Minimizing energy consumption is critical for the long-range (LoRa) Internet of Things (IoT) networks, to extend the battery lifetime of end devices (EDs) while reducing the maintenance cost. To overcome the excessive computational complexity required by the traditional optimization methods, learning approaches such as deep neural network (DNN) and reinforcement learning (RL) have been researched in wireless networks, where star topologies are widely employed. In contrast, LoRa usually deploys relays to assist the connection between the EDs and the gateway (GW), leading to a much more complex network topology. Consequently, DNN and RL would become less effective in LoRa, since those methods are difficult to learn the complex network topology constructed by LoRa. In this paper, we propose a learning method based on graph neural network (GNN), which is known for its prominent ability to capture and represent intricate graph-structural dependencies, to tackle the energy optimization problem for relay-assisted LoRa. Specifically, the multi-hop LoRa network is modeled as a directed graph, with channel state information (CSI) defined as node features, while the spreading factor and transmission power are deemed labels. A hierarchical message aggregation mechanism is proposed to effectively capture the multi-hop structural dependencies, followed by the process of inductive learning. Results show that against conventional optimization algorithms, the proposed method can achieve near-optimal energy consumption with a gap of 10% to 16%, while reducing the runtime by about six orders of magnitude. Compared to DNN, the GNN-based model can provide an energy saving of up to 32%, at a similar level of inference time.

Energy optimization is crucial for extending battery life and reducing maintenance costs in long-range (LoRa) Internet of Things (IoT) networks. Traditional optimization methods usually need excessive computational complexity, limiting the practicability. This drives the recent development of machine learning (ML)-based optimization, such as deep neural networks (DNNs) and reinforcement learning (RL), in wireless local area networks (WLANs). However, compared to WLANs, LoRa owns a more complicated network topology due to the engagement of multi-hop, which is difficult for the existing ML methods to learn. Motivated by this, we propose a topology-aware graph neural network (GNN) learning method, which is specially tailored to tackle the energy optimization problem in multi-hop LoRa networks. By leveraging each node’s topological position to adaptively determine the optimal message-passing depth, the model better integrates the neighborhood information with node feature representations, enhancing the prediction of transmission parameter and overall energy efficiency. Also, a closed-form model of collision probability is derived for the nodes in LoRa, to measure the energy consumption due to retransmissions. Simulation results show that against traditional optimization methods such as game theory, the proposed method can reduce the runtime by five orders of magnitude, with an energy optimization gap below 13%. Compared to existing GNN-based methods, it achieves up to 50% lower energy consumption 20% fewer outage probability, at a similar level of inference time.

Abstract-As a narrowband communication technology, Long Range (LoRa) contributes to the long development of Internet of Things (IoT) applications. The LoRa gateway plays an important role in the IoT transport layer, and security and efficiency are the key issues of current research. On the one hand, in the centralized working model of IoT systems built by traditional LoRa gateways, all the dat a generated and reported by end devices are processed and stored in cloud servers, which are susceptible to security issues such as data loss and data falsification. On the other hand, edge computing, as an innovative approach that brings data processing and storage closer to the endpoints, can create a decentralized security infrastructure for LoRa gateway systems, resulting in an edge intelligent IoT working model. Although this paradigm delivers unique features and improved quality of service (QoS), installing IoT applications at LoRa gateways with limited computing and memory capabilities presents considerable obstacles. To address this challenge, an edge intelligent LoRa gatew ay is designed and implemented in this paper. Firstly, we developed an edge intelligent LoRa gateway prototype on an FPGA-based embedded hardware system. Then, we proposed Latency-Aware Algorithm (LAA) can greatly improve the reliability of the network system by using a distributed edge computing network technology that can achieve maintenance operations such as detection, repair, and replacement of failures of edge nodes in the network. Finally, many experiments were conducted to evaluate the performance of the proposed edgeintelligent LoRa gat eway. The results indicate that the proposed edge intelligent LoRa gateway is more effective in latency-aware in IoT applications, while greatly ensuring system availability and IoT network reliability.

With the development of the Internet of Things (IoT), the amount of global data is increasing rapidly. However, due to the increased burden of the cloud network, it is difficult to implement low-latency and high-efficiency video analysis in the cloud computing mode. To address this issue, this paper proposes an Intelligent Video Analysis System (IVAS) that can execute deep learning algorithms on low-power edge IoT devices, such as face detection and face recognition algorithms. IVAS enables fast, accurate, and real-time inference calculations of intelligent video analysis algorithms, providing an evaluation platform for performance testing, key parameter selection, and result analysis. The experiments based on real-world data confirm that IVAS can achieve good performance in people counting under an edge computing environment.

