Considering the dynamic resource allocation and energy management problem in the 5G Heterogeneous Cloud Radio Access Networks(H-CRANs) architecture for hybrid energy supply, a dynamic network resource allocation and energy management algorithm based on deep reinforcement learning is proposed. Firstly, due to the volatility of renewable energy and the randomness of user data service arrival, taking into account the stability of the system, the sustainability of energy and the Quality of Service(QoS) requirements of users, the resource allocation and energy management issues in the H-CRANs network as a Constrained infinite time Markov Decision Process (CMDP) are modeled with the goal of maximizing the average net profit of service providers. Then, the Lagrange multiplier method is used to transform the proposed CMDP problem into an unconstrained Markov Decision Process (MDP) problem. Finally, because the action space and the state space are both continuous value sets, the deep reinforcement learning is used to solve the above MDP problem. The simulation results show that the proposed algorithm can effectively guarantee the QoS and energy sustainability of the system, while improving the average net income of the service provider and reducing energy consumption.

Lossless data compression system is prone to bit error and causes error spread during communication transmission, which affects its application to file system and wireless communication. For the lossless data compression algorithm Lempel-Ziv-Welch (LZW), which is widely used in the field of general coding, analyzes and utilizes the redundancy of LZW compressed data, carries the check code by selecting part of the codeword and dynamically adjusting the length of its corresponding compressed string. A lossless data compression method Carrier-LZW(CLZW) with error correction capability is proposed. This method does not need additional data, does not change the data specification and coding rules, and is compatible with the standard LZW algorithm. The experimental results show that the file compressed by this method can still be decompressed by the standard LZW decoder. In the range of error correction capability, the method can effectively correct the error of LZW compressed data.

Due to the popularity of vehicle applications and the increase of the number of vehicles, the physical resources of roadside infrastructure are limited. When a large number of vehicles are connected to the vehicle networks, the energy consumption and latency are simultaneously increased. The framework for integrating the Content Delivery Network (CDN) and Mobile Edge Computing (MEC) can reduce the latency and energy consumption. In vehicle network, vehicle mobility poses a major challenge to the continuity of cloud services. Therefore, Mobility Management (MM) is proposed to deal with this problem. The Dynamic Channel Allocation algorithm with Overhead selection (ODCA) is used to avoid  the ping-pong effect and reduces the handover time of vehicles between cells. The cooperative game algorithm based on RoadSide Unit (RSU) is used for virtual machine migration and a learning-based price control mechanism is developed to process vehicular computation resources efficiently. The simulation results show that the proposed algorithm can improve resource utilization and reduce overhead compared with the existing algorithms.

Since real-time processing scenarios for ever-increasing amount and type of streaming data caused by the development of the Internet of Things (IoT) keep increasing, and strategies based on empirical knowledge for checkpoint configuration are deficiencies, the strategy faces huge challenges, such as time-consuming, labor-intensive, causing system anomalies, etc. To address these challenges, regression algorithm-based prediction is proposed for checkpoint performance. Firstly, six kinds of features, which have a huge influence on the performance, are analyzed, and then feature vectors of the training set are input into the regression algorithms for training, finally, test sets are used for the checkpoint performance prediction. Compared with other machine learning algorithms, the experimental results illustrat that the Random Forest (RF) has lower errors, higher accuracy and faster execution on CPU intensive benchmark, memory intensive benchmark and network intensive benchmark.

In virtualized network slicing scenario, one anomaly Physical Node (PN) or Physical Link (PL) in substrate networks will cause performance degradation of multiple network slices. For new measurements are achieved in each period, two online anomaly detection algorithms to monitor the working states of substrate networks in real time are designed. An online One-Class Support Vector Machine (OCSVM) algorithm is first proposed in this paper to detect the working states of PNs. Without requiring any labeled data, the model parameters of OCSVM can be updated based on the new measurements of Virtual Nodes (VNs) in each iteration. Then, an online Canonical Correlation Analysis (CCA) based PL anomaly detection algorithm is proposed according to the natural correlation of measurements between neighboring VNs of virtual links. With a small amount of labeled data, the algorithm can accurately analyze the working states of PLs. The simulation results verify the effectiveness and robustness of the proposed online anomaly detection algorithms for the virtualized network slicing.

