Matrix Completion-Based Channel Estimation for MmWave Communication Systems With Array-Inherent Impairments

TL;DR

Numerical results demonstrate the advantages of the proposed MC-based channel estimator in terms of estimation performance, computational complexity, and robustness against array-inherent impairments over the orthogonal matching pursuit-based CS channel estimators.

Abstract

Hybrid massive MIMO structures with reduced hardware complexity and power consumption have been widely studied as a potential candidate for millimeter wave (mmWave) communications. Channel estimators that require knowledge of the array response, such as those using compressive sensing (CS) methods, may suffer from performance degradation when array-inherent impairments bring unknown phase errors and gain errors to the antenna elements. In this paper, we design matrix completion (MC)-based channel estimation schemes which are robust against the array-inherent impairments. We first design an open-loop training scheme that can sample entries from the effective channel matrix randomly and is compatible with the phase shifter-based hybrid system. Leveraging the low-rank property of the effective channel matrix, we then design a channel estimator based on the generalized conditional gradient (GCG) framework and the alternating minimization (AltMin) approach. The resulting estimator is immune to array-inherent impairments and can be implemented to systems with any array shapes for its independence of the array response. In addition, we extend our design to sample a transformed channel matrix following the concept of inductive matrix completion (IMC), which can be solved efficiently using our proposed estimator and achieve similar performance with a lower requirement of the dynamic range of the transmission power per antenna. Numerical results demonstrate the advantages of our proposed MC-based channel estimators in terms of estimation performance, computational complexity and robustness against array-inherent impairments over the orthogonal matching pursuit (OMP)-based CS channel estimator.

Figures (14)

• Figure 1: The fully connected hybrid system
• Figure 2: NMSE of the channel estimation in the ULA system with $N_t=128, N_r=32, K_t=16, K_r=4$, different training steps, ${\rm{PNR}}=20$ dB, and perfectly calibrated arrays, i.e., $\varkappa^t=\varkappa^r=0, \varrho^t=\varrho^r=0$.
• Figure 3: NMSE of the channel estimation in the ULA system with $N_t=128, N_r=32, K_t=16, K_r=4$, $512$ training steps, different $\rm PNR$s and perfectly calibrated arrays, i.e., $\varkappa^t=\varkappa^r=0, \varrho^t=\varrho^r=0$.
• Figure 4: NMSE of the channel estimation in the ULA system with $N_t=128, N_r=32, K_t=16, K_r=4$, $MS=512$ training steps, different phase error levels, ${\rm{PNR}}=20$ dB and $\varrho^t=\varrho^r=0$. The BS and MS phase error levels are assumed the same, i.e., $\varkappa^t=\varkappa^r$.
• Figure 5: NMSE of the channel estimation in the ULA system with $N_t=128, N_r=32, K_t=16, K_r=4$, $MS=512$ training steps, different gain error levels, ${\rm{PNR}}=20$ dB and $\varkappa^t=\varkappa^r=0$. The BS and MS gain error levels are assumed the same, i.e., $\varrho^t=\varrho^r$.