Stable Diffusion (SD) often produces degraded outputs when the training dataset contains adversarial noise. Adversarial purification offers a promising solution by removing adversarial noise from contaminated data. However, existing purification methods are primarily designed for classification tasks and fail to address SD-specific adversarial strategies, such as attacks targeting the VAE encoder, UNet denoiser, or both. To address the gap in SD security, we propose Universal Diffusion Adversarial Purification (UDAP), a novel framework tailored for defending adversarial attacks targeting SD models. UDAP leverages the distinct reconstruction behaviors of clean and adversarial images during Denoising Diffusion Implicit Models (DDIM) inversion to optimize the purification process. By minimizing the DDIM metric loss, UDAP can effectively remove adversarial noise. Additionally, we introduce a dynamic epoch adjustment strategy that adapts optimization iterations based on reconstruction errors, significantly improving efficiency without sacrificing purification quality. Experiments demonstrate UDAP’s robustness against diverse adversarial methods, including PID (VAE-targeted), Anti-DreamBooth (UNet-targeted), MIST (hybrid), and robustness-enhanced variants like Anti-Diffusion (Anti-DF) and MetaCloak. UDAP also generalizes well across SD versions and text prompts, showcasing its practical applicability in real-world scenarios.

Deep learning models have been widely applied in many fields, including medical image analysis and computer-aided disease diagnosis. However, these models are easily fooled by adversarial attacks from some created adversarial examples which are hardly distinguished by humans. In this paper, we implement a comprehensive assessment of six popular adversarial attacks on four multimodal medical image datasets using two main deep learning-based target models. Moreover, in order to evaluate the capability of defense, two new defense methods are leveraged to cope with medical adversarial attacks. More importantly, we also build and release a big multimodal medical adversarial database (including four medical adversarial datasets) with 712,596 examples to facilitate future research of adversarial attacks and defenses in the multimodal medical image field. Extensive experiments indicate that all-sided adversarial attacks like BIM are still scarce under different evaluation metrics and defenses are not universally successful.

In recent years, cooperative jamming (CJ) is introduced as a promising method to improve the security performance in the presence of eavesdroppers. By masking the confidential signal with the help of the cooperative jammer, the signal-to-noise ratio of the eavesdropper can be selectively reduced without prior information, whereas the authorized receiver is unaffected since the CJ can be suppressed by some means. Unfortunately, with the impacts of the clock jitter caused by non-ideal crystals, the received CJ may suffer sampling clock offset, carrier frequency offset and phase noise, which deteriorates the performance of the CJ cancellation at the authorized receiver. In this paper, the impacts of clock jitter on CJ cancellation are investigated. First, the deterioration of CJ caused by sampling clock offset is modeled as inter-symbol interference (ISI) in the time domain. Then, the impacts of the sampling clock offset, carrier frequency offset and phase noise on the spectral purity of CJ are modeled as inter-frequency interference (IFI). Furthermore, the expression of the power of residual CJ (PRCJ) is derived to analyze the impacts of clock jitter on CJ cancellation. Finally, numerical results prove the theoretical analysis, indicating that the PRCJ increases as the carrier frequency offset and the 3 dB coherence bandwidth of phase noise increases, and is more sensitive to phase noise.

Cooperative jamming (CJ) can mask signal-of-interest (SOI) by additional jamming signal in the presence of eavesdroppers, and thus improves the security performance from the perspective of physical layer. In exchange, an additional CJ canceller should be designed for the destination node to restore the SOI, whose performance is possibly degraded by the nonideal hardware features. Typically, the inherent nonlinearity of power amplifier is unavoidably distorted the CJ as the transmit power increases, leading to the limit of the CJ cancellation performance. Therefore, nonlinear CJ cancellation is suggested to alleviate the impact of nonlinear CJ components on the SOI. Besides, the short-term instability of the oscillator is characterized by phase noise in the time domain, resulting in a random phase mismatch between the received CJ and the reference signal, which further degrades the nonlinear CJ cancellation performance. Inspired by the above problems, this paper explicitly analyzes the impact of phase noise on nonlinear CJ cancellation, where the expression of CJ cancellation ratio (CJCR) in the presence of phase noise and nonlinear distortion is derived, where the coefficients of the nonlinear canceller are estimated by LS estimator, and the deterioration of the spectral purity of CJ caused by phase noise is modeled as inter frequency interference (IFI) additional noise. Theoretical analyses and simulation results show that the CJCR decreases as the normalized 3 dB coherence bandwidth of phase noise increases.

Nonlinear self-interference (SI) cancellation is crucial for eliminating the impact of the transmitter-side nonlinearity on the comprehensive SI cancellation performance in full-duplex radios. However, in the propagation environment with rich reflection paths of signals, the number of the delayed taps required for time-domain nonlinear SI cancellation booms exponentially with the increase of the multi-path number, leading to unacceptable complexity. In this paper, a novel frequency-domain nonlinear SI canceller based on successive interference cancellation (SIC) technique is proposed for orthogonal frequency division multiplexing (OFDM) modulated full-duplex systems subjected to both transmitter nonlinearity and receiver nonlinearity. The proposed frequency-domain nonlinear SIC (FD-NSIC) cancels nonlinear components one by one via separately estimating the channel frequency response suffered by each order nonlinear component, which significantly relaxes the computational cost. Besides, to avoid the influence of the strong correlation between nonlinear components, orthogonalization is operated before SIC to white basis functions, which accelerates the convergence speed of the proposed FD-NSIC. Simulations are explicitly performed, showing that compared with the conventional frequency-domain nonlinear canceller based on recursive least squares (RLS) channel estimation, the proposed FD-NSIC is capable of canceling the SI to the noise floor with only 52% computational cost under 50 dB interference-to-noise ratio.

In full-duplex (FD) systems, time alignment error between the received self-interference (SI) and the reconstructed one results in performance degradation of the SI cancellation. As a crucial part of SI cancellation, nonlinear interference (NI) cancellation is usually invested in the ideal situation, and the impact of the time alignment error on NI cancellation performance is rarely considered. In this paper, a robust NI cancellation algorithm is proposed to tackle the time alignment error in FD systems. At first, the impact of the time alignment error on NI cancellation is analyzed. The approximate expression of the residual NI power with time alignment error is derived to predict the NI cancellation performance. Then, an improved memory polynomial (IMP) model is derived to describe both time alignment error and nonlinearity. Finally, based on the IMP model, a novel NI cancellation algorithm with robustness to the time alignment error is proposed. Simulations present the impact of the time alignment error on NI cancellation capability, in which the theoretical value is coincident with the actual value. Besides, the proposed algorithm significantly enhances the NI cancellation performance. Specifically, when the time alignment error is 10% of the sampling interval and the interference to noise ratio is 60 dB, the proposed algorithm outperforms the conventional algorithm by 33 dB.