Wireless Sensor Networks

LoRa technology has emerged as an interesting solution for Low Power Wide Area applications. To support a massive amount of devices in large-scale networks, it is necessary to design an appropriate parameter allocation scheme for device. LoRa devices provide high flexibility in choosing settings of communication parameters (including spreading factors, bandwidth, coding rate, transmission power, etc), which results in there are over 6000 settings for choosing. However, the existing methods mainly focus on the same parameter setting for network deployment. To this aim, the impact of different parameter selections on communication performance is analyzed first. Then, channel collision and link budget model are established and implemented in the NS3 simulator. A dynamic parameter selection method based on orthogonal genetic algorithm (OGA) is introduced to solve the model, ultimately according to link budget, each device selects its parameter setting, which minimized collision probability. Finally, simulation results show that the OGA algorithm proposed in this paper can improve the packet delivery rate by 30%. Knowing different packet sizes have an impact on network performance, the experiment also evaluated the impact of different packet sizes on network transmission reliability under different parameter setting methods, the introduced OGA has significantly improved adaptability and scalability of the network in the case of high payloads.

For the node deployment problem in three-dimensional heterogeneous sensor networks, the traditional virtual force method is prone to local optimization and the parameters required for calculation are uncertain. A spatial deployment algorithm for 3D mobile wireless sensor networks based on weighted Voronoi diagram is proposed (TDWVADA) to solve the problem. Based on the positions and weights of all nodes in the monitoring area, a three-dimensional weighted Voronoi diagram is constructed. Next, the central position of each node’s the Voronoi region is calculated and the position is regarded as target position of the node movement. Each node moves from the original position to the target position to complete one iteration. After multiple iterations, each node is moved to the optimal deployment location and network coverage is improved. In view of the initial centralized placement of sensor nodes, the addition of virtual force factors is added to the TWDVDA algorithm. An improved algorithm TDWVADA-I was proposed. The algorithm enables nodes that are centrally placed to spread quickly and speeds up deployment. The simulation results show that TDWVADA and TDWVADA-I effectively improve the network coverage of the monitored area compared to the virtual force algorithm and the unweighted Voronoi method. Compared with the virtual force method, the coverage of TDWVADA has increased from 90.53% to 96.70%, and the coverage of TDWVADA-I has increased from 81.12% to 96.56%. Compared with the Voronoi diagram method, the coverage of TDWVADA has increased from 85.01% to 96.70%, and the coverage of TDWVADA-I has increased from 80.82% to 96.56%. TDWVADA and TDWVADA-I also greatly reduce the energy consumption of the network. Experimental results demonstrate the effectiveness of the algorithms.

We study the joint uplink and downlink (JUAD) resource allocation problem in D2D networks, where D2D sender communicates with D2D recipient by reusing the channel of cellular users (CUs). In order to maximize the throughput of D2D networks, we model the JUAD problem as a mixed integer nonlinear programming problem (MINLP). Since the problem is NP hard, to solve it better, we divide it into two sub-problems by analyzing the structure of the primal problem, including channel assignment and power allocation. Then, we turn the sub-problem of power allocation into convex problem by the Lagrangian dual theory for getting the optimal power value of CUs and D2D pair. Next, an improved Hopcroft-Karp algorithm is proposed to solve the sub-problem of channel allocation, which has lower complexity compared with the traditional channel allocation approaches. Finally, extensive simulations show that our proposed approach achieves a near optimal solution.

Accurate link quality estimation is a prerequisite for efficient routing in wireless sensor networks. Good link quality estimators should provide agility and stability simultaneously, which not only filter out transient link quality fluctuations but also respond quickly when sudden changes arise. However, only stability or agility is considered as optimization goal in existing estimators, so their performance is always below expectations. In this paper, a fluctuation adaptive link quality estimator is proposed, which adjusts smoothing factor of the estimation dynamically according to the degree of link quality fluctuations and achieves equilibrium of stability and agility. Experimental results show that stability of the proposed estimator is same as that of existing stable estimators when there are transient fluctuations, and agility of the proposed estimator is same as that of existing agile estimators when sudden changes arise. More importantly, compared with existing estimators, the estimate error of the proposed one is reduced by 22.5%–31.8% for different link characteristics.

