Device-free gesture recognition using mobile communication signals is a convenient and efficient technology with broad application prospects in smart homes and human-computer interaction. It utilizes the effect of gestures on surrounding signals to achieve gesture recognition. The cell-specific reference signals (CRS) information can be used to achieve the task in close-range training scenarios. However, when gestures are performed at long-range or in non-training scenarios, the recognition performance will significantly degrade. To enhance device-free gesture recognition capability in arbitrary scenarios, we propose the signal quality enhancement algorithm and the gesture spectrogram construction method to solve this problem. Specifically, we superimpose the CRS information from multiple carriers to improve the gesture signal-to-noise ratio and increase the gesture sensing range. Then, we extract the gesture dynamic components from the CRS information and construct gesture spectrograms to represent scenario-independent gesture motion patterns. Using the gesture spectrogram features, we design a deep network to accomplish the gesture recognition task. We built a prototype system on a software-defined radio platform. Experimental results show that our proposed method can effectively increase the gesture sensing range from 30m² to 228m² and achieve an average recognition accuracy of 82.5% for five types of gestures in arbitrary scenarios.

With the explosive growth of smart devices and applications, the demand for mobile service with higher data rate and better quality of service is growing rapidly. Ultra-dense networks, capable of providing higher network throughput, remain one of the key technologies for next-generation mobile communications. However, the densification of network further reduces the coverage of base stations and the distance between each other, which in turn leads to unnecessary and frequent handovers (HOs), affects the stability and reliability of communication links. HO failures can even occur due to the improper HO control parameter (HCP) values. To this end, a HO type prediction and parameter optimization method based on machine learning is proposed. First, the HO is divided into four categories: successful handover (SHO), ping-pong handover (PPHO), too-late handover (TLHO), and too-early HO (TEHO). Second, we combine reinforcement learning with supervised learning and propose a novel adaptive HCP adjusting scheme. Specifically, deep Q-network dynamically selects HCP values through environmental information and supervised learning-based HO prediction results. Simulation results demonstrate that our proposed scheme achieves a prediction accuracy of 94.83%, while reducing the PPHO rate by 15%, the TEHO rate by 2%, and the TLHO rate by 3%.

User-centric clustering is a valid solution to enhance the coverage and throughput for future mobile communication networks. However, the size of clusters, the location of nodes, and the number of cooperating nodes within the cluster can all have an impact on the data rate of the typical user. In this paper, the user-centric clustering with limited cooperation (LC) in downlink cell-free (CF) architecture is considered, and the effect of composite channels and intra-cluster cooperation on the data rate of the typical user is analyzed from a theoretical derivation level. Specifically, the user classification, the distributions of distances between the serving nodes, and the average data rates of each type of user are given, respectively. The approximate expressions of the Laplace transform (LT) of interfering power for different types of users are obtained with the Gauss-Hermitian integral approximation, and the long-term average data rate of the typical user is derived. Finally, Monte Carlo simulations are executed to verify the accuracy of the theory. The results show that shadowing fading should not be ignored for accurately evaluating user performance, and it is particularly important to reasonably select the radius of the cluster and the cooperation threshold that controls whether the access points cooperate or not.

If the future traffic of small base stations (SBSs) can be foreseen, we can systematically adjust the system resources to meet the quality of service (QoS) of users and realize the effective assignment of network resources. To this end, a dynamic channel allocation schemes based on traffic prediction is proposed. First, the machine learning is adopted to extract temporal and spatial features of the service traffic or load in a certain region, and then the prediction results and graph theory are both used to realize the division of a given frequency band and bandwidth allocation, in order to achieve the purpose of coordinating the interference between SBSs and improve the system throughput; Secondly, oriented to the fluctuation of service traffic in the region, a dynamic channel allocation method based on user satisfaction is proposed to fulfill the dynamic adjustment of the total system bandwidth and channel allocation, so as to utilize the frequency band resources more effectively. Simulations show that our proposed method improves the number of bandwidths allocated per SBS by a factor of 4.932 and 1.225 on average compared to OCA and CA-CM, and it realizes the data rate requirement of most users with less bandwidth.

