Flexible Configurable Modular Neural Network-Based OFDM Receiver

Introduction. Orthogonal frequency division multiplexing (OFDM) is the dominant modulation scheme in mobile communications. OFDM systems should be capable of operating across a wide range of multipath fading channel conditions. The recent research focus in this field has been on the design of OFDM receivers based on machine learning, including artificial neural networks. Neural networks in such receivers are typically trained for one specific communication system configuration. This complicates the use of neural network-based receivers in real-world systems, thus rendering development of more flexible schemes highly relevant.
Aim. To obtain and optimize the structure of an OFDM receiver based on an artificial neural network and consisting of separate modules that can be combined depending on the configuration of the pilot signals and the modulation used.
Materials and methods. Computer simulation in the MATLAB environment.
Results. The proposed architecture of a neural network-based OFDM receiver uses a combination of two multilayer perceptrons, one of which implicitly implements channel state information estimation and equalization, and the other performs demodulation. The first perceptron forms intermediate representations of data symbols, for which there were no specific references during network training, while the instances of the second perceptron work with these representations for individual data symbols. Variants of the second perceptron were trained for three quadrature modulation (QAM) constellations: 4QAM, 16QAM, and 64QAM.
Conclusion. The proposed OFDM receiver for all considered modulation types provided error rates comparable to those of the baseline algorithms under favorable channel conditions (moderate delay spread with low Doppler spread) and outperformed baseline algorithms in severe conditions (channel with a large delay spread and high Doppler spread). Further research directions involve neural network-based generation of soft decisions of the demodulator and development of specialized layers of the neural network that would facilitate approximation of the necessary operations.

1. LTE; Evolved Universal Terrestrial Radio Access (E-UTRA); Physical channels and modulation (3GPP TS 36.211 version 18.0.1 Release 18). ETSI, 2024. Available at: https://www.etsi.org/deliver/etsi_ts/136200_136299/136211/18.00.01_60/ts_136211v180001p.pdf (accessed 09.06.2025).

2. G; NR; Physical channels and modulation (3GPP TS 38.211 version 18.2.0 Release 18). ETSI, 2024. Available at: https://www.etsi.org/deliver/etsi_ts/138200_138299/138211/18.02.00_60/ts_138211v180200p.pdf (accessed 09.06.2025).

3. G RAN – key building blocks for new 6G radio access networks. Available at: https://www.ericsson.com/ en/blog/2024/5/future-6g-radio-access-network-designchoices (accessed 04.12.2024).

4. Soltani M., Pourahmadi V., Mirzaei A., Sheikhzadeh H. Deep Learning-Based Channel Estimation. IEEE Communications Let. 2019, vol. 23, no. 4, pp. 652–655. doi: 10.1109/LCOMM.2019.2898944

5. Luan D., Thompson J. Low Complexity Channel Estimation with Neural Network Solutions. 25th Intern. ITG Workshop on Smart Antennas, French Riviera, France, 10– 12 Nov. 2021. IEEE, 2021, pp. 1–6.

6. Zhao S., Fang Y., Qiu L. Deep Learning-Based Channel Estimation with SRGAN in OFDM Systems. IEEE Wireless Communications and Networking Conf. (WCNC), Nanjing, China, 29 March–01 Apr. 2021. IEEE, 2021, pp. 1–6. doi: 10.1109/WCNC49053.2021.9417242

7. Mohammed A. S. M., Taman A. I. A., Hassan A. M., Zekry A. Deep Learning Channel Estimation for OFDM 5G Systems with Different Channel Models. Wireless Personal Communications. 2023, vol. 128, no. 4, pp. 2891–2912. doi: 10.1007/s11277-022-10077-6

8. Peng Q., Li J., Shi H. Deep Learning Based Channel Estimation for OFDM Systems with Doubly Selective Channel. IEEE Communications Let. 2022, vol. 26, no. 9, pp. 2067–2071. doi: 10.1109/LCOMM.2022.3187161

9. Soltani N., Cheng H., Belgiovine M., Li Y., Li H., Azari B., D'Oro S., Imbiriba T., Melodia T., Closas P., Wang Y., Erdogmus D., Chowdhury K. Neural Network-Based OFDM Receiver for Resource Constrained IoT Devices. IEEE Internet of Things Magazine. 2022, vol. 5, no. 3, pp. 158–164. doi: 10.1109/IOTM.001.2200051

10. Ye H., Li G. Y., Juang B.-H. Power of Deep Learning for Channel Estimation and Signal Detection in OFDM Systems. IEEE Wireless Communications Let. 2018, vol. 7, no. 1, pp. 114–117. doi: 10.1109/LWC.2017.2757490

11. Fischer M. B., Dörner S., Cammerer S., Shimizu T., Lu H., Brink S. T. Adaptive Neural NetworkBased OFDM Receivers. IEEE 23rd Intern. Workshop on Signal Processing Advances in Wireless Communication (SPAWC). 2022, pp. 1–5. doi: 10.1109/SPAWC51304.2022.9833970

12. Wang B., Dai H., Xu K., Sun Y., Zhang Y., Li P. A Signal Processing Method of OFDM Communication Receiver Based on CNN. Physical Communication. 2023, vol. 59, art. no. 102055. doi: 10.1016/j.phycom.2023.102055

13. Sergienko A. B., Apalina P. V., Lebedinskaya A. D. Reception of OFDM Signals in Narrow Subbands by a Neural Network-Based Receiver with Batch Reshape. IEEE 3rd Intern. Conf. on Problems of Informatics, Electronics and Radio Engineering (PIERE), Novosibirsk, 15–17 Nov. 2024. IEEE, 2024, pp. 70–75. doi: 10.1109/PIERE62470.2024.10804946

14. G; Study on channel model for frequencies from to 100 GHz (3GPP TR 38.901 version 16.1.0 Release 16). ETSI, 2020. Available at: https://www.etsi.org/ deliver/etsi_tr/138900_138999/138901/16.01.00_60/ tr_138901v160100p.pdf (accessed 09.06.2025).

15. LTE; Evolved Universal Terrestrial Radio Access (E-UTRA); User Equipment (UE) radio transmission and reception (3GPP TS 36.101 version 17.12.0 Release 17). ETSI, 2024. Available at: https://www.etsi.org/ deliver/etsi_ts/136100_136199/136101/17.12.00_60/ ts_136101v171200p.pdf (accessed 09.06.2025).

16. Kingma D. P., Ba J. Adam: A Method for Stochastic Optimization. Intern. Conf. on Learning Representations (ICLR). 2015, pp. 1–15. doi: 10.48550/arXiv.1412.6980

17. MATLAB 5G Toolbox, Documentation for nrChannelEstimate function. Available at: https://www.mathworks.com/help/5g/ref/nrchannelesti mate.html (accessed 04.12.2024).