This paper presents Rate Adaptive Broadcast (RAB), a novel wireless design that enables the rate adaptive broadcast in Internet of things (IoT). Broadcast is common in IoT due to the ubiquitous tree topologies. Channel resource is usually underused in broadcast because there is no rate adaptation in conventional broadcast and the data rate is always set as the lowest one by default. Existing rate adaptation methods work only for unicast or multicast, relying on information interaction between senders and receivers. These methods cannot directly apply in broadcast, which is a one-way transmission without acknowledgement (ACK). It is also impractical to transplant conventional ACK into broadcast, otherwise, massive ACKs will lead to a heavy overhead. To tackle this dilemma, we propose RAB, which allows the sender to broadcast data ceaselessly while adjusting the data rate according to real-time channel states. The core contribution is the subtly designed feedbacks that can be concurrently delivered and do not affect any reception. We implement RAB on USRPs and establish a 20-node IoT testbed. Experiment results demonstrate that the throughput is largely improved. The throughput of RAB is 2.8x of the standard WiFi and 1.3x of MuDRA, the state-of-the-art multicast rate adaptation method.

The large-scale of data collection from IoT devices to central cloud brings several challenges that need to be overcome, especially for needs of real-time collection and bandwidth restrictions. In order to address this issue, we proposed a data collection method that combine cloud with edge node by using deep learning technology to provide the data collection service. The cloud is responsible for storing the large amount of historical sensor data, training the deep learning model, and deploying the model to the edge side. The edge node will receive the model of data prediction and then determines whether the real data will be uploaded to the cloud to optimize the model. Experiments show that the method we proposed can not only increase the speed of data collection, but also reduces the network traffic and eliminates bandwidth load effectively.

Many applications ask for information of nodes in wireless sensor networks (WSNs), among which the node location is an important type of information. In WSNs, localization of target node is often obtained with aid of anchors with providing distance-related information in many algorithms. However some anchors may be attacked in real environments. In order to guarantee the accuracy of localization, it is necessary to identify the malicious anchors under attack. In this paper, we propose a localization detection algorithm with malicious anchors existing in WSNs. The algorithm firstly utilizes Isolation Forest to confirm the reference anchors. After obtaining the initial position estimate of the target nodes, we establish a detection model using the consistency of the measuring distance and the Euclidean distance to the initial position estimate. Finally sequential probability ratio testing (SPRT) is carried out on the remained anchors. The simulation results demonstrate the proposed algorithm can efficiently identify the malicious anchors and outperforms other existing algorithms.

In wireless sensor networks (WSNs), sensor data are usually corrupted by the noise. Meanwhile, it is inevitable to face the problems of node energy in WSNs. For both of these questions, this paper proposes a data gathering method via compressed sensing combined with cross validation. In the proposed method, data gathering via CS can save and balance energy consumption of sensor nodes due to the features of CS, and CV technique is used to judge whether stable reconstruction have been obtained. This method is essentially an adaptive intelligent method. Unlike the existing methods, the proposed method does not need the knowledge of signal sparsity, noise information and/or regularization parameter while those knowledge is expensive to acquire, especially in adaptive systems. That is to say, the method proposed in this paper is not sensitive to signal sparsity, noise, regularization parameters and/or other information when it is used for WSNs data collection for noise case, but the existing methods rely heavily on the prior information. Experimental results show that the proposed data gathering method can obtain stable reconstruction results for noisy WSNs in the case of unknown signal sparsity, noise and/or regularization parameters.

