The distributed deep learning architecture between front-deployed sensors and edge-deployed gateways attracts increasing interest. However, the inference performance of distributed deep models is also impacted by the delivery loss of intermediate representation in the wireless link, especially in the harsh industrial fading environments. Traditional communication systems usually focus on transmission errors at bit level, which treat all bits in the packets equally and fail to suit the varying importance in distributed deep models, which urges the essential evolution of the communication method to form a joint co-design paradigm for distributed deep models. This article then proposes to optimize the Mean Time To First Failure (MTTFF) of wireless link instead of traditional bit error rate, which enables a guaranteed transmission window. This paper first derives the analytical model of MTTFF under MIMO systems, then utilizes the kernel mixture distribution to obtain a closed-form solution of MTTFF, which forms a optimization algorithm minimizing the transmitted power while achieving the aiming MTTFF. Extensive reallife experiments show more than 70% satisfaction rate of MTTFF, which leads to more than 10 times higher inference accuracy than the original deep model.

Accurate online link quality metrics represented by the Rician parameter are critical to enhancing the reliability of industrial wireless networks subject to temporal fading channels. The Rician parameters can be estimated by fitting the received I/Q symbols with GMM (Gaussian Mixture Model). However, the classical Expectation-Maximization estimations of GMM rely on the preset hyper-parameter of kernel numbers to guarantee the convergence, making it hard to work under adaptive modulation schemes. To address this challenge, we first reveal that the derivative of likelihood is less capable of representing the global optimal, which leads to the well-known local optimal problem and the failure to recognize the false convergence caused by incorrectly configured kernel numbers. A new empirical metric derived from KLD (Kullback-Leibler divergence) has been proposed to identify the local optimal convergence, as well as a new metric tuple to discriminate redundant kernels. A novel estimation algorithm has then been designed to shift the number of kernels from the preset hyper-parameter to the adjustable parameter. This improvement guarantees the global optimal convergence of the GMM with any initial number of kernels. Extensive experiments demonstrate that the proposed method achieves over ten times better accuracy, while requires less than half the iterations.

The distributed deep learning architecture can support the front-deployment of deep learning systems in resource constrained IoT devices and is attracting increasing interest. However, most ready-to-use deep models are designed for centralized deployment without considering the transmission loss of the intermediate representation inside the distributed architecture. This oversight significantly affects the inference performance of distributed deployed deep models. To alleviate this problem, a state-of-the-art work chooses to retrain the original model to form an intermediate representation with ordered importance and yields better inference accuracy under constrained transmission bandwidth. This paper first reveals that this solution is essentially a pruning-like solution, where unimportant information is adaptively pruned to fit within the limited bandwidth. With this understanding, a novel scheme named Naturally Aggregated Intermediate Representation (NAIR) has been proposed, which aims to naturally amplify the difference of importance embedded in the intermediate representation from a mature deep model and reassemble the intermediate representation into a hierarchy of importance from high-to-low to accommodate the transmission loss. As a result, this method shows further improved performance in various scenarios, avoids compromising the overall inference performance of the system, and saves astronomical retraining and storage costs. The effectiveness of NAIR has been validated through extensive experiments, achieving a 112% improvement in performance compared to the state-of-the-art work.

In recent years, we have witnessed a significant rise in both the popularity and capability of Connected and Autonomous Vehicles (CAVs). This progress has been facilitated, in part, by advancements in CAV simulators that enable researchers to efficiently, safely, and cost-effectively test their vehicles. However, many current simulators do not directly address one of the most pressing challenges that CAVs encounter: cybersecurity. In this paper, we present a step towards resolving this issue. We have developed a simulator with the ability to simulate various Vehicle-to-Everything (V2X) attacks in real time. Our approach involves co-simulation of three simulators: CARLA, SUMO, and Artery. Utilizing the V2X communication capabilities of these simulators, our attacks involve injecting malicious Cooperative Awareness Messages (CAMs) and Decentralized Environmental Notification Messages (DENMs) into the simulation.

The recent proposed deep learning system is a promising technology to extract essential information from vision based high dimension sensors. The current intelligent vehicle systems heavily rely on vision sensors to be aware of the environment. Even though a single vehicle may have several vision sensors installed, view and range restrictions may result in information loss, leading to incorrect autopilot system decisions. To ease this limitation, a cooperative deep learning architecture for intelligent vehicle systems has been proposed in this paper. In this architecture, the observed raw data from vehicles will be processed by a partitioned deep learning system with convolution layers deployed in geographically distributed computation units. With the proposed cooperative deep learning architecture, numerous observations of a single object from various views can be utilized to increase detection accuracy. As to be expected, almost all of the currently available intelligent vehicle systems have been designed in the context of a single entity deep learning system. This means that in order to deploy a cooperative deep learning system, which has been discussed in this paper, various aspects such as computation, communication, and storage may need to be modified. A simple but representative experiment has been deployed to demonstrate the feasibility of the proposed system.

