Location awareness is crucial for numerous emerging wireless indoor applications. Deep learning algorithms have demonstrated the potential for achieving the required level of positioning accuracy in indoor environments. However, obtaining abundant labels for data-driven machine learning is costly in practical situations. As an effective solution to alleviating the insufficiency of labeled data for deep learning-based indoor positioning, deep semi-supervised learning (DSSL) can be employed to lessen the dependency on labeled data by exploiting potential patterns in unlabeled samples. In this paper, we propose an Adapted Mean Teacher (AMT) model within the DSSL paradigm for indoor fingerprint positioning by using a channel impulse response. To enhance the generalization of the trained model, we design an efficient implicit augmentation scheme for the training process in the AMT model. Furthermore, we develop a tailored residual network to efficiently extract location characteristics in the AMT framework. We conduct extensive simulation experiments for indoor scenarios with heavy non-line-of-sight conditions based on open datasets to demonstrate the effectiveness of our proposed AMT model. Numerical results indicate that the AMT model outperforms several consistency regularization methods and the pseudo-label method in terms of positioning accuracy and lower positioning latency, achieving a mean error of 90cm when using a small number of labels.

With the emerging Internet of Things paradigm, massive Machine-Type Communication (mMTC) has been identified as one of the prominent services that enables a broad range of applications with various Quality of Service (QoS) requirements for 5G-and-beyond networks. However, it is very difficult to employ a monolithic physical network to support various mMTC applications with differentiated QoS requirements. Moreover, in ultra-dense mobile networks, the scarcity of the preamble and Physical Downlink Control CHannel (PDCCH) resources may easily lead to resource collisions when a large number of devices access the network simultaneously. To tackle these issues, in this paper, we propose a network slicing-enabled intelligent random access framework for mMTC. First, by tailoring a gigantic physical network into multiple lightweight network slices, fine-grained QoS provisioning can be accomplished, and the collision domain of Random Access (RA) can be effectively reduced. In addition, we propose a novel concept of sliced preambles (sPreambles), based on which the transitional RA procedure is optimized, and the issue of preamble shortage is effectively relieved. Furthermore, with the aim of alleviating PDCCH resource shortage and improving transmission efficiency, we propose a learning-based resource-sharing scheme that can intelligently multiplex the PDCCH resources in the naturally dynamic environment. Simulation results show that the proposed framework can efficiently allocate resources to individual mMTC devices while guaranteeing their QoS requirements in random access processes.

The emerging paradigm of perceptive networks has attracted much recent attention to the integration of communication and sensing functionalities. In this paper, we mainly consider an integrated sensing and communication (ISAC) system comprised of a target and multiple energy harvesting (EH) enabled low-power sensors. To support a monitoring task by charging the sensors with radiated radio-frequency power while keeping sensing the target, a transmit beamforming-based scheme is designed for focusing energy in the directions of sensors and the target. Given a quality-of-service (QoS) constraint on the sensing task, a unified stochastic scheduling scheme is presented to assure the freshness of the sensor information at a monitor. Under the stochastic scheduling scheme, we minimize the average Age of Information (AoI) of all sensors by dynamically adjusting transmission power and beamforming weights in use. Given a power constraint, the optimal scheduling policy is obtained by a constrained Markov Decision Process (CMDP) approach. More specifically, an optimal tradeoff between the average power and average AoI is revealed through linear programming (LP) formulation, which is further demonstrated by extensive numerical results.

With the development of sensor technology and computer vision, the speed of video generation is increasing exponentially. The workload of video browsing and the amount of stored data are increasing. In order to achieve fast browsing and efficient storage, video structure technology is increasingly important. This paper focuses on the video structure technology in the construction of smart cities, outlines the system process of video structure, and studies the extraction algorithm of human and vehicle target attributes based on deep learning. And combined with the actual video monitoring scene, the experiment is carried out. Compared with traditional image processing methods, the video structure target attribute extraction model based on deep learning has high detection accuracy and strong algorithm generalization ability. The intelligent analysis and processing of data by video structure technology is conducive to quickly increasing the speed of target search, reducing storage capacity, solving the problem of long-term storage, and realizing full automation of surveillance video.