The rapid development of wireless sensor networks (WSN) in various fields has brought great convenience. Since most wireless sensor nodes are powered by batteries, energy efficiency is very important for WSN, and many existing routing protocols aim to reduce energy consumption. At present, the emergence of cross technology communication (CTC) enables direct communication between heterogeneous nodes at the physical layer. Therefore, new routing algorithms need to be designed for WSN based on CTC, which considers the heterogeneous characteristics of sensor nodes such as the energy heterogeneity, the mobility of LTE nodes, etc. In this paper, we propose an energy efficient fuzzy clustering routing protocol based on CTC, which is named as Fuzzy-K. Different from other protocols, Fuzzy-K first uses k-means algorithm to form balanced clusters and then select CH (cluster head). In the proposed protocol, the Mamdani fuzzy inference system (FIS) is used twice to select the initial cluster center and the final CH. The input parameters of these two systems are obviously different, which considers the differences in frame length, mobility and other heterogenous characters between nodes. Simulation results of three different network topologies show that compared with LEACH, EEHCCP, TEAR and DUCF, Fuzzy-K protocol has better performance in extending network life cycle, balancing network load and improving network throughput. And the average value of the rounds when first node dies could be 27% higher than the protocols mentioned above. What’s more, the proposed protocol is scalable across a range of situation by changing parameters of the FIS.

In Named Data Networking (NDN), Pending Interest Table (PIT) is proposed to record the forwarding information of interest packets forwarded but not responded. Each incoming interest packet or data packet needs to be queried and processed in PIT, and the overhead would rise as the scale of PIT increases. Therefore, PIT is required to have a very high processing speed. To effectively improve the forwarding efficiency of PIT in NDN, a new architecture of the PIT using a hot table to achieve prefix grading is designed and implemented. The concept of “prefix value” is proposed to determine the value of a prefix carrying the information content of an interest packet, and to store and prioritize the prefix information with a higher value. The results of the comparison experiment show that the architecture of the PIT with a hot table can significantly improve processing speed of the PIT and accelerate forwarding efficiency of the NDN node.

Detecting structural damage in real time is important and challenging for bridge structural health monitoring systems, especially when large amount of time series monitoring data are collected for continuous monitoring and evaluation of abnormal conditions. Conventional approaches fail to efficiently process such large-scale data in real time due to high storage and processing cost. In this paper, we present an efficient real-time bridge structural condition evaluation based on data compression. We introduce an efficient time series representation to compress sensor data into symbol streams by applying symbolic aggregate approximation (SAX), which transforms sensing data into symbolic representation to reduce dimension while preserving important features and guaranteeing low-bounding distance. Upon receiving sensing data in real time, we compress raw data into SAX representation before evaluation. Then, we evaluate bridge structural condition by performing classification based on compressed data efficiently. The proposed method is evaluated using a typical real bridge data set from SMC. Compared with the prediction results on original data using existing methods, our approach reduces the processing time from hours to several seconds with improved accuracy, showing that the proposed method is effective in improving both efficiency and accuracy of bridge structural condition evaluation in real time.

In the mobile crowd sensing (MCS) system, task assignment is a core and common research issue. Based on the traditional MCS platform, there is a cold start problem. This paper introduces social networks and communication networks to solve the cold start problem. Therefore, this paper draws on social influence to propose a greedy task assignment algorithm H-GTA. The core idea of the algorithm is to first use the communication network to select seed workers in a heuristic manner according to the recruitment probability and then the seed workers spread the task on social networks and communication networks simultaneously. The publisher selects workers to assign the task in a greedy way to maximize the task’s spatial coverage. When calculating the probability of recruitment, this paper considers various factors such as worker’s ability, stay time and worker’s movement to improve the accuracy of recruitment probability. Considering the influence of worker’s movement on recruitment probability, a worker’s movement prediction algorithm based on meta-path is proposed to analyze worker’s movement. The experimental results show that compared with the existing algorithms, the algorithm in this paper can guarantee the time constraint of the task, and have better performance in terms of spatial coverage and running time.

