This project aims to track locations of certain devices using Wi-Fi rssi measurements. We chose NodeMCU's to monitor Wi-Fi signals, but later we realized the signal strength measurements were very unstable. It is infeasible to use NodeMCU's for this project.
This project has many sub-applications. We will try to go through the steps to use this software as much as possible.
There two main steps:
- Training
- Production
Training step essentially uses 4 applications:
- android_data_labelling_client
- android_server
- data_labelling_mqtt_client
- node_software
node_software must be installed on NodeMCU's. We used 3 NodeMCU's in our example, but any number of NodeMCU's can be used. Every NodeMCU must be connected to a computer with android_server software installed on them. These computers and a mobile phone with android_data_labelling_client application installed must be on the same network.
In one of the machines, (or in an external machine on the same network) data_labelling_mqtt_client and mosquitto must be running.
android_server applications must be customized to have different id numbers on each PC. The port which NodeMCU is connected to and the ip address of the MQTT broker must also be defined in the program's constants.
After setting up everything, you may start using the Android application. In the first text field, the IP address and port of every machine on the network that runs android_server application must be written in the following format:
[host1]:[port1],[host2]:[port2],
On the second field, the MAC address of the target device (usually the mobile phone itself), must be typed in. Newer Android versions do not allow us to read MAC address of the phone directly, so we have to give it manually.
After this, we use the connect button to connect to the servers on the network. We may then set location of the tracked device, and then start capturing packets from the target device.
data_labelling_mqtt_client application writes the collected data into a file named db.sqlite. This file can be used to for the machine learning algorithms. Any tool can be used for the machine learning part. We have placed our machine learning code into the machine_learning directory.
Production uses 5 of the applications:
- android_server
- cloud_server
- frontend
- rpi_edge_server
- node_software
In the android_server application the script named only_serial_port.js is used. This script transfers the received packet information to the MQTT server.
rpi_edge_server sends 3 matched packets into the cloud for measurements.
cloud_server receives packets and stores predictions in memory, also sends the the predictions via socket.io.
frontend connects to the cloud_server, and receives predictions via socket.io.
- Node/android_data_labelling_client: Small native Android application used for data labelling.
- Node/android_server: Node.js application that is used to connect Android client to computer. Used open source libraries:
- Node/data_labelling_mqtt_client: Node.js application to store received data on MQTT to an sqlite3 database.
- Node/node_software: The code that runs on NodeMCU's.
- Node/rpi_edge_server: The code that runs on Raspberry Pi, on production. This is a Node.js application that listens to MQTT and sends received information to cloud.
- mqtt (https://github.com/mqttjs/MQTT.js)
- axios (https://github.com/axios/axios)
- socket.io (https://socket.io/)
- UI/frontend: This is the UI of the application, it is written using the React.js framework.
- Cloud/cloud_server: This is the piece of software that runs on cloud. This software receives measurements from the nodes and sends predictions.
- react.js (https://reactjs.org/)
- socket.io (https://socket.io/)
- ML: This directory shows what we have done for machine learning purposes.
- scikit-learn (https://scikit-learn.org/stable/index.html)
- numpy (https://www.numpy.org/)
- pandas (https://pandas.pydata.org/)
| Component | Name | Action |
|---|---|---|
| Embedded HW (monitoring) x3 | NodeMCU | Integration |
| Embedded HW (communication) x3 | NodeMCU | Integration |
| Gateway + Edge Computing Unit | Raspberry Pi | Integration |
| Embedded HW (communication) | Realtek USB Wi-Fi Adapter | Integration |
| Development UI (Android App for Data Collection) | Local, (Native Android) | Development; we receive data labels from the application. |
| Embedded SW (Server Software for Android Application) | Local, (node.js) | Development; collects and sends data to MQTT broker |
| Connectivity | mosquitto | Integration & Development; Used both in production and in development(data collection) of the solution |
| Storage | sqlite3 | Integration & Development; Used both in production and in development of the application |
| Embedded SW | Local, (Lua script for NodeMCU) | Configuration; Provides interface to configure NodeMCU's on start-up |
| Cloud Service | DigitalOcean | Integration |
| Software (ML Tools) | Various Python Libraries (scikit-learn, pandas, numpy) | Integration & Development; Trained a machine learning model on development, used that model in production |
| Cloud SW | Local (node.js server application) | Integration |
| Web Client | Local (react.js application) | Integration |