Signal Processing Networks Research Articles

Real-time Optimisation for Industrial Internet of Things (IIoT): Overview, Challenges and Opportunities

Industrial Internet-of-Things (IIoT) with massive data transfers and huge numbers of connected devices, incombination with the high demand for greater quality-of-services, signal processing is no longer producing small data sets but rather, very large ones (measured in gigabytes or terabytes), or even higher. This has posed critical challenges in the context of optimisation. Communication scenarios such as online applications come with the need for real-time optimisation. In such scenarios, often under a dynamic environment, a strict real-time deadline is the most important requirement to be met. To this end, embedded convex optimisation, which can be redesigned and updated within a fast time-scale given sufficient computing power, is a candidate to deal with the challenges in real-time optimisation applications. Real-time optimisation is now becoming a reality in signal processing and wireless networks of IIoT. Research into new technologies to meet future demands is receiving urgent attention on a global scale, especially when 5G networks are expected to be in place in 2020. This work addresses the fundamentals, technologies and practically relevant questions related to the many challenges arising from real-time optimisation communications for industrial IoT.

Image Noise Reduction Based on a Fixed Wavelet Frame and CNNs Applied to CT.

Radiation exposure in CT imaging leads to increased patient risk. This motivates the pursuit of reduced-dose scanning protocols, in which noise reduction processing is indispensable to warrant clinically acceptable image quality. Convolutional Neural Networks (CNNs) have received significant attention as an alternative for conventional noise reduction and are able to achieve state-of-the art results. However, the internal signal processing in such networks is often unknown, leading to sub-optimal network architectures. The need for better signal preservation and more transparency motivates the use of Wavelet Shrinkage Networks (WSNs), in which the Encoding-Decoding (ED) path is the fixed wavelet frame known as Overcomplete Haar Wavelet Transform (OHWT) and the noise reduction stage is data-driven. In this work, we considerably extend the WSN framework by focusing on three main improvements. First, we simplify the computation of the OHWT that can be easily reproduced. Second, we update the architecture of the shrinkage stage by further incorporating knowledge of conventional wavelet shrinkage methods. Finally, we extensively test its performance and generalization, by comparing it with the RED and FBPConvNet CNNs. Our results show that the proposed architecture achieves similar performance to the reference in terms of MSSIM (0.667, 0.662 and 0.657 for DHSN2, FBPConvNet and RED, respectively) and achieves excellent quality when visualizing patches of clinically important structures. Furthermore, we demonstrate the enhanced generalization and further advantages of the signal flow, by showing two additional potential applications, in which the new DHSN2 is used as regularizer: (1) iterative reconstruction and (2) ground-truth free training of the proposed noise reduction architecture. The presented results prove that the tight integration of signal processing and deep learning leads to simpler models with improved generalization.

LLISP: Low-Light Image Signal Processing Net via Two-Stage Network

Images taken in extremely low light suffer from various problems such as heavy noise, blur, and color distortion. Assuming the low-light images contain a good representation of the scene content, current enhancement methods focus on finding a suitable illumination adjustment but often fail to deal with heavy noise and color distortion. Recently, some works try to suppress noise and reconstruct low-light images from raw data. But these works apply a network instead of an image signal processing pipeline (ISP) to map the raw data to enhanced results which leads to heavy learning burden for the network and get unsatisfactory results. In order to remove heavy noise, correct color bias and enhance details more effectively, we propose a two-stage Low Light Image Signal Processing Network named LLISP. The design of our network is inspired by the traditional ISP: processing the images in multiple stages according to the attributes of different tasks. In the first stage, a simple denoising module is introduced to reduce heavy noise. In the second stage, we propose a two-branch network to reconstruct the low-light images and enhance texture details. One branch aims at correcting color distortion and restoring image content, while another branch focuses on recovering realistic texture. Experimental results demonstrate that the proposed method can reconstruct high-quality images from low-light raw data and replace the traditional ISP.

