robotics404 / csil

The channel separation incremental learning algorithm for wireless device identification. With mathematical proofs of the accuracy of orthogonal memory representation in Artificial Neural Networks.

This is the public repository of our paper: Class-Incremental Learning for Wireless Device Identification in IoT, which is available HERE and IEEE Internet of Things Journal

More importantly, we mathematically proved and verified the effect orthogonal memory representation within artificial neural network.

We are delighted to know that recent advancement in neuroscience also shows the biological evidence of orthogonal memory representations. But we have totally different storylines.

Our raw dataset is available at https://ieee-dataport.org/documents/ads-b-signals-records-non-cryptographic-identification-and-incremental-learning. It's a raw ADS-B signal dataset with labels, the dataset is captured using a BladeRF2 SDR receiver @ 1090MHz with a sample rate of 10MHz.

Please goto IEEE Dataport for the dataset (adsb_bladerf2_10M_qt0.mat) and preprocessed data (adsb-107loaded.mat).

Sample code for data preprocessing and incremental learning are in ContinualLearning

The code that drives the discovery of Remark 1 is in the numerical simulation folder

Comparison of various incremental learning algorithms are in ContinualLearning/WorkStage