Low-end Devices Research Articles

Designing a SHA-256 processor for blockchain-based IoT applications

An increasing number of Internet of Things applications have been recently considering the use of the blockchain technology, calling for dedicated computational capabilities of the underlying devices. This poses an inherent challenge in an application domain notoriously characterised by several, often stringent limitations in terms of available resources. This paper focuses on a key component of the blockchain technology, the SHA-256 hash function, proposing a hardware accelerator which can be adapted to different requirements according to the application context. Experimental results show how the proposed accelerator can be used to satisfy different sets of constraints, effectively providing low-end devices with the computational capabilities required to undertake complex tasks such as blockchain protocols.

A Lightweight Speaker Recognition System Using Timbre Properties

Speaker recognition is an active research area that contains notable usage in biometric security and authentication system. Currently, there exist many well-performing models in the speaker recognition domain. However, most of the advanced models implement deep learning that requires GPU support for real-time speech recognition, and it is not suitable for low-end devices. In this paper, we propose a lightweight text-independent speaker recognition model based on random forest classifier. It also introduces new features that are used for both speaker verification and identification tasks. The proposed model uses human speech based timbral properties as features that are classified using random forest. Timbre refers to the very basic properties of sound that allow listeners to discriminate among them. The prototype uses seven most actively searched timbre properties, boominess, brightness, depth, hardness, roughness, sharpness, and warmth as features of our speaker recognition model. The experiment is carried out on speaker verification and speaker identification tasks and shows the achievements and drawbacks of the proposed model. In the speaker identification phase, it achieves a maximum accuracy of 78%. On the contrary, in the speaker verification phase, the model maintains an accuracy of 80% having an equal error rate (ERR) of 0.24.

Lightweight Broadcast Authentication Protocol for Edge-Based Applications

In this article, we propose a lightweight authentication protocol that provides forward secrecy for edge-based applications. Motivated by the general consensus that centralized authentication solutions are not suitable for an expanding Internet of Things (IoT), our edge-based authentication reduces latency for critical applications, lowers cloud dependency, and employs cryptographic primitives, which are efficiently implemented on resource-constrained low-end devices. Moreover, the edge entity broadcast messages using session keys that are derived securely from a hash function. The protocol utilizes hash chains and authenticated encryption which makes it resilient to quantum attacks. Moreover, entities are not required to hold a permanent master key, and all session keys are derived securely from a hash function. As a use case, we present a smart emergency system where an edge application broadcasts alert messages for individual responder groups when specific events occur. We formally define and prove the main security properties of our protocol, and compare it to other lightweight protocols in terms of security and performance. The computational complexity of our protocol comprises of three decryption operations, two HMAC, and five hash computations. The required storage for each node is 96 B and the communication overhead is only 56 B per session.

Block-term tensor neural networks

Deep neural networks (DNNs) have achieved outstanding performance in a wide range of applications, e.g., image classification, natural language processing, etc. Despite the good performance, the huge number of parameters in DNNs brings challenges to efficient training of DNNs and also their deployment in low-end devices with limited computing resources. In this paper, we explore the correlations in the weight matrices, and approximate the weight matrices with the low-rank block-term tensors. We name the new corresponding structure as block-term tensor layers (BT-layers), which can be easily adapted to neural network models, such as CNNs and RNNs. In particular, the inputs and the outputs in BT-layers are reshaped into low-dimensional high-order tensors with a similar or improved representation power. Sufficient experiments have demonstrated that BT-layers in CNNs and RNNs can achieve a very large compression ratio on the number of parameters while preserving or improving the representation power of the original DNNs.

Adapting Recursive Sinusoidal Software Oscillators for Low-power Fixed-point Processors

The growing field of the Internet of Things relies at the bottom on components with very scarce computing resources that currently do not allow complex processing of sensed data. Any computation involving Fast Fourier Transforms (FFT), Wavelet Transforms (WT), or simple sines and cosines is considered impractical on low-end devices due to the lack of floating point and math libraries. This article presents new techniques that make it possible to use these functions also on severely constrained target platforms. Current literature abounds with schemes to compute sine and cosine functions, with focus on speed, hardware footprint, software size, target type, or precision. Even so, there is no practical exploration of the design space available for embedded devices with limited resources, in particular when only integer operations are possible. We select an efficient set of recursive sine and cosine generators and measure the frequency, amplitude, and phase error over a wide parameter range. We show that their simplicity allows them to be implemented on the most bare targets with good precision, reducing power consumption and size while being the fastest on integer-only processors. We also introduce specially tailored FFT and WT algorithms and show that they are usable in practice while having an extremely small code footprint, good precision, and high speed.

