Radio Frequency Fingerprinting (RFF) enables passive physical-layer device authentication by exploiting unintentional hardware variations in wireless transmitters. Neuromorphic implementations are attractive, given their potential for low-latency, energy-efficient inference capability under Size, Weight, and Power (SWaP) constraints at the edge. A new RFF capability is demonstrated here using recently introduced Radio Frequency Resonate-and-Fire (RF-RAF) neurons and eight WirelessHART devices. Performance is evaluated for RF-RAF-generated fingerprints against the established Gabor Transform (GTX) baseline using three classifier architectures: Random Forest (RndF), Convolutional Neural Network (CNN), and a Time-Incremented Spiking Neural Network (TI-SNN). The results show that RF-RAF fingerprints achieve an average classification accuracy of 96.7% across all three classifier types and consistently outperform GTX fingerprints at all evaluated fingerprint sizes. This performance persists under time-span-matched conditions, and the RF-RAF versus GTX benefit is not solely attributable to input data utilization. The TI-SNN surpasses 94% classification accuracy using M = 4 time step RF-RAF fingerprints with approximately 100 spikes per inference-a 4 & times; larger GTX fingerprint requires approximately 1000 spikes to achieve the same classification accuracy. RF-RAF fingerprints offer two additional benefits: they are natively non-negative, which supports efficient neuromorphic hardware implementation, and they provide greater flexibility in fingerprint size selection. It is concluded that RF-RAF neurons provide an efficient neuromorphic-native encoding pathway for device RFF discrimination and offer improved accuracy-efficiency tradeoffs in training and inference for various classifier architectures.

Recent advances in Radio Frequency (RF)-based device classification have shown promise in enabling secure and efficient wireless communications. However, the energy efficiency and low-latency processing capabilities of neuromorphic computing have yet to be fully leveraged in this domain. This paper is a first step toward enabling an end-to-end neuromorphic system for RF device classification, specifically supporting development of a neuromorphic classifier that enforces temporal causality without requiring non-neuromorphic classifier pre-training. This Spiking Neural Network (SNN) classifier streamlines the development of an end-to-end neuromorphic device classification system, further expanding the energy efficiency gains of neuromorphic processing to the realm of RF fingerprinting. Using experimentally collected WirelessHART transmissions, the TI-SNN achieves classification accuracy above 90% while reducing fingerprint density by nearly seven-fold and spike activity by over an order of magnitude compared to a baseline Rate-Encoded SNN (RE-SNN). These reductions translate to significant potential energy savings while maintaining competitive accuracy relative to Random Forest and CNN baselines. The results position the TI-SNN as a step toward a fully neuromorphic “RF Event Radio” capable of low-latency, energy-efficient device discrimination at the edge.

This paper provides details for the most recent step taken in RndF-to-CNN-to-SNN classifier transition activity supporting an envisioned RF "event radio" concept. Successful results here include the transition from CNNs to neuromorphic-friendly CNN-derived SNNs and pique sufficient interest for pursuing next-step hardware demonstrations. Consistent with earlier RndF and CNN works that used the same experimentally collected WirelessHART signals, SNN results here show that two-dimensional event-based fingerprinting is best overall using events detected in burst Gabor transform responses. The approximate %C-Delta approximate to -2% decrease in average percent correct classification performance resulting from RF eventization encoding is effectively offset by a complementary %C-Delta approximate to +2% to +3% increase that occurs with the CNN-to-SNN transition. This level of neuromorphic-friendly SNN performance is promising when considering the potential 10X-100X energy efficiencies that remain to be demonstrated.