This work presents the deep learning networks-based method using fine-tuning for classification and search of a diversity of action videos. First, a 3D convolutional neural networks (3D CNN) model which performs pre-training operation and fine-tuning strategy is employed to extract the spatiotemporal features of videos. It is first pre-trained on UCF-101 datasets to train model with initial parameters. Then, a small new dataset is employed to fine-tune the initial model for the training of the new model. Once features are extracted by the final CNNs model, distance measure can be adopted to calculate the similarities between the query video and the test dataset for the video search. The searched video is returned and ranked according to the priority when it has higher similarity with the query video. The comparison results in the experiment shows that the search method using fine-tuning obtains better performance than the method without using fine-tuning. Second, the classification results based on the 3D CNN model using fine-tuning are also presented for the consideration of a query by keyword. Accuracy result obtained using the model with the help of fine-tuning is approximately 2.8% higher than that without using fine-tuning.

RFID is one of the indispensable technologies employed to implement the Internet of Things, where it facilitates the identification and awareness of objects in our daily lives by integrating backscatter tags. The interference synthetic aperture radar (InSAR)-based localization scheme provides high precision in 3-D conditions, although it still has some practical problems such as its strict sampling operation and it is also highly time consuming. In this article, we developed E3DinSAR as an optimized InSAR-based 3-D localization approach for pinpointing tags rapidly and easily with high accuracy. E3DinSAR does not use any other infrastructure or reference tags, and only one movable reader with one antenna is sufficient to implement the localization method, where it measures the phase value of the tag at different positions throughout its movement in order to create multiple holographic images. In contrast to the previously proposed InSAR-based technique, we extend the sampling trajectory of the reader to an arbitrary curve by restricting it to several approximate linear apertures, which allows more flexible operations and increase the accuracy. We also employ a rapid computing method to calculate the results for the holographic images by estimating the direction of arrival for the tag. Experiments were performed where E3DinSAR was implemented with commercial off-the-shelf (COTS) RFID products and our systematic evaluation showed that E3DinSAR achieved a mean locating accuracy of 18.4 cm in a 3-D space.

The prohibitive cost of deploying a sophisticated DNS-based CDN makes anycast-based CDN an attractive alternative for new or small CDN operators. In anycast-based CDNs, user requests are naturally routed to the “closest” server determined by Internet routing. For the operators, however, this comes at a cost—loss of control—how the traffic is routed is entirely at the mercy of BGP routing. The “closest” server may be overloaded, or simply not the best choice. This “loss of control” undermines thescalabilityof anycast-based CDN architectures. To have control over how traffic is routed, existing work either requires adding a large amount of complexity to the system (high Capex/Opex) or is unable to achieve precise and fine-grained control. This paper proposes T-SAC, a scalable anycast-based CDN architecture that capitalizes on the programmability and flexibility of SDN/NFV, enabling fine-grained traffic redirection among CDN servers. T-SAC achieves precise control by leveraging a load-based redirection algorithm and a single 1-bit no-redirect flag. We implement T-SAC in the real system and evaluate its performance from various aspects using DASH and web applications. The results show that T-SAC is capable of redirecting the right amount of traffic at the right time to the right servers, making the system highly scalable.

Compared with traditional methods that employ inertial sensors or wireless sensors, device-free approaches do not require that people carry devices, and they are considered a useful technique for indoor navigation and posture recognition. However, few existing methods can detect the trajectory and movements of humans at the same time. In this study, we propose a scheme called PADAR for addressing these two problems simultaneously by using passive radio frequency identification (RFID) tags but without attaching them to the human body. The idea is based on the principle of radio tomographic imaging, where the variance in a tag’s backscattered radio frequency signal strength is influenced by human movement. We integrated a commodity off-the-shelf RFID reader with a two-dimensional phased array antenna and a matrix of passive tags to evaluate the performance of our scheme. We conducted experiments in a simulated indoor environment. The experimental results showed that PADAR achieved an accuracy of over 70%.

Computing performance is one of the key problems in embedded systems for high-resolution face detection applications. To improve the computing performance of embedded high-resolution face detection systems, a novel parallel implementation of embedded face detection system was established based on a low power CPU-Accelerator heterogeneous many-core architecture. First, a basic CPU version of face detection prototype was implemented based on the cascade classifier and Local Binary Patterns operator. Second, the prototype was extended to a specified embedded parallel computing platform that is called Parallella and consists of Xilinx Zynq and Adapteva Epiphany. Third, the face detection algorithm was optimized to adapt to the Parallella architecture to improve the detection speed and the utilization of computing resources. Finally, a face detection experiment was conducted to evaluate the computing performance of the proposal in this paper. The experimental results show that the proposed implementation obtained a very consistent accuracy as that of the dual-core ARM, and achieved 7.8 times speedup than that of the dual-core ARM. Experiment results prove that the proposed implementation has significant advantages on computing performance.