Barrier gap repair in wireless sensor networks is an inevitable problem in barrier coverage research. This paper studies the repair of the barrier gap of wireless sensor networks in 3D environment and improves the coverage performance. This paper proposes a distributed barrier gap repair method in 3D environment. Firstly, look for the repair chain in a three-dimensional environment through an improved ant colony algorithm. Then, construct a relationship model between the mobile node and the patch chain, and then select the optimal barrier gap repair location. Finally, the priority model of gap filling is set by the moving distance of the node and whether it is the key area, and the corresponding repair of the gap is performed. The results of real experiments and simulation experiments show that the proposed algorithm can reduce the number of nodes used and reduce the mobile energy consumption when the mobile node repairs the gap.

The sensor-cloud technology alleviates the restrictions of the traditional wireless sensor networks (WSNs) in terms of storage, computation, and scalability by integrating WSNs with cloud computing. In recent years, sensor-cloud technology is increasingly applied to various real-world applications, especially in agriculture irrigation. With the powerful computing and storage sources, the sensor-cloud enables the massive on-field sensing data to be processed efficiently. Furthermore, the virtualization technology allows multiple clients, typically farmers, to share the same infrastructure resources at a low cost. In this paper, we propose a novel agriculture irrigation system by applying the sensor-cloud technology into the traditional sprinkler irrigation. Targeting the practical irrigation scenes, we illustrate the specific work pattern of the proposed system. Finally, compared with the conventional WSN-based scheme, the simulation results show that our system achieves about 31.06%–41.24% decrease in energy consumption.

Sentiment analysis is a hot topic and has various application scenarios. The polarity recognition of implied sentiment in a sentence can be achieved by the way of statistic and prediction. However, the polarity of sentiment is influenced by funny, humorous, and ironic Internet cultures, therefore it is hard to be verified. In this paper, we use a deep memory network with the auxiliary sequence to obtain the text feature vectors. Then the Emoji set and the special word set from the internet are imported, which are combined with the formal text feature vectors to form the classification feature vectors. At last a binary classifier is designed to get the final polarity prediction. Besides, an incremental online learning method with feedback adjustment is introduced to update the Emoji set and the special word set. Experiment results show that, on the IMDB datasets the prediction accuracy is about 85% and on the Chinese implied sentiment evaluation datasets the prediction accuracy is about 96%, which prove the effectiveness of the model.

With the development of the globalization, the living standard has been improved. The increased number of vehicles on road made the method of controlling the traffic light which is traditional and empirical with poor efficiency. The default traffic light system can not satisfy peoples travel demand especially on the congested intersection. The intelligent traffic light system can adapt to the flow at the intersection and change the Traffic Light Duration Cycle (TLDC) in order to reduce travel time of all vehicles. Moreover, there are High Priority Vehicles (HPV) on road, designed to reach the destination on time. So they should be given privileges to avoid traffic jams. The proposed work, based on the priority of vehicles, aims at providing an intelligent traffic light system which the HPV can send request to after be loaded at junction. According to the highest priority, System would turn traffic light green to clear the Road Segment (RS) for saving travel time of HPV. The system is tested on Simulation of Urban Mobility (SUMO) and use the Traffic Control Interface (TraCI) of Python. The results show the effectiveness of the intelligent traffic light system. It may has significant theoretical as well as practical value for Intelligent Transportation System (ITS) in the future.

Heterogeneous information network (HIN) is widely used in recommendation system because of its superiority in complex information modeling. However, the existing HIN-based methods ignore two issues. First, low-quality information may cause users to be dissatisfied with the recommendation results. Secondly, HIN-based recommendations are difficult to predict the user’s attitude toward the item. Therefore, this paper proposes two improvement strategies: (1) propose a semantic information filtering strategy to filter low-quality information and improve recommendation efficiency; (2) integrate tag information into HIN-based recommendation system to achieve personalization recommend. This paper verifies the validity of the proposed model on two real data sets.