Novel Arithmetics in Deep Neural Networks Signal Processing for Autonomous Driving: Challenges and Opportunities

This article focuses on the trends, opportunities, and challenges of novel arithmetic for deep neural network (DNN) signal processing, with particular reference to assisted- and autonomous driving applications. Due to strict constraints in terms of the latency, dependability, and security of autonomous driving, machine perception (i.e., detection and decision tasks) based on DNNs cannot be implemented by relying on remote cloud access. These tasks must be performed in real time in embedded systems on board the vehicle, particularly for the inference phase (considering the use of DNNs pretrained during an offline step). When developing a DNN computing platform, the choice of the computing arithmetic matters. Moreover, functional safe applications, such as autonomous driving, impose severe constraints on the effect that signal processing accuracy has on the final rate of wrong detection/decisions. Hence, after reviewing the different choices and tradeoffs concerning arithmetic, both in academia and industry, we highlight the issues in implementing DNN accelerators to achieve accurate and lowcomplexity processing of automotive sensor signals (the latter coming from diverse sources, such as cameras, radar, lidar, and ultrasonics). The focus is on both general-purpose operations massively used in DNNs, such as multiplying, accumulating, and comparing, and on specific functions, including, for example, sigmoid or hyperbolic tangents used for neuron activation.

Supervised Machine Learning Classifiers for Diversity Combined Signals in 6G Massive MIMO Receivers

Support Vector Machine (SVM) is a statistical learning tool that was initially developed by Vapnik in 1979 and later developed to a more complex concept of structural risk minimization (SRM). SVM, as a supervised machine learning tool, is playing an increasing role in applications to detection problems in various engineering problems, notably in statistical signal processing, pattern recognition, image analysis, and 6G wireless communication networks. In this paper, SVM is applied to signal detection in 6G communication systems in the presence of channel noise in the form of fully developed Rayleigh multipath fading and receiver noise generalized as additive color Gaussian noise (ACGN). The structure and performance of SVM in terms of the bit error rate (BER) metric is derived and simulated for these advanced stochastic noise models and the computational complexity of the implementation, in terms of average computational time per bit, is also presented. The performance of SVM is then compared to conventional M-ary signaling optimal model-based detector driven by M-ary phase shift keying (MPSK) modulation. We show that the SVM performance is superior to that of conventional detectors which require as much as 7 bits-coding (M≥128) to produce comparable results to those of SVM. Finally, the SVM-based detector is implemented in an uplink SIMO system using both Equal Gain Combiner (EGC) technique and Root Mean Square Gain Combiner (RMSGC) technique in which the later technique will be proven to be superior to the earlier.

Cell-to-cell diversification in ERBB-RAS-MAPK signal transduction that produces cell-type specific growth factor responses

Growth factors regulate cell fates, including their proliferation, differentiation, survival, and death, according to the cell type. Even when the response to a specific growth factor is deterministic for collective cell behavior, significant levels of fluctuation are often observed between single cells. Statistical analyses of single-cell responses provide insights into the mechanism of cell fate decisions but very little is known about the distributions of the internal states of cells responding to growth factors. Using multi-color immunofluorescent staining, we have here detected the phosphorylation of seven elements in the early response of the ERBB–RAS–MAPK system to two growth factors. Among these seven elements, five were analyzed simultaneously in distinct combinations in the same single cells. Although principle component analysis suggested cell-type and input specific phosphorylation patterns, cell-to-cell fluctuation was large. Mutual information analysis suggested that each cell type uses multitrack (bush-like) signal transduction pathways under conditions in which clear fate changes have been reported. The clustering of single-cell response patterns indicated that the fate change in a cell population correlates with the large entropy of the response, suggesting a bet-hedging strategy is used in decision making. A comparison of true and randomized datasets further indicated that this large variation is not produced by simple reaction noise, but is defined by the properties of the signal-processing network.