Vehicle to Vehicle (V2V) communication opens a new way to make vehicles directly communicate with each other, providing faster responses for time-sensitive tasks than cellular networks. Effective V2V routing protocols are essential yet challenging, as the high dynamic road environment makes communication easy to break. Many prediction methods proposed in the existing protocols to address this issue are either flawed or have a poor effect. In this paper, to cope with the two aspects of the problems that cause communication interrupt, i.e., link breaks and route quality degradation, we design an acceleration-based trajectory prediction algorithm to estimate the link lifetime, and a machine learning model to predict route quality. Based on the prediction algorithms, we propose PQR, a Prediction-supported Quality-aware Routing protocol, which can proactively switch to a better route before the current link breaks or the route quality degrades. Especially, considering the limitations of the current routing protocols, we elaborate a new hybrid routing protocol that integrates the topology-based method and location-based method to achieve instant communication. Simulation results show that PQR outperforms the existing protocols in Packet Delivery Ratio (PDR), Roundtrip Time (RTT), and Normalized Routing Overhead (NRO). Specifically, we have also implemented a vehicular testbed to demonstrate PQR’s real-world performance, and results show that PQR achieves almost no packet loss with latency less than 10ms during route handoff for topology change.

Intelligent Transport System (ITS) raises the increasing demand on accurate vehicle trajectory prediction for navigation efficiency. The rapidly developing 5G networks provides communications with high transmission bandwidth and super-low latency, paving the way for Mobile Edge Computing (MEC) to calculate more accurate trajectory prediction for vehicles, as the MEC server holds more comprehensive vehicular information. However, the current methods for trajectory prediction are not efficient due to the dynamical environment. To address this issue, we propose GlobalInsight, a Long Short-Term Memory (LSTM) based model, which runs on the MEC to perform accurate trajectory prediction for multiple vehicles no matter how scenario changes. In particular, we use three auxiliary layers to respectively capture the principal component of vehicle features, social interaction of adjacent vehicles, and the cross-vehicle correlation of similar vehicles. We further integrate the above information into LSTM in the main layer to enhance the trajectory learning and prediction. We evaluate our model under the NGSIM dataset, and experimental results exhibit that our model outperforms the state-of-the-art approaches.

Traffic monitoring is one of the key challenges in Intelligent Transportation Systems (ITS). In this paper, we propose to build a crowdsourcing application for traffic monitoring. The novelty of the proposed approach is that visual data is collected to enable automatic event inference with the recent advance in Computer Vision. The challenge is that mobile devices are not capable of handling visual task processing in high accuracy. We propose to build a networked system so that mobile devices can offload data via available wireless access interfaces (e.g., 4G LTE, WiFi, DSRC) to edge servers, e.g., GENI Rack. We plan to use the testbed at Kettering University to validate the proposed approach.

Mobile crowdsensing is a new paradigm in which a crowd of mobile users exploit their carried smart phones to conduct complex sensing tasks. In this paper, we focus on the makespan sensitive task assignment problems for the crowdsensing in mobile social networks, where the mobility model is predicable, and the time of sending tasks and recycling results is non-negligible. To solve the problems, we propose an Average makespan sensitive Online Task Assignment (AOTA) algorithm and a Largest makespan sensitive Online Task Assignment (LOTA) algorithm. In AOTA and LOTA, the online task assignments are viewed as multiple rounds of virtual offline task assignments. Moreover, a greedy strategy of small-task-first-assignment and earliest-idle-user-receive-task is adopted for each round of virtual offline task assignment in AOTA, while the greedy strategy of large-task-first-assignment and earliest-idle-user-receive-task is adopted for the virtual offline task assignments in LOTA. Based on the two greedy strategies, both AOTA and LOTA can achieve nearly optimal online decision performances. We prove this and give the competitive ratios of the two algorithms. In addition, we also demonstrate the significant performance of the two algorithms through extensive simulations, based on four real MSN traces and a synthetic MSN trace.