RETRACTED ARTICLE: Preserving learnability and intelligibility at the point of care with assimilation of different speech recognition techniques

The neoteric introduction of 5G technology in mobile internet is transforming Internet of mobile Things (IoMT) massively by addressing low latency, support for a large number of IoMT devices, and less power consumption, thereby delivering cost-effective solutions to low-end devices. This transformational technology enables a ubiquitous connected critical communication network between the healthcare system and IoT, as they largely depend on low-end devices for gathering data at the point of care. Gathering and interpolation of data from things are moved over to the cloud, making the extraction of knowledge and decision-making capabilities more robust. Vocal signals form the basis of communication between human beings with the transfer of complex data with variations in thrust, pitch, and tones. The representation and recognition of these analog signals by digital systems prove to be quite exciting and challenging. The spoken language models are converted to digital signals to be identifiable based on different cues like phonetic, prosodic, phonotactic, and lexical features. Voice patterns tend to be specific for every individual with a slight orientation towards the language spoken by the individual of a particular region. While speech patterns tend to alter the meaning of words with tones, high and low pitches in the utterance of the words, NLP tends to learn specific associations of words through vectors. The focus on learned networks in solving the problem of speech synthesis to text with minimal loss and high predictability of syllable of word, sentence, and paraphrase is needed. The creation of a knowledge base corpus of learned variable prosody of features helps in the learnability of interestingness directly without any perturbations. The learning algorithms to realize the degree of understandability of speech with the word, sentence identified, and transcription with substantial noise interference. The transfer of the acoustic features learned by algorithms proves to be quiet challenging as they are distorted by sudden environmental changes. Syllable extracted from the speech translation may or may not represent the Sentiment of the word, with different phonetical modulation. Utilization of the MobileNets and DistillBERT to transfer the language extraction and the edge reducing the time of processing and reducing the corpus of the size, reducing the adversial learning of the voice features and the patterns, reducing the Transfer of learned corpus and patterns.

Cloud Gaming with Foveated Video Encoding

Cloud gaming enables playing high-end games, originally designed for PC or game console setups, on low-end devices such as netbooks and smartphones, by offloading graphics rendering to GPU-powered cloud servers. However, transmitting the high-resolution video requires a large amount of network bandwidth, even though it is a compressed video stream. Foveated video encoding (FVE) reduces the bandwidth requirement by taking advantage of the non-uniform acuity of human visual system and by knowing where the user is looking. Based on a consumer-grade real-time eye tracker and an open source cloud gaming platform, we provide a cloud gaming FVE prototype that is game-agnostic and requires no modifications to the underlying game engine. In this article, we describe the prototype and its evaluation through measurements with representative games from different genres to understand the effect of parametrization of the FVE scheme on bandwidth requirements and to understand its feasibility from the latency perspective. We also present results from a user study on first-person shooter games. The results suggest that it is possible to find a “sweet spot” for the encoding parameters so the users hardly notice the presence of foveated encoding but at the same time the scheme yields most of the achievable bandwidth savings.

Current optimal active feedback and stealing response mechanism for low-end device constrained defocused iris certification

Addressing the problem of iris certification security and unclear textures caused by defocused scanning, the constrained defocused state iris scanned by a low-end device is taken as the research object. A codeless mode certification algorithm for label statistical learning based on current optimal active feedback is proposed. The certification model is improved based on the convolutional neural network structure. Guided by statistical learning, the features from the same category irises are set to the feature label of one category, and the iris certification function is designed. In addition, a current optimal active feedback mechanism is proposed. The certification effects of the certification model will be actively reflected in real time based on the current training data and the accumulated certification data, and the user can adjust the model structure and parameters in real time, to maintain the current optimality of the certification model structure. In addition, a result mapping layer included a stealing response mechanism is set between the certification result and the user. The stealing behavior that occurs can be made meaningless by changing the result output mapping relationship. Under the algorithm prerequisites, the JLU iris library is used to conduct experiments, and the accuracy of the algorithm can reach more than 99%, which indicates that this algorithm can effectively improve the accuracy of constrained state defocused iris scans with a low-end device, the model can be adjusted dynamically, and stealing attack behavior can be responded to effectively.

