Deep Learning-Aided Multicarrier Systems

Thien Van Luong, Youngwook Ko, Michail Matthaiou, Ngo Anh Vien, Minh-Tuan Le, Vu-Duc Ngo

Abstract

This paper proposes a deep learning (DL)-aided multicarrier (MC) system operating on fading channels, where both modulation and demodulation blocks are modeled by deep neural networks (DNNs), regarded as the encoder and decoder of an autoencoder (AE) architecture, respectively. Unlike existing AE-based systems, which incorporate domain knowledge of a channel equalizer to suppress the effects of wireless channels, the proposed scheme, termed as MC-AE, directly feeds the decoder with the channel state information and received signal, which are then processed in a fully data-driven manner. This new approach enables MC-AE to jointly learn the encoder and decoder to optimize the diversity and coding gains over fading channels. In particular, the block error rate of MC-AE is analyzed to show its higher performance gains than existing hand-crafted baselines, such as various recent index modulation-based MC schemes. We then extend MC-AE to multiuser scenarios, wherein the resultant system is termed as MU-MC-AE. Accordingly, two novel DNN structures for uplink and downlink MU-MC-AE transmissions are proposed, along with a novel cost function that ensures a fast training convergence and fairness among users. Finally, simulation results are provided to show the superiority of the proposed DL-based schemes over current baselines, in terms of both the error performance and receiver complexity.

Introduction

Multicarrier modulation (MCM), with OFDM being the primary example, serves as the backbone of modern wireless standards like 5G and Wi-Fi. Traditionally, these systems are designed using "hand-crafted" components—mathematical models for modulation and demodulation optimized in isolation. However, these classical designs often fail to reach the global optimum for specific, complex fading channels. While recent "model-driven" deep learning approaches have attempted to improve performance, they typically incorporate standard channel equalizers (like Zero Forcing) into the architecture. This research argues that such constraints limit the neural network's potential to discover more efficient ways to harness the diversity inherent in wireless channels.

The Multicarrier Autoencoder (MC-AE) Architecture

The proposed Multicarrier Autoencoder (MC-AE) adopts a fully data-driven approach. The system is modeled as an end-to-end autoencoder where the Encoder (transmitter) maps data bits to complex symbols, and the Decoder (receiver) maps the noisy, faded signals back to bits.

The architecture of the proposed fully data-driven Multicarrier Autoencoder (MC-AE).

A critical innovation in MC-AE is the direct feeding of raw Channel State Information (CSI) alongside the received signal into the decoder. By avoiding pre-defined equalizers, the network learns to perform equalization and detection simultaneously. This allows the system to optimize diversity and coding gains directly from the data. For high-data-rate scenarios, an embedding layer (Emb-MC-AE) is utilized to manage the computational complexity associated with one-hot encoding.

Multiuser Extensions and Fairness

The framework extends to multiuser scenarios (MU-MC-AE) for both uplink and downlink transmissions. To ensure fairness among users, a novel loss function was introduced. Standard Mean Squared Error (MSE) training can be "unfair" in multiuser settings, as the network might optimize for one user while ignoring another. The custom loss function includes a penalty for the variance between users' error rates, ensuring both fast training convergence and equitable performance across the network.

Performance Evaluation

The system was evaluated using synthetic data generated from Rayleigh fading and Additive White Gaussian Noise (AWGN) channels. Training typically involved $5 \times 10^4$ to $10^5$ samples, with evaluation on $10^6$ testing samples to ensure statistical significance.

Comparison of constellations for traditional OFDM-IM, S-OFDM-IM, and the proposed MC-AE.

The MC-AE demonstrated significant performance gains, achieving SNR improvements of 5 dB to 12 dB over state-of-the-art baselines like Spread OFDM with Index Modulation (S-OFDM-IM) at a Block Error Rate (BLER) of $10^{-2}$. As shown in the constellation visualizations, the AE learns a geometry where points are better separated than in traditional schemes, increasing the minimum Euclidean distance and reducing error rates. Beyond error performance, the MU-MC-AE showed a 60-fold speedup in multiuser uplink detection compared to Sparse Code Multiple Access (SCMA) using traditional Message Passing Algorithm (MPA) detectors.

Conclusion

The MC-AE framework represents a significant shift toward fully data-driven wireless communication. By removing the constraints of traditional equalizers and letting the DNN process raw channel coefficients, the system achieves superior error performance and computational efficiency. Future work will address the overhead of CSI estimation in highly mobile environments and scaling to handle massive connectivity in extremely overloaded 6G networks.

Summary updated

Abstract

This paper proposes a deep learning (DL)-aided multicarrier (MC) system operating on fading channels, where both modulation and demodulation blocks are modeled by deep neural networks (DNNs), regarded as the encoder and decoder of an autoencoder (AE) architecture, respectively. Unlike existing AE-based systems, which incorporate domain knowledge of a channel equalizer to suppress the effects of wireless channels, the proposed scheme, termed as MC-AE, directly feeds the decoder with the channel state information and received signal, which are then processed in a fully data-driven manner. This new approach enables MC-AE to jointly learn the encoder and decoder to optimize the diversity and coding gains over fading channels. In particular, the block error rate of MC-AE is analyzed to show its higher performance gains than existing hand-crafted baselines, such as various recent index modulation-based MC schemes. We then extend MC-AE to multiuser scenarios, wherein the resultant system is termed as MU-MC-AE. Accordingly, two novel DNN structures for uplink and downlink MU-MC-AE transmissions are proposed, along with a novel cost function that ensures a fast training convergence and fairness among users. Finally, simulation results are provided to show the superiority of the proposed DL-based schemes over current baselines, in terms of both the error performance and receiver complexity.
BibTeX
@article{luong2020deep,
  author = {Thien Van Luong and Youngwook Ko and Michail Matthaiou and Ngo Anh Vien and Minh-Tuan Le and Vu-Duc Ngo},
  title = {{Deep Learning-Aided Multicarrier Systems}},
  journal = {IEEE Transactions on Wireless Communications},
  year = {2020},
  doi = {10.1109/twc.2020.3039180},
}
Figure