Feasibility study of range verification based on proton-induced acoustic signals and recurrent neural network

Range verification in proton therapy is a critical quality assurance task. We studied the feasibility of online range verification based on proton-induced acoustic signals, using a bidirectional long-short-term-memory recurrent neural network and various signal processing techniques. Dose distribution of 1D pencil proton beams inside a CT image-based phantom was analytically calculated. The propagation of acoustic signal inside the phantom was modeled using the k-Wave toolbox. For signal processing, five methods were investigated: down-sampling (DS), DS + HT (Hilbert transform), Wavelet decomposition (Wavedec db1, db4 and db20). The performances were quantitatively evaluated in terms of mean absolute error, mean relative error (MRE) and the Bragg peak localization error (). In addition, the study analyzed the impact of noise levels, the number of sensors, as well as the location of sensors. For the noiseless case (32 sensors), the Wavedec db1 method demonstrates the best performance: is less than one pixel and the dose accuracy over the region adjacent to the Bragg peak (MRE50) is ∼3.04%. With the presence of noise, the Wavedec db1 method demonstrates the best noise immunity, achieving less than 1 mm and an MRE50 of ∼12%. The proposed machine learning framework may become a useful tool allowing for online range verification in proton therapy.

A new binary to gray code converter based on quantum-dot cellular automata nanotechnology

Quantum-dot cellular automata (QCA) is one of the most prominent technologies in nanometer-scale with appreciable reduction of size and power consumption and high switching frequency to overcome the scaling limitations of complementary metal-oxide semiconductor. On the other hand, code converters play a key role in signal processing and efficient network designs. The researchers have focused on emerging nano-devices that can identify errors throughout information transfer. Therefore, in this research, a new QCA-based 4-bit binary to gray converter circuit employing the appropriate configuration of the XOR gate as a basic building block has been suggested. The layout has been generated using the QCADesigner simulation tool to test the functionality of the code converter. The performance results indicated that the proposed converter works properly and has optimum performance parameters such as latency, complexity, and consumed area as compared to the current schemes.

An Optical Switch Based on Electro-Optic Mode Deflection in Lithium Niobate Waveguide

We propose a 1 × 2 electro-optic miniature optical switch based on mode deflection of lithium-niobate (LN) annealed proton exchange waveguides with microstructured electrodes. Our fabricated device consists of LN waveguides covered with isosceles-triangle-shaped array electrodes, a horn-shaped input waveguide, and two tapered output waveguides. At a driving voltage of 40 V, the two-modes can switch between two output ports with an extinction ratio of 17.39 dB at the wavelength of 1530 nm. The switching speed is 22 ns. This miniature mode switch could find applications in signal processing and high-speed optical communication networks.

Spectral Efficiency of Wireless Relay Network in Frequency Non-Selective Channel

Introduction. A wireless communication system based on a relay network where a link between a source and a destination is carried out through a network of relay nodes have been considered. Relay networks operate according with amplifier-and-forward protocol where each relay node performs reception, amplifying, phase shifting and retranslation of a signal to the destination node. As a result a task of powers and phases optimization in the relay nodes (i.e. the complex weighted coefficients optimization) becomes actual. Complex weighted coefficients of the relay nodes are optimized in such a way as to ensure the maximum signal to noise ratio at the receiver while limiting a power emitted by the relay nodes. In the paper, optimization of spatial processing with different a priori channel state information (i.e. instantaneous channel state information and the second order statistics) have been considered. Aim. Spectral efficiency analysis of a relay network in a multipath channel where the relay network was optimized by using of two types a priori information: an instantaneous channel state information and second order statistics. Materials and methods. Optimization of spatial signal processing in the relay network was based on methods of statistical theory and optimization using analytics of linear algebra and methods of mathematical programming. Performances of the relay network were analyzed using Monte Carlo simulation. The simulation was performed in MATLAB program environment using CVX toolbox for solving convex optimization task. Results. In the paper optimal solutions for spatial signal processing in the relay network were presented. The solutions were based on maximum of signal to noise ratio while limiting total relay power and individual power of relay nodes. Monte Carlo simulation was performed to provide performances of the relay network for different types of channel state information and channel parameters. Mean capacities versus mean source power, a budget of relay nodes power and a ratio between random and deterministic power of the channel were gained for the Rayleigh model of multipath channel. Conclusions. The results have a practical application. Thus, the use of the second order statistics is possible in relay networks when direct visibility with a low level of background from local objects is provided. In urban areas, where shading and multipath propagation of signals occur, it is possible to use only an approach based on the knowledge of channel instantaneous state.