Modeling and analyzing architectural vulnerability factor (AVF) of microprocessors is helpful in reliability-aware microarchitecture design. This paper derives formulas for computing AVF of critical pipeline structures by calculating the instruction occupancy of a structure based on instruction-level parallelism. A mechanistic relationship is shown to exist between instruction fetch width and basic block size to AVF. There is an upper bound threshold impact of variable fetch width and basic block size on AVF, and the value of this threshold can be calculated. There is a quadratic relationship for AVF from the minimum value between fetch width and basic block size. Experimental results prove the above conclusion. Based on this mechanistic relationship, a design architect can rapidly evaluate reliability, choose corresponding parameters at an early design stage, and obtain a better trade-off between reliability, performance, and cost.

Issue queue is a critical structure in the pipeline and more vulnerable to soft-error strikes. Reducing soft error vulnerability of issue queue cannot be ignored in microprocessor reliability design. Instructions clog in issue queue caused by instruction flow mix and functional unit configuration mismatch. That makes the soft error vulnerability of issue queue increases. This paper proposed a vulnerability mitigation method which does not need to change functional unit configuration. It adjusted the instruction flow mix in issue queue, reduced the waiting time of instruction in issue queue to reduce the architectural vulnerability factor. The experiment result shows that average reduces of architectural vulnerability factor are 2.8% and improvement of reliability-performance is 4.9%.

Radio Frequency IDentification (RFID) system has been wildly used in recent ten years. A Handhold reader and some passive tags compose an efficient system for retail scenarios, such as storage management and item inventory. However, the heavily energy consumption of the reader battery, which due to the frequent tag collisions, is a main limitation for such a RFID system in large scale item-level scenario. To avoid the collisions, Dynamic Frame Slotted ALOHA anti-collision algorithms (DFSA) are popularly being used. But unreliable tag number estimation and non-optimized frame size selection may lead to more energy consumption. In this paper, we propose a novel tag population estimation and frame length selection method based on tag location-aware scheme to optimize the DFSA tag anti-collision algorithm. In our method, the handhold reader firstly collects the tags' location information and divides them into different clusters based on their distance from the reader. Then by estimating the tag population for each cluster, the reader utilizes this value to achieve an optimum frame size and automatically adjusting the transmission power to scan the tags in corresponding tag cluster. We also give a quantitative energy consumption model for the passive RFID system. According to the simulation results, the passive RFID system throughput and energy efficiency can be highly increased by using our scheme.

In LTE/LTE-A system, low frequency reuse factor is used to improve the spectral efficiency. In order to mitigate the inter-cell interference (ICI), inter-cell interference coordination (ICIC) is introduced which divides the bandwidth into a 1-reuse sub-bands for cell center users and -reuse sub-bands for cell edge users. ICIC technique has been considered as a promising technology for alleviating the degradation caused by ICI and improving throughput of cell edge users. Coordinated multipoint transmission/reception (CoMP) technique has been proposed in 3GPP release 11 to avoid ICI and enhance both system average and cell edge throughput. In this paper, we combine the ICIC with joint reception receiver together in uplink CoMP system to futher remove the multiuser and inter-cell interference. According to the requirement for exchanging received signals' information among cooperation base stations (BSs), two detection methods are proposed for uplink CoMP system. Simulation results illustate the performance of the proposed methods compared to that obained through ICIC and CoMP techniques.

When two or more sound detectors are available, interaural time differences may be used to determine the direction of a sound's origin. This process, known as sound localization, is performed in mammals via the auditory pathways of the head and by computation in the brain. The Jeffress Model successfully describes the mechanism by exploiting coincidence detector neurons in conjunction with delay lines. However, one of the difficulties of using this model on neural simulators is that it requires timing accuracies which are much finer than the typical 1 ms resolution provided by simulation platforms. One solution is clearly to reduce the simulation's time step, but in this paper we also explore the use of population coding to represent more precise timing information without changing the simulation's timing resolution. The implementation of both the Jeffress and population coded models are contrasted, together with their results, which show that population coding is indeed able to provide successful sound localization.

Alternative interconnects are attractive for scaling on-chip communication bandwidth in a power-efficient manner. However, efficient utilization of the bandwidth provided by these emerging interconnects still remains an open problem due to the spatial and temporal communication heterogeneity. In this article, a Stream Arbitration scheme is proposed, where at runtime any source can compete for any communication channel of the interconnect to talk to any destination. We apply stream arbitration to radio frequency interconnect (RF-I). Experimental results show that compared to the representative token arbitration scheme, stream arbitration can provide an average 20% performance improvement and 12% power reduction.