Boosted Metaheuristic Algorithms for QoE-Aware Server Selection in Multiplayer Cloud Gaming

Cloud Gaming (CG) provides a high performance and cost-effective solution where players with low-end devices can play high-end games without the need for advanced hardware. A cloud-based video game system offloads all the computational tasks to the cloud. Considering the dynamic nature of game workloads and resource capacity, resource management is still a significant challenge. Since CG is a real-time gaming service, graphics processing units (GPUs) are necessary to accelerate game scene rendering. GPUs are one of the most expensive resources in a CG platform. Therefore, service providers have a strong incentive to utilize GPUs efficiently to maximize their economic profit. In addition, players' quality of game experience (QoE) is a crucial parameter that can directly affect a service provider's profit and must be taken into account in any resource scheduling optimization. To satisfy both parties, in this paper we propose two efficient methods for GPU based server selection in CG. The proposed methods are an improved version of two well-known metaheuristic algorithms called Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), which we refer to as Boosted-PSO and Boosted-GA, respectively. The proposed methods consider service providers' profits and players' experience simultaneously. Our objective is to maximize GPU utilization, which will not only lead to the service provider's economic benefit, but also increase the player's QoE. Our simulation results show that compared to the existing methods to solve such an NP-Hard optimization problem, our Boosted-PSO method, followed by Boosted-GA, achieves the highest efficiency in terms of GPU utilization, capacity wastage, and player's QoE.

Early Disease Classification of Mango Leaves Using Feed-Forward Neural Network and Hybrid Metaheuristic Feature Selection

Plant disease, especially crop plants, is a major threat to global food security since many diseases directly affect the quality of the fruits, grains, and so on, leading to a decrease in agricultural productivity. Farmers have to observe and determine whether a leaf was infected by naked eyes. This process is unreliable, inconsistent, and error prone. Several works on deep learning techniques for detecting leaf diseases had been proposed. Most of them built their models based on limited resolution images using convolutional neural networks (CNNs). In this research, we aim at detecting early disease on plant leaves with small disease blobs, which can only be detected with higher resolution images, by an artificial neural network (ANN) approach. After a pre-processing step using a contrast enhancement method, all the infested blobs are segmented for the whole dataset. A list of several measurement-based features that represents the blobs are chosen and then selected based on their influences on the model's performance using a wrapper-based feature selection algorithm, which is built based on a hybrid metaheuristic. The chosen features are used as inputs for an ANN. We compare the results obtained using our methods with another approach using popular CNN models (AlexNet, VGG16, ResNet-50) enhanced with transfer learning. The ANN's results are better than those of CNNs using a simpler network structure (89.41% vs 78.64%, 79.92%, and 84.88%, respectively). This shows that our approach can be implemented on low-end devices such as smartphones, which will be of great assistance to farmers on the field.

Optimal Policies for Status Update Generation in an IoT Device With Heterogeneous Traffic

A large body of applications that involve monitoring, decision making, and forecasting require timely status updates for their efficient operation. Age of Information (AoI) is a newly proposed metric that effectively captures this requirement. Recent research on the subject has derived AoI optimal policies for the generation of status updates and AoI optimal packet queueing disciplines. Unlike previous research, we focus on low-end devices that typically support monitoring applications in the context of the Internet of Things. We acknowledge that these devices host a diverse set of applications some of which are AoI sensitive while others are not. Furthermore, due to their limited computational resources, they typically utilize a simple first-in–first-out (FIFO) queueing discipline. We consider the problem of optimally controlling the status update generation process for a system with a source–destination pair that communicates via a wireless link, whereby the source node is composed of a FIFO queue and serves two applications, one that is AoI sensitive and one that is not. We formulate this problem as a dynamic programming problem and utilize the framework of Markov decision processes to derive the optimal policy for the generation of status update packets. Due to the lack of comparable methods in the literature, we compare the derived optimal policies against baseline policies such as the zero-wait policy. Results indicate that the baseline policy fails to capture the complex system dynamics that determine the relationship between the frequency of status update generation and the resulting queueing delay and thus perform poorly. To the best of our knowledge, the derived optimal policy does not exhibit a simple structure; thus, we utilized the baseline policies, whose operation is intuitive, to gain insight into the inner workings of the optimal policy.