Capacity Analysis of Hybrid MIMO Using Sparse Signal Processing in mmW 5G Heterogeneous Wireless Networks

The mmW cellular systems of large bandwidths offer multiple times rise in capacity as compared to existing 4G networks with comparable cell density. These avoid the unnecessary cell splitting by enlarging the capacity of individual tiny cells significantly in a scenario of very high-density cell deployments. This work will provide an opportunity to achieve a particular capacity value by varying the mmW channel gain in the 5G and 6G wireless networks. The OMP algorithm is modified and the existing sparse signal processing concept is utilized for the capacity analysis of hybrid MIMO here. The capacity (b/s/Hz) of the conventional and hybrid MIMOs are calculated and compared against given SNR range (dB) in a 5G mmW heterogeneous network under different values of mmW channel gain. It has been found that capacity analysis curves of conventional and hybrid MIMOs both show a descending trend with the increase in SNR range as the channel gain is increased due to over-saturation of the used sparse mmW channel. These curves exhibit local variation, time dependence, frequency selectivity, reliable communication rate, and diversity on both ends and the use of MIMO has a gain in degrees of freedom. The hybrid MIMO due to hardware limitations of conventional MIMO utilized a large number of antennas with lesser radio frequency chains. At some point of channel gain, its capacity curve approached the capacity curve of conventional MIMO in a moderate SNR range, and approached very closely for all channel gains in low and high SNR ranges.

UltraTrail: A Configurable Ultralow-Power TC-ResNet AI Accelerator for Efficient Keyword Spotting

Recent advances in machine learning show the superior behavior of temporal convolutional networks (TCNs) and especially their combination with residual networks (TC-ResNet) for intelligent sensor signal processing in comparison to classical CNNs and LSTMs. In this article, we propose UltraTrail, a configurable, ultralow-power TC-ResNet AI accelerator for sensor signal processing and its application to efficient keyword spotting (KWS). Following a strict hardware/model co-design approach, we have derived an optimized low-power hardware architecture for generalized TC-ResNet topologies consisting of a configurable array of processing elements and a distributed memory with dynamic content reallocation. We additionally extend the network with conditional computing to reduce the number of operations during inference and to provide the possibility for power-gating. The final accelerator implementation in Globalfoundries' 22FDX technology achieves a power consumption of 8.2 μW for the task of always-on KWS meeting the real-time requirement of 100 ms per inference with an accuracy of 93% on the Google Speech Command Dataset.

Elliptical-Core Highly Nonlinear Few-Mode Fiber Based OXC for WDM-MDM Networks

In order to realize an optical cross-connect (OXC) converting wavelengths and spatial modes into one-dimensional switching ports, we propose an active mode selective conversion without parasitic wavelength conversion, based on the intermodal four-wave mixing (FWM) arising in a few-mode fiber (FMF). First, we design a dispersion-engineered elliptical-core highly nonlinear FMF (e-HNL-FMF) with a graded refractive index (RI) profile, which can independently guide 3 linearly polarized (LP) spatial modes. Meanwhile, a high doping concentration of germanium in the core leads to relatively high intermodal nonlinear coefficients of 3.23 (W·km) <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> between LP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">01</sub> and LP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">11a</sub> modes and 3.14 (W·km) <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> between LP01 and LP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">11b</sub> modes. Next, we propose an e-HNL-FMF based OXC scheme for wavelength division multiplexing-mode division multiplexing (WDM-MDM) networks. After optimizing both the e-HNL-FMF length and pump power, we can realize either active mode selective conversion over the designated wavelength-band or three-wavelength to three-mode superchannel conversion for 100 Gbaud 16-quadratic-amplitude modulation (16-QAM) signals over the C-band. Due to excellent characteristics of the e-HNL-FMF, both cost and configuration complexity of the OXC can be reduced, showing great potentials for all-optical signal processing in the future WDM-MDM networks.

