Pruning the Pilots: Deep Learning-Based Pilot Design and Channel Estimation for MIMO-OFDM Systems

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 curtail the potential benefits of massive MIMO. In this paper, we propose a deep neural network (DNN)-based scheme, consisting of convolutional and dense layers, for joint pilot design and downlink channel estimation for frequency division duplex (FDD) MIMO orthogonal frequency division duplex (OFDM) systems. We also propose an effective pilot reduction technique by gradually pruning less significant neurons from the dense neural network (NN) layers during training. Our proposed NN-based scheme exploits the frequency-specific features of the propagation environment to design more efficient pilots. It outperforms linear minimum mean square error (LMMSE) estimation by learning the channel statistics over the training data and exploiting it to estimate the channel without prior assumptions on the channel distribution. Finally, our pruning-based pilot reduction technique effectively reduces the overhead by allocating pilots across subcarriers non-uniformly; allowing less pilot transmissions on subcarriers that can be satisfactorily reconstructed by the subsequent convolutional layers utilizing inter-frequency correlations.

Deep Learning-Aided 6G Wireless Networks: A Comprehensive Survey of Revolutionary PHY Architectures

“…In order to increase the throughput considerably, authors of [193] addressed both CP and pilot reduction issues and demonstrated that it is possible to eliminate CP and pilot entirely by exploiting E2E learning. In [194], for frequency division duplex massive MIMO-OFDM systems, a NN-based combined downlink pilot design and channel estimation approach is presented. An efficient pilot reduction approach is also suggested for reducing pilot overhead and saving time-frequency resources for data transmission by progressively pruning less important neurons from dense layers.…”

“…The mentioned controlled interaction is realized with the adaptive meta elements on intelligent surfaces. It is aimed to obtain the optimum phase configuration [193] N/A CP&Pilot Overhead AE, Residual CNN [194] N/A Pilot Overhead CNN [195] N/A Pilot Overhead AE, CNN, DNN [196] ChanEstNet High Mobility CNN, BiLSTM [197] N/A High Mobility CNN, BiLSTM [198] ICINet High Mobility DNN, Residual CNN [199] Cascade-Net High Mobility Unfolding [200 and to optimize the wireless communication performance by manipulating the incoming signals thanks to these meta elements. Here, the incoming signals from the BS are controlled over-the-air in real-time and reflected to the receivers.…”

Channel estimation of OFDM in C-band communication systems under different distribution conditions

“…The channel estimation strategies for OFDM frameworks dependent on the pilot course of action are explored. The channel estimation dependent on brush type pilot course of action is considered through various calculations for both evaluating channels at pilot frequencies and inserting the channel [21]- [23]. The estimation of the channel at pilot frequencies depends on LS and LMS while the channel insertion is finished utilizing straight interposition, second solicitation inclusion, low-pass presentation, spline cubic interjection, also, time-space addition.…”

Deep Learning based Channel Estimation for Massive MIMO with Hybrid Transceivers

“…The aforementioned methods either suffer from unsatisfactory performance or high complexity, hence channel estimation algorithms with better performance-complexity tradeoffs are urgently required for practical HAD massive MIMO systems. Recently, deep learning (DL) has been successfully applied to many areas in wireless communication [18][19][20], including spectrum sensing [21], resource management [22][23][24], beamforming [25][26][27], signal detection [28][29][30], and channel estimation [31][32][33][34][35][36][37]. Thanks to the simple forward computation and the acceleration brought by dedicated hardware like graphics processing unit (GPU), the DL-based approaches are usually more computationally efficient than the conventional iterative optimization algorithms [23].…”

“…In [33] and [34], the pilot and channel estimator are jointly optimized for downlink massive MIMO with autoencoders, where the pilot matrix and channel estimator are modeled as the encoder and decoder, respectively. In [36] and [37], different attention modules are inserted to the deep neural networks to better exploit the channel features and improve estimation performance. Apart from channel estimation, DL can also be applied to channel tracking and prediction.…”