Wifi_Activity_Recognition using LSTM

Latest dataset & Tensorflow code for IEEE Communication Magazine.
Title: A Survey on Behaviour Recognition Using WiFi Channel State Information

Work by Siamak Yousefi, Hirokazu Narui, Sankalp Dayal, Stefano Ermon, Shahrokh Valaee

Prerequisite

Tensorflow >= 1.0
numpy
pandas
matplotlib
scikit-learn

How to run

0. Download dataset from here
   -> Notice: Dataset size is ~4GB
   -> Please replace this fileannotation_sankalp_sitdown_6.csv instead of Dataset one.

1. "git clone" this repository.

2. Run the cross_vali_data_convert_merge.py
   -> This script makes csv files(input features & label) of each activity in "input_files" folder.　　

3. Run the cross_vali_recurrent_network_wifi_activity.py -> This script makes learning curve images & confusion matrix in a new folder.　　

Dataset

We collect dataset using Linux 802.11n CSI Tool.

The files with "input_" prefix are WiFi Channel State Information data.
-> 1st column shows timestamp.
-> 2nd - 91st column shows (30 subcarrier * 3 antenna) amplitude.
-> 92nd - 181st column shows (30 subcarrier * 3 antenna) phase.

The files with "annotation_" prefix are annotation data.

Jupyter notebook

PCA_STFT file visualize the data from .csv file. This code refers to CARM.