The minimum mean-squared error (MMSE) detector is widely used in massive multiple-input multiple-output communication systems for its suboptimal performance and simple deployment. However, the computational complexity of the MMSE detector is extremely high due to its matrix calculation when the number of antennas increases. In this paper, a learning vector quantization method is introduced to significantly accelerate specific matrix multiplication at the expense of acceptable performance decay and memory cost. Specifically, exact matrix multiplication is replaced by a series of simple operations involving comparison, querying the lookup table (LUT) and integer additions. The LUT saving the results of vector multiplication is constructed offline. When performing matrix multiplication online for signal detection, received signals are encoded into address indices for querying the LUT. Numerical results prove that the proposed method can achieve similar performance compared to exact matrix calculation for the MMSE detector with much lower computational complexity and acceptable memory cost.
MADDNESS Detector for MIMO Systems with Learning Vector Quantization
2024-06-24
782770 byte
Conference paper
Electronic Resource
English
Learning Vector Quantization with Alternative Distance Criteria
British Library Conference Proceedings | 1999
|Structural Identification Using Learning Vector Quantization Neural Network
British Library Online Contents | 2001
|Predictive residual vector quantization
IEEE | 1994
|