Spectrum scarcity and low utilization constrain the development of wireless communication. Cognitive radio enables sensor nodes (SNs) to monitor primary user (PU) channel usage, thereby enabling access to idle channels and improving spectrum efficiency. To address sensing impairments in wireless environments, cooperative spectrum sensing (CSS) has been implemented in cognitive wireless sensor networks (CWSNs) to ensure reliable spectrum detection. However, malicious SNs (MSNs) launching Byzantine attack can cause severe security threats to CSS. To ensure the effective operation of CSS against MSNs, this paper proposes a support vector machine (SVM) based on improved sand cat swarm optimization (SCSO), denoted as SCSO-SVM, for accurate MSN identification. The algorithm leverages the small-sample superiority of SVM and employs an enhanced SCSO to adaptively optimize its kernel and penalty parameters. In contrast to existing isolation forest (IF)-based algorithms and SVM methods based on other meta-heuristic algorithms, the proposed algorithm significantly reduces time consumption. Furthermore, it achieves high classification accuracy and F1-score with limited training samples. At last, simulation results demonstrate that across various Byzantine attack scenarios, the proposed SCSO-SVM algorithm reduces the time consumption by an order of magnitude compared to IF and spectral clustering -based fusion algorithm (IFSC), improved IF algorithm, particle swarm optimization (PSO)-based SVM (PSO-SVM), and grey wolf optimizer (GWO)-based SVM (GWO-SVM), while maintaining high accuracy.

Cooperative Spectrum Sensing (CSS) is vulnerable to hybrid Byzantine attack (HBA) from malicious users (MUs). Existing defense mechanisms, such as hard/soft fusion and reputation-based models, often struggle to adapt to dynamic attack patterns and fluctuating noise environments. For this aim, this paper proposes a self-parameterized and adaptive defense framework, termed dynamic Byzantine detection (DBD) to achieve robust Byzantine identification in an unsupervised manner. Following a state-change detection paradigm, we analyze the relationship between energy measurements from consecutive sensing periods and transform the problem into detecting distribution shifts to eliminate the reliance on prior “clean” data or ground-truth labels. Then, we further employ an improved hierarchical density-based clustering algorithm, with parameters self-determined via ak-distance analysis, to identify and effectively remove MUs across multiple levels. In a complex time-varying attack sequence involving independent and collusive MUs, DBD consistently achieves high detection accuracy and strong stability, while most benchmarks exhibit significant performance degradation. Under severe noise power fluctuation, DBD demonstrates superior robustness compared to an online support vector machine (SVM) that relies on ground-truth training data. Furthermore, a series of numerical simulation results demonstrate that DBD achieves notably superior performance over traditional methods while operating more efficiently than an online SVM, presenting a practical and effective solution for securing CSS.

Recently, Radio Frequency Fingerprint (RFF) technology has emerged as a promising technique for physical layer authentication. However, overcoming the interference from wireless multipath channels remains a key challenge for robust RFF extraction. Existing channel-robust studies often suppress channel effects at the expense of intrinsic RFF information, and the retained RFF features may lack sufficient discrimination. This paper proposes a channel-robust RFF extraction scheme based on channel reciprocity. Firstly, a Channel State Information (CSI) feedback mechanism is proposed to collect reciprocal uplink and downlink CSI. Then, CSI preprocessing methods and the channel reciprocity judging method based on Mean Absolute Distance (MAD), are exploited to further enhance CSI reciprocity. Finally, a novel RFF extraction algorithm eliminates reciprocal channel components by calculating the quotient of uplink and downlink CSI, thereby retaining the differentiated device-specific RFF features. A unified identification framework supports both in-library classification and unknown device detection. Extensive experiments using 41 ESP32 development kits under various Wi-Fi channel environments show that the extracted RFF features are channel-robust, highly distinctive, and stable over time. Specifically, in a static scenario with strong multipath effects, the in-library classification accuracy reaches 96.86%. And the same metric in a dynamic scenario with significant interference remains 93.48%. Furthermore, new-device recognition accuracy remains above 94% across all scenarios, with a False Negative Rate (FNR) below 1.3%.