Image forgery is becoming more difficult to detect due to advances in AI image generation. As such, the usefulness - and even requirement - for detection techniques that are affordable (computationally and monetarily) as well as intuitive and simple are equally increasing. This work demonstrates the first adoption of Distinct Native Attribute (DNA) Fingerprinting to image and forgery detection to achieve similar results while mitigating the cost of implementation. General image classification results with accuracy of %C = 98.8% support the overall utility while the ability to detect within-category image forgeries produce an average of %C = 81.8%. Using an intuitive and small set of features, preliminary results show an approximate average classification accuracy difference of only %C-Delta = -9% from more complex solutions. This work demonstrates the ability to adopt DNA Fingerprinting for image classification, and image forgery using Image Domain DNA (ID-DNA) that is holistically less resource intensive while requiring less time, money, and expert knowledge.

This work supports development of an envisioned “RF Event Radio” capability by extending previous communication signal demonstrations into the radar arena. Promising methods in recent communications-based RF eventization works are adopted here and adapted for radar signal demonstration. As a matter of convenience, 13-bit Barker coded radar signals from collection archives are used here for demonstration. Multiple Discriminant Analysis (MDA) and Random Forest (RndF) classifiers are used to discriminate four different radar signal channels. Emphasis is on RndF discrimination performance using non-eventized and eventized fingerprint features generated from pulse two-dimensional Gabor transform (2D-GTX) responses. Resultant classification performance losses $(\%\mathrm{C}_{\Delta})$ due to eventization span $-5.26 \%<\% \mathrm{C}_{\Delta}<+0.02 \%$ using low, medium and high frequency resolution GTX responses. As with previous communication signal eventization, it is expected that radar signal RF eventization will benefit from using more robust convolutional (CNN) and spiking (SNN) neural network classifiers. The use of these classifiers is expected to reduce radar $\% \mathrm{C}_{\Delta}$ eventization losses that will ultimately be traded-off as potential 1000X improvements are realized in neuromorphic processing systems.

The results of first-step research activity are presented for realizing an envisioned “event radio” capability that mimics neuromorphic event-based camera processing. The energy efficiency of neuromorphic processing is orders of magnitude higher than traditional von Neumann-based processing and is realized through synergistic design of brain-inspired software and hardware computing elements. Relative to event-based cameras, the development of event-based hardware devices supporting Radio Frequency (RF) applications is severely lagging and considerable interest remains in obtaining neuromorphic efficiency through event-based RF signal processing. In the Operational Technology (OT) protection arena, this includes efficient software computing capability to provide reliable device classification. A Random Forest (RndF) classifier is considered here as a reliable precursor to obtaining Spiking Neural Network (SNN) benefits. Both 1D and 2D eventized RF fingerprints are generated for bursts from NDev = 8 WirelessHART devices. Average correct classification (%C) results show that 2D fingerprinting is best overall using detected events in burst Gabor transform responses. This includes %C ≥ 90% under multiple access interference conditions using an average of NEPB ≥ 400 detected events per burst. This is sufficiently promising to motivate next-step activity aimed at (1) reducing fingerprint dimensionality and minimizing the required computational resources, and (2) transitioning to a neuromorphic-friendly SNN classifier—two significant steps toward developing the necessary computing elements to achieve the full benefits of neuromorphic processing in the envisioned RF event radio.

First-step demonstration activity is presented here for an envisioned "event radio" capability that mimics neuromorphic event-based camera concepts. Radio Frequency (RF) eventization is introduced and its impact on Convolutional Neural Network (CNN) classification is assessed. Classification of eight commercial WirelessHART adapters is performed using sparsely populated event-based fingerprints containing 200-of-896 possible event detections-this is an important first-step toward realizing an envisioned neuromorphic-friendly Spiking Neural Network (SNN) capability supporting edge RF sensing. Superiority of Gabor Transform (GTX) features from collected bursts is evident in CNN classification results that include 1) classification accuracy greater than 90 percent using an average of 200 detected events per burst, versus 300 events per burst required for GTX-Direct eventization, and 2) a non-eventized versus RF eventized classification loss in classification accuracy of 4.11 percent for GTX-Derivative eventization, versus a loss of 5.42 percent in accuracy for GTX-Direct eventization. The CNN performance here motivates next-step event radio research aimed at demonstrating a neuromorphic-friendly SNN RF sensing capability using RF eventized fingerprints. Future demonstration objectives include completing the CNN-to-SNN transition, characterizing SNN classification performance, and performing hardware demonstrations. These objectives support achievement of an envisioned 1000X performance improvement that includes a 10X reduction in required power and 100X improvement in overall processing speed.

