GM-LAMP with Residual Learning Network for Millimetre Wave MIMO Architectures

Abstract

Gaussian Mixture Learned Approximate Message Passing with Residual Learning Network (GM-LAMP ResNet) for Millimeter-Wave (mmWave) Massive Multiple-Input Multiple Output (MIMO) system is presented. In mmWave systems, channels are sparse in nature. The increased sparsity in mmWave massive MIMO system, increases computation in Orthogonal Matching Pursuit (OMP) and cannot achieve high estimation accuracy. The gap between OMP scheme and Simultaneously OMP (SOMP) is filled by usage of classical iterative algorithm, which gives low computational complexity. Approximate Message Passing (AMP) algorithm, along equivalent version named as learned AMP (LAMP), are realized by Deep Neural Network (DNN). But algorithms in use are not provided with lower estimation error and higher achievable rates. To improve the higher achievable rates and low estimation error, GM-LAMP with ResNet is used to find the channel estimation. The proposed algorithm attain low computational complexity compared to the existing algorithms.