In the campus scenario, a basic community of collaborative teams is formed among the nodes participating in the collaborative learning interaction in the mobile opportunistic network. Due to the existing research does not consider the weak connection, node influence and the contact characteristics between nodes. In this paper, a routing method using a collaborative group as a communication unit is proposed. The route mainly counts the contact characteristics among the groups according to the characteristics of the node movement and predicts the subsequent contact situation. Combined with the weak connection relationship and the node’s influence, the optimal node to be transmitted is selected. It has been experimentally verified that the routing method can greatly improve the speed of message dissemination and avoid unnecessary message redundancy and waste of contact opportunities.

Indoor intrusion detection technology has been widely used in smart home management, public safety, disaster relief, and other fields. In recent years, with the rapid deployment of Wireless Local Area Network (WLAN) and general support of the IEEE 802.11 protocol by mobile devices, indoor intrusion detection can be realized conveniently. Most of the existing indoor intrusion detection algorithms have large computational and storage overheads and do not consider the instability of signals in the indoor environment. In response to this compelling problem, this paper proposes a new integrated redundant Access Points (APs) reduction and transfer learning for indoor WLAN intrusion detection via link-layer data transformation. First, the detection technology for mobile APs based on a fuzzy rough set is exploited to filter the redundant APs in the indoor environment. Second, the target domain and the source domain are constructed through the link-layer data of the online phase and the offline phase. Then, the Maximum Mean Deviation (MMD) minimum value corresponding to the two domains is worked out by the mathematical statistics method to obtain the optimized migration matrix, and the link-layer information of the two domains is transferred into the same subspace by using the matrix. Finally, the optimal intrusion detection classifiers are obtained by training the transferred link-layer data. This method not only has better robustness in the complex indoor environment but also reduces time and labor costs.

We study the problem of heterogeneous task assignment in mobile crowdsensing (MCS) scenarios where the opportunistic mode and participatory mode coexist. Workers in opportunistic mode complete tasks during their daily routines while workers in participatory mode complete tasks by moving to designated locations. This problem can be simplified into a Knapsack problem which is NP-hard. Then, to solve this problem, we propose a two-phase task assignment algorithm MSHTA based on the workers’ mobility and historical information which leverage the advantages of two sensing modes in sensing quality and sensing cost of tasks. Specifically, a task is optimally assigned to workers who meet their sensing requirements (e.g., sensing time, sensing sensor) at each phase. Extensive simulation results show the effectiveness of our proposed algorithm in terms of tasks’ sensing quality and tasks’ sensing cost.

This paper studies distributed range-based localization in arbitrarily deployed wireless ad hoc networks. Existing range-based localization approaches depend on specially deployed anchors or require dense network deployment. Our algorithm is a distributed paradigm that only requires local information of each node. Therefore, it is applicable to the resource-limited embedded sensors. Specifically, our algorithm performs a three-stage optimization through coarse-grained, middle-grained, and fine-grained levels. We designed an efficient but accurate neural network to learn the hidden relations between the distances of nodes and their positions. Simulations show that our proposed algorithm works in many more types of network deployments than the existing approaches. Furthermore, our algorithm achieves the highest localization accuracy on average.

With the ever-increasing demand on location-based services (LBS), fingerprint-based methods have attracted more and more attention in indoor localization. However, the considerable overhead of fingerprint is still a problem which hinders the practicability of such technology. Due to the prevalent of Wi-Fi access points (APs) and the high location accuracy of Ultra-Wideband (UWB), in this paper, we propose a hybrid system which utilizes UWB and Wi-Fi technologies to alleviate the offline overhead and improve the localization accuracy. Specifically, we employ UWB to determine the coordinate of each reference point (RP) instead of traditional manual measurement. Meanwhile, the Received Signal Strength Indicator (RSSI) of Wi-Fi is collected by a customized software installed in the mobile device. Then, a timestamp matching scheme is proposed to fuse these data coming from different devices and construct the offline fingerprint database. Besides, in order to better map the online data with offline database, an AP weight assignment scheme is proposed, which allocates APs with different weights based on the RSSI characteristic in each RP. We implement the system in real-world environment and the experimental results demonstrate the effectiveness of the proposed method.