In the realm of cognitive radio (CR), the soft and hard combinations are commonly applied for the fusion rule of cooperative spectrum sensing (CSS) to detect the primary user (PU) signal for available vacant spectrum resources and allow cooperative secondary users (SUs) to opportunistically access the channel without harmful interference to the PU’s normal communication. However, in the process of submitting the sensing information to the fusion center (FC), the soft combination requires great communication overhead from the SU to the FC, and the hard combination reports only one bit of the local decision resulting in unsatisfactory detection performance. In view of this, in this paper we make an in-depth investigation on the quantization of raw measurement data and propose a quantization-based multi-bit combination method that utilizes the information of each sampling point. On the basis of the proposed multi-bit combination approach, each SU quantifies the information obtained from the local sensing information and transmits the quantified sensing data in a few bits to the FC. Furthermore, we formulate an optimization problem for the error probability of the multi-bit combination to achieve the optimal CSS performance. Finally, numerical simulation results confirm the effectiveness and robustness of the proposed multi-bit combination method, establishing its superiority in various environments, in terms of the overall error probability.

Physical-layer key generation (PKG) leveraging the reciprocity of wireless channel provides an effective approach for key agreement among resource-constrained Internet of Things devices. However, current researches on PKG predominantly focus on pairwise communication scenarios, and there remain challenges in achieving group key generation for multiuser scenarios. In this article, we propose a novel recovery mechanism-driven wireless group key generation protocol to facilitate key sharing in the star network typology. Specifically, the root node will assign each member node its unique group key component before initiating group key distribution. Subsequently, all group key components are distributed to member nodes using a forward error correction mechanism, which helps reduce system overhead. Finally, all member nodes utilize a recovery mechanism and their respective group key component to obtain the same complete group key, thereby achieving group key distribution. Compared to existing schemes, our protocol can avoid the significant information leakage caused by repeated distribution of the same group key, thereby enhancing security. We further design and implement a practical wireless group key generation system using ESP32. Additionally, a group channel state information (CSI) extraction tool for multiuser channel measurements is developed. Experimental results demonstrate that our protocol can generate the group key with high randomness while benefiting from good channel reciprocity, making it suitable for cryptographic applications in multiuser communication scenarios.

Radio Frequency Fingerprint (RFF) identification is a promising technique for physical layer identification that can enhance wireless security. However, interference of wireless channel characteristics is a key challenge hindering its robustness. To solve this problem, we propose a channel-robust RFF extraction method. First, we design a challenge-response mechanism-based framework to satisfy the uplink and downlink channel reciprocity. Then, we propose a novel RFF extraction method named Quotient of the Estimated Channel State Information (QoECSI) that exploits channel reciprocity to eliminate channel effects. We implemented the QoECSI with the ESP32 development kits that support 2.4GHz Wi-Fi, Experimental results show that the extracted RFF features have high discrimination and long-term stability, and are robust to channel variations and noise. The accuracy rate is higher than 98% when the Signal-to-Noise Ratio (SNR) exceeds 25 dB. Specifically, in a Non-Line-of-Sight (NLOS) scenario with SNR = 35 dB, the average recognition accuracy of cross-validation is 98.56%. In a dynamic scenario where the terminal moves slowly indoors along a fixed route, the highest cross-time-validation accuracy is 98.57%.

Deep learning (DL) applied to a device’s radio-frequency fingerprint (RFF) has attracted significant attention in physical-layer authentication due to its extraordinary classification performance. Conventional DL-RFF techniques are trained by adopting maximum likelihood estimation (MLE). Although their discriminability has recently been extended to unknown devices in open-set scenarios, they still tend to overfit the channel statistics embedded in the training dataset. This restricts their practical applications as it is challenging to collect sufficient training data capturing the characteristics of all possible wireless channel environments. To address this challenge, we propose a DL framework of disentangled representation (DR) learning that first learns to factor the signals into a device-relevant component and a device-irrelevant component via adversarial learning. Then, it shuffles these two parts within a dataset for implicit data augmentation, which imposes a strong regularization on RFF extractor learning to avoid the possible overfitting of device-irrelevant channel statistics, without collecting additional data from unknown channels. Experiments validate that the proposed approach, referred to as DR-based RFF, outperforms conventional methods in terms of generalizability to unknown devices under unknown complicated propagation environments, e.g., dispersive multipath fading channels, even though all the training data are collected in a simple environment with dominated direct line-of-sight (LoS) propagation paths.

