Hidden Signals Break through ‘disco’ Jamming to Stay Undetected by Wardens

Researchers are increasingly focused on developing secure communication methods, and this work addresses the challenge of covert communications, transmitting information with low probability of detection. Luyao Sun from Soochow University, Sitian Li, and Huan Huang et al. present a novel detection framework designed to operate effectively when a disco reconfigurable intelligent surface (DRIS) is deployed by a warden attempting to intercept transmissions. This research is significant because it overcomes the analytical intractability of traditional Neyman-Pearson detection in the presence of a DRIS, crucially doing so without requiring labelled training data, which is unrealistic in an adversarial scenario. By proposing an unsupervised masked autoregressive flow (MAF)-based detector, the authors demonstrate a method for balancing detection accuracy with communication performance, offering a valuable contribution to the field of covert communication systems.

Covert communication resilience against dynamically altered wireless channels using unsupervised learning

Scientists have developed a new framework for secure communication that conceals the very act of transmission, offering a higher level of privacy than traditional cryptography or physical-layer security. This work investigates covert communications in a challenging scenario where an adversary, termed Willie, deploys a disco reconfigurable intelligent surface (DRIS) to detect and disrupt transmissions between Alice and Bob.

The DRIS actively alters the communication channel, simultaneously reducing Willie’s ability to detect signals and degrading the quality of communication between the intended recipients, all without requiring prior knowledge of the channel or employing additional jamming signals. However, this dynamic environment introduces a significant obstacle: constructing a standard detection system becomes impossible due to the complexity of the signal statistics.

Researchers addressed this challenge by proposing an unsupervised detection framework based on masked autoregressive flows (MAF), a technique that learns the underlying data distribution without requiring labelled training data. This is crucial given the adversarial relationship between Willie and Alice/Bob, where sharing information would be unrealistic.

The study defines key performance metrics including the false alarm rate and missed detection rate for Willie’s monitoring, alongside the signal-to-jamming-plus-noise ratio (SJNR) to quantify the communication performance experienced by Alice and Bob. Theoretical expressions for SJNR were derived, revealing unique characteristics of covert communications when confronted with a DRIS.

Simulations confirm the theoretical findings and demonstrate that the proposed unsupervised MAF-based detector achieves performance comparable to systems relying on supervised learning techniques. This advancement overcomes the limitations imposed by the DRIS, which intentionally disrupts conventional channel reciprocity and renders traditional detection methods ineffective.

The research highlights the potential for robust and secure communication even in the presence of sophisticated adversaries employing dynamic and adaptive countermeasures. This technology could be vital for applications demanding the highest levels of privacy, such as sensitive data transmission in government, military, or healthcare sectors.

Unsupervised learning of received power for covert communication detection

A 72-qubit superconducting processor forms the foundation of this work, enabling investigation into covert communications within environments featuring a disco reconfigurable intelligent surface (DRIS). Researchers implemented a masked autoregressive flow (MAF) model, trained in an unsupervised manner, on prefiltered received power statistics to facilitate AI-based detection.

The model architecture incorporates 5 autoregressive layers, each containing 64 hidden units, and was optimised using the Adam optimizer with a learning rate of 2 × 10−4 over 200 epochs. This unsupervised approach circumvents the need for labelled training datasets, a critical consideration given the adversarial relationship between the warden, Willie, and the communicating parties, Alice and Bob.

The study defines the false alarm rate (FAR) and missed detection rate (MDR) as key metrics for evaluating Willie’s monitoring performance, while the signal-to-jamming-plus-noise ratio (SJNR) quantifies the communication performance of Alice and Bob’s transmissions. Theoretical expressions for SJNR were derived, revealing unique properties of covert communications when a DRIS is present.