Mobile crowdsensing is a new paradigm in which a crowd of mobile users exploit their carried smart devices to conduct complex computation and sensing tasks in mobile social networks (MSNs). In this paper, we focus on the task assignment problem in mobile crowdsensing. Unlike traditional task scheduling problems, the task assignment in mobile crowdsensing must follow the mobility model of users in MSNs. To solve this problem, we propose an oFfline Task Assignment (FTA) algorithm and an oNline Task Assignment (NTA) algorithm. Both FTA and NTA adopt a greedy task assignment strategy. Moreover, we prove that the FTA algorithm is an optimal offline task assignment algorithm, and give a competitive ratio of the NTA algorithm. In addition, we demonstrate the significant performance of our algorithms through extensive simulations, based on four real MSN traces and a synthetic MSN trace.

The Internet of Things (IoT) opens up tremendous opportunities to location-based industrial applications that leverage both Internet-resident resources and phones' processing power and sensors to provide location information. Location-based service is one of the vital applications in commercial, economic, and public domains. In this paper, we propose a novel localization scheme called NextMe, which is based on cellular phone traces. We find that the mobile call patterns are strongly correlated with the co-locate patterns. We extract such correlation as social interplay from cellular calls, and use it for location prediction from temporal and spatial perspectives. NextMe consists of data preprocessing, call pattern recognition, and a hybrid predictor. To design the call pattern recognition module, we introduce the notions of critical calls and corresponding patterns. In addition, NextMe does not require that the cell tower addresses should be bounded with concrete coordinates, e.g., global positioning system (GPS) coordinates. We validate NextMe across MIT Reality Mining Dataset, involving 500 000 h of continuous behavior information and 112 508 cellular calls. Experimental results show that NextMe achieves fine-grained prediction accuracy at cell tower level in the forthcoming 1-6 h with 12% accuracy enhancement averagely from cellular calls.

A mobile opportunistic social network (MOSN) is a new type of delay tolerant network (DTN), in which the mobile users contact each other opportunistically. While cooperative caching in the Internet has been studied extensively, cooperative caching in MOSNs is a considerably different and challenging problem due to the probabilistic nature of contact among the mobile users in MOSNs. In order to reduce the total access delay, we let the mobile users cooperatively cache these data items in their limited buffer space. We balance between selfishness (caching the data items according to its own preference) and unselfishness (helping other nodes to cache). The friends with higher contact frequency may share similar interests, hence, caching the data items for friend users can lead to some benefit. In this paper, we present a hierarchical cooperative caching scheme, which divides the buffer space into three components: self, friends, and strangers. In the self component, mobile users cache the data items according to their preference. In the friends component, mobile users help their friends to cache some data items. In the strangers component, mobile users randomly cache the remaining data items. We formally analyze the access delay of the proposed scheme. The effectiveness of our approach is verified through extensive real world trace-driven simulations.

With the recent technical advances and popularization of smartphones, which are able to store, display, and transmit various types of media content, message forwarding in opportunistic mobile social networks has become a hot topic. In this paper, we propose a social-aware single-copy routing approach, which leverages the internal social feature information to resolve the social distance between the source and destination step-by-step. We introduce an optimal social feature forwarding set selection scheme for routing guidance, which is proven to achieve small message forwarding delay. We convert the routing process into an iterative 2-hop routing scheme, with a novel transition probability to measure the delivery delay of each hop. The transition probability is according to the community structure in each social feature space. Extensive simulations on both real and synthetic traces are conducted in comparison with several existing state-of-art approaches.

In this paper, an opportunistic multi-copy two-hop routing algorithm is proposed for mobile social networks (MSNs) to minimize the expected data delivery delay, using local information. For each source-destination pair, the source dynamically maintains a forwarding set consisting of relay nodes. The forwarding set selection is based on the number of remaining message copies, as well as the number and quality of relays that have not received a message copy. The source only forwards its message to the relay nodes in its forwarding set, which will in turn forward the message to the destination directly. We propose a greedy approach to select the forwarding set with n message copies at the source, in an MSN with m (m>n) relays. All forwarding sets can be determined with a time complexity of O(m log m+nm). Then, the proposed multi-copy two-hop routing algorithm is applied to a feature space routing scheme, where the contact frequencies are estimated by social feature distances. Finally, the competitive performance of the proposed schemes are shown in real trace-driven simulations.

