameter representation using QR decomposition, changing the parameter space from the original channel coefficients to {r11, r22, phase(r12), |r12|}. This parameter transformation enables accurate MIB estimation for MLD while maintaining computational efficiency, resolving the contradiction between estimation speed and accuracy.
2Adaptability or versatility
If GMM model is used to represent MIB for MIMO channels, then MIB can be estimated for high constellation and MIMO channels, but the model is difficult to calibrate and achieves poor performance when MIMO layers are highly correlated
Solution Approach 1:
The patent uses QR decomposition to transform the channel matrix into a simplified four-parameter form, eliminating the need for complex GMM calibration. The parameters {r11, r22, phase(r12), |r12|} directly capture the essential channel characteristics without requiring difficult calibration procedures, thus resolving the contradiction between versatility and calibration complexity.
3Measurement precision
If accurate MIB estimation for MLD is achieved using traditional methods, then BLER prediction improves, but computational complexity increases significantly
Solution Approach 1:
The patent segments the MIB estimation problem into two parts: first performing QR decomposition to obtain the four-parameter representation, then using this simplified representation for MIB calculation. This segmentation reduces computational complexity while maintaining accuracy, resolving the contradiction between precision and device complexity.
Solution Approach 2:
By changing from the full channel matrix parameters to the four-parameter QR decomposition representation, the patent reduces computational complexity while preserving the essential information needed for accurate MIB estimation for MLD.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system and method for MIB estimation including generating a signal model for rank=2, based on the reference signals of a received wireless signal; converting the signal model to a four-parameter representation, where each parameter is associated with the coefficients of an upper triangular square matrix of size two times two, obtained from QR decomposition; determining, for values of parameters derived from the four-parameter representation, whether mutual information per bit (MIB) values depend on a single parameter or on a plurality of parameters; if the MIB values depend on the single parameter, calculating MIB values based on the single parameter; and if the MIB values depend on the plurality of parameters, calculating MIB values based on the plurality of parameters. Calculating MIB values based on the single parameter, determining, whether MIB values depend on a single parameter or on a plurality of parameters and, calculating MIB values based on the plurality of parameters, are performed using a machine learning algorithm.