A.V. Eponeshnikov1, A.I. Sulimov2

1,2 Kazan (Volga Region) Federal University, Institute of Physics (Kazan, Russia)

1 sashah275@gmail.com; 2 asulimo@gmail.com

This paper presents a robust methodology for automatic detection and classification of partial (multipath) components in the impulse response of wireless radio channels under low signal-to-noise ratio (SNR) conditions. The proposed approach synergistically combines three stages: correlation-based matched filtering using Gold codes of length 2047 chips with BPSK modulation, adaptive wavelet denoising based on a modified VisuShrink algorithm applied separately to in-phase (I) and quadrature (Q) signal components, and machine learning–based classification using XGBoost with SHAP interpretability. The system operates reliably down to SNR = –10 dB, while other methods typically require SNR > +10 dB. Temporal resolution reaches 0.2–0.4 ns, equivalent to 0.1–0.2 chip periods at a symbol rate of 500 MHz.

The classifier was trained on a dataset comprising 1.35 million detected components extracted from 19,680 channel realizations, covering five standardized TDL profiles (A–E), RMS delay spreads from 0.1 to 0.9 µs, and SNR values from –10 to +10 dB. At a classification confidence threshold of 0.50, the method correctly identifies 50.1% of all multipath components, which collectively account for 90.5% of the total received signal power-demonstrating effective capture of energetically dominant paths.

Failure analysis reveals that 36.4% of missed components are due to the 17 dB registration threshold limit inherent to Gold code autocorrelation sidelobes, 3.9% are lost due to temporal resolution constraints (minimum resolvable delay separation of 3 ns), and 7.8% result from classification errors, rising to 14.3% when the decision threshold is increased to 0.95 for stricter false-alarm control. The mean absolute error in delay estimation decreases monotonically with per-ray SNR, ranging from ≈ 0. 9 ns at –35 dB to ≈ 0.2 ns at 0 dB. False alarm rates drop from 5.8% at SNR = –10 dB (threshold = 0.5) to <0.2% under high-SNR conditions (SNR = +10 dB) with a strict threshold (0.95).

Crucially, the method requires no environment-specific retraining, unlike many deep learning alternatives, and provides interpretable decisions via SHAP analysis—enabling physical validation of feature contributions (e.g., peak prominence, width, local SNR). This makes the approach suitable for safety-critical or verifiable applications such as physical-layer key generation, massive MIMO beam management, and high-precision time-of-arrival positioning.

Eponeshnikov A.V., Sulimov A.I. Method for automatic analysis of partial components of the impulse response of a multi-beam radio channel // Radiotekhnika. 2026. V. 90. № 8. P. 118−130. DOI: https://doi.org/10.18127/j00338486-202608-11

1. Pérez Fontán F., Mariño Espiñeira P. Modeling the wireless propagation channel: a simulation approach with MATLAB®. Modeling the wireless propagation channel. Wiley. 2008.
2. Huang J., Jiang T. Secret key generation exploiting Ultra-wideband indoor wireless channel characteristics: UWB, secret key generation, multipath relative delay, reciprocity. Security and Communication Networks. 2015. V. 8. Secret key generation exploiting Ultra-wideband indoor wireless channel characteristics. № 13. P. 2329-2337.
3. Zhang J., Duong T.Q., Marshall A., Woods R. Key generation from wireless channels: a review. IEEE Access. 2016. V. 4. P. 614-626.
4. Sui Y., Gao H., He Y., Jiang G. Local predictability analysis-based significant multipath component identification for OFDM systems. IEEE Wireless Communications Letters. 2024. V. 13. № 9. C. 2482-2486.
5. Nam W., Kong S.-H. Least-squares-based iterative multipath super-resolution technique. IEEE Transactions on Signal Processing. 2013. V. 61. № 3. C. 519-529.
6. Aboaba O.A., Chung K. Pathlet: A new wavelet for resolving the constituent components in a multipath channel. 2006 Asia-Pacific Conference on Communications. 2006. Pathlet. P. 1-5.
7. Quinquis A., Boulinguez D. Multipath channel identification with wavelet packets. IEEE Journal of Oceanic Engineering. 1997. V. 22. № 2. P. 342-346.
8. Artai H.A., Bedi J.S. Determination of multipath channel parameters using wavelet decomposition. Integrated Computer-Aided Engineering. 2001. V. 8. № 2. P. 119-133.
9. Niitsoo A., Edelhäußer T., Eberlein E., Hadaschik N., Mutschler C. A deep learning approach to position estimation from channel impulse responses. Sensors. 2019. V. 19. № 5. P. 1064.
10. Lee G., An S., Jang B.-J., Lee S. Deep learning for counting people from UWB channel impulse response signals. Sensors. 2023. V. 23. № 16. P. 7093.
11. Wang K., Yang C. Analysis of machine learning-based NLOS signal identification algorithm for UWB indoor localization using CIR waveform features. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 2024. V. XLVIII-4-2024. P. 705-710.
12. Li B., Zheng Q., Tian X., Yang M., Gui G., Jiang W., Lei H., Jiang J., Shu F., Elhanashi A., Saponara S. A survey of artificial intelligence enabled channel estimation methods: recent advance, performance, and outlook. Artificial Intelligence Review. 2025. V. 58. A Survey of Artificial Intelligence Enabled Channel Estimation Methods. № 6. P. 187.
13. Chen T., Guestrin C. XGBoost: a scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. San Francisco California USA: ACM. 2016. XGBoost. P. 785-794.
14. Hamilton R.I., Papadopoulos P.N. Using SHAP values and machine learning to understand trends in the transient stability limit. IEEE Transactions on Power Systems. 2024. V. 39. № 1. P. 1384-1397.
15. Zhang C., Zou X., Lin C. Fusing XGBoost and SHAP models for maritime accident prediction and causality interpretability analysis. Journal of Marine Science and Engineering. 2022. V. 10. № 8. P. 1154.
16. Shatilov A.Yu. Harakteristiki radiosignalov global'nyh sputnikovyh radionavigacionnyh sistem GLONASS, GPS, Galileo, Beidou i funkcional'nyh dopolnenij SBAS: uchebnoe posobie dlya studentov, obuchayushchihsya po napravleniyu «Radioelektronnye sistemy i kompleksy» [Elektronnyj resurs]. M.: MEI, 2016. 36 s. (in Russian).
17. Fedorovich T.E., Sergeevich F.V. Svojstva kodov golda i neodnoznachnost' v RSA. Radiotekhnicheskie i telekommunikacionnye sistemy. 2023. № 4. S. 41-50 (in Russian).
18. TR 138 901-V16.1.0-5G; Study on channel model for frequencies from 0.5 to 100 GHz (3GPP TR38.901 version 16.1.0 Release 16).
19. Abedi O., Yagoub M.C.E. Performance comparison of UWB Pulse modulation schemes under white gaussian noise channels.
20. Fourati W., Bouhlel M. Visushrink pretreatment for image compression. International Journal of Computer Applications. 2011. V. 23.