Development of Security Analytics for Software Defined Communications using Deep Learning

Software Defined Communications are becoming an essential part of military communications providing higher bandwidth, frequency agility, security, and other desirable attributes.  However, the added software functionality and complexity brings the possibility of unique security vulnerabilities not fully addressed by traditional security analysis and testing tools.  Cyber/EW attacks are an emerging threat which many communications systems have not been designed to fight through.  These combined attacks can target advanced waveform software and firmware in Software Defined Radios (SDR). Identifying the attack profiles that will exploit a Cyber/EW vulnerability can be complex making them difficult to mitigate.  As a result, many existing and future software defined communications systems may be vulnerable to this new mode of electronic attack.   Secmation developed CyRF, a security analysis and testing tool which specifically targets these unique vulnerabilities in software defined communication systems such as Software Defined Radios (SDR). Originally developed for the Air Force Research Laboratory, CyRF uses a guided search method to identify communications signal parameters (e.g., frequency, phase, offset) which create candidate malicious signals that, when received by an SDR, will cause a software processing failure.  This software processing failure can lead to denial of communications to the warfighter or an exploitable vulnerability than can be used to affect the communications network.  This technology was recently granted a US Patent. CyRF uses a unique combination of signal and computational analytics to drive the search for vulnerabilities.  The search is implemented by a guided mutation fuzzer which targets the vulnerability search using metrics derived from the analytics.  CyRF focuses on finding software vulnerabilities in the Physical/Signal Processing Layer (Layer 1) and Data Link/Waveform Layer (Layer 2) as defined in the OSI model.  This “out of sight, out of mind” software is not regularly targeted by security researchers or security test plans. The proposed work will develop integrated computational and communication AI/ML based analytics which can drive a fuzzing framework to determine the security robustness of software defined communications. These robustness metrics will be used to identify possible regions of vulnerability in software defined communications and characteristics of malicious signals that can drive the receiver software into the regions of vulnerability.  The additional of AI/ML technology promises to find vulnerabilities quicker and be more rapidly adaptable to different types of SDR.  The technology will be demonstrated both in simulation and through Over the Air Testing (OTA) using commercial SDRs.