ESP32 CSI for localization / gesture recognition

[reneprins]

Hi,

I'm looking for a solution and hopefully someone can give me an answer. I want to develop a number of edge devices where I can very accurately determine what the movements are in a room, that's how I ended up on the Atheros CSI and Linux 802.11n CSI tool. Now my search has ended on the ESP32 chip that can provide me with CSI (channel state information) data. The tools from Atheros and Linux are pretty accurate to get CSI data because they have multiple antennas, among other things.

Now I was wondering, is the ESP32 chip accurate / reliable enough to do a similar measurement like other tools with the Linux 802.11n CSI tool and is it sufficient in terms of range for gesture recognition?

As an example: If I make a number of edge devices based on the ESP32 chip and I put them throughout my house. Would I theoretically have reach through all the rooms where I put an edge device where I can get the CSI data (of course depending on the setup, etc.)

I have already looked at the following github page: https://github.com/jonathanmuller/ESP32 ... ation-CSI-

If there are people who have experimented with this before, I would also like to hear it.

Thank you so much for your helps.

[ricksondpenha]

Hi, I was researching on this too, Esp32 doesnt have multiple antennas so i dont know how accurate the data will be, in addition theres esp wifi mesh available, may be we could stream the csi data from the the mesh connected devices?

Then theres tensorflow lite available on esp32, which can do further inferencing on the esp32 itself. This would be an interesting project, If esp32 can do localisation and gesture recognition using human activity detection algorithms, this would be a game changer.

[fatihoe]

Hi Guys,

I am also doing indoor localization. If you gained some knowledge, can you share it to me fatih.oezdemir@tum.de .
I want to get AGC info for the CSI Packets. Do you know how to get them? We can also work together on it

Greetings

[Efisio]

Hello

I'm working on a similar project, the idea is to use CSI as a complement to a passive (or connectionless) sensing solution. This is my understanding (here is wher I need your feedback) that CSI is computed for each packet for each AP-station connection is that it?

I present the general configuration of my tests: I have 2 to 3 ESP sniffing CSI packets from a near AP, each of them at different position since I should have mostly the same values anywhere. All this happen over a single channel.

As you said ESP32 (off the shelf ones) has only 1 antenna I am trying simply to identify the CSI typical values for an empty room and eventually any change on amplitude (I'm not ready to go for the phase ).

You can see, if the preambles mentioned are correct, even if I am able to recognize human presence, I won't be able to know where is this person as I only have sender's MAC address. I tried to use set an ESP as AP but I got a very poor range.

Here are some of my references
https://academic.oup.com/jcde/article/d ... 48/5837600
http://tns.thss.tsinghua.edu.cn/wifirad ... CC2017.pdf

[ZhangBinbin@qq.com]

Hello，I read your references carefully.I want to collect CSI by using Wi-ESP，but I don't know how to operate it. Could you give me some help? Thank you.