Abstract
With the large number of antennas and subcarriers the overhead due to pilot transmission for channel estimation can be prohibitive in wideband massive multiple-input multiple-output (MIMO) systems. This can degrade the overall spectral efficiency significantly, and as a result, curtail the potential benefits of massive MIMO. In this paper, we propose a neural network (NN)-based joint pilot design and downlink channel estimation scheme for frequency division duplex (FDD) MIMO orthogonal frequency division multiplex (OFDM) systems. The proposed NN architecture uses fully connected layers for frequency-aware pilot design, and outperforms linear minimum mean square error (LMMSE) estimation by exploiting inherent correlations in MIMO channel matrices utilizing convolutional NN layers. Our proposed NN architecture uses a non-local attention module to learn longer range correlations in the channel matrix to further improve the channel estimation performance.We also propose an effective pilot reduction technique by gradually pruning less significant neurons from the dense NN layers during training. This constitutes a novel application of NN pruning to reduce the pilot transmission overhead. Our pruning-based pilot reduction technique reduces the overhead by allocating pilots across subcarriers non-uniformly and exploiting the inter-frequency and inter-antenna correlations in the channel matrix efficiently through convolutional layers and attention module.
Highlights
INTRODUCTIONMassive multiple-input multiple-output (MIMO) systems are considered as the main enabler of 5G and future wireless networks thanks to their ability to serve a large number of users
We propose a pilot reduction technique based on neural network (NN) pruning, which effectively reduces the pilot overhead by allocating pilots across subcarriers non-uniformly; fewer pilots are transmitted on subcarriers that can be satisfactorily reconstructed by the subsequent convolutional layers utilizing inter-frequency correlations
Note that we have presented the results for the outdoor scenario in Subsection V.B, but as the simulations revealed very similar results and conclusions for both the indoor and outdoor scenarios, we have included the results only for the indoor scenario in the subsequent subsections to avoid tedious discussions of similar results
Summary
Massive multiple-input multiple-output (MIMO) systems are considered as the main enabler of 5G and future wireless networks thanks to their ability to serve a large number of users. Deep learning (DL)-based approaches have been used for massive MIMO CSI acquisition and showed significant improvements in comparison with their counterparts based on sparsity and compressive sensing (refer to [8], [9], and references therein) In these works, neural network (NN) architectures are trained over large CSI datasets to learn complex distributions, structures, and correlations, and exploit them for data-driven pilot design [10], channel estimation [11]–[15], compression [16]–[21] and feedback [22], [23]. Our proposed NN architecture uses dense layers to design frequency-aware pilot signals followed by convolutional layers to learn the inherent correlations in the MIMO-OFDM channel, and to exploit them for efficient channel estimation.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
