Wideband Millimeter-Wave Massive MIMO Channel Training via Compressed Sensing

In this work, a compressed sensing-aided wideband MIMO-OFDM channel training framework is proposed to reduce the training overhead in slowly-varying channels with frequency- and spatial-wideband (dual-wideband) effects. To combat the beam squint effect, a set of frequency-dependent array response matrices are constructed, enabling the recovery of the sparse beamspace channel from multiple observations across OFDM subcarriers, via multiple measurement vectors (MMV). A channel training algorithm (MMV-LS-CS) is proposed to estimate slowly-varying multipath channel parameters: MMV least squares (MMV-LS) is first used to estimate the channel on the previous beam index support, followed by MMV compressed sensing (MMV-CS) on the residual to estimate the time-varying multipath components. Finally, a channel refining algorithm is proposed to estimate the gains and time delays of the dominant channel paths jointly on pilot subcarriers. Numerical results show that MMV-LS-CS achieves more accurate and robust channel estimation than the state-of-the-art approach on slowly-varying dual-wideband MIMO-OFDM: given a moderate SNR of 20 dB, our algorithm attains text{NMSE}=0.15, as opposed to the state-of-the-art which attains text{NMSE}=0.43 in the same configuration. Besides, MMV-LS-CS necessitates text{SNR} =14 text{dB} to achieve the spectral efficiency of 6 bit/s/Hz/stream, while the state-of-the-art scheme needs text{SNR}=17 text{dB} to attain the same spectral efficiency.