Radio Frequency Fingerprinting (RFF) is the attribution of uniquely identifiable signal distortions to emitters via Machine Learning (ML) classifiers. RFF approaches relying on pre-determined expert features lack generalizability, and state-of-the-art approaches based on Convolutional Neural Networks (CNNs) can be too demanding for endpoint devices to train. This work presents Fingerprint Extraction through Distortion Reconstruction (FEDR), a best-of-both-worlds technique which employs a pre-trained CNN to identify and extract a small, salient set of unique features, amenable for use in lightweight machine learning models. Given a received distorted signal, the FEDR network encodes signal distortions into "fingerprints," which can be used by lightweight ML classifiers to perform RFF with minimal resource consumption at the endpoint. FEDR learns by transforming generated signals into reconstructions of received signals, relying solely on the fingerprints as representations of the distortions - as the reconstructions improve, the fingerprints better encode the distortions. The FEDR technique was evaluated on synthetic IQ-imbalanced IEEE 802.11a/g data, where FEDR fingerprints were shown to encode actual IQ imbalance parameters, signifying successful isolation of distortion information and validating the FEDR technique. FEDR was further evaluated on a representative real-world WiFi dataset, where extracted fingerprints were coupled with a lightweight two-layer dense network. When compared against two common RFF techniques, the FEDR-based approach achieved state-of-the-art performance with Matthews Correlation Coefficient ranging from 0.984 (5 classes) to 0.851 (100 classes), using nearly 73% fewer training parameters than the next-best technique.

Reliable detection of counterfeit electronic, electrical, and electromechanical devices within critical information and communications technology systems ensures that operational integrity and resiliency are maintained. Counterfeit detection extends the device’s service life that spans manufacture and pre-installation to removal and disposition activity. This is addressed here using Distinct Native Attribute (DNA) fingerprinting while considering the effects of sub-Nyquist sampling on DNA-based discrimination. The sub-Nyquist sampled signals were obtained using factor-of-205 decimation on Nyquist-compliant WirelessHART response signals. The DNA is extracted from actively stimulated responses of eight commercial WirelessHART adapters and metrics introduced to characterize classifier performance. Adverse effects of sub-Nyquist decimation on active DNA fingerprinting are first demonstrated using a Multiple Discriminant Analysis (MDA) classifier. Relative to Nyquist feature performance, MDA sub-Nyquist performance included decreases in classification of %CΔ ≈ 35.2% and counterfeit detection of %CDRΔ ≈ 36.9% at SNR = −9 dB. Benefits of Convolutional Neural Network (CNN) processing are demonstrated and include a majority of this degradation being recovered. This includes an increase of %CΔ ≈ 26.2% at SNR = −9 dB and average CNN counterfeit detection, precision, and recall rates all exceeding 90%.

Distinct native attribute fingerprinting is considered as a means to ensure the longevity of WirelessHART communication devices used in industrial automation and control systems. The aim is for these devices to reach full life expectancy using a technical cradle-to-grave lifecycle protection strategy. The protection addressed here includes pre-deployment near-cradle counterfeit device detection using active fingerprinting and operational mid-life rogue device detection using passive fingerprinting. The counterfeit and rogue device detection rates are estimated for 56 five-class multiple discriminant analysis models. Detection demonstrations include using three non-modeled devices to complete a total of 3 × 5 × 56 = 840 individual rogue and counterfeit device identity (ID) verification demonstrations. The device ID verification process uses binary accept/reject decisions with false positive outcomes used to estimate rogue and counterfeit device detection rates. For device ID verification using active fingerprints, the demonstrated counterfeit detection rate approached 99 % using only 15-of-99 available features—an approximate 85 % dimensional-reduction. Device ID verification using passive fingerprints was more challenging and the rogue detection rate approached 94 % using 120-of-243 available features—an approximate 50 % dimensional-reduction.. Collectively, the dimensionally-reduced implementations support efficiency improvement objectives required for providing near-cradle counterfeit device and mid-life operational rogue device detection in critical industrial automation and control systems.