Abstract Deep convolutional neural networks (CNN) have become the main method for face recognition (FR). To deploy deep CNN models on embedded and mobile devices, several lightweight FR models have been proposed. However, multi‐scale facial features are seldom considered in these approaches. To overcome this limitation, an attention‐based hierarchical pyramid feature fusion (AHPF) structure was proposed in this paper. Specifically, hierarchical multi‐scale features were directly extracted from the backbone based on its pyramidal hierarchy, and the bidirectional cross‐scale connection was used to better combine the high‐level global features with low‐level local features. In addition, instead of simple concatenation or summation, an attention‐based feature fusion mechanism was used to highlight the most recognizable facial patches, and to address the unequal contribution to the output during the fusing process. Based on the AHPF structure and efficient backbones, multiple sizes of lightweight FR models were presented, called HSFNet. After an extensive experimental evaluation involving 10 mainstream benchmarks, the proposed models consistently achieved state‐of‐the‐art FR performance compared to other lightweight FR models with same level of model complexity. With only 0.659M parameters and 94.94M FLOPs, our HSFNet‐05‐M exhibited a performance competitive with recent top‐ranked FR models containing up to 4M parameters and 500M FLOPs.

In this paper, the issue of user-centric clustering in cloud radio access networks is investigated. Specially, the outage probability of the typical user is derived by taking void cell and composite fading into account. The locations of remote radio heads and users working on the same resource block are modeled as the Poisson point process (PPP) and Mat$\acute {e}$rn hard-core point process of type II (MHCPP), respectively. Due to intractability of MHCPP, the PPP-based approximation is adopted. Then the closed expression of Laplace transform of the probability density function of the interfering power under composite fading channels is derived by Gauss-Hermite quadrature. Based on the approximated PPP, we obtain the outage probability of the typical user with user-centric clustering in the presence of void cell. Finally, the outage probability with or without considering void cell is verified and analyzed under different system parameters (i.g., the density of nodes, the radius of cluster, and the threshold of signal quality) and channel fading conditions (i.g., the pathloss exponent, shadowing standard deviation) through extensive simulations. Simulation results show that the effect of void cell on the system performance should be considered, especially when the density of nodes or the size of cluster is limited.

Increasing the deployment density of small base stations (SBS) is a key method designed to satisfy high data traffic in 5th generation mobile network (5G). However, a large number of SBSs in such ultra-dense network (UDN) may cause ping-pong handovers (HOs), accompanied by increased delay and HO failure. In addition, because of the separation of control and data signaling in 5G, the HO procedure must be performed in both layers. In this paper, we introduce an SDN-based intelligent dynamic HO parameter optimization strategy to minimize both HO failures and ping-pong HOs together. The goal of the proposed strategy is to reduce the HO failure rate and redundant HO (i.e. ping-pong HO) while enabling user equipment (UE) to make full use of the benefits of dense deployment of BSs. Simulation results present that the method proposed in this paper effectively suppresses the ping-pong effect and keeps it at a low level in all of the investigated scenes. In addition, compared with the other algorithms, the HO failure rate is significantly reduced and the throughput of UE is greatly increased, especially in the case of high BS density. Therefore, the benefits of intensive BS deployment are retained.

With the rapid increase of demand on mobile data traffic of user equipment (UE), network operators have begun to deploy abundant heterogeneous base stations (BSs) to ensure the quality of service (QoS) of UEs, which will cause new problems such as network congestion and load imbalance. If the pattern of user association (UA) can be adjusted in accordance with the results of traffic prediction, the performance of system will be greatly improved. Therefore, a new neural network approach based on spatial and temporal characteristics of traffic data is proposed for traffic prediction. The fluctuations of traffic in the future week are predicted by the proposed method. Then, UA is represented as a problem of maximizing the utility function of load balancing index, and a dynamic user association based on load prediction algorithm (DUALP) which aims to achieve a proactive load balancing is proposed. The QoS of UEs is ensured and the long-term stability of the system is achieved by DUALP. Experimental results show that compared to the classic UA strategies, the most optimal load distribution is realized by DUALP.

This article focuses on the joint optimization of beamforming and baseband unit (BBU) computing resource allocation, with the goal of minimizing the network power consumption for the downlink cloud radio access network (C-RAN). To reduce the computational complexity, we split the joint optimization problem into two subproblems for transmission and computation respectively. The subproblem of minimizing power consumption for transmission is a network-wide beamforming design problem. By using positive semi-definite relaxation (SDR) technology, we transform it into a convex positive semi-definite programming (SDP) problem, which can be solved with effect. For the second subproblem, which intends to minimize the power consumption for computation, a computing resource allocation scheme based on the Simulated Annealing algorithm is proposed, which minimizes the active servers in the BBU pool to save power while meeting each user’s computing resource requirement. The simulation results demonstrate that the algorithm proposed in this paper has a better performance compared with the existing algorithms.

