SignFi icon indicating copy to clipboard operation
SignFi copied to clipboard

Query about the dataset, Is the dataset in the link has been preprocessed?

Open Guolin-Yin opened this issue 4 years ago • 3 comments

Dear Yongsen

I am really interested in your paper and your dataset, I am just wondering is this .mat that I can download after preprocessed or before preprocessed?

Guolin-Yin avatar Jul 18 '21 13:07 Guolin-Yin

The dataset is not preprocessed. It contains raw amplitude and phase using the get_csi.m function of the 802.11n CSI tool.

yongsen avatar Aug 05 '21 00:08 yongsen

The dataset is not preprocessed. It contains raw amplitude and phase using the get_csi.m function of the 802.11n CSI tool.

Thanks for replying, all the length of CSI the dataset is 200, I did you perform interpolate to the CSI stream? since, in reality, the CSI stream should be around 200, not extractly 200 for every single packet.

Guolin-Yin avatar Aug 07 '21 09:08 Guolin-Yin

The dataset is not preprocessed. It contains raw amplitude and phase using the get_csi.m function of the 802.11n CSI tool.

Thanks for replying, all the length of CSI the dataset is 200, I did you perform interpolate to the CSI stream? since, in reality, the CSI stream should be around 200, not extractly 200 for every single packet.

There is no interpolation. Instead, there might be downsampling for some CSI segments. The sampling frequency of raw CSI measurements is 200Hz (about 0.5us between two neighboring CSI packets). The time window length of different users varies from 1s to 2s. For 2s CSI segments, they will be downsampled to 100Hz to form a CSI tensor with the length of 200. Another thing is that some CSI segments might contain silent parts in the tail for some short gestures that can only fit into the beginning of the 1s or 2s time window.

yongsen avatar Aug 18 '21 00:08 yongsen