This paper addresses the inherent unreliability and instability of the multiple uncertain characteristics of complex embedded multiprocessor systems, such as MPSoC (multi-processor system on chip) systems. In this work, we propose a fuzzy sets description for the multiple uncertain characteristics of system, and using fuzzy set membership calculation to determine the scheduling priorities of tasks and resources, in order to improve the capability of concurrent executions of tasks. And also we present a method estimation of comprehensive analysis on the multiple performances in order to increase utilization factor and balancing loads on processors. Through simulation, we demonstrate that the fuzzy dynamic scheduling algorithm can handle wide variety requirements of multiple performances, and we also show an improved approach to solve its demerit of partial adjustment. We implemented a prototype MPSoC system, which utilizes the proposed fuzzy dynamic scheduling algorithm to allocating tasks onto the multiprocessors. And in our case studies, we design a parallel architecture h.264 encoder, running on a multicore MPSoC system on FPGA. The speedup ratio of this prototype application system is up to 12.69.

Position information is vital for wireless sensor networks in many applications. In this paper, based on the barycentric coordinate system, we incorporate a term constrains sensors to remain the intrinsic structure revealed by range measurements between neighboring nodes into the STRESS function, which is the cost function optimized by the distributed weighted-multidimensional scaling algorithm (dwMDS). By minimizing the modified cost function, we derive a distributed localization algorithm called the Multidimensional Scaling and Barycentric Coordinates based Distributed Localization Algorithm (MDS_BC_DLA). Experimental results on four different types of WSNs show MDS_BC_DLA outperforms dwMDS.

Utterance verification is increasingly essential for robustness and better performance of speech recognition systems. In this paper, we use overlapping one-class SVMs to verify utterances and propose a K-means based training algorithm for the overlapping one-class SVMs. The training algorithm first divides the training data into several clusters based on the K-means algorithm and then expands each cluster by inserting some nearest outside data. Then it iteratively trains the overlapping one-class SVMs on the expanded clusters and constructs the train clusters based on the learned overlapping one-class SVMs until the train clusters remain unchanged. Experimental results on a real dataset show the overlapping one-class SVMs can greatly improve the recall of the speech recognition systems.

General reference models are essential for the ambient intelligence (AmI) system design, and they should be clear and simple both in terms of physical structure and software architecture. In this paper, we introduce a conceptual system model. According to this model, we present a physical structure model and a multiagent system based software architecture in the following step. We also give a data management mechanism model for the AmI environment. To evaluate our models, we implement them in a prototype ldquoSmart-Ovenrdquo system. Our models are suitable for rapid system modeling and development; they can help the designers and the programmers develop new AmI applications more rapidly and conveniently. Using the simplified AmI physical structure model, the physical AmI supporting environment can be established easily and rapidly. Via the software architecture and data management model, AmI application systems achieve flexibility, service discovery, and interoperability, which are the most common technical limitations in other related systems. Furthermore, our models can also help the terminal users configure their own AmI system by themselves. As a result of the works carried out in this paper, one can simplify the procedures and enhance the efficiency of AmI applications system design and development.

A simple and clear framework for a local Ambient Intelligence (Am1) Space application system is very important, because it can help the designers develop new applications in Am1 Space, even will help the users configure their future home Am1 system. In this paper, first, the open problems in the Am1research field are discussed. Then, a framework of a typical Am1 Space application system is presented, including the local Am1 Space model, Multi-agent based software architecture and the system network configuration. Next, based on the above framework, a prototype system Am1-Space was established. After that, a whole appearance of our Am1-Space project is given, and the services provided by Am1-Space are introduced briefly. Finally, future works of Am1-Space are discussed. Via the works of this paper, a research platform is established, which will allow experiments on other local Am1Space applications using this platform in future research.

An adaptive threshold-based call admission control (CAC) scheme used in wireless/mobile network for multi-class services is proposed. In the scheme, each class' CAC thresholds are solved through establishing a reward-penalty model which tries to maximize network's revenue in terms of each class's average new call arrival rate and average handoff call arrival rate, the reward or penalty when network accepts or rejects one class's call etc. To guarantee the real time running of CAC algorithm, an enhanced genetic algorithm is designed. Analyses show that the CAC thresholds indeed change adaptively with the average call arrival rate. The performance comparison between the proposed scheme and mobile IP reservation (MIR) scheme shows that with the increase of average call arrival rate, the average new call blocking probability (CBP) and the average handoff dropping probability (HDP) within 2000 simulation intervals of the proposed scheme are confined to lower levels, and they show approximatively periodical trends of first rise and then decline. While these two performance metrics of MIR always increase. At last, the analysis shows the proposed scheme outperforms MIR in terms of network's revenue