Hybrid neural network architecture ResBiLSTM-BiGRU for efficient noise suppression in radio channels with digitally modulated signals

Objectives. The work sets out to develop an effective method for suppressing additive Gaussian noise in radio channels to ensure the high-quality reconstruction of digitally modulated signals at low signal-to-noise ratio (SNR) levels. 

Methods. To address this problem, a hybrid architecture is proposed that combines Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) recurrent neural networks. The ResBiLSTM-BiGRU model enables more stable training and improved reconstruction quality, as well as more accurate modeling of the temporal structure of signals. The performance of the model was evaluated using a multi-modulation dataset with various input SNR levels. 

Results. As confirmed by the experimental results, the proposed model achieves an average improvement of 2–3 dB in output SNR compared with the Relativistic Average Generative Adversarial Network (RaGAN) under identical simulation conditions, as well as a more than a 10-dB improvement as compared with the wavelet-transform-based method. The model is characterized by lower mean squared error values and higher correlation coefficients across the entire input SNR range from –10 to 10 dB. The peak SNR values also indicate high signal reconstruction quality, especially at low input SNR levels. Analysis of architectural parameters shows that the best performance is achieved with 128 hidden units and a deep ResBiLSTM-BiGRU architecture.

Conclusions. The obtained results confirm the effectiveness and practical applicability of the proposed ResBiLSTM-BiGRU hybrid architecture for suppressing additive Gaussian noise in radio channels and reconstructing digitally modulated signals, particularly under low-SNR conditions.

1. Wang W. Evaluation of Adaptive Filtering Algorithms in Signal Denoising: Insights from Simulation and Algorithmic Enhancement. Theoretical and Natural Science . 2025;101(1):61–69. https://doi.org/10.54254/2753-8818/2025.CH22293

2. Lei J., Zhou J., Jian Z., Liu H., Zhang L., Feng Y., Wu Z. A Wiener Filter Denoising Based Intelligent Modulation Recognition. In: 2022 IEEE/CIC International Conference on Communications in China ( ICCC ). IEEE. 2022. P. 88–93. https://doi.org/10.1109/ICCC55456.2022.9880789

3. Guo T., Zhang T., Lim E., López-Benítez M., Ma F., Yu L. A Review of Wavelet Analysis and Its Applications: Challenges and Opportunities. IEEE Access . 2022;10:58869–58903. https://doi.org/10.1109/ACCESS.2022.3179517

4. Akkaya S. Wavelet-Based Denoising Strategies for Non-Stationary Signals in Electrical Power Systems: An Optimization Perspective. Electronics . 2025;14(16):3190. https://doi.org/10.3390/electronics14163190

5. Facciolo G., Pierazzo N., Morel J.-M. Conservative Scale Recomposition for Multiscale Denoising. (The Devil is in the High Frequency Detail). SIAM J. Imaging Sci . 2017;10(3):1603–1626. https://doi.org/10.1137/17M1111826

6. Wahab M.F., Gritti F., C. O’Haver T. Discrete Fourier transform techniques for noise reduction and digital enhancement of analytical signals. TrAC Trends in Analytical Chemistry. 2021;143(4):116354. https://doi.org/10.1016/j.trac.2021.116354

7. Zhao S., Ma L., Xu L., Liu M., Chen X. A Study of Fault Signal Noise Reduction Based on Improved CEEMDAN-SVD. Appl. Sci. 2023;13(19):10713. https://doi.org/10.3390/app131910713

8. Wang H., Yao Y., Zhang G., Pan J., Gao L., Jin H., Wang C. Enhanced Noise Suppression in Partial Discharge Signals via SVD and VMD with Wavelet Thresholding. Modelling and Simulation in Engineering . 2024;2024:5676986. https://doi.org/10.1155/2024/5676986

9. Almazrouei E., Gianini G., Almoosa N., Damiani E. A Deep Learning Approach to Radio Signal Denoising. In: 2019 IEEE Wireless Communications and Networking Conference Workshop ( WCNCW ). IEEE. 2019. https://doi.org/10.1109/WCNCW.2019.8902756

10. Duan R., Chen Z., Zhang H., Wang X., Meng W., Sun G. Dual Residual Denoising Autoencoder with Channel Attention Mechanism for Modulation of Signals. Sensors . 2023;23(2):1023. https://doi.org/10.3390/s23021023

11. He S., Zhu L., Yao C., Wang L., Qin Z. A Novel Approach Based on Generative Adversarial Network for Signal Enhancement in Wireless Communications. Wireless Communications and Mobile Computing . 2022. https://doi.org/10.1155/2022/7050573

12. Peng L., Fang S., Fan Y., Wang M., Ma Z. A Method of Noise Reduction for Radio Communication Signal Based on RaGAN. Sensors . 2023;23(1):475. https://doi.org/10.3390/s23010475

13. Yao H., Zhou Q., Chen Z., Huang L., Xu D., Xuan Q. Denoising of Radio Modulation Signal Based on Deep Learning. In: Xuan Q., Xiang Y., Xu D. (Eds.). Deep Learning Applications . 2023. Ch. 7. P. 125–143. https://doi.org/10.1142/9789811266911_0007

14. Wu T., Chen Z., He D., Qian L., Xu Y., Tao M. CDDM: Channel Denoising Diffusion Models for Wireless Semantic Communications. IEEE Trans. on Wireless Commun . 2024;23(9):11168–11183. https://doi.org/10.1109/TWC.2024.3379244

15. Letafati M., Ali S., Latva-Aho M. Conditional Denoising Diffusion Probabilistic Models for Data Reconstruction Enhancement in Wireless Communications. IEEE Trans. on Machine Learning in Communications and Networking . 2025;3:133–146. https://doi.org/10.1109/TMLCN.2024.3522872

16. Ji Z., Wang S., Yang K., Zhang Q., Ye P. Transfer Learning Guided Noise Reduction for Automatic Modulation Classification. In: 2025 IEEE 102nd Vehicular Technology Conference ( VTC2025-Fall ). IEEE. 2025. https://doi.org/10.1109/VTC2025-Fall65116.2025.11310099

17. Chung J., Gulcehre C., Cho K.H., Bengio Y. Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling. arXiv preprint. arXiv:1412.3555. 2014. http://arxiv.org/abs/1412.3555

18. O’Shea T.J., Roy T., Clancy T.C. Over-the-Air Deep Learning Based Radio Signal Classification. IEEE J. Selected Topics in Signal Process . 2018;12(1):168–179. https://doi.org/10.1109/JSTSP.2018.2797022