I would like to thank the authors for their interesting work entitled “Addressing Preprocessing for Spectrum Sensing Using Image Processing”. However, I have a few questions regarding the study. Please note that the following comments are intended solely to initiate a constructive discussion about your paper and, hopefully, contribute to advancements in the field:
The manuscript utilizes bilateral Gaussian filtering for spectrogram denoising, which is indeed a well-established technique. However, its effectiveness remains unconvincing without a comparative analysis against more advanced state-of-the-art deep learning methods, such as GANs or U-Net architectures. Such a comparison would substantiate the claim of its superiority or suitability over alternative approaches. Furthermore, the study heavily relies on synthetic spectrogram data generated via MATLAB, raising concerns about the method’s practical applicability to real-world signals. Without empirical validation using actual spectrogram data, the robustness and real-world relevance of the findings may be significantly compromised.
The presented simulations are confined to a limited range of noise environments, primarily focusing on AWGN channels. However, real-world conditions are far more complex, often involving challenges such as fading channels, shadowing, and multi-path interference. By neglecting these scenarios, the paper risks overstating the general applicability of its proposed approach in practical spectrum sensing applications. Expanding the scope of testing to include such conditions would enhance the reliability and broader relevance of the study’s conclusions.
Looking forward to hearing from you.
