Iterative Linear Detection for Massive MIMO Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Massive MIMO systems face computational inefficiencies and high pipeline latency in exact linear detection methods, such as Cholesky decomposition, which limit data throughput and hardware efficiency.
Innovation Solution
An iterative approximate scheme for linear detection is introduced, involving matrix inversion, which reduces the number of operations and latency by converging in fewer iterations, allowing for parallel computation and hardware reuse.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Cholesky decomposition is used for exact linear detection, then detection accuracy is improved, but computational complexity increases cubically with the number of transmitting antennas
Solution Approach 1:
The patent changes the parameter of detection precision by using approximate linear detection methods instead of exact Cholesky decomposition. This allows the system to operate with controlled precision loss in exchange for dramatically reduced computational complexity, making massive MIMO detection feasible with practical hardware resources.
Solution Approach 2:
The patent employs low-complexity approximate detection algorithms that can be implemented with simpler, more cost-effective hardware. These approximate methods sacrifice some detection accuracy but provide sufficient performance for practical applications while enabling hardware-efficient implementations.
2Measurement precision
If Cholesky decomposition pipeline is used, then exact detection is achieved, but pipeline latency increases due to sequential execution of 3 stages
Solution Approach 1:
The patent segments the detection process into independent parallelizable operations by using iterative approximate methods. Unlike the sequential 3-stage Cholesky decomposition, the iterative approach allows multiple operations to be performed in parallel across different antenna elements and iteration steps, significantly reducing pipeline latency.
Solution Approach 2:
The patent performs preliminary computations that can be executed in parallel before the final detection step. By pre-computing correlation matrices and other intermediate values that can be calculated independently, the system reduces the critical path latency of the detection pipeline.
3Measurement precision
If Cholesky decomposition is used, then exact linear detection is achieved, but data throughput is limited due to high computational requirements
Solution Approach 1:
The patent changes the computational parameter from exact to approximate detection, enabling higher data throughput. The approximate methods require fewer floating-point operations per detection, allowing the system to process more data streams simultaneously and increase overall productivity while maintaining acceptable detection accuracy.
Solution Approach 2:
The patent applies partial action by using a limited number of iterations in the approximate detection algorithm. Instead of performing the complete exact decomposition, the system performs a sufficient number of iterative steps to achieve acceptable accuracy, then stops early to maximize throughput for time-varying channels where channel conditions change faster than the detection can converge.
Data Source
AI summary
Per given time instance, K samples b are acquired from a signal r, which is based at least on K transmitted symbols x and a transfer matrix H of a communication channel, and a linear detection matrix A of a size K×K is acquired, which is based at least on the transfer matrix H (S101). For the K samples b and the linear detection matrix A, at most K(K−1) tentative parameters b{tilde over ( )} and at most K(K−1) tentative parameters A{tilde over ( )} are iteratively calculated (S102). It is checked whether or not the tentative parameters b{tilde over ( )} and A{tilde over ( )} have converged (S103). If b{tilde over ( )} and A{tilde over ( )} have converged, K estimation values x{circumflex over ( )} are decided for the K transmitted symbols x based on b{tilde over ( )} and A{tilde over ( )} (S104). If b{tilde over ( )} and A{tilde over ( )} have not converged, it is returned to the iteratively calculating b{tilde over ( )} and A{tilde over ( )} for the K samples b.