Optimal Time-Frequency Distribution Selection for LPI Radar Pulse Classification

The work presented in this paper shows the performance of various time-frequency distributions when gathering ELectronic INTelligence (ELINT) from an electromagnetic environment that contains transmissions from radars operating in a Low Probability of Interception (LPI) mode. A radar device varying waveform parameters on a pulse-by-pulse basis to enhance sensing capabilities and/or to avoid interception warrants a method that can assign a unique Pulse Descriptor Word (PDW) to each pulse detected. The simulations presented here makes use of a Deep Learning classifier that is fed by time-frequency representations of noisy LFM and FMCW pulses that each have unique signal parameters. The performance of the radar pulse classifier is conveyed for multiple time-frequency methods. The results show that the time-frequency representation requirements for accurate PDW generation varies for each signal parameter being estimated whilst also having a dependence on the SNR of the intercepted signal.