Most routing protocols for delay tolerant networks resort to the sufficient state information, including trajectory and contact information, to ensure routing efficiency. However, state information tends to be dynamic and hard to obtain without a global and/or long-term collection process. In this paper, we use the internal social features of each node in the network to perform the routing process. In this way, feature-based routing converts a routing problem in a highly mobile and unstructured contact space to a static and structured feature space. This approach is motivated from several human contact networks, such as the Infocom 2006 trace and MIT reality mining data, where people contact each other more frequently if they have more social features in common. Our approach includes two unique processes: social feature extraction and multipath routing. In social feature extraction, we use entropy to extract the m most informative social features to create a feature space (F-space): (F1, F2,..., Fm), where Fi corresponds to a feature. The routing method then becomes a hypercube-based feature matching process, where the routing process is a step-by-step feature difference resolving process. We offer two special multipath routing schemes: node-disjoint-based routing and delegation-based routing. Extensive simulations on both real and synthetic traces are conducted in comparison with several existing approaches, including spray-and-wait routing, spray-and-focus routing, and social-aware routing based on betweenness centrality and similarity. In addition, the effectiveness of multipath routing is evaluated and compared to that of single-path routing.

A mobile opportunistic social network (MOSN) is a new type of delay tolerant network (DTN), in which the mobile users contact each other opportunistically. Information dissemination is a challenging problem in MOSNs, due to uncertainty and intermittent connectivity. In this paper, we propose a distributed social tie strength calculation mechanism to identify the relationship between each set of pairwise mobile nodes. Following arguments originally proposed by Mark Granovetter's seminal 1973 paper, The Strength of Weak Ties, the majority of the novel information dissemination is generated by weak ties. We first evaluate the strength of weak ties in MIT reality mining data. Then, a social-tie-based information dissemination protocol is presented, which is a token-based information dissemination scheme, including two phases: weak tie-driven forwarding and strong tie-driven forwarding. In the weak tie-driven forwarding phase, the susceptible nodes with more weak ties will receive more tokens for future forwarding. The number of forwarding tokens is related to the number of weak ties of two encountered nodes. After a while, the information will have been spread to multiple communities. Our scheme switches to a strong tie driven forwarding phase, in which the influential nodes are more important. The number of forwarding tokens is proportional to the number of strong ties of two encountered nodes. Extensive simulations are conducted in comparison to several approaches in real world mobile traces.

Social behavior plays a more and more important role in delay tolerant networks (DTNs). In this paper, we present an analytical model for a hypercube-based social feature multipath routing protocol in DTNs. In this routing protocol, we use the internal social features of each node (individual) in the network for routing guidance. This approach is motivated from several real social contact networks, which show that people contact each other more when they have more social features in common. This routing scheme converts a routing problem in a highly mobile and unstructured contact space (M-space) to a static and structured feature space (F-space). The multipath routing process is a hypercube-based feature matching process where the social feature differences are resolved step-by-step. A feature matching shortcut algorithm for fast searching is presented where more than one feature difference is resolved at one time. The multiple paths for the routing process are node-disjoint. We formally analyze the delivery rate and latency by using hypercube-based routing. The solutions for the expected values of latency and delivery rate are given under different path conditions: single-/multipath and feature difference resolutions with/without shortcuts. Extensive simulations on both real and synthetic traces are conducted in comparison to several existing state-of-the-art DTN routing protocols.

With the rapid growth of smartphone usage, mobile social networks (MSNs) are becoming increasingly popular. MSN can be considered as a type of delay tolerant network (DTN) which lacks continuous end-to-end connections between nodes, due to the node mobility and limited transmission range. Inspired by the homophily of social networks that friends are usually similar in characteristics, we present a novel concept - cloud, where the nodes in frequent contact with the destinations will form destination clouds. Neighbors in the destination cloud have a special status that can forward the message to the destination directly. We propose a cloud-based multicast scheme with feedback in MSNs with two phases: pre-cloud and inside-cloud. In the pre-cloud process, the message holder will forward the copy of the multicast message to the encountered node, based on a given forwarding metric. The forwarding metric can be iteratively refined from a feedback control mechanism. In the inside-cloud process, the message holder will wait until it meets with the destinations. We analytically formulate the multicast problem into a continuous Markov chain problem, and formally analyze the latency in this model. Extensive trace-driven simulations show that our scheme significantly improves the performance compared to existing schemes.

Delay tolerant networks (DTNs) use mobility-assisted routing, where nodes carry, store, and forward data to each other in order to overcome the intermittent connectivity and limited network capacity of this type of network. In this paper, we propose a routing protocol that includes two mechanisms: message replication and message migration. Each mechanism has two steps: message selection and node selection. In message replication, we choose the smallest hop-count message to replicate. The hop-count threshold is used to control the replication speed. We propose a metric called 2-hop activity level to measure the relay node's transmission capacity, which is used in node selection. Our protocol includes a novel message migration policy that is used to overcome the limited buffer space and bandwidth of DTN nodes. We validate our protocol via extensive simulation experiments; we use a combination of synthetic and real mobility traces.

