ICACT20260241 Question.1
Questioner: tranthithanhbg1995@gmail.com 2026-02-09 오전 10:32:50 |
ICACT20260241 Answer.1
Answer by Auhor md.t.a.812@ms.saitama-u.ac.jp 2026-02-09 오전 10:32:50
Chrome  Click!! |
| Why did you choose 4-QAM signals and the SNR range from −19 to −11 dB for your dataset? |
We use a 4-QAM signal since it provides an applicable yet practical signal model, while the low SNR range (−19 to −11 dB) challenges the ML models in conditions where spectrum sensing is most difficult—thereby offering a meaningful comparison of classification and regression algorithms for real-world cognitive radio applications. |
ICACT20260241 Question.2
Questioner: tranthithanhbg1995@gmail.com 2026-02-09 오전 10:35:01 |
ICACT20260241 Answer.2
Answer by Auhor md.t.a.812@ms.saitama-u.ac.jp 2026-02-09 오전 10:35:01
Chrome Click!! |
| Among all the tested models, which one would you recommend for real-time spectrum sensing and why? |
We will recommend the BLC (Binary Linear Classifier with Logistic Learner), which provides the best balance of accuracy, speed, and resource efficiency for real-time spectrum sensing, especially in low-SNR environments where computational constraints are as important as detection performance. |
ICACT20260241 Question.3
Questioner: tranthithanhbg1995@gmail.com 2026-02-09 오전 10:36:52 |
ICACT20260241 Answer.3
Answer by Auhor md.t.a.812@ms.saitama-u.ac.jp 2026-02-09 오전 10:36:52
Chrome Click!! |
| How does the performance of machine learning–based spectrum sensing compare with traditional energy detection at low SNR? |
At low SNR, machine learning–based spectrum sensing significantly outperforms traditional energy detection (ED). While ED relies on a fixed threshold and suffers from poor detection reliability and high false alarm rates in noisy conditions, ML models such as neural networks (NN), support vector machines (SVM), and logistic-based classifiers adapt to the non-linear characteristics of signals and noise. They achieve higher accuracy, better probability of detection, and lower probability of false alarm without needing a predefined threshold. This makes ML-based approaches more robust and reliable for spectrum sensing in challenging low-SNR environments. |
ICACT20260241 Question.4
Questioner: tranthithanhbg1995@gmail.com 2026-02-09 오전 10:39:44 |
ICACT20260241 Answer.4
Answer by Auhor md.t.a.812@ms.saitama-u.ac.jp 2026-02-09 오전 10:39:44
Chrome Click!! |
| Do you think the trained models can generalize well to other modulation schemes or real-world channel conditions? |
While the extracted features (energy and standard deviation of FFT components) are somewhat general, the models performance could degrade if the statistical characteristics of the input data change significantly. To ensure robustness, the models would likely require retraining or fine-tuning with diverse, real-world datasets that include various modulations, channel impairments, and operational conditions. |
ICACT20260241 Question.5
Questioner: arjunvirvashisht@gmail.com 2026-02-10 오전 2:07:03 |
ICACT20260241 Answer.5
Answer by Auhor md.t.a.812@ms.saitama-u.ac.jp 2026-02-10 오전 2:07:03
Chrome Click!! |
| What is the next planned step for this research project? |
The next step for this project will probably include testing more realistic and varied wireless channel models (like fading channels such as Rayleigh or Rician), trying out more modulation schemes beyond 4-QAM, and gathering or creating real-world spectrum data to confirm the results. Finally, we will attempt to use various deep learning techniques and compare their performance with that of machine learning models. |
ICACT20260241 Question.6
Questioner: hominpark@etri.re.kr 2026-02-10 오전 11:49:57 |
ICACT20260241 Answer.6
Answer by Auhor md.t.a.812@ms.saitama-u.ac.jp 2026-02-10 오전 11:49:57
Chrome Click!! |
| You used simulation data based on an AWGN channel generated in MATLAB. Do you expect the SVM Regression model, which showed the highest performance, to remain effective in real wireless environments where complex interferences like fading or shadowing exist? |
By carefully retraining the model using various real-world or realistic simulated data that accounts for these channel effects, along with possibly adjusting features or adapting to the specific area, SVM Regression could still work well. In practice, the model must be validated under conditions such as Rayleigh or Rician fading, impulsive noise, and co-channel interference to ensure its robustness before being deployed in real cognitive radio systems. |
