Diffusion-Based Enhancement of Low-SNR Time- Frequency Signals

Notice

This is an unedited manuscript accepted for publication and provided as an Article in Press for early access at the author’s request. The article will undergo copyediting, typesetting, and galley proof review before final publication. Please be aware that errors may be identified during production that could affect the content. All legal disclaimers of the journal apply.

Year : 2026 | Volume : 17 | 02 | Page :
    By

    Pradnya Kharde,

  • Sunil Kharde,

  • Jyoti Lokhande,

  • Sangita Wamane,

  1. Assistant Professor, Department of Electronics & Telecommunication, Dr. Vitthalrao Vikhe Patil College of Engineering, Ahilyanagar, Maharashtra, India
  2. Assistant Professor, Department of Electronics & Telecommunication, Dr. Vitthalrao Vikhe Patil College of Engineering, Ahilyanagar, Maharashtra, India
  3. Assistant Professor, Department of Electronics & Telecommunication, Dr. Vitthalrao Vikhe Patil College of Engineering, Ahilyanagar, Maharashtra, India
  4. Assistant Professor, Department of Electronics & Telecommunication, Dr. Vitthalrao Vikhe Patil College of Engineering, Ahilyanagar, Maharashtra, India

Abstract

Traditional enhancing techniques are useless in low signal-to-noise ratio (LSNR) situations because noise drastically interferes with communication signals. Based on an enhanced DiffBIR model, this paper suggests a dual-stage signal improvement approach that combines diffusion with deep learning. By combining the Inception module for multi-scale feature extraction with the Pixel Fusion Attention (PFA) module for significant region highlighting, the model improves signal recovery in the time- frequency domain. Experiments show that the enhanced DiffBIR model works better than conventional techniques in terms of noise reduction and time-frequency characteristic retention, with a wide range © STM Journals 2021. All Rights Reserved P A G E 1 of potential applications in acoustic processing, radar, and communication. The DiffBIR method first converts communication signals into time-frequency representations using Short-Time Fourier Transform (STFT), enabling joint analysis of temporal and spectral characteristics. The enhancement framework consists of two stages. In the first stage, an improved restoration module based on SwinIR is employed, where standard convolution layers are replaced with multi-scale Inception modules to capture diverse frequency components effectively. Additionally, a Pixel Fusion Attention (PFA) mechanism is introduced to emphasize signal-dominant regions while suppressing noise at the pixel level, thereby improving structural preservation and feature clarity. In the second stage, a diffusion generation module refines the restored output by iteratively denoising latent representations, enabling recovery of high-frequency details and improving perceptual quality. A conditional guidance mechanism further enhances region-specific restoration, balancing global consistency and local detail reconstruction. Experimental results on synthetic datasets with multiple modulation schemes and noise types demonstrate that the proposed model significantly outperforms existing methods in terms of PSNR, SSIM, and LPIPS metrics. Moreover, it achieves superior performance in signal detection tasks, improving precision, recall, and F1-score across varying SNR levels. These findings demonstrate the suggested framework’s resilience, effectiveness, and usefulness for actual LSNR communication systems.

Keywords: DiffBIR model LSNR, Signal recovery, Time-frequency images Communication signals, Diffusion Models, Noise Suppression

How to cite this article:
Pradnya Kharde, Sunil Kharde, Jyoti Lokhande, Sangita Wamane. Diffusion-Based Enhancement of Low-SNR Time- Frequency Signals. Current Trends in Signal Processing. 2026; 17(02):-.
How to cite this URL:
Pradnya Kharde, Sunil Kharde, Jyoti Lokhande, Sangita Wamane. Diffusion-Based Enhancement of Low-SNR Time- Frequency Signals. Current Trends in Signal Processing. 2026; 17(02):-. Available from: https://journals.stmjournals.com/ctsp/article=2026/view=247556


