Signal Sorter for Advanced Multifunction Radio Frequency Concept (AMRF-C) Using Neural Networks and Advanced Statistical Techniques

V.C. Kowtha,¹ M.J. Thompson,¹ T.N. Reynolds,¹ J.R. Connell,² A.E. Spezio,¹ and J.C. Sciortino, Jr.¹

¹Tactical Electronic Warfare Division
²Varilog Research, Inc.

We describe an artificial intelligence (AI)-based electronic surveillance processing signal sorter currently being developed for AMRF-C that uses emitter characteristics as its input. It has been evaluated on two different data sets. The clustering quality and the processing time were found to be comparable to that achieved by experienced human analysts.

AMRF-C is an Office of Naval Research program that addresses the increasing challenges of shipboard topside RF functions, Electronic Warfare (EW), radar, and RF communications, in the context of a proof-of-concept demonstration, sharing a common receive and transmit antenna (Fig. 9). Each antenna uses phased-array technology organized into softwareprogrammable subarray apertures that can be dynamically allocated to selected combinations of EW, radar or communications functions. The initial AMRF-C demonstration covers the broad frequency range of H, I, and J bands. It requires broadband Electronic Support (ES) receiver assets to provide the timely warning and surveillance necessary for ship-self protection. EW includes ES passive receive functions and the Electronic Attack (EA) active countermeasures transmit function.

Electronic Warfare Support (ES) Functions: In the AMRF-C demonstration, ES performs two functions, High Probability of Intercept (HPOI) and High Gain High Sensitivity (HGHS) and uses state-of-theart Wideband Digital Channelized Receiver System (WBDCRS) technology. Figure 10 illustrates the ES functional hardware. HPOI uses nine dedicated elements in the receive array to perform pulsed radar intercept, which includes a bearing measurement obtained using interferometric techniques. State-of-theart downconverters and fiber optic links process and transmit each analog radar pulse to the WBDCRS for digital conversion and Pulse Descriptor (PD) records-generation, for subsequent HPOI software processing. HGHS uses a state-of-the-art digital beamformer and a fast Fourier transform processor to convert the radar pulse energy collected over the entire phased array into both time-domain and frequency- domain digital data streams. Radar pulses in the time domain are converted into PD records. Frequency domain data are processed to detect LPI radar.

FIGURE 9
AMRF-C concept testbed architecture.

Fig10 Image

FIGURE 10
Electronic surveillance functions in AMRFC.

ES Signal Sorter (SS) Components: Electronic Warfare Support Measures (ESM) includes collection and analysis of radar signals. A typical environment may contain multiple signals with pulses from one emitter interleaved with pulses from others. In baseline systems, pulses are collected, processed, and then deinterleaved into their separate pulse trains. The AMRF-C SS module receives Pulse Descriptors (PD) into a modular AI-based electronic surveillance processor. The ES Signal Sorter (Fig. 11) contains several artificial neural network (NN) components: the Pruner, a rapid statistical signal sorter; the Signal Processing (SP) Manager; the NN Toolbox, Rapid Emitter Multiple Clustering Algorithm (REMCAM); and the Correlator, Rapid Emitter Clustering Expert System Software (RECESS).

FIGURE 11
Signal processing signal flow.

Pruner: The Pruner isolates and extracts the wellbehaved PD records from the interleaved incoming stream. It is a fast, computationally efficient module that is used in the first-stage sorting process. Figure 12 shows the Pruner module processing time for a simulated 2.5 million pulses from a typical AMRF-C scenario (Fig. 13) on different systems. The Pruner module passes the isolated signal descriptors and the residue PD records to the SP Manager. The SP Manager controls the overall data flow. It first directs the interleaved residue to the NN-based REMCAM for deinterleaving. It then passes the isolated PD records to the library correlator (RECESS) for identification. Finally, it reports situational awareness (SA) information to an ESM functional graphical user interface.

FIGURE 12
Evaluation of pruner processing time on four systems.

Fig13 Image

FIGURE 13
Simulated scenario.

Neural Network Toolbox (REMCAM): The AMRF-C REMCAM process uses four NN clustering algorithms and isolates PD records from the residue submitted by the SP Manager. Seven clustering algorithms have been evaluated for REMCAM; the optimal four are to be inserted. They are: Cellular Network Classifier (CNC), Fuzzy Adaptive Resonant Theory (FA), K-Means, Learning Vector Quantization (LVQ), Radial Basis Function (RBF), Self-Organizing Feature Maps (SOFM), and Supervised Piriform Hierarchical Clusterer (SuperPHC). Each algorithm outputs a matrix of binned pulses that form labeled pulse trains. The outputs are combined, and a consensus report of the residue analysis is transmitted to RECESS via the SP Manager for identification.

Expert System (RECESS): RECESS implements Dempster-Shafer reasoning to associate the isolated PD records to emitter types. It correlates received radar pulses with single emitting sources (i.e., isolated PD records), uses the pulse information to measure characteristic parameters of the source, uses the measured parameters to identify the source against an emitter library, and correlates emitters to their platforms. RECESS provides a plausibility ranking showing the likelihood that an emitter is any one of the library types present in the scenario. Figure 14 depicts the classification of the isolated PD records to the emitter types. The SP Manager uses the plausibility reports to decimate the emitter types in the situational awareness report (highlighted structures).

FIGURE 14
Plausibility description from RECESS.

Summary: This program demonstrates the use of advanced automation techniques to perform realtime ES functions in an integrated modern RF system including EW and communications. Use of such enabling technologies supports the Navy mission in reducing emitter ambiguities and processing complex tasks. Modern EW/ES systems will increasingly use the AI-based technology demonstrated here in integrated systems such as AMRF-C and become commonplace.

[Sponsored by ONR]