Abstract:
Specific Emitter Identification (SEI) refers to the identification of transmitters based on their unique
characteristics, known as RF fingerprints. SEI can be performed using two main methods: manual
feature-based and deep learning-based approaches. In this project, we are developing a novel SEI
algorithm that uniquely combines both approaches.
First, the signal is acquired using a software-defined radio. The received signal is then passed through
the SEI algorithm, which extracts RF fingerprints from it. The signal is decomposed into five modes
using Variational Mode Decomposition (VMD). From each mode, two types of features are extracted:
RF-DNA and time-frequency spectrograms. We refer to this combination of RF-DNA with VMD as
modified RF-DNA, and it serves as the input to the XGBoost classifier, which classifies the signal
among known classes.
If an unknown transmitter is detected, it is sent to the Siamese Neural Network (SNN). The SNN uses
both modified RF-DNA and time-frequency spectrograms to convert the inputs into a shared
representation space using a 2-channel Convolutional Neural Network (CNN) with bimodal feature
fusion. The similarity scores are then calculated by comparing the input with other signals. If a match
is found, the transmitter is labeled accordingly; otherwise, it is assigned a new label as an unknown
transmitter