Radio Frequency Fingerprinting (RFF) is often proposed as an authentication mechanism for wireless device security, but application of existing techniques in multi-channel scenarios is limited because prior models were created and evaluated using bursts from a single frequency channel without considering the effects of multi-channel operation. Our research evaluated the multi-channel performance of four single-channel models with increasing complexity, to include a simple discriminant analysis model and three neural networks. Performance characterization using the multi-class Matthews Correlation Coefficient (MCC) revealed that using frequency channels other than those used to train the models can lead to a deterioration in performance from MCC > 0.9 (excellent) down to MCC < 0.05 (random guess), indicating that single-channel models may not maintain performance across all channels used by the transmitter in realistic operation. We proposed a training data selection technique to create multi-channel models which outperform single-channel models, improving the cross-channel average MCC from 0.657 to 0.957 and achieving frequency channel-agnostic performance. When evaluated in the presence of noise, multi-channel discriminant analysis models showed reduced performance, but multi-channel neural networks maintained or surpassed single-channel neural network model performance, indicating additional robustness of multi-channel neural networks in the presence of noise.

The need for reliable communications in industrial systems becomes more evident as industries strive to increase reliance on automation. This trend has sustained the adoption of WirelessHART communications as a key enabling technology and its operational integrity must be ensured. This paper focuses on demonstrating pre-deployment counterfeit detection using active 2D Distinct Native Attribute (2D-DNA) fingerprinting. Counterfeit detection is demonstrated using experimentally collected signals from eight commercial WirelessHART adapters. Adapter fingerprints are used to train 56 Multiple Discriminant Analysis (MDA) models with each representing five authentic network devices. The three non-modeled devices are introduced as counterfeits and a total of 840 individual authentic (modeled) versus counterfeit (non-modeled) ID verification assessments performed. Counterfeit detection is performed on a fingerprint-by-fingerprint basis with best case per-device Counterfeit Detection Rate (%CDR) estimates including 87.6% < %CDR < 99.9% and yielding an average cross-device %CDR ≈ 92.5%. This full-dimensional feature set performance was echoed by dimensionally reduced feature set performance that included per-device 87.0% < %CDR < 99.7% and average cross-device %CDR ≈ 91.4% using only 18-of-291 features—the demonstrated %CDR > 90% with an approximate 92% reduction in the number of fingerprint features is sufficiently promising for small-scale network applications and warrants further consideration.

This work supports a technical cradle-to-grave protection strategy aimed at extending the useful lifespan of Critical Infrastructure (CI) elements. This is done by improving mid-life operational protection measures through integration of reliable physical (PHY) layer security mechanisms. The goal is to improve existing protection that is heavily reliant on higher-layer mechanisms that are commonly targeted by cyberattack. Relative to prior device ID discrimination works, results herein reinforce the exploitability of constellation-based PHY layer features and the ability for those features to be practically implemented to enhance CI security. Prior work is extended by formalizing a device ID verification process that enables rogue device detection demonstration under physical access attack conditions that include unauthorised devices mimicking bit-level credentials of authorized network devices. The work transitions from distance-based to probability-based measures of similarity derived from empirical Multi-Variate Normal Probability Density Function (MVNPDF) statistics of multiple discriminant analysis radio frequency fingerprint projections. Demonstration results for Constellation-Based Distinct Native Attribute (CB-DNA) fingerprinting of WirelessHART adapters from two manufacturers includes 1) average cross-class percent correct classification of %C > 90% across 28 different networks comprised of six authorized devices, and 2) average rogue rejection rate of 83.4% ≤ RRR ≤ 99.9% based on two held-out devices serving as attacking rogue devices for each network (a total of 120 individual rogue attacks). Using the MVNPDF measure proved most effective and yielded nearly 12% RRR improvement over a Euclidean distance measure.

