Analysis of various neural network algorithms for estimating radio signal duration with unknown amplitude, frequency and initial phase

Abstract

This study presents a comparative evaluation of several neural network–based approaches for estimating the duration of radio signals with arbitrary envelope shapes under conditions of a priori uncertainty in signal amplitude, carrier frequency, and initial phase. First, the statistical performance of a quasi-maximum likelihood algorithm for radio signal duration estimation is analyzed and employed as a benchmark for comparison. Subsequently, the performance of different neural network architectures is assessed under various signal processing scenarios. The results indicate that the accuracy of duration estimation depends not only on the selected neural network architecture but also on the strategy used to construct the training dataset. Among the investigated models, convolutional neural networks achieve the highest estimation accuracy, particularly when the input data are represented in the time–frequency domain using discrete time–frequency representations. In contrast, fully connected neural networks demonstrate limited capability in extracting informative signal features. Furthermore, incorporating an autoencoder as a preprocessing stage for white noise suppression significantly enhances the accuracy of radio signal duration estimation.

• quasi-maximum likelihood algorithm
• neural network–based algorithm
• duration estimation
• convolutional neural network
• multilayer perceptron
• autoencoder
• unknown frequency
• duration
• initial phase
• radio signal

Аннотация (на русском)

Анализ различных нейросетевых алгоритмов для оценки длительности радиосигнала с неизвестными амплитудой, частотой и начальной фазой

В работе выполнен сравнительный анализ ряда нейросетевых методов, предназначенных для оценки длительности радиосигналов с огибающей произвольной формы в условиях априорной неопределенности по амплитуде, начальной фазе и несущей частоте. Получены статистические характеристики квазиправдоподобной оценки длительности радиосигнала. Проведена оценка эффективности различных архитектур нейронных сетей при изменении условий обработки сигнала. Установлено, что точность оценки длительности определяется не только особенностями архитектуры нейронной сети, но и способом формирования обучающей выборки. Наиболее высокие показатели показали сверточные нейронные сети, особенно при использовании дискретного представления входных данных в частотно-временной области. В то же время полносвязная нейронная сеть характеризуется ограниченной способностью к извлечению информативных признаков сигнала. Показано что применение автоэнкодера в качестве средства предварительного подавления белого шума способствует повышению точности оценки длительности радиосигнала.

Funding

The study was supported by a grant from the Russian Science Foundation Project No. 24-19-00891, https://rscf.ru/project/24-19-00891/

For citation

References

1. Kulikov E.I., Trifonov A.P. Estimation of signal parameters against a background of interference. – M.: Sov. radio, 1978. (In Russian)
2. Korchagin Yu.E. Estimation of Radio Signal Duration with Unknown Amplitude and Phase // Radioengineering. – 2013. – No. 9. – P. 011-019. (In Russian)
3. Korchagin Yu.E. Estimation of the radio pulse duration with unknown phase // Radioelectronics and Communications Systems. – 2013. – Vol. 56. – No. 7. – P. 345-355.
4. Trifonov A.P., Shinakov Yu.S. Joint discrimination of signals and assessment of their parameters against a background of interference. – M.: Radio and Communications, 1986. (In Russian)
5. Zhashkova T.V., Renskova A.Yu. A Comparative Analysis of the Effectiveness of Convolutional and Recurrent Neural Networks for Complex Signal Identification Tasks // Modern Information Technologies. – 2025. – No. 42. – P. 127-130. (In Russian)
6. Zhou Y., Jiao X. Intelligent analysis system for signal processing tasks based on LSTM recurrent neural network algorithm // Neural Computing and Applications. – 2022. – Т. 34. – №. 15. – С. 12257-12269. https://doi.org/10.1007/s00521-021-06478-6
7. Jiao J. et al. An overview of wireless communication technology using deep learning // China Communications. – 2021. – Т. 18. – №. 12. – С. 1-36. https://doi.org/10.23919/JCC.2021.12.001
8. Korchagin Yu.E, Nguyen Van Thuy. Analysis of various neural network algorithms for estimating the amplitude of a radio signal with unknown initial phase, duration, and frequency // Journal of Radio Electronics. – 2026. – No. 5. https://doi.org/10.30898/1684-1719.2026.5.15 (In Russian)
9. Pan X., Wang Y., Qi Y. Artificial neural network model and its application in signal processing // Asian Journal of Advanced Research and Reports. – 2023. – Т. 17. – №. 1. – С. 1-8. https://doi.org/10.9734/AJARR/2023/v17i1459
10. Alkhafaji M.J.A. et al. A Lightweight 1D CNN for unified real-time communication signal classification and denoising in low-SNR edge environments // Buletin Ilmiah Sarjana Teknik Elektro. – 2025. – Т. 7. – №. 3. – С. 496-508. https://doi.org/10.12928/biste.v7i3.13789
11. Schneider N., Lermer M., Reich C. Investigating the performance of CNN feature extractor-based LSTM and GRU variants for time series classification // Procedia Computer Science. – 2024. – Т. 246. – С. 1070-1079. https://doi.org/10.1016/j.procs.2024.09.526
12. Yang T.Y. et al. A radio signal recognition approach based on complex-valued CNN and self-attention mechanism // IEEE Transactions on Cognitive Communications and Networking. – 2022. – Т. 8. – №. 3. – С. 1358-1373. https://doi.org/10.1109/TCCN.2022.3179450