MIMO ML Receiver Using NT×NT Inputs for Massive Antenna Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ML-based receivers, such as the DeepRx receiver, face challenges in scaling to larger, massive MIMO setups due to high computational resource requirements.
Innovation Solution
The proposed ML-based receiver operates based on data representations defined by the number of MIMO layers, eliminating the antenna dimension consideration, using a pre-trained ML model to process a modified matrix and array of symbols, reducing computational resources through techniques like Hermitian conjugate operations and CNNs with residual connections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing ML-based receivers (e.g., DeepRx) are used to achieve high radio performance, then radio performance is improved, but computational resource requirements increase significantly
Solution Approach 1:
The patent extracts and eliminates the antenna dimension (NR) from the computational model by using a modified matrix formulation. Instead of processing the full NR×NT CSI matrix, the invention uses a modified NT×NT matrix that captures essential channel information without depending on the number of antennas, thereby reducing computational complexity while maintaining detection performance.
Solution Approach 2:
The patent changes the dimensional parameters of the processing matrices from NR×NT to NT×NT. This parameter transformation reduces the computational burden from O(NR×NT) to O(NT²), making the system scalable to massive MIMO configurations where NR can be very large, while NT remains relatively small.
2Productivity
If existing ML-based receivers are designed to handle more antennas, then MIMO system capacity increases, but device complexity increases and scalability deteriorates
Solution Approach 1:
The invention extracts the essential channel state information needed for detection by using a modified matrix that eliminates the antenna dimension. This allows the receiver complexity to remain constant (dependent only on NT) while the system can support any number of antennas NR, enabling scalable MIMO configurations.
Solution Approach 2:
The modified ML model is designed to be universal across different MIMO configurations. By formulating the processing based on NT×NT matrices rather than NR×NT matrices, the same receiver architecture can handle various antenna configurations without redesign, making it adaptable to both conventional and massive MIMO systems.
3Measurement precision
If the ML model processes data with antenna dimension (NR) to maintain accuracy, then detection precision is improved, but the model cannot scale to massive MIMO setups
Solution Approach 1:
The patent transforms the data representation parameters from including antenna dimension (NR×NT) to excluding it (NT×NT). This parameter change enables the model to maintain detection precision by preserving essential channel information while becoming adaptable to any antenna configuration, including massive MIMO where NR can be very large.
Solution Approach 2:
The invention changes the dimensional structure of the input data to the ML model by eliminating the antenna dimension. Instead of feeding the model data with dimensions dependent on both NR and NT, the model receives processed data with dimensions dependent only on NT, effectively reducing the problem from two-dimensional scaling to one-dimensional scaling.
Data Source
AI summary
The present disclosure relates to a machine learning (ML)-based receiver that is computationally efficient, irrespective of a number NR of receiver antennas used in a Multiple Input Multiple Output (MIMO) scenario. To achieve this, the ML-based receiver is configured to obtain a modified matrix and a modified array of symbols based on a channel state information (CSI) matrix and a received array of symbols. The modified matrix has a dimension NT×NT, where NT is a number of MIMO layers. The modified array of symbols has a dimension NT. After that the ML-based receiver is configured to restore a transmitted array of symbols from the received array of symbols by applying a pre-trained ML model that receives the modified matrix and the modified array of symbols as input data and outputs a set of bit log-likelihood ratio (LLR) estimates for the transmitted array of symbols.