References

  1. Li C, Guo H, Tong S, Zeng X, Cao Z, Zhang M, Yan Q, Xiao L, Wang J, Liu Y. NELoRa: Towards ultra-low SNR LoRa communication with neural-enhanced demodulation. InProceedings of the 19th ACM Conference on Embedded Networked Sensor Systems 2021 Nov 15 (pp. 56-68).
  2. Wang J, Huang S, Huo Z, Zhao S, Qiao Y. Bilateral enhancement network with signal-to- noise ratio fusion for lightweight generalizable low-light image enhancement. Scientific Reports. 2024 Nov 30;14(1):29832.
  3. Kumar R, Patil M. Improved the Image Enhancement Using Filtering and Wavelet Transformation Methodologies. Available at SSRN 4182372. 2022 Jul 22.
  4. Jiang K, Wang Q, An Z, Wang Z, Zhang C, Lin CW. Mutual retinex: Combining transformer and CNN for image enhancement. IEEE Transactions on Emerging Topics in Computational Intelligence. 2024 Mar 12;8(3):2240-52.
  5. Xiao H, Wang X, Wang J, Cai JY, Deng JH, Yan JK, Tang YD. Single image super- resolution with denoising diffusion GANS. Scientific Reports. 2024 Feb 21;14(1):4272.
  6. Lin X, He J, Chen Z, Lyu Z, Dai B, Yu F, Qiao Y, Ouyang W, Dong C. Diffbir: Toward blind image restoration with generative diffusion prior. InEuropean conference on computer vision 2024 Sep 29 (pp. 430-448). Cham: Springer Nature Switzerland.
  7. Xia B, Zhang Y, Wang S, Wang Y, Wu X, Tian Y, Yang W, Van Gool L. Diffir: Efficient diffusion model for image restoration. InProceedings of the IEEE/CVF international conference on computer vision 2023 (pp. 13095-13105).
  8. Zheng D, Wu XM, Yang S, Zhang J, Hu JF, Zheng WS. Selective hourglass mapping for universal image restoration based on diffusion model. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition 2024 (pp. 25445-25455).
  9. Yue Z, Loy CC. Difface: Blind face restoration with diffused error contraction. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2024 Jul 23;46(12):9991-10004.
  10. Liang J, Cao J, Sun G, Zhang K, Van Gool L, Timofte R. Swinir: Image restoration using swin transformer. InProceedings of the IEEE/CVF international conference on computer vision 2021 (pp. 1833-1844).
  11. Si C, Yu W, Zhou P, Zhou Y, Wang X, Yan S. Inception transformer. Advances in neural information processing systems. 2022 Dec 6;35:23495-509. © STM Journals 2021. All Rights Reserved P A G E 1
  12. Zhang K, Li Y, Liang J, Cao J, Zhang Y, Tang H, Fan DP, Timofte R, Gool LV. Practical blind image denoising via Swin-Conv-UNet and data synthesis. Machine Intelligence Research. 2023 Dec;20(6):822-36.
  13. Xia B, Zhang Y, Wang Y, Tian Y, Yang W, Timofte R, Van Gool L. Basic binary convolution unit for binarized image restoration network. arXiv preprint arXiv:2210.00405. 2022 Oct 2.
  14. Li Z, Liu Y, Chen X, Cai H, Gu J, Qiao Y, Dong C. Blueprint separable residual network for efficient image super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition 2022 (pp. 833-843).
  15. Zamir SW, Arora A, Khan S, Hayat M, Khan FS, Yang MH. Restormer: Efficient transformer for high-resolution image restoration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition 2022 (pp. 5728-5739).
  16. Conde MV, Choi UJ, Burchi M, Timofte R. Swin2sr: Swinv2 transformer for compressed image super-resolution and restoration. InEuropean conference on computer vision 2022 Oct 23 (pp. 669-687). Cham: Springer Nature Switzerland.
  17. Cao J, Liang J, Zhang K, Li Y, Zhang Y, Wang W, Gool LV. Reference-based image super-resolution with deformable attention transformer. InEuropean conference on computer vision 2022 Oct 23 (pp. 325-342). Cham: Springer Nature Switzerland.
  18. Wang Z, Cun X, Bao J, Zhou W, Liu J, Li H. Uformer: A general u-shaped transformer for image restoration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition 2022 (pp. 17683-17693).
  19. Yue Z, Wang J, Loy CC. Resshift: Efficient diffusion model for image super-resolution by residual shifting. Advances in neural information processing systems. 2023 Dec 15;36:13294- 307.
Ahead of Print Subscription Review Article
Volume 17
02
Received 10/04/2026
Accepted 24/06/2026
Published 25/06/2026
Publication Time 76 Days
59

Login


My IP

PlumX Metrics