NilLlisterri / MQTT-FL

Getting Started

Create the secrets.h file in the node directory to define the WiFi SSID and password:

#define WIFI_SSID "XXX"
#define WIFI_PASSWORD "XXX"

Set the MQTT broker IP address modifying the MQTT_SERVER constant in node/config.h to your computer's IP.

Upload the code to the nodes:
pio run --target upload -e ttgo-lora32-v1 --upload-port com3; pio run --target upload -e ttgo-lora32-v1 --upload-port com20;

Install the required Python libraries:
pip install -r host/requirements.txt

Start the MQTT server:
docker-compose up -d

Start the experiment:
python host/experiment.py

Project structure

Host

For the host machine there are 3 important files under the host folder.

• fl_server.py connects to the MQTT server and communicates with the nodes to organize the FL rounds.
• node_manager.py communicates with the nodes via Serial port, and is used to create reproductble experiments. It handles the initialization of the nodes NN, sends training samples to the devices, etc.
• experiment.py is used to automate the experiments. It uses both previous files to start the FL server and then execute the experiment.

Nodes

• main.cpp is the main file and it initializes the LoRaMesher library and listens to the Serial port for the experiment commands.
• fl.h and fl.cpp contain the KWS application, encapsulated in a LoRaMesher app format.
• NN.h and NN.cpp are the files that contain the Neural Network.