Radio-frequency fingerprinting (RFF) exploiting hardware characteristics has been employed for device recognition to enhance the overall security. However, the performance unreliability in long-term experiments, channel fading interference, and unauthorized devices verification are three open problems that restrict the development of RFF recognition. To address these issues, a robust RFF extraction scheme based on three corresponding algorithms is studied. For the first problem, a long-term stacking of repetitive symbols (LSRSs) algorithm is proposed to reduce the acquired signal variance, which contributes to the identification accuracy and long-term stability. For the second issue, we propose an artificial noise adding (ANA) algorithm to enhance the recognition robustness through regularization and channel adaptation. For the third issue, a verification algorithm based on the generative Gaussian probabilistic linear discriminant analysis (GPLDA) model is developed to handle unauthorized devices. Our robust RFF extraction scheme is verified in the experiments with 54 CC2530 ZigBee devices. It enables reliable node identification with the accuracy of 99.50% in the short rang line-of-sight (SLOS) scenarios for signals collected over 18 months, and 95.52% in the extensive multipath fading experiments. The equal error rate (EER) of the verification experiments with six authorized devices versus six unseen unauthorized devices is as low as 0.63%.

Radio-frequency fingerprints (RFFs) are promising solutions for realizing low-cost physical layer authentication. Machine learning-based methods have been proposed for RFF extraction and discrimination. However, most existing methods are designed for the closed-set scenario where the set of devices is remains unchanged. These methods can not be generalized to the RFF discrimination of unknown devices. To enable the discrimination of RFF from both known and unknown devices, we propose a new end-to-end deep learning framework for extracting RFFs from raw received signals. The proposed framework comprises a novel preprocessing module, called neural synchronization (NS), which incorporates the data-driven learning with signal processing priors as an inductive bias from communication-model based processing. Compared to traditional carrier synchronization techniques, which are static, this module estimates offsets by two learnable deep neural networks jointly trained by the RFF extractor. Additionally, a hypersphere representation is proposed to further improve the discrimination of RFF. Theoretical analysis shows that such a data-and-model framework can better optimize the mutual information between device identity and the RFF, which naturally leads to better performance. Experimental results verify that the proposed RFF significantly outperforms purely data-driven DNN-design and existing handcrafted RFF methods in terms of both discrimination and network generalizability.

Radar is an indispensable part of the Internet of Things (IoT). Specific emitter identification is essential to identify the legitimate radars and, more importantly, to reject the malicious radars. Conventional methods rely on pulse parameters that are not capable to identify the specific emitter as two radars may have the same configuration or a malicious radar can perform spoofing attacks. Radio-frequency fingerprint (RFF) is the unique and intrinsic hardware characteristic of devices resulted from hardware imperfection, which can be used as the device identity. This article proposes a robust and reliable radar identification scheme based on the RFF, taking linear frequency modulation (LFM) radar as a case study. This scheme first classifies the operation mode of the pulses, then eliminates the noise effect, and finally identifies the radar emitters based on the transient and modulation-based RFF features. The experimental results verify the effectiveness of our radar identification scheme among three real LFM radars (same model) operating at four modes, each mode with 2000 pulses from each radar. The identification rates of the four modes are all higher than 90% when the signal-to-noise ratio (SNR) is about 5 dB. In addition, mode 3 achieves almost 100% identification accuracy even when the SNR is as low as -10 dB.