In order to meet the demand of the substantial increase of wireless data traffic, the resource optimization of the Heterogeneous Cloud Radio Access Network (H-CRAN) is still an important problem that needs to be solved urgently. In this paper, under the H-CRAN downlink scenario, a wireless resource allocation algorithm based on Deep Reinforcement Learning (DRL) is proposed. Firstly, a stochastic optimization model for maximizing the total network throughput is established to jointly optimize the congestion control, the user association, subcarrier allocation and the power allocation under the constraint of queue stability. Secondly, considering the complexity of scheduling problem, the DRL algorithm uses neural network as nonlinear approximate function to solve the dimensional disaster problem efficiently. Finally, considering the complexity and dynamic variability of the wireless network environment, the Transfer Learning(TL) algorithm is introduced to make use of the small sample learning characteristics of TL so that the DRL algorithm can obtain the optimal resource allocation strategy in the case of insufficient samples. In addition, TL further accelerates the convergence rate of DRL algorithm by transferring the weight parameters of DRL model. Simulation results show that the proposed algorithm can effectively increase network throughput and improve network stability.

Considering the problems of low resource utilization and poor reliability of traditional network slice embedding, a Reliability-aware Network Slice (NS) Reconfiguration and Embedding (RNSRE) strategy is proposed. Firstly, a utility function of reliable embedding oriented reliability and available resources is established. Then, considering the resource requirements and the location constraints of Virtual Network Function (VNF), a method is proposed to quantify the reliability requirement of VNF. Based on the above works, the reliable network slice embedding problem is formulated as an integer linear programming which maximizes the profits of reliable VNF deployment while minimizing the consumption of link bandwidth resource. Finally, according to different types of network slices, a network slice reliable embedding algorithm based on neighborhood search and a network slice reconfiguration embedding algorithm based on key VNF backup are proposed. Simulation results show that the proposed algorithms improve the resources utilization and reduce the embedding cost while meeting the reliability of VNF.

Pedestrian detector performance is damaged because occlusion often leads to missed detection. In order to improve the detector's ability to detect pedestrian, a single-stage detector based on feature-guided attention mechanism is proposed. Firstly, a feature attention module is designed, which preserves the association between the feature channels while retaining spatial information, and guides the model to focus on visible region. Secondly, the attention module is used to fuse shallow and deep features, then high-level semantic features of pedestrians are extracted. Finally, pedestrian detection is treated as a high-level semantic feature detection problem. Pedestrian location and scale are obtained through heat map prediction, then the final prediction bounding box is generated. This way, the proposed method avoids the extra parameter settings of the traditional anchor-based method. Experiments show that the proposed method is superior to other comparison algorithms for the accuracy of occlusion target detection on CityPersons and Caltech pedestrian database. At the same time, the proposed method achieves a faster detection speed and a better balance between detection accuracy and speed.

Hilbert curve is an important method for high-dimensional reduction to one-dimensional. It has good characteristics of spatial aggregation and spatial continuity and is widely used in geographic information system, spatial databases and information retrieval. Existing Hilbert encoding or decoding algorithms do not consider the differences between input data, thus treating them equally. To this end, an efficient Hilbert coding algorithm Front-Zero-Free Hilbert Encoding(FZF-HE) and an efficient decoding algorithm Front-Zero-Free Hilbert Decoding(FZF-HD) are proposed. FZF-HE and FZF-HD can quickly identify the 0 s of the front part of input data before iterative calculation by combining efficient state views and first bit-1 detection algorithm, thus reducing the number of iterations and the complexity of the algorithm, and improving the encoding and decoding efficiency. The experimental results show that efficiencies of these two algorithms are slightly higher than existing algorithms for uniform distributed data, and are much higher than existing algorithms for skew distributed data.

With an increasing diversity in modern architectural design, the inner structure of buildings is much more complex than before, which makes the traditional fire emergency escape indication system fail to provide people with real-time instructions because of its inflexibility of changing direction. These failures always lead people to dangerous areas during a fire emergency, which is actual a threaten to people in buildings. A combined algorithm to find a path dynamically during a fire emergency based on Dijkstra and Ant Colony Optimization (ACO) algorithm is presented in this article. This new algorithm shortens the programming time by getting a globally optimal path based on Dijkstra algorithm and operates every single point with ACO algorithm in sequence to get a best path. The combined algorithm is tested by a simulation, in which it is proved effective in adjusting evacuation path depending on the point of ignition. The changeable real-time indication will extend the escaping time with people in a burning building, which is quite precious for saving lives.