Most routing protocols for delay tolerant networks resort to the sufficient state information, including trajectory and contact information, to ensure routing efficiency. However, state information tends to be dynamic and hard to obtain without a global and/or long-term collection process. In this paper, we use the internal social features of each node in the network to perform the routing process. This approach is motivated from several social contact networks, such as the Infocom 2006 trace, where people contact each other more frequently if they have more social features in common. Our approach includes two unique processes: social feature extraction and multi-path routing. In social feature extraction, we use entropy to extract the m most informative social features to create a feature space (F-space): (F1, F2, ..., Fm), where Ficorresponds to a feature. The routing method then becomes a hypercube-based feature matching process where the routing process is a step-by-step feature difference resolving process. We offer two special multi-path routing schemes: node-disjoint-based routing and delegation-based routing. Extensive simulations on both real and synthetic traces are conducted in comparison with several existing approaches, including spray-and-wait routing and spray-and-focus routing.

The increase in multimedia content makes providing good quality of service in wireless networks a challenging problem. Consider a set of users, with different content interests, connected to the same base station. The base station can only broadcast a limited amount of content, but wishes to satisfy the largest number of users. We approach this problem by considering each user as a point in a 2-D space, and each type of broadcast content as a circle. A point that is covered by a circle will be satisfied, and the closer the point is to the center of the circle, the higher the satisfaction. In this paper, we first formulate this problem as an optimal content distribution problem and show that it is NP-hard. The optimal problem can also be extended into an m-dimensional (m-D) space, and distance measurements can be expressed in a general p-norm. We then introduce three local greedy algorithms and compare their complexity. The approximation ratio of our greedy algorithms to the optimization problem is also formally analyzed in this paper. We perform extensive simulations using various conditions to evaluate our greedy algorithms. The results demonstrate that our solutions perform well and reflect our analytical results.

Delay tolerant networks (DTNs) are a kind of wireless mobile network which may lack continuous network connectivity. Multicast supports the distribution of data to a group of users, a service needed for many potential DTNs applications. While multicasting in the Internet and mobile ad hoc networks has been studied extensively, due to the unique characteristic of frequent partitioning in DTNs, multicasting in DTNs is a considerably different and challenging problem. It not only requires new destinations of multicast semantics, but also brings new issues to the design of routing algorithms. In this paper, we propose new forwarding models for DTNs multicast and develop several multicast forwarding algorithms. We use delegation forwarding (DF) in DTNs multicast and compare it with single and multiple copy multicast models, which are also designed by us. The effectiveness of our approach is verified through extensive simulation.

Delay tolerant networks (DTNs) are a special type of wireless mobile networks which may lack continuous network connectivity. Multicast is an important routing function that supports the distribution of data to a group of users, a service needed for many potential DTNs applications. While multicasting in the Internet and mobile ad hoc networks has been studied extensively, efficient multicasting in DTNs is a considerably different and challenging problem due to the probabilistic nature of contact among nodes. This paper aims to provide a non-replication multicasting scheme in DTNs while keeping the number of forwardings low. The address of each destination is not replicated, but is assigned to a particular node based on its contact probability level and node active level. Our scheme is based on a dynamic multicast tree where each leaf node corresponds to a destination. Each tree branch is generated at a contact based on the compare-split rule proposed in this paper. The compare part determines when a new search branch is needed, and the split part decides how the destination set should be partitioned. When only one destination is left in the destination set, we use either wait (no further relay) or focus (with further relay) to reach the final destination. The effectiveness of our approach is verified through extensive simulation.

Many applications of sensor networks require batteries as the energy source, and hence critically rely on energy optimisation of sensor batteries. But as often neglected by the networking community, most batteries are non-ideal energy reservoirs and can exhibit battery recovery effect — the deliverable energy in batteries can be replenished per se, if left idling for sufficient duration. We made several contributions towards harnessing battery recovery effect in sensor networks. First, we empirically examine the gain of battery runtime due to battery recovery effect, and found this effect significant and duration-dependent. Second, based on our findings, we model the battery recovery effect in the presence of random sensing activities by a Markov chain model, and study the effect of duty cycling and buffering to harness battery recovery effect. Third, we propose a more energy-efficient duty cycling scheme that is aware of battery recovery effect, and analyse its performance with respect to the latency of data delivery.