Detecting rogue attacks on commercial wireless Insteon home automation systems

DNA Feature Selection for Discriminating WirelessHART IIoT Devices

“…TD-DNA Fingerprinting utilizes various machine learning algorithms and concepts such as feature selection via DRA which has been the subject of prior related works [9,18,28]. The demonstration here focuses on examining DRA methods for use with an MDA/ML classifier, given the MDA/ML classification process has been shown to be computationally efficient while reliably discriminating IIoT signals [9,13].…”

“…Results here are based on TD-DNA fingerprints generated from WirelessHART burst preamble responses and are generated using a methodology consistent with prior related works [9][10][11][12][13][14][15][16][17][18]. The instantaneous amplitude (AMP), phase (PHZ), and frequency (FRQ) responses of the PreAmbRgn ROI are divided into NR = 26 subregions.…”

“…The MDA/ML processing used here is a readily implementable, computationally efficient process that has provided reliable device discrimination in prior TD-DNA works [9,[11][12][13][14][15][16][17][18]. As detailed in [18], MDA is effectively a feature selection process that performs best when input class features and their corresponding projections are Gaussian distributed.…”

“…One PHY-based method supporting offensive, defensive, and exploitive network operations is Time Domain Distinct Native Attribute (TD-DNA) Fingerprinting which has been successfully used to discriminate IoT and IIoT communication devices and their operating states [9][10][11][12][13][14][15][16][17][18]. The TD-DNA fingerprinting methodology therein is well-suited for consideration here given 1) the observed ZigBee-like signal characteristics of WirelessHART signals, and 2) the ability to perform Dimensional Reduction Analysis (DRA) and identify the minimal subset of features required to achieve a given level of discrimination performance.…”

Industrial IoT cross‐layer forensic investigation

“…While the cross‐device uniqueness of electronic device fingerprints may not be totally on par with cross‐human fingerprint uniqueness, results such as provided in Deng et al (), Hall et al (), Huang and Zheng (); Lopez Jr. et al, ; Mirowski et al, ; Rehman et al, ; Reising et al, ; Rondeau et al, ; Suski et al, ; Talbot et al, ; Zhuo et al, ) routinely demonstrate near 100% discrimination for selected scenarios and have been sufficiently promising to sustain progressive RDD over the past 10 years. Collectively, these and other related RFF works have addressed nearly all common communication signaling schemes, including Bluetooth (Hall et al, ), automation (Lopez Jr. et al, ; Talbot et al, ) and ZigBee (Rondeau et al, ) Personal Area Networks (PANs); WiFi (Huang & Zheng, ; Rehman et al, ; Suski et al, ; Zhuo et al, ) Wireless Local Area Network (WLANs), and WiMAX (Deng et al, ; Reising et al, ) Wide Area Networks (WANs), to name a few. For the references provided, the unique RFF features have been reliably extracted from various signal domains, including (a) time (Deng et al, ; Hall et al, ; Lopez Jr. et al, ; Rehman et al, ; Suski et al, ), (b) frequency (Lopez Jr. et al, ; Suski et al, ; Talbot et al, ), (c) joint time–frequency (Reising et al, ; Zhuo et al, ), and (d) constellation (Huang & Zheng, ; Rondeau et al, ).…”