Specific Emitter Identification for Comm Signals

As known, the wording “Signal-In-Space” (SIS) refers the timing and frequency characteristic of the aerial transmitted signal.
The SIS, at the source, mathematically defines the formal time-depending function representing the theoretical signal an emitter should transmit.
The SIS provides also the emitter with a fingerprint of the transmitted signal.

Concerning the Radar, two type of time-depending functions have adopted to classify the radars’ SIS: pulsed or Continuous Wave (CW).
Both the radars’ SIS are specifically devised in order to measure the target’s distance and the relative radial speed, as the pulsed SIS is for medium or long range radars, while the FM-CW SIS is mainly adopted for short range/high resolution radars.
The relevant Radars’ SIS parameters affect the timing behaviour through specific modulations of the frequency carrier F_o, as the pulse-width, the intra-pulse modulation and the Pulse Repetition Interval (PRI), where the last one can be dynamically changed in order that both a sort of ECCM and the range ambiguity resolution can be performed.
For the FM-CW radars, the frequency span and the sweep time allow to tune the range resolution.

There is a strong difference between Radar and COMMS signals, which affects the Specific Emitter Identification problem: COMMS signals are required to support information routing and relaying, so they imply an infrastructure as Network layers superimposed to the link and physical layers (access layer).
Overall, this is known as “Waveform”, being it either a wireless communication standard or a legacy radio access technology.
The challenge is that there is a plethora of waveforms and networking protocols being each of them applied depending on a specific operational requirement.
So, in order to use the same AI-based approach as for Radar signals, is necessary to define correctly the features of the signal over which the classifier has to be trained and understand if those features are peculiar for each communication signal or can be generalized (or a mix).

Currently, some SEI methods based on transient feature extraction are mainly researched [1-5].

A technique for classifying radio devices was first implemented utilizing the transient features [1], and literature [2] proposed a method based on transient features for classifying different models of radio stations [2].

In addition, the multi-fractal model [3], the amplitude and phase of transient signals [4] and frequency fingerprinting [5] were further researched.
However, classification based on transient features has strict requirements for integrity of transient signals.
Since transient signals have very short duration, feature extraction is usually difficult, especially when communication is non-cooperative.

For emitters with the same model, individual difference between transmitted signals will grow out of the nuances of electrics device and uncertainty during the device manufacturing procedure, such as the stray features and the nuances of carrier frequency.
Both features in steady state have longer duration and are easy to be seized, to extract these steady features for classification is more practical.

The extension to communication signals is a new edge of research.
The result is useful to increase the performance of the decision making process in ISR missions, once an anomalous behaviour is detected and classified in the electromagnetic scenario.
Then, the radio emitter’s features and both the networking features and the topics relevant to the information payload, make specific the used AI-based approach to the SEI for the COMMS moving the analysis beyond the SIS identification, unlikely performed against radar signals.
Adopting Machine-Learning based procedures and algorithms, SEI provides a method aimed to detect and identify the specific RF emitters [6].
Hence, SEI allows the control of the RF emissions, providing a general capability of authentication in the range of potential threats.