Radiation-Hardened Sensor Interface Circuit for Monitoring Severe Accidents in Nuclear Power Plants

Radiation sensor interface circuits can be mounted on sensor nodes in wireless sensor networks to detect radiation signals. They respond quickly to operators and command centers, allowing appropriate decisions to be made to mitigate nuclear power plant accidents. For instance, high-level radiation leaked from a reactor building during a nuclear plant accident can degrade the performance of the radiation sensor interface implemented by integrated circuits due to threshold voltage shift, leakage current change, and noise increase induced by radiation impact events. To obtain reliable signal measurements in the high radiation environment, radiation-hardening technologies must be applied to radiation sensor interface circuits. The radiation-hardened sensor interface circuit proposed in this article exploits a pair of charge-sensitive amplifiers (CSAs) in dual paths to equivalently cancel out the distorted output signals caused by radiation impacts on the circuits. In addition, a low-power, fully differential successive approximation register (SAR) analog-to-digital converter (ADC) is used to reduce the radiation effects by subtracting the radiation-induced noise of one input from the other. The implemented radiation sensor interface circuit can maintain its performance up to 1.0 mrad (SiO2) of irradiation with only 2.5 mW power consumption, while successfully sensing and transferring the radiation information to the downstream signal-processing network.

Performance analysis of all-optical NAND, NOR, and XNOR logic gates using photonic crystal waveguide for optical computing applications

All-optical photonic integrated devices have gained great attention in the field of optical computing and large-scale integration. All-optical logic gates are useful for optical signal processing and optical communication network. The flexible devices presented here satisfy the functionality of NAND (NOT-AND), NOR (NOT-OR), and XNOR (exclusive NOR) logic gates using only one structure with proper changes in the phase of an applied light signal. The design of all-optical logic gates is implemented with photonic crystal waveguides using square lattice silicon rods. The performance of the structure is simulated, verified, and analyzed by the finite-difference time-domain method, with the principle of interference effect at a wavelength of 1550 nm. The contrast ratio (CR) of NAND, NOR, and XNOR logic gates is 17.59, 14.3, and 10.52 dB, respectively, with an optimized size of 7.2 μm × 5.4 μm.

An Integrated System of Artificial Intelligence and Signal Processing Techniques for the Sorting and Grading of Nuts

The existence of conversion industries to sort and grade hazelnuts with modern technology plays a vital role in export. Since most of the hazelnuts produced in Iran are exported to domestic and foreign markets without sorting and grading, it is necessary to have a well-functioning smart system to create added value, reduce waste, increase shelf life, and provide a better product delivery. In this study, a method is introduced to sort and grade hazelnuts by integrating audio signal processing and artificial neural network techniques. A system was designed and developed in which the produced sound, due to the collision of the hazelnut with a steel disk, was taken by the microphone placed under the steel disk and transferred to a PC via a sound card. Then, it was stored and processed by a program written in MATLAB software. A piezoelectric sensor and a circuit were used to eliminate additional ambient noise. The time-domain and wavelet domain features of the data were extracted using MATLAB software and were analyzed using Artificial Neural Network Toolbox. Seventy percent of the extracted data signals were used for training, 15% for validation, and the rest of the data was used to test the artificial neural network (Multilayer Perceptron network with Levenberg-Marquardt Learning algorithm). The model optimization and the number of neurons in the hidden layer were conducted based on mean square error (MSE) and prediction accuracy (PA). A total of 2400 hazelnuts were used to evaluate the system. The optimal neural network structure for sorting and grading hazelnuts was 4-21-3 (four neurons in input layers, 21 neurons in the hidden layer, and three outputs which are the desired classification). This neural network (NN) was used to classify hazelnut as big, small, hollow, or damaged. Results showed 96.1%, 89.3%, and 93.1% accuracy for big/small, hollow, or damaged hazelnuts were obtained, respectively.

