Dear Kitoo,
I am attempting to replicate and adapt your framework using an ADALM-Pluto SDR (instead of the USRP X310). Since Pluto has lower sampling rates and bandwidth, I would greatly appreciate your guidance on a few technical points to ensure my implementation aligns with your methodology:
Dataset – Is it possible to access the raw IQ or spectrogram dataset used in your experiments? If sharing is not possible, could you provide detailed guidelines on how you structured your dataset collection and labeling?
Spectrogram preprocessing – Could you confirm the final spectrogram resolution you used (e.g., 224×224, 640×640) and whether normalization was applied globally or per sample?
YOLO setup – Which YOLO version and input size did you use for the detection stage, and what were the key hyperparameters (batch size, epochs, learning rate)?
ViT classifier – Which ViT variant did you use (e.g., ViT-L/16) and what fine-tuning strategy (frozen layers vs full fine-tuning)?
SNR augmentation – Did you inject AWGN in the IQ domain or spectrogram domain? What was the SNR range and step size used for augmentation?
Evaluation protocol – How were accuracy and detection metrics computed across different SNR bins? Did you average over multiple random noise realizations?
Hardware adaptation – Since Pluto provides ~61.44 MS/s with narrower bandwidth compared to the USRP X310, do you foresee any limitations or modifications necessary to adapt your framework to Pluto?
Implementation details – Are the training scripts, configurations, or pretrained weights for YOLO and ViT available? If not, could you kindly share your recommended hyperparameters?
Your guidance would be extremely valuable in helping me replicate your excellent work on more accessible hardware, making RFUAV-style approaches more widely usable.
This is my email: waseemarshad94@gmail.com , if it is possible to share the scripts for recording and analysis.
Thank you for your time and for this important contribution to UAV detection research. I look forward to your response.
Regards
Dear Kitoo,
I am attempting to replicate and adapt your framework using an ADALM-Pluto SDR (instead of the USRP X310). Since Pluto has lower sampling rates and bandwidth, I would greatly appreciate your guidance on a few technical points to ensure my implementation aligns with your methodology:
Dataset – Is it possible to access the raw IQ or spectrogram dataset used in your experiments? If sharing is not possible, could you provide detailed guidelines on how you structured your dataset collection and labeling?
Spectrogram preprocessing – Could you confirm the final spectrogram resolution you used (e.g., 224×224, 640×640) and whether normalization was applied globally or per sample?
YOLO setup – Which YOLO version and input size did you use for the detection stage, and what were the key hyperparameters (batch size, epochs, learning rate)?
ViT classifier – Which ViT variant did you use (e.g., ViT-L/16) and what fine-tuning strategy (frozen layers vs full fine-tuning)?
SNR augmentation – Did you inject AWGN in the IQ domain or spectrogram domain? What was the SNR range and step size used for augmentation?
Evaluation protocol – How were accuracy and detection metrics computed across different SNR bins? Did you average over multiple random noise realizations?
Hardware adaptation – Since Pluto provides ~61.44 MS/s with narrower bandwidth compared to the USRP X310, do you foresee any limitations or modifications necessary to adapt your framework to Pluto?
Implementation details – Are the training scripts, configurations, or pretrained weights for YOLO and ViT available? If not, could you kindly share your recommended hyperparameters?
Your guidance would be extremely valuable in helping me replicate your excellent work on more accessible hardware, making RFUAV-style approaches more widely usable.
This is my email: waseemarshad94@gmail.com , if it is possible to share the scripts for recording and analysis.
Thank you for your time and for this important contribution to UAV detection research. I look forward to your response.
Regards