Abstract:
In modern electronic warfare and spectrum monitoring scenarios, the increased complexity of electromagnetic spectrum has posed serious challenges for early warning receivers. One of the most critical tasks is to identify and separate multiple radiation sources emitting overlapped waveform signatures. Conventional Signal Processing techniques often fails to adapt and learn in a dense and noisy environment. This thesis explores the role of Artificial Intelligence at different levels to optimize the performance of existing signal processing algorithms. Proposed framework uses a modular-hybrid approach; frst to identify emitters using classical signal processing techniques i-e FFT, then using unsupervised classifcation clustering algorithms to form Pulse Descriptive words (PDWs), subsequently employing supervised classifcation models for deinterleaving the interleaved pulsed stream coming from different radars.Used models are Bi-directional Gated Recurrent Unit(BGRU),Bi-Long Short Memory(BiLSTM) and Transformer Model.Our proposed model is validated and tested on multiple datasets;comprising synthetic dataset and semi real datasets,simulating the extreme dense, overlapping noisy scenarios possible in IF domain. The results are highly encouraging and opening new venues for researchers for full scale deployment of AI in Emitter Identifcation and Deinterleaving in modern ELINT warfare.
