A Novel Blind mmWave Channel Estimation Algorithm Based on ML-ELM

Abstract

The millimeter wave (mmWave) with wide frequency spectrum can fulfill the demand of escalating communication system capacity. However, the large bandwidth of the mmWave channel leads to a large increase in the dimension of the received signal, which results to the increased computational complexity for channel estimation. In this letter, we propose a novel blind channel estimation algorithm based on Manifold Learning-Extreme Learning Machine (ML-ELM) in mmWave communication system. In the proposed ML-ELM algorithm, Manifold Learning (ML) is employed to reduce the feature dimension of received signal, and Extreme Learning Machine (ELM) with one-shot training is applied for the estimated Channel State Information (CSI). The ELM channel estimator is trained using the channel fading features extracted by ML algorithm. The CSI estimated by ML-ELM algorithm is more accurate, and the computational complexity satisfied the real-time requirements. Simulation results show that the proposed ML-ELM algorithm without any pilot aided achieves better MSE performance of CSI compared with the non-blind channel estimation algorithms in higher SNR scenarios, and better compared with pilot-based LS algorithm in lower SNR scenarios.