Big Mac: A Distributed PaaS Framework for on Demand Big Data Processing Using Machine Learning Techniques

In the age of start-ups and technical research, the demand for high-end computing power and loads of space is ever increasing. Machine learning techniques have become an inseparable part of the big data analytics. Setting up one’s own infrastructure to deal with all this vastness is usually not feasible due to high expenses and lack of desired expertise. As a solution to this problem, this paper proposes a system for Big-Data Analytics and Machine Learning based on Hadoop and Spark frameworks that also supports Operating System (OS) Rental Services. Machine Learning (ML) services provide option to use both existing inbuilt popular models or create one’s own model. OS Rental services provide users with high end infrastructure on their low-end devices on rent. The entire implementation has been made open source for ease of access and facilitating extensibility.

SARA: Secure Asynchronous Remote Attestation for IoT Systems

Remote attestation has emerged as a valuable security mechanism which aims to verify remotely whether or not a potentially untrusted device has been compromised. The protocols of Remote attestation are particularly important for securing Internet of Things (IoT) systems which, due to the large number of interconnected devices and limited security protections, are susceptible to a wide variety of cyber attacks. To guarantee the integrity of a software running on a single device, remote attestation is usually executed as an uninterrupted procedure: at the attestation time, a device stops the normal operation and executes the attestation of the entire device without interruption. The remote attestation protocols that aim to attest a large number of devices also follow the assumption on uninterrupted execution: when a device attests its network neighbours, each device verified in the neighborhood suspends its normal operation until the attestation protocol is completed. To avoid unnecessary suspension of the normal operation of the devices, this paper proposes a novel Secure Asynchronous Remote Attestation (SARA) protocol that releases the constraint of synchronous interaction among devices. In particular, SARA is an attestation protocol that exploits asynchronous communication capabilities among IoT devices in order to attest a distributed IoT service executed by them. SARA verifies both that each IoT device is not compromised (device trustworthiness), and that the exchanged communication data have not maliciously influence the communicating devices (legitimate operations). By tracing the execution order of each service invocation of an asynchronous distributed service, SARA allows each service to collect accurately historical data of its interactions, and transmits asynchronously such historical data to other interacting services. We have implemented and validated SARA through a realistic simulation on the Contiki emulator that demonstrates the functionality and efficiency of our protocol. The results confirm the suitability of SARA for low-end devices.

Modeling Secure Home Area Network Based on IoT for Resource Constraints Environment

The Internet of things (IoT) can be used in our daily life. Home area network (HAN) is one of the applications of loT. However, deploying HAN for underdeveloped countries in Africa is not easy due to the lack of Internet connectivity and stable electricity. First of all, Internet in Africa is growing fast but the majority of countries in Africa have Internet penetration of less than 10%. Furthermore, the problem of accessing electricity varies greatly across African countries, which is still less than 30% of the average usage rate. The aim of this paper is to model a secure HAN based on IoT for resource constraints environment, which are focused on both network connectivity and energy stability. Poor connectivity, the prevalence of low-end devices and several other obstacles should be considered for the secure HAN deployment in developing and underdeveloped countries. The proposed HAN model will provide the resource constraints module composed of network connectivity function and energy stability function to support the two constraints based on the IoT basic module and the security and privacy module. The proposed HAN model could be securely applied to various countries HAN applications with poor electricity supply and network connectivity.

Lightweight Security Architecture for Iot Device Communication

Internet of Things (IoT) becomes part of our daily life. IoT has greatly uplifted the human life and has touched many aspect in our life style. IoT devices are sophisticated low-end device having limited computational and energy resources. Most of the cryptographic algorithms are based on complex mathematical calculation which is not feasible to be computed on IoT devices. Hence presently IoT devices lack strong security features. Security and privacy are becoming the real concern for IoT devices. In this paper we are exploring the various cryptographic algorithm which can be used for IoT device authentication and secure communication. The overall system is designed considering the light weight factor, scalability, time complexity and ease of implementation