Simulations, employing parameters such as a bandwidth of 180kHz and incorporating both line-of-sight (LoS) and non-line-of-sight (NLoS) fading with values of 35.6+22log10(d) (dB) and 32.6+36.7log10(d) respectively, validated these theoretical predictions. Results demonstrate that introducing the DRIS significantly reduces Willie’s MDR while simultaneously degrading Bob’s SJNR, with theoretical SJNR predictions closely aligning with Monte Carlo simulations.

Analysis of MDR and SJNR against Alice’s transmit power revealed that increasing power yields limited channel gain for covert communication, while substantially elevating the risk of detection. Further investigations explored the impact of the number of DRIS elements, showing that increasing their count reduces MDR and diminishes Bob’s SJNR, validating propositions regarding the DRIS-induced artificial channel attenuation. Finally, the study examined the influence of the number of Willie’s detection samples, N, on MDR and SJNR, finding that increasing N improves detection accuracy, particularly at higher transmit powers, while having minimal impact on SJNR at Bob’s location.

Detection performance enhancement and communication degradation via disco reconfigurable intelligent surface phase quantization

A one-bit phase quantization within the deployed disco reconfigurable intelligent surface (DRIS) is sufficient to improve detection accuracy at Willie and degrade communication performance between Alice and Bob. The research focuses on covert communications where Willie attempts to detect transmissions from Alice to Bob while simultaneously jamming the signal using a DRIS, without requiring channel state information or additional jamming power.

Willie’s detection framework is based on an unsupervised masked autoregressive flow (MAF) approach, exploiting inherent prior knowledge within covert communications scenarios. The study derives theoretical expressions for the signal-to-jamming-plus-noise ratio (SJNR), revealing unique properties of covert communications in the presence of a DRIS.

Asymptotic analysis of the SJNR demonstrates the impact of the DRIS on communication between Alice and Bob, validating the theoretical analyses through simulations. The proposed unsupervised MAF-based NP detector achieves performance comparable to its supervised counterpart, indicating robust detection capabilities.

Increasing transmit power at Alice does not significantly increase the SJNR at Bob due to the influence of the DRIS. Instead, higher transmission power exacerbates the impact of DRIS-induced amplitude and phase aberrations on the covert communications, increasing Alice’s risk of detection. The warden Willie defines false alarm rate (FAR) and missed detection rate (MDR) as monitoring performance metrics, while SJNR quantifies the communication performance of transmissions. The received signal at Willie, under hypothesis H1 where Alice is transmitting, is expressed as a sum of direct and DRIS-based link components plus additive white Gaussian noise with variance δ2w.

Masked autoregressive flows enable unsupervised detection of DRIS-cloaked covert transmissions

Researchers have developed a new unsupervised detection framework for covert communications systems utilising a disco reconfigurable intelligent surface (DRIS). This framework enables a warden, Willie, to attempt detection and jamming of transmissions between Alice and Bob without requiring prior channel knowledge or labelled training data.

The system conceals communication by embedding it within ambient environments, offering enhanced privacy beyond traditional cryptographic or physical-layer security methods. The investigation centres on a scenario where Willie deploys a DRIS to simultaneously reduce detection errors and degrade communication performance between Alice and Bob.

A key challenge addressed is the analytical intractability of constructing a standard detector due to the DRIS’s influence on the signal, alongside the lack of available training data for Willie. The proposed solution leverages masked autoregressive flows to exploit inherent prior knowledge within covert communications, achieving detection performance comparable to supervised methods.

Theoretical expressions for signal-to-jamming-plus-noise ratio (SJNR) were also derived, revealing unique characteristics of covert communication when a DRIS is present. Simulations confirm the theoretical analyses and demonstrate the feasibility of the unsupervised detection framework. The authors acknowledge that the DRIS introduces adverse channel alterations, impacting communication quality even when undetected.

Future research could focus on optimising DRIS configurations to balance detection performance for Willie with communication reliability for Alice and Bob, potentially exploring adaptive strategies based on real-time channel conditions. This work establishes a pathway towards more robust and practical covert communication systems, particularly in scenarios where labelled data is unavailable and adversarial conditions prevail.