Standalone Global Navigation Satellite System (GNSS) applications demand higher precision than is typically achieved using differential processing. While differential processing removes the effects of most common-mode error sources, it provides limited compensation for distortions caused by multi-path or pseudorange measurement biases from dissimilar receiver hardware due to subtle space vehicle-borne signal deformations. These naturally occurring phenomena directly impact system integrity and lead to ranging error in the GNSS receiver solution. Signal Quality Monitoring (SQM) has the objective of providing confidence in the GNSS Positioning, Navigation, and Timing (PNT) solution, and aims to offer timely warning in the event that SV signal conditions degrade to unsafe levels. Several methods of SQM have been previously introduced and implemented to augment civilian Safety-of-Life (SoL) applications. The methods considered in this work focus on implementing effective SQM using low-cost Commercial Off-theShelf (COTS) equipment, a Software-Defined Radio (SDR), and a typical software GNSS receiver architecture that tracks the Galileo E1 signals and the Global Positioning System (GPS) L1 Coarse-Acquisition (C/A) signals. The techniques here are centered on acquiring and discriminating signal chip shapes with a goal of identifying both `clean' and `deformed' signals. The demonstrated identification method is relevant to the growing significance of SQM for SoL applications while providing benefit for confidently monitoring received GNSS signal integrity without requiring specialized receiver hardware.

This paper summarizes demonstration activity aimed at applying Distinct Native Attribute (DNA) feature selection methods to improve the computational efficiency of time domain fingerprinting methods used to discriminate Wireless Highway Addressable Remote Transducer (WirelessHART) devices being used in Industrial (IIoT) applications. Efficiency is achieved through Dimensional Reduction Analysis (DRA) performed here using both pre-classification analytic (WRS and ReliefF) and post-classification relevance (RndF and GRLVQI) feature selection methods. Comparative assessments are based on statistical fingerprint features extracted from experimentally collected WirelessHART signals, with Multiple Discrimination Analysis, Maximum Likelihood (MDA/ML) estimation showing that pre-classification methods are collectively superior to post-classification methods. Specific DRA results show that an average cross-class percent correct classification differential of 8% ≤ %CD ≤ 1% can be maintained using DRA selected feature sets containing as few as 24 (10%) of the 243 full-dimensional features. Reducing fingerprint dimensionality reduces computational efficiency and improves the potential for operational implementation.

Radio frequency (RF) fingerprinting extracts fingerprint features from RF signals to protect against masquerade attacks by enabling reliable authentication of communication devices at the “serial number” level. Facilitating the reliable authentication of communication devices are machine learning (ML) algorithms which find meaningful statistical differences between measured data. The Generalized Relevance Learning Vector Quantization-Improved (GRLVQI) classifier is one ML algorithm which has shown efficacy for RF fingerprinting device discrimination. GRLVQI extends the Learning Vector Quantization (LVQ) family of “winner take all” classifiers that develop prototype vectors (PVs) which represent data. In LVQ algorithms, distances are computed between exemplars and PVs, and PVs are iteratively moved to accurately represent the data. GRLVQI extends LVQ with a sigmoidal cost function, relevance learning, and PV update logic improvements. However, both LVQ and GRLVQI are limited due to a reliance on squared Euclidean distance measures and a seemingly complex algorithm structure if changes are made to the underlying distance measure. Herein, the authors (1) develop GRLVQI-D (distance), an extension of GRLVQI to consider alternative distance measures and (2) present the Cosine GRLVQI classifier using this framework. To evaluate this framework, the authors consider experimentally collected Z-wave RF signals and develop RF fingerprints to identify devices. Z-wave devices are low-cost, low-power communication technologies seen increasingly in critical infrastructure. Both classification and verification, claimed identity, and performance comparisons are made with the new Cosine GRLVQI algorithm. The results show more robust performance when using the Cosine GRLVQI algorithm when compared with four algorithms in the literature. Additionally, the methodology used to create Cosine GRLVQI is generalizable to alternative measures.