Aiming at the problems of high cost and low tracking performance of mobile target tracking, this paper proposes a CSI-based moving target trajectory tracking method. This method combines velocity estimation and hidden Markov model to achieve tracking of moving target trajectories. Firstly, the collected channel state information (CSI) in the offline phase, after preprocessing, is stored in the fingerprint database. Secondly, in the online stage, the model proposed in this paper is used for real-time matching, so as to realize real-time trajectory tracking of the target. Set up contrast experiments is carried out to verify the moving target trajectory tracking method proposed in this paper. The CSI-based moving target trajectory tracking method can track moving targets more accurately, has universality to different environments and targets, and has stability and robustness.

In this paper, we propose a method for indoor intrusion detection and localization that makes use of channel state information (CSI), which consists of an offline phase and an online phase. In the former, we collect CSI in different scenarios, and at different times, for more comprehensive characterization of signal propagation. To reduce the redundancy and dimensionality of CSI data, we employ the principal component analysis algorithm to extract the main features of CSI, and build the fingerprint database for localization. In the online phase, we first apply the earth mover’s distance algorithm to detect the presence of the person in the test area. Following this, we determine the approximate location of the target according to the change of CSI measurements, and compare this to the fingerprint database, to select reference points to build the sub-fingerprint database. Finally, we evaluate the actual position of this target using the improved k-Nearest Neighbor algorithm.

In the indoor environment, the monitoring of personnel activity behavior becomes more and more important. Although the traditional camera monitoring method has good performance, due to the limitation of deployment mode, there are monitoring blind spots and its deployment scope involves privacy issues. The non-device personnel acquisition and motion recognition through WiFi equipment, as a new type of highly promising technology, has received more attention and research. In this paper, we propose a human motion recognition method based on channel state information (CSI) amplitude phase mixing information, and classify the different activities of people. Different from the traditional single-person daily activity behavior recognition, this program focuses on the human exercise behavior of different people with different intensity, and promotes to the related sports behavior recognition of two people. Compared with the single person situation, the strength, amplitude and regularity of the two people exercising at the same time are very different. We experimentally tested the effects of different activities of single and double on CSI in two real environments, extracted relevant amplitude and phase information, and used machine learning to summarize the change patterns classification. At the same time, consideration of the line-of-sight factor has improved the overall flexibility of the system and improved the condition of motion recognition.

Data quality assurance is one of the most critical challenges in the context of Mobile CrowdSensing (MCS). How to effectively select appropriate participants from large-scale candidates to perform sensing tasks while satisfying certain constraint is a problem to be solved. Motivated by this, this paper studies the problem of data-quality-aware participant selection for MCS. Firstly, we propose a quality-aware participant reputation model by introducing active factor to lay a theoretical foundation. Secondly, we present a Multi-Stage Decision solution based on Greedy strategy (MSD-G) to optimize the pending problem while satisfying certain data quality constraint. Extensive simulations over a real dataset verify that our proposed MSD-G can effectively realize participant selection with ideal recruitment cost and sensing data quality.

How to detect small targets accurately under complex background and low signal-to-clutter ratio is of great significance to the development of precision guided weapons and infrared early warning. The traditional local contrast method is difficult to detect small and dim targets in complex background. In this paper, in order to improve the traditional local contrast method and detect small targets effectively under complex background conditions, a novel method base on Facet-kernel filtering local contrast measure (FFLCM) is proposed for small target detection. Initially, a nest sliding window structure of the central layer and the surrounding background layer is given. Then, the Facet-kernel filter is used to enhance the target in the center layer, the gray similarity difference between the central layer and the surrounding layer is calculated to suppress the background. Finally, a threshold operation is used to extract target. Experimental results demonstrate that our proposed method could effectively enhance small targets and suppress complex background clutters simultaneously.