General Questions about the CSI data

[olliec28]

Can anyone tell me what one row of the CSI data represents is it 1 row = 1 packet of data or does one row = 1 second for example. I am just trying to understand the data in order to feed it into a relevant ML algorithm. If anyone could tell me what algorithms they used as Im struggling to identify suitable ML algorithms at the moment. Thanks

[StevenMHernandez]

Each row indicates 1 packet of data. So if you are receiving packets at 100Hz, you should receive 100 rows per second. Each packet contains a certain number of subcarriers which are explained in (https://github.com/StevenMHernandez/ESP32-CSI-Tool/issues/12). I hope this helps.

[olliec28]

Each row indicates 1 packet of data. So if you are receiving packets at 100Hz, you should receive 100 rows per second. Each packet contains a certain number of subcarriers which are explained in (#12). I hope this helps.

Thanks. I'm assuming in order to find the rate at which im receiving packets, that's the Tick rate In your tutorial?

[StevenMHernandez]

That is a common question. I should probably update the documentation to mention this.

It is actually in ESP32 CSI Tool Config > *** for the active_sta (the transmitter).

[olliec28]

That is a common question. I should probably update the documentation to mention this.

It is actually in ESP32 CSI Tool Config > *** for the active_sta (the transmitter).

Perfect thanks for your reply. You have any ideas on suitable mL algorithms. By reading around it looks like Recurrent Neural networks are worth attempting? Thanks

[StevenMHernandez]

There are many different algorithms to try, it really depends on the task.

In our survey (Figure 4 and Figure 1: https://www.people.vcu.edu/~ebulut/COMST22_WiFi_Sensing_Survey.pdf), we showed that signal preprocessing steps can be very important for achieving good accuracy. Then, after signal processing, you can input the model into something like an RNN or CNN.

[olliec28]

Could I also ask you, how this imaginary part is calculated and if it is an estimation should the imaginary part be removed during data pre-processing so that only the real CSI data is used?

[StevenMHernandez]

This script shows how I convert the real/imaginary data to amplitude/phase (https://github.com/StevenMHernandez/ESP32-CSI-Tool/blob/master/python_utils/parse_csi.py). My work so far focuses on amplitude specifically. I would recommend that you also focus on amplitude initially. Good luck!

[StevenMHernandez]

Because this is not a specific bug for the codebase, I am going to "close" this issue. However, if you still have questions, feel free to leave additional comments here.