A wireless nondestructive fault detection test for loose or damaged connectors is demonstrated. An architecture known as the conditioned multiclassification of stimulated emissions (CMSE) is pretrained on simulated and empirical radar outputs, and transfer learning is applied to classify connected and disconnected coaxial interconnections. The two main data conditioning methods of this architecture, a statistical signal analysis tool and a convolutional filter bank, are evaluated in order to determine the cost-value proposition of each component. Novel contributions of this technique include the use of two simulation-aided convolutional filter banks to generate a multinetwork ensemble and transfer learning from artificial neural networks trained on two primitive datasets revolving around the electromagnetic phenomena of reflection and filtering. A total of 560 different neural network topologies across four different signal conditioning configurations are considered, with all results compared against the current standard for measurement of cable and connection faults, time-domain reflectometry. Metrics used for comparison are time (training and evaluation), detection (connector engagement at state change detection), and clustering (projection space performance, used as a measure of transfer learning potential). It is determined that the full CMSE architecture performs best, with nearly any neural network topology of this configuration displaying an early detection improvement of 113% and requiring 30% less time to execute an individual classification versus the current standard, all while meeting the most stringent definitions of nondestructive evaluation (NDE).

The Industrial Internet of Things (IIoT) market is skyrocketing towards 100 billion deployed devices and cybersecurity remains a top priority. This includes security of ZigBee communication devices that are widely used in industrial control system applications. IIoT device security is addressed using ConstellationBased Distinct Native Attribute (CB-DNA) Fingerprinting to augment conventional bit-level security mechanisms. This work expands upon recent CB-DNA “discovery” activity by identifying reduced dimensional fingerprints that increase the computational efficiency and effectiveness of device discrimination methods. The methods considered include Multiple Discriminant Analysis (MDA) and Random Forest (RndF) classification. RndF deficiencies in classification and post-classification feature selection are highlighted and addressed using a pre-classification feature selection method based on a Wilcoxon Rank Sum (WRS) test. Feature down-selection based on WRS testing proves to very reliable, with reduced feature subsets yielding cross-device discrimination performance consistent with full-dimensional feature sets, while being more computationally efficient.

A novel method is presented for remotely assessing microwave system health using environmental monitoring sensors that employ a low-power random noise radar with artificial neural network-based machine learning processing. The method expands prior stimulated unintended radiated emission (SURE) research in [M. W. Lukacs, A. J. Zeqolari, P. J. Collins, and M. A. Temple, “ ‘RF-DNA fingerprinting’ for antenna classification,” IEEE Antennas Wireless Propag. Lett., vol. 14, pp. 1455–1458, 2015] by adding a new hybrid expert-empirical concept-forming technique called matched filter replication (MFR), the outputs of which are used in ensemble learning. Also by comparison with baseline performance, classification improvement is demonstrated using a single iteration of MFR ensemble learning that improves antenna termination state classification by $\%C_\Delta >\text{19}\%$. A nested ensemble learning architecture is also introduced that enables classification of truly unknown devices with no learner in the ensemble being trained. This exploits the concept of multiple iterations using the MFR process by establishing an enlarged hypothesis space that is subsequently collapsed to form a final classification decision. The new SURE architecture enables the assessment of unknown device capabilities using a network trained on selected primitive traits (in this case, reflectivity).