Han Zhu

Integrated sensing and communication (ISAC) enabled by backscatter communication is emerging as a promising paradigm for dynamic vehicular networks. However, the inherent decision latency in such systems poses a critical performance challenge, causing a mismatch between the executed decision and the actual channel state. To tackle this challenge, we propose a novel deep learning (DL)-based predictive framework utilizing a customized long short-term memory (LSTM) network to accurately forecast the continuous kinematic evolution of the system states from sequential historical measurements. These predictions then enable proactive transmit beamforming, which in turn effectively compensates for system latency to maximize end-to-end performance. Comprehensive simulation results demonstrate that the proposed predictive method achieves near-optimal performance and exhibits significant superiority in dynamic vehicular scenarios.

Visible light communication (VLC) is a promising solution for indoor wireless access owing to its unlicensed optical spectrum and compatibility with existing light-emitting diode (LED) infrastructure. However, practical VLC systems suffer from channel variability, multipath interference, and partial blockages. To address these challenges, we propose an intelligent adaptive deep neural network (IADNN) framework for generalized spatial modulation (GSM)-based VLC systems. The proposed framework comprises three innovations: a residual ensemble DNN architecture to capture nonlinear channel effects, a dynamic weight fusion mechanism to stabilize inference, and a multi-scenario data augmentation strategy to enhance robustness. Simulation results demonstrate that the proposed framework achieves near-optimal bit error rate (BER) performance across a wide signal-to-noise ratio (SNR) range, outperforming conventional zero-forcing (ZF) and minimum mean square error (MMSE) detectors with BER reductions of up to 87% and 66%, respectively. At the practical operating point of $10^{-3}$ BER, the IADNN incurs only a 0.5 dB SNR penalty versus the maximum-likelihood (ML) bound while cutting inference latency by 21 times. The ensemble strategy further lowers the coefficient of variation (CoV) by 58 % and improves reliability by 21 % over a single-DNN baseline. Although IADNN introduces moderate additional parameters, it offers a favorable complexity–performance trade-off, especially for large LED arrays where ML detection is computationally prohibitive. These properties make IADNN a practical enabler for real-time Light Fidelity (LiFi) hot-spots and dense Internet of Things (IoT) illumination networks in next-generation indoor environments.

Dense retail scenes pose unique challenges for both product detection and vacancy identification, where items are tightly packed on shelves and spaces between products carry critical inventory management implications. However, the existing detection frameworks, including state-of-the-art (SOTA) YOLO models, face limitations in feature fusion, particularly when integrating multi-level features and interpreting complex shelf arrangements. To address these challenges, we propose the RetailDet detection architecture and establish the RPV1K retail scene dataset. Our main contributions include: (1) A multi-attention module architecture that effectively fuses RGB and depth information for both product and vacancy detection; (2) A novel implicit gradient regulation (IGR) mechanism is proposed to address the redundant decision modules and feature modularization problems in traditional detection models. The proposed IGR mechanism dynamically regulates the gradient flow, optimizing the feature fusion path only during the training phase and not participating in calculations during the inference phase, thereby reducing computational overhead and enhancing model generalization performance; and (3) the RPV1K dataset, which is specifically tailored for retail detection tasks featuring product-vacancy co-occurrence scenarios. Experimental results show RetailDet significantly outperforms existing methods on the RPV1K dataset. Compared with YOLOv11-nano, RetailDet-nano achieves 10.6% mAP improvement while reducing inference latency by 17.2 ms. RetailDet-large achieves 85.2% precision with 20.79 M parameters. These innovations provide strong technical support for automatic shelf management and inventory replenishment in smart retail environments. Project resources can be obtained from https://github.com/bilychen88/RetailDet.

The Internet of Bio-Nano Things (IoBNT) is collaborative cell biology and nanodevice technology interacting through Molecular Communication (MC). The IoBNT can be accomplished by using the Information and Communication Theory (ICT) study of biological networks. Various technologies such as the Internet of Nano Things (IoNT), the Internet of Bio-degradable Things (IoBDT), and the Internet of Ingestible Things (IoIT) contribute to the development of IoBNT. Our survey discussed the Bio-Nano network and the role of IoT-based technologies along with a comparative study from various literature. We surveyed the various applications of IoNT in which the drug delivery for Insulin-Glucose system is prominent. Our survey aims to provide information about the Insulin-Glucose system involving the IoBNT and MC. We described the details of various factors for a diabetes analysis. We surveyed the diffusion coefficients of Insulin, Glucose, and the various parameters that influence insulin production in the body. Our survey identifies the security aspects of IoBNT such as attacks in nanonetworks, bio-cyber interface, Insulin-Glucose system, and their possible mitigation techniques. Our survey also provides a hierarchical model of the Bio-Nano network collaborating all the related technologies. Finally, our survey includes the research challenges involved in the proper handling of the IoBNT.

Ambient backscatter communication (AmBC) is emerging as a promising energy-saving and spectrum-efficient passive Internet of Things (IoT) technology that can be used for battery-less communication devices due to its low power consumption and cost constraints. However, in AmBC systems, recovering tag information at the reader is a challenging problem due to the difficulty in obtaining relevant channel state information (CSI). In this paper, we propose a variational Bayesian inference and machine learning (VBI-ML) based blind signal detection method, which can automatically recover tag information in AmBC systems. Firstly, two known labels are transmitted from the tag to the reader before valid data is transmitted, thus eliminating the need for CSI estimation. Secondly, we use the VBI approach with the Gaussian mixture model to obtain the real constellation information, which does not require a priori signal modulation technology to recover the tag information at the reader automatically. Using real constellation information, the signal detection problem is converted into a clustering problem. Finally, we cluster all the received signals using an improved expectation maximization algorithm in ML to learn the parameters in labeled and unlabeled signals and recover the signals. Thorough simulation results demonstrate that our proposed method performs similarly to the optimal detector with perfect CSI and outperforms traditional constellation learning methods. More critically, ML algorithms can mitigate direct link interference and simplify the number of estimated parameters.

In the rapidly evolving landscape of Internet of Things (IoT), Zigbee networks have emerged as a critical component for enabling wireless communication in a variety of applications. Despite their widespread adoption, Zigbee networks face significant security challenges, particularly in key management and network resilience against cyber attacks like distributed denial of service (DDoS). Traditional key rotation strategies often fall short in dynamically adapting to the ever-changing network conditions, leading to vulnerabilities in network security and efficiency. To address these challenges, this paper proposes a novel approach by implementing a reinforcement learning (RL) model for adaptive key rotation in Zigbee networks. We developed and tested this model against traditional periodic, anomaly detection-based, heuristic-based, and static key rotation methods in a simulated Zigbee network environment. Our comprehensive evaluation over a 30-day period focused on key performance metrics such as network efficiency, response to DDoS attacks, network resilience under various simulated attacks, latency, and packet loss in fluctuating traffic conditions. The results indicate that the RL model significantly outperforms traditional methods, demonstrating improved network efficiency, higher intrusion detection rates, faster response times, and superior resource management. The study underscores the potential of using artificial intelligence (AI)-driven, adaptive strategies for enhancing network security in IoT environments, paving the way for more robust and intelligent Zigbee network security solutions.