Deep Learning for Signal Demodulation in Physical Layer Wireless Communications: Prototype Platform, Open Dataset, and Analytics

Hongmei Wang, Zhenzhen Wu, Shuai Ma, Songtao Lu, Han Zhang, Guoru Ding, Shiyin Li

Abstract

In this paper, we investigate deep learning (DL)-enabled signal demodulation methods and establish the first open dataset of real modulated signals for wireless communication systems. Specifically, we propose a flexible communication prototype platform for measuring real modulation dataset. Then, based on the measured dataset, two DL-based demodulators, called deep belief network (DBN)-support vector machine (SVM) demodulator and adaptive boosting (AdaBoost) based demodulator, are proposed. The proposed DBN-SVM based demodulator exploits the advantages of both DBN and SVM, i.e., the advantage of DBN as a feature extractor and SVM as a feature classifier. In DBN-SVM based demodulator, the received signals are normalized before being fed to the DBN network. Furthermore, an AdaBoost based demodulator is developed, which employs the $k$-Nearest Neighbor (KNN) as a weak classifier to form a strong combined classifier. Finally, experimental results indicate that the proposed DBN-SVM based demodulator and AdaBoost based demodulator are superior to the single classification method using DBN, SVM, and maximum likelihood (MLD) based demodulator.

Introduction

In the physical layer of wireless communication, demodulation is the critical process of recovering original data from a received waveform. Traditionally, this relies on rigorous mathematical models, such as AWGN or Rayleigh fading, and requires accurate Channel State Information (CSI). However, real-world environments are complex, involving multipath fading, impulse noise, and hardware impairments that are difficult to capture in a single tractable equation.

While deep learning (DL) offers a "model-free" alternative that can learn these impairments directly from data, most existing research relies on synthetic, simulated datasets. This paper bridges the gap between theory and practice by introducing the first open dataset of real-world modulated signals and proposing robust DL architectures designed to handle the unpredictability of actual physical hardware.

Proposed DL-based Demodulators

The authors propose two primary hybrid architectures to improve demodulation performance in real-world channels:

1. DBN-SVM Hybrid Architecture

This architecture combines a Deep Belief Network (DBN) with a Support Vector Machine (SVM). The DBN, composed of stacked Restricted Boltzmann Machines, acts as an unsupervised feature extractor, reducing raw, noisy signals into high-level features. These features are then fed into a One-Versus-One SVM (OVO-SVM), which serves as a robust classifier to determine the transmitted symbol.

Architecture of the proposed DBN-SVM based demodulator.

2. AdaBoost-KNN Ensemble

This approach utilizes Adaptive Boosting to combine multiple "weak" k-Nearest Neighbor (KNN) classifiers into a single "strong" classifier. During training, the algorithm iteratively increases the weights of samples that were incorrectly demodulated, forcing subsequent KNNs to focus on the more ambiguous regions of the signal space. This method is particularly effective in low Signal-to-Noise Ratio (SNR) regions.

Real-world Dataset and Experimental Setup

A significant contribution of this work is the establishment of an open dataset of real modulated signals. Using a hardware prototype with Keysight signal generators and analyzers, data was collected in a 15m x 5m x 3m indoor office environment.

  • Modulation Schemes: BPSK and M-QAM (M = 4, 8, 16, 32, 64, 128, 256).
  • Signal Specs: Carrier frequency of 2.4 GHz and sampling rate of 100 MHz/s.
  • Volume: 10,000 transmit signal periods per modulation scheme, with SNRs ranging from 3 dB to 25 dB.

Results and Analysis

Experimental results demonstrate that the proposed hybrid models consistently outperform standalone DL models and traditional methods. Notably, the traditional Maximum Likelihood Detector (MLD) showed significant performance fluctuations in real-world settings due to its inability to account for "unmodeled" interferences present in actual hardware.

Demodulation accuracy vs. SNR for 16-QAM signals in a real-world environment.

As shown in the performance comparison for 16-QAM, both the DBN-SVM and AdaBoost models achieve nearly 100% accuracy once the SNR reaches 15-17 dB. The hybrid DBN-SVM approach proves superior to standalone DBN or SVM models, suggesting that while deep networks excel at feature extraction, specialized algorithms like SVMs are better suited for defining final decision boundaries.

Conclusion and Future Work

This study confirms that DL-based hybrid architectures can significantly enhance demodulation robustness in complex physical layer environments. Future work will focus on reducing the computational complexity for real-time implementation on low-power IoT devices and expanding the dataset to include high-mobility scenarios like vehicle-to-everything (V2X) communications.

Summary updated

Abstract

In this paper, we investigate deep learning (DL)-enabled signal demodulation methods and establish the first open dataset of real modulated signals for wireless communication systems. Specifically, we propose a flexible communication prototype platform for measuring real modulation dataset. Then, based on the measured dataset, two DL-based demodulators, called deep belief network (DBN)-support vector machine (SVM) demodulator and adaptive boosting (AdaBoost) based demodulator, are proposed. The proposed DBN-SVM based demodulator exploits the advantages of both DBN and SVM, i.e., the advantage of DBN as a feature extractor and SVM as a feature classifier. In DBN-SVM based demodulator, the received signals are normalized before being fed to the DBN network. Furthermore, an AdaBoost based demodulator is developed, which employs the $k$-Nearest Neighbor (KNN) as a weak classifier to form a strong combined classifier. Finally, experimental results indicate that the proposed DBN-SVM based demodulator and AdaBoost based demodulator are superior to the single classification method using DBN, SVM, and maximum likelihood (MLD) based demodulator.
BibTeX
@inproceedings{wang2019deep,
  author = {Hongmei Wang and Zhenzhen Wu and Shuai Ma and Songtao Lu and Han Zhang and Guoru Ding and Shiyin Li},
  title = {{Deep Learning for Signal Demodulation in Physical Layer Wireless Communications: Prototype Platform, Open Dataset, and Analytics}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=LpBPft_9HqN3},
}
Figure