Non-intrusive load disaggregation solutions for very low-rate smart meter data

With the active large-scale roll-out of smart metering worldwide, details about the type of smart meter data that will be available for analysis are emerging. Consequently, focus has steadily been shifting from analysis of high-rate power readings (usually in kHz to MHz) to low-rate power readings (sampled at 1–60 s) and very low-rate meter readings of the order of 15–60 min. This has triggered renewed research into practical non-intrusive load disaggregation of low- to very-low granularity meter readings to address challenges not addressed by existing disaggregation approaches, namely, indistinct appliance ON/OFF transitions, increased likelihood of overlapping appliance usage within a sample and noise due to unknown appliances. In this paper, focusing on smart meter readings at hourly resolution, three load disaggregation solutions are proposed based on: (i) optimisation (minimisation of error between aggregate and disaggregated loads), (ii) graph signal processing and (iii) convolutional neural network. These are benchmarked with state-of-the-art approaches, based on factorial hidden Markov model and combinatorial optimisation implemented in the NILMTK toolbox, and discriminative disaggregation sparse coding. The hourly electricity profile data is obtained from real-world active power readings from the REFIT dataset11The REFIT dataset (cleaned version) used for validation in this paper can be accessed via doi: 10.15129/9ab14b0e-19ac-4279-938f-27f643078cec. over a period of longer than one year. All proposed disaggregation approaches outperform benchmarking methods for labelled appliances in terms of both energy performance metrics and faster execution time. The proposed approaches succeed in disaggregating, at very low resolutions, a wide range of loads including white goods even when there are unlabelled loads contributing to the meter readings.

Learning Convolution Neural Network with Shift Pitching based Data Augmentation for Vibration Analysis

Data augmentation is a common approach that been implemented in order to increase the training data quantity for Convolutional Neural Networks in signal processing, image recognition and speech recognition. However, the conventional data augmentation methods usually implement the window slicing and overlap window slicing methods in the bearing fault analysis. Meanwhile, the audio deformation approach such as time stretching and pitch shifting methods have been commonly used as data augmentation approach in speech recognition. Thus, this paper proposed a data augmentation based on shift pitching technique for the vibration signal. The relationship between the audio and the vibration signal is evaluated for a bearing fault analysis using Convolution Neural Networks. The new dataset produce by the data augmentation is used to increase the number of training dataset and to improve the Convolutional Neural Networks training performance. The result shows that the shift pitching based data augmentation method able to achieve higher training accuracy compared to the window sliding data augmentation. The combinations of all ratio pitch obtain 93% accuracy whilst the accuracy for a single rate pitch are between 81% to 91%.Thus, the proposed method is competent and able to improve the performance of bearing fault classification

Software module development for non-invasive blood glucose measurement using an ultra-wide band and machine learning

Diabetes is a chronic disease and in uprising trend worldwide. There is no remedy, hence, blood glucose management is essential by screening blood glucose concentration levels (BGCL) regularly to maintain a healthy life. However, the present way of measuring BGCL is invasive by using a glucometer and drawing a blood sample directly from the human body. To overcome this discomfort-problem, a non-invasive device to measure BGCL is in demand. This paper presents an autonomous software module with a user-friendly graphical user interface (GUI) based on digital signal processing (DSP) and artificial neural network (ANN) to process, classify and recognize the BGL signature from captured ultra-wideband (UWB) signal through human blood medium. To capture the signal, a pair of UWB bio-antenna is placed in between the human earlobe. Received signals are captured and processed through GUI and undergo signal processing, ANN training, testing, and validation. An interface is developed to integrate the hardware (UWB transceiver, bio-antenna, etc.) and the developed software module to make a system. The initial system showed a consistent result with reliability and demonstrated 90.6% accuracy to detect the BGCL. The detection accuracy is 9.6% improved compared to existing work. Besides, this proposed system is cost-effective, user-friendly and suitable to be used by both doctors and home users.

The Prism: Recursive FIR Signal Processing for Instrumentation Applications

This paper provides the mathematical background for the precise, repeat integral signal monitor (Prism), a signal processing network node used in a variety of sensing and instrumentation applications. The key operation is a double Fourier-style integration, which can be implemented recursively using sliding windows and precisely using Romberg integration. The Prism generates one or two outputs; if two are generated, they are orthogonal, analogous to an analytic signal, from which sinusoid properties, such as frequency, phase, and amplitude, can readily be derived. The Prism outputs are finite impulse response (FIR), but the calculation is recursive, resulting in a low computational cost which is independent of filter length. This paper compares the Prism’s computational efficiency with both a least squares FIR filter design and an equivalent Prism filter implemented as a conventional convolution. The advantages of the Prism include design simplicity, low computational cost, and a linear phase response, making it a useful network node for a wide range of instrumentation and signal processing tasks.