Source node location protection is critical to the Marine Wireless Sensor Networks (MWSNs), especially for unattended environment. However, due to most of the static deployment and the limitations in energy, storage and communication capabilities of the sensors, MWSNs are vulnerable to various location (and derivative) attacks. In this work, the node location privacy protection issues are studied from both aspects of attacks and defenses. First, a new two-phase location attack is proposed for two important types of nodes (including base station and source node). It can locate a base station node within few amounts of local wireless transmission monitoring, and then reversely traces the location of the source node. Different from existing methods, the proposed attack determines the node location based on the transmission direction, which can break through existing defenses. Then, to defend against such attack, a Hilbert-filling-curve-based Location-privacy Protection Scheme (HLPS) is designed for MWSNs. The theory analysis and confrontation experiment of attack and defense show that the proposed scheme owns capable of protecting the location privacy of the target node with moderate communication and computation overhead.

The lightweight block cipher algorithm PUFFIN based on substitution-permutation network structure is widely used in resource-constrained hardware environments. Differential fault attack is a more effective attack method for hardware cryptographic algorithms. The multi-bit fault model for PUFFIN algorithm is improved. By constructing the relationship between the output difference and the possible input values, the single input value of a single S-box can be determined by injecting 5 faults. The probability of successfully recovering the round key is 78.64%, and the initial key can be recovered.

In order to increase the classification speed of Aggregated Bit Vector (ABV) algorithm, an Improved Aggregated Bit Vector (IABV) algorithm is proposed, which is connection-oriented. Based on the characteristic that the packets which belong to the same connection have similar classification results, IABV establishes a Hash table-rule set two-level searching structure. It first searches in the Hash table to check the packet classification rule and then finds the matching rule in the rule set when the Hash table lookup fails. To avoid the accumulation of rules in the table, a collision handling mechanism is proposed. It judges whether to overwrite the Hash table entry which is collision according to the last hit time of the entry; Secondly, for the purpose of accelerate rule set searching, IABV divides each dimension into multiple intervals equally and employs array to index these intervals; Finally, the prefix in the rule is converted into range to reduce the complexity of the search structure, so that the time and memory consumption of the algorithm can be decreased. The experiment result shows that the performance of the algorithm can be improved by converting prefix into range and the time performance of IABV algorithm is significantly improved compared with the ABV algorithm under the same conditions.

In the view of the integrity verification problem of data sharing on the cloud platform, a Shared Data auditing scheme for efficient Revocation of group Members via multi-participation (SDRM) is proposed. First, through the Shamir secret sharing method, multiple group members participate in revoking the illegal group members, ensuring the equal rights between the group members. Second, this scheme combines with algebraic signature technology, the file identifier identifies the data owner’s upload data record and the normal group member’s access record, enabling the data owner to update efficiently all of its data. Finally, theoretical analysis and experimental verification of the correctness, safety and effectiveness of the scheme show that the scheme meets the requirement of efficient cancellation of group members, at the same time, as the number of data owners increases, the efficiency of updating data in this scheme is significantly higher than that of NPP.

Parameter estimation is essential for SAR imaging of moving targets. The existing algorithms mainly estimate the radial velocity and azimuth velocity of the moving target, but the normal velocity of the three-dimensional moving target can not be estimated. In this paper, a joint estimation algorithm of azimuth velocity and normal velocity is proposed by using an airborne multi-channel SAR system with L-shaped baseline. The algorithm extracts the moving target signal in Range-Doppler domain, and estimates the azimuth and normal velocity jointly using the phase differences between multiple SAR images. The algorithm does not rely on image registration, does not need to solve Doppler ambiguity. Therefore, the algorithm has high estimation accuracy and robustness, and has strong practical significance and application value.

A simulation model of total power microwave radiometer is developed for the microwave humidity and temperature sounder onboard the FY-3 satellite. The key components such as mixer, low noise amplifier, local oscillator, filter and detector are parametrically modeled. The model is studied from the aspect of signal processing, and dynamic range, sensitivity and linearity of the simulation system are evaluated and analyzed. The correctness of the simulation model is verified by comparing them with the test results of the actual system.