Software Protection Using Dynamic PUFs

Low-end computing devices are becoming increasingly ubiquitous, especially due to the widespread deployment of Internet-of-Things products. There is, however, much concern about sensitive data being processed on these low-end devices which have limited protection mechanisms in place. This paper proposes a Hardware-Entangled Software Protection (HESP) scheme that leverages hardware features to protect software code from malicious modification before or during run-time. It also enables implicit hardware authentication. Thus, the software will execute correctly only on an authorized device and if the timing of the software, e.g., control flow, was not changed through malicious modifications. The proposed ideas are based on the new concept of Dynamic Physically Unclonable Functions (PUFs). Dynamic PUFs have time-varying responses and can be used to tie the software execution to the timing of software and the physical properties of a hardware device. It is further combined with existing approaches for code self-checksumming, software obfuscation, and call graph and register value scrambling to create the HESP scheme. HESP is demonstrated on commodity, off-the-shelf computing devices, where a DRAM PUF is used as an instance of a Dynamic PUF. The protection scheme can be applied automatically to LLVM Intermediate Representation (IR) code through an AutoPatcher written in Python. For a sample program containing AES encryption and decryption routine, HESP introduces 48% execution time overhead and increases the binary file size by 32.5%, which is moderate compared to other schemes such as software obfuscation. It takes about 2.6 seconds on average for the tested programs to be patched and compiled through the modified compilation flow and scripts.

Operating Systems for Internet of Things Low-End Devices: Analysis and Benchmarking

In the era of the Internet of Things (IoT), billions of wirelessly connected embedded devices rapidly became part of our daily lives. As a key tool for each Internet-enabled object, embedded operating systems (OSes) provide a set of services and abstractions which eases the development and speedups the deployment of IoT solutions at scale. This article starts by discussing the requirements of an IoT-enabled OS, taking into consideration the major concerns when developing solutions at the network edge, followed by a deep comparative analysis and benchmarking on Contiki-NG, RIOT, and Zephyr. Such OSes were considered as the best representative of their class considering the main key-points that best define an OS for resource-constrained IoT devices: low-power consumption, real-time capabilities, security awareness, interoperability, and connectivity. While evaluating each OS under different network conditions, the gathered results revealed distinct behaviors for each OS feature, mainly due to differences in kernel and network stack implementations.

Architectural Considerations for Next Generation IoT Processors

Currently, the growth of the first generation Internet of Things (IoT) system for low-end devices having less memory and infrastructure is challenged due to the lack of engineering support in terms of architecture, technology, hardware, privacy, security, and business. This paper discusses system engineering issues relevant in IoT processors and builds a case for exploring rough set as an approximate computing technique in next generation IoT hardware. The proposed approach has a potential of dealing with inconsistencies arising at layers in hardware infrastructure for IoT applications with a dominance at cloud and fog processing. The authors bring out a research direction in IoT as an engineering discipline with an emphasis on the impact of processor hardware extension.

RELIGHT: A compact and accurate RTI representation for the web

Relightable images have been widely used as a valuable tool in Cultural Heritage (CH) artifacts, including coins, bas-reliefs, paintings, and epigraphs. Reflection Transformation Imaging (RTI), a commonly used type of relightable images, consists of a per-pixel function which encodes the reflection behavior, estimated from a set of digital photographs acquired from a fixed view. Web visualisation tools for RTI images currently require to transmit substantial quantities of data in order to achieve high fidelity renderings. We propose a web-friendly compact representation for RTI images based on a joint interpolation-compression scheme that combines a PCA-based data reduction with a Gaussian Radial Basis Function (RBF) interpolation exhibiting superior performance in terms of quality/size ratio. This approach can be adapted also to other data interpolation schemes, and it is not limited to Gaussian RBF. The rendering part is simple to implement and computationally efficient allowing real-time rendering on low-end devices.

Tensorizing Restricted Boltzmann Machine

Restricted Boltzmann machine (RBM) is a famous model for feature extraction and can be used as an initializer for neural networks. When applying the classic RBM to multidimensional data such as 2D/3D tensors, one needs to vectorize such as high-order data. Vectorizing will result in dimensional disaster and valuable spatial information loss. As RBM is a model with fully connected layers, it requires a large amount of memory. Therefore, it is difficult to use RBM with high-order data on low-end devices. In this article, to utilize classic RBM on tensorial data directly, we propose a new tensorial RBM model parameterized by the tensor train format (TTRBM). In this model, both visible and hidden variables are in tensorial form, which are connected by a parameter matrix in tensor train format. The biggest advantage of the proposed model is that TTRBM can obtain comparable performance compared with the classic RBM with much fewer model parameters and faster training process. To demonstrate the advantages of TTRBM, we conduct three real-world applications, face reconstruction, handwritten digit recognition, and image super-resolution in the experiments.