ABSTRACT Maintaining strict synchronization among receive channels, as well as ensuring persistent automatic calibration, represents a key challenge in coherent receiver architectures. In this paper, we propose an automatic technique for inferring fractional delay mismatches across the receive channels of cognitive radio (CR) receivers. The method leverages space‐time adaptive processing algorithms to track and compensate inter‐channel delay offsets with high precision. The proposed approach enables robust synchronization, improved receiver coherence and reliable operation in wideband CR systems where dynamic channel access is beneficial for CR systems, as it enables higher spectral efficiency. Moreover, the proposed method has a very low computational complexity and can be easily implemented on processing hardware.
ABSTRACT The large‐scale deployment of the Internet of Things (IoT) and the advancement of intelligent service management impose stringent delay requirements, as stable low latency and synchronous data arrival are critical. Traditional microservice data processing suffers from prominent challenges: over‐reliance on single‐edge servers, lack of backpressure awareness, and neglect of dynamic edge‐to‐edge offloading, leading to severe load imbalance. To address these issues, this paper proposes a backpressure matching learning‐based intelligent service management algorithm. First, an intelligent service management architecture for multimodal communication networks is constructed. Next, a service data processing optimisation problem is presented to minimise the weighted sum of service processing delay and delay section deviation. The problem is transformed into three sub‐problems: microservice embedding, edge‐to‐edge offloading, and computing resource allocation. Then, a backpressure matching learning‐based optimisation algorithm is introduced to solve the microservice embedding sub‐problem and edge‐to‐edge offloading sub‐problem. On this basis, a convex optimisation method is applied to derive the optimal computing resource allocation strategy. Simulation results show that the proposed method can effectively improve microservice data processing efficiency and boost the overall delay performance.
ABSTRACT AI‐driven medical sensor networks require sensing tasks, distributed learning, and communication resources to be coordinated across heterogeneous and potentially unreliable devices. We present RIFT, a round‐based architecture that links medical crowdsensing with federated learning through quality‐aware aggregation within a unified workflow. RIFT uses a medical sensing market for feasible task–agent matching and formulates cross‐round resource and learning control as a constrained Markov decision process. The architecture further integrates truth and quality estimation with anomaly‐aware federated aggregation, Top‐ update sparsification, and lightweight audit records. In the aggregation process, sensing‐quality and reliability information is combined with update‐level anomaly measures to limit the influence of unreliable participants. We implement the federated learning, quality‐gating, communication‐compression, and audit paths using Flower and PyTorch and evaluate them on CIFAR‐10 and NEU‐CLS as controlled proxy benchmarks. Experiments consider non‐IID partitions, composite poisoning attacks, optimiser changes, and different communication sparsification ratios. In the system‐level comparison, RIFT maintains higher accuracy than FedSense under the evaluated poisoning configurations, with particularly large performance differences on NEU‐CLS. The compression experiments further characterise the trade‐off between communication reduction and predictive performance. These results demonstrate the robustness and communication behaviour of the implemented RIFT components.
ABSTRACT To address the low‐altitude security risks posed by the illegal use of unmanned aerial vehicles (UAVs), this paper investigates the problem of radio frequency (RF)‐based UAV monitoring and identification in open environments with limited device resources. Traditional RF identification networks often suffer from excessive parameter counts, making them difficult to be deployed on resource‐constrained monitoring devices. To mitigate this issue, we propose HybridNet, which follows a global‐to‐local feature extraction paradigm, progressively capturing global dependencies and refining local signal structures. The model integrates multi‐scale convolution, dual‐path Transformer modelling and gated local refinement to achieve efficient and discriminative RF representation. To enable reliable unknown signal detection, a class‐adaptive hybrid metric combining Euclidean and Mahalanobis distances is introduced to construct distribution‐aware decision boundaries without requiring prior knowledge of unseen classes. A progressive multi‐loss optimization strategy further improves feature compactness and inter‐class separability. Experimental results demonstrate that HybridNet reduces model parameters by 97.4% and computational complexity by 6.3%, while improving known‐class accuracy by 5.64% and unknown‐class rejection accuracy by 1.26% compared with S3R. These results validate that the proposed method achieves a favourable balance between efficiency and recognition performance, making it suitable for deployment in resource‐constrained environments.
ABSTRACT Electronic health records (EHRs), as sensitive and critical healthcare data, are commonly encrypted before being uploaded to cloud servers. Attribute‐based searchable encryption (ABSE) has been extensively studied to enable fine‐grained search capabilities over encrypted data. However, existing ABSE schemes for medical data sharing remain limited by vulnerabilities such as single point of failure and potential leakage of user identity information. To address these limitations, we propose a novel decentralized attribute‐based searchable encryption scheme deployed in a cloud–edge collaborative environment. In the proposed scheme, the system master key is collaboratively managed by multiple attribute authorities (AA), each holding only a share of the master key, ensuring recoverability even if some Attribute Authorities are compromised. Data users obtain shares of their private keys from multiple authorities and reconstruct the complete private key via Lagrange interpolation. During data search requests, user identities are not disclosed to the AA, thereby enhancing privacy protection. Security analysis demonstrates that the proposed scheme can resist chosen keyword attack and chosen plaintext attack under the decisional bilinear Diffie–Hellman and decisional linear Diffie–Hellman assumption. Furthermore, performance evaluation shows that the computational, communicational and storage overheads of the proposed scheme are significantly lower compared to existing approaches.