Along with the dramatic rising of data rate, user-centric network (UCN) is regarded as a promising concept. However, densely deployed same-frequency small base stations (SBSs) in UCN deteriorates the user performance. Coordinated multipoint (CoMP) is an important technique for alleviating inter-cell interference. In this paper, to combat the severe interference, we proposed a user-centric cooperation scheme-Voronoi CoMP in a cell-less architecture. In this scheme, users are assigned their own clusters of SBSs respectively and the clusters for different users are not overlapped. This CoMP structure ensures that each user served by the base stations which are the closest ones of all SBSs to it and this will maximize the useful power of each user and in the meantime minimize the interference. The numerical results and simulations show that the proposed CoMP scheme can indeed promote the performance of dense small cell network (DSCN) corresponding to the designed metrics.

Software defined network (SDN) is considered as one of the most promising network architectures in the next generation mobile networks. SDN-enabled ultra dense network (UDN) has a simpler and more flexible network architecture, but its mobility management is still a challenging task. The major problem is the occurrence of frequent handover (FHO). Therefore, a SDN-enabled UDN architecture is firstly proposed to make the network more agile. Then, a deep Q-learning (DQN) method is used to control the handover (HO) procedure of the user equipments (UEs) by well capturing the characteristics of wireless signals/interferences and network load. In details, we use the SINR and the access rate per node to characterize the state of the UE. Thanks to the generalization ability of deep neural network (DNN), newly arrived UEs can use the trained neural network to avoid possible bad initial points. Experimental results show that the proposed scheme can reduce HO rate and guarantee the system throughput, which is better than the traditional HO scheme.

The cell association based on the maximum downlink received power is optimal in a conventional homogeneous network, but this association is not particularly suitable for heterogeneous networks (HetNets). Therefore, the concept of downlink and uplink decoupling (DUDE) is proposed for improving uplink performances of HetNets. However, in DUDE association scheme, macro user equipment (MUE) will suffer more interference from the offloaded users in uplink. In this paper, we investigate uplink interference mitigation through frequency reuse and power control schemes. The reverse frequency allocation (RFA) scheme is adopted to alleviate the cross-tier interference by increasing the distance between users of the same frequency. Then, a dynamic distributed power control (DDPC) with user admission control is proposed to further improve the uplink users performance and system spectrum utilization. The DDPC mitigates co-channel interference by dynamically updating transmitting power of active users and new access users, meanwhile it can maintain the link quality of active users above given signal to interference plus noise ratio (SINR) thresholds at all times. The dynamics and performance of the network are investigated through simulation experiments in terms of average SINR, evolution of uplink transmitting power, uplink SINR, and dynamic adjusting factor. The simulation results show that, in comparison with the DUDE scheme and DUDE with RFA scheme, DUDE with RFA and DDPC scheme can achieve a better quality of service (QoS) and uplink data rate performance.

With the exponential explosion of high traffic demand in wireless networks, the heterogeneous network (HetNet) is considered as one promoting method for improving the throughput of network and quality of service (QoS) of users. The HetNet consists of different cells, which operate at different transmit power. The small cells can be used to offload traffic from macro cells. Nevertheless, the uses of small cells also bring some problems. One problem is how efficiently an handover (HO) of small cells is performed in mobility management, when small cells are deployed in coverage area of a macro cell. In this condition, the various problems faced by users include frequent and unnecessary HO, ping-pong effect, which severely degrade the quality of traffic and users experience. Many existing works propose optimal solutions to improve the performance of long-term evolution (LTE) HetNets, which only consider the one of users' speeds and traffic classifications. In this paper, we propose an optimal HO algorithm to meet the margin coverage problems. Our proposed algorithm considers the different conditions of user speed and traffic classification. Our proposed scheme avoids users with high speed in handing off to home-evolved nodeB (HeNB) and makes them limited to evolved nodeB (eNB). Simulation results show that our proposed algorithm improves the performance in in terms of the downlink throughput, physical downlink shared channel (PDSCH) utilization, system delay, voice end-to-end delay and voice jitter.