Radar transmitter identification technology based on pulse descriptor word (PDW) is broadly used in military and civilian applications. However, as the complexity of the electromagnetic environment has increased, radar identification has been challenging. Radio frequency fingerprint (RFF) is an intrinsic hardware characteristic and has been widely employed for device identification. In this paper, we propose a robust RFF identification scheme for linear frequency modulation (LFM) pulse radars. The scheme includes a proposed piecewise curve fitting based denoising (PCFD) algorithm and a hybrid RFF identification algorithm. The PCFD algorithm can reduce the noise of LFM pulses without undermining RFF features. The hybrid RFF identification algorithm extracts both transient-based and modulation-based RFF features. Experimental results demonstrate that the proposed radar identification scheme can achieve a 100% identification accuracy when the SNR is about 0 dB.

Radio Frequency Fingerprinting (RFF) is one of the promising passive authentication approaches for improving the security of the Internet of Things (IoT). However, with the proliferation of low-power IoT devices, it becomes imperative to improve the identification accuracy at low SNR scenarios. To address this problem, this paper proposes a general Denoising AutoEncoder (DAE)-based model for deep learning RFF techniques. Besides, a partially stacking method is designed to appropriately combine the semi-steady and steady-state RFFs of ZigBee devices. The proposed Partially Stacking-based Convolutional DAE (PSC-DAE) aims at reconstructing a high-SNR signal as well as device identification. Experimental results demonstrate that compared to Convolutional Neural Network (CNN), PSCDAE can improve the identification accuracy by 14% to 23.5% at low SNRs (from -10 dB to 5 dB) under Additive White Gaussian Noise (AWGN) corrupted channels. Even at SNR = 10 dB, the identification accuracy is as high as 97.5%.

Radio frequency (RF) fingerprint is the inherent hardware characteristics and has been employed to classify and identify wireless devices in many Internet of Things applications. This paper extracts novel RF fingerprint features, designs a hybrid and adaptive classification scheme adjusting to the environment conditions, and carries out extensive experiments to evaluate the performance. In particular, four modulation features, namely differential constellation trace figure, carrier frequency offset, modulation offset and I/Q offset extracted from constellation trace figure, are employed. The feature weights under different channel conditions are calculated at the training stage. These features are combined smartly with the weights selected according to the estimated signal to noise ratio at the classification stage. We construct a testbed using universal software radio peripheral platform as the receiver and 54 ZigBee nodes as the candidate devices to be classified, which are the most ZigBee devices ever tested. Extensive experiments are carried out to evaluate the classification performance under different channel conditions, namely line-of-sight (LOS) and nonline-of-sight scenarios. We then validate the robustness by carrying out the classification process 18 months after the training, which is the longest time gap. We also use a different receiver platform for classification for the first time. The classification error rate is as low as 0.048 in LOS scenario, and 0.1105 even when a different receiver is used for classification 18 months after the training. Our hybrid classification scheme has thus been demonstrated effective in classifying a large amount of ZigBee devices.

With the increasing popularity of the Internet of Things (IoT), device identification, and authentication has become a critical security issue. Recently, radio frequency (RF) fingerprint-based identification schemes have attracted wide attention as they extract the inherent characteristics of hardware circuits which is very hard to forge. However, existing RF fingerprint-based approaches face the problems of unstable region of interest (ROI), high-cost feature design, and incomplete automation. To address these problems, this paper proposes a multisampling convolutional neural network (MSCNN) to extract RF fingerprint from the selected ROI for classifying ZigBee devices. A signal-to-noise ratio (SNR) adaptive ROI selection algorithm is also developed to alleviate the effect of semi-steady behavior of ZigBee devices owing to sleep mode switching. The proposed MSCNN uses multiple downsampling transformations for multiscale feature extraction and classification automatically. To validate and evaluate the performance of our proposed method, we design a testbed consisting of one low-cost universal software radio peripheral (USRP) as the receiver and 54 CC2530 devices as targets for identification. Extensive experiments are conducted to demonstrate the feasibility and reliability of MSCNN both in the line-of-sight (LOS) scenarios and non-LOS (NLOS) scenarios. The classification accuracy is as high as 97% under the LOS scenarios around SNR = 30 dB. Our scheme is robust over a wide range of SNRs under the LOS scenarios as well as under the NLOS scenarios.