Physical-Layer Fingerprinting of LoRa devices using Supervised and Zero-Shot Learning

This dataset contains all raw signals (complex float I/Q samples) used in the LoRa fingerprinting experiments of the paper entitled "Physical-Layer Fingerprinting of LoRa devices using Supervised and Zero-Shot Learning". There are 4 databases included: lora1msps, lora2msps, lora5msps, and lora10msps. Each document in the databases is a symbol extracted from a 4-byte random payload LoRa frame, transmitted by a RN2483 radio and received by a USRP B210 sampling at a rate corresponding to the database name. A total of 22 different transmitters were used. For more information, please consult the paper. The document structure is as follows:

_id: Unique MongoDB document ID

chirp: Base 64 encoded binary float complex I/Q data

field: Symbol location inside a LoRa frame

tag: Name of the device that sent the frame

date: Time and date of reception

fn: Frame number

rand: Random number for sorting

How to import

Extract the tar archive. Inside the directory, run the following command to import the lora2msps database:

mongorestore --gzip -d lora2msps ./lora2msps

This process can be repeated for each dataset. Alternatively, all datasets can be imported automatically by executing:

mongorestore --gzip .

How to use

After the data has been imported, an experiment can be run by simply providing the corresponding config file to tf_train (see https://github.com/rpp0/lora-phy-fingerprinting), e.g.:

./tf_train.py train conf/experiment_lora2msps_mlp.conf