Sage Journals: Discover world-class research

Abstract

In digital communication, data transmission is subject to distortion caused by physical characteristics such as channel multipath propagation and Doppler frequency shift, as well as noise interference. Among these, inter-symbol interference (ISI) induced by channel dispersion leads to energy overlap between adjacent symbols, severely limiting the reliability and rate of transmission. To enhance the reliability of digital communication and support high-speed transmission, it is essential to overcome the performance limitations of traditional equalization techniques in time-varying multipath channels and effectively suppress distortions such as intersymbol interference and Doppler frequency shift. This paper proposes a hybrid equalizer architecture combining a CNN-Transformer-GRU framework with a cyclic shift preprocessing technique. This model generates an 18-dimensional feature matrix by performing nine groups of cyclic shifts on the received signal, transforming the time-varying channel features into a spatial dimension representation. By combining CNN local feature extraction, Transformer global dependency modelling, and GRU dynamic temporal adjustment, efficient channel equalization is achieved. Experimental results carried out under the SUI-3 channel model indicate that, on the premise of the same bit error rate (BER), the proposed model achieves a signal-to-noise ratio (SNR) gain of approximately 1–8 dB and 1–6 dB in comparison with the traditional decision feedback equalizer (DFE) and the temporal model LSTM, respectively, and improves the performance by 1–3 dB compared with the CRNN-RES (Convolutional recurrent neural network with ResNet) model, while maintaining a robustness of 2–8 dB. The parallel computation of transformers improves training efficiency by 30%, while the GRU architecture reduces the number of parameters by 20%. This architecture provides a high-precision, low-complexity solution for high-speed dynamic channel equalization in 5 G/6G systems.

Keywords

digital communication channel equalization inter symbol interference deep learning hybrid neural network architecture self-attention mechanism

Get full access to this article

View all access options for this article.

References

Ospina

RSB

Hout

van der Heide

, et al. MDG and SNR estimation in SDM transmission based on artificial neural networks. J Lightwave Technol 2022; 40(15): 5021–5030.

Lee

. Deep learning-aided coherent direction-of-arrival estimation with the FTMR algorithm. IEEE Trans Signal Process 2022; 70: 1118–1130.

Tong

Wang

, et al. Deep learning compressed sensing-based beamspace channel estimation in mmWave massive MIMO systems. IEEE Wirel Commun Lett 2022; 11(9): 1935–1939.

Peng

Shi

. Deep-learning based channel estimation for OFDM systems with doubly selective channel. IEEE Commun Lett 2022; 26(9): 2067–2071.

Farhati

Feki

Shaiek

, et al. Deep-learning for generic blind-Joint channel equalization and power amplifier Post-distortion. IEEE Access 2023; 11: 104754–104762.

Qaddus

Hassan

Minhas

. An efficient radio frequency channel distribution in 5g heterogeneous cellular networks for avoiding cross-tier interference in macro and small cells. Electr Control Commun Eng 2018; 14 (1): 30–38.

Carrera

Vargas-Rosales

Azpilicueta

, et al. Comparative study of channel estimators for massive MIMO 5G NR systems. IET Commun 2020; 14 (7): 1175–1184.

Xizheng

Qingyang

Huanhuan

. CNN neural network temporal feature storage structure fusion for the visible channel equalization algorithm. Appl Opt 2023; 62 (35): 9238–9252.

Liu

Sun

Guobao

. A blind equalization algorithm for short-wave time-varying channels based on a convolutional recurrent neural network. J Wuhan Univ (Nat Sci Ed) 2021; 67 (03): 241–246.

10.

Zhang

. Rapidly varying channel State estimation employing deep learning techniques for predictive Analysis. In: 2024 IEEE 4th international conference on data science and computer application (ICDSCA). IEEE, 2024, p. 197.

11.

. Research on fading channel equalization techniques based on deep learning. In: Harbin Institute of Technology, 2022.

12.

Yang

Han

. Channel equalizer research based on deep neural Networks//2024 IEEE 4th international conference on data science and computer application (ICDSCA): IEEE, 2024, pp. 396–402.

13.

Chen

Zhuang

, et al. SupportNet: a deep learning based channel equalization network for multi-type multipath fading. Mobile Network Appl 2023; 29(6):1–14.

14.

Huang

Chen

, et al. Wireless channel estimation and equalization based on deep learning. In: 2024 10th IEEE international conference on intelligent data and security (IDS): NYC, 2024, pp. 47–52. DOI: 10.1109/IDS62739.2024.00016.

High-speed channel equalization based on a hybrid neural network architecture

Abstract

Keywords

Get full access to this article

References