During the online learning process of support vector machines (SVMs), when a newly added sample is violating the Karush–Kuhn–Tucker) conditions, the new sample should be a new SV and transfer the old samples between the SVs and the non-SVs. Normally, the performance of an SVM model is decided by the SVs, and the model should be updated by the newly added SVs; therefore, the selection of high-quality candidate SVs will lead to a better learning accuracy, whereas low-quality candidate SVs may result in low learning efficiency and unnecessary updating. A new strategy is proposed to select the candidate SVs. SVs are selected according to two new criteria: the importance and the informativeness criteria. Furthermore, a mixed local–global regularisation method is applied during the online learning process to improve the penalty coefficients. Experiment results show that the proposed algorithm can achieve a better performance with a faster speed and a higher accuracy when compared with traditional methods.

A new segmentation method for mammography imaging system is proposed. Segmentation of masses is always a difficult problem in radiological image interpretation. Conventional methods such as region growing suffer from their computational complexity and hence can hardly be used for segmentation of high-resolution images. In order to achieve efficiency in both computational complexity and accuracy, a novel digital cellular neural network (CeNN) based approach is presented for segmentation. The approach is featured with quantisation to significantly reduce the computational complexity and non-linear template for robustness. After segmentation, a multilayer perceptron classifier is used for feature extraction and classification. Compared with other prior works, the proposed work is able to reduce resource overhead up to 63% and energy consumption up to 41% on FPGA while maintaining only up to 1.5 and 0.6% accuracy deviations for mediolateral-oblique and cranial-caudal views, respectively.

Transdermal active drug delivery is one way to propose new disease treatment as an improvement to passive patches diffusion based on concentration gradient. Although skin limitations due to low permeability active transdermal delivery approach are investigated, by the proposition of an iontophoretic reusable and controllable patch. Making such an e-health care device offers active control and permanent actuating link able to control the amount of drug delivered. In vitro experiment is conducted to investigate the current effect on the permeation. Modelling and simulations are carried out to optimise the design and the electrical stimulation and finally a global overview of the architecture of the patch is presented.

A 40 Gbps 4-tap 4-level pulse amplitude modulation closed-loop decision feedback equaliser (DFE) is proposed. The DFE adopts a novel high-speed comparator to resolve the critical timing constraints of the first tap. The comparator decreases the slicing delay by shortening the gap between initial and target voltages. Compared with the existing closed-loop DFE designs, the proposed scheme relieves timing constraints without complex clock distribution circuits and extra area. Simulations based on the RF-MOS model verify that the delay of the comparator is improved by 32.8% and the output swing is increased by more than 2.8 times. The proposed DFE which can compensate −9.5 dB channel loss is designed in 55 nm CMOS technology. The power consumption is 67 mW from a 1.2 V supply and the circuit occupies an active area of 0.021 mm², achieving 1.68 pJ/bit energy efficiency.

A high-multiplication factor multiplying delay-locked loop (MDLL) with infinite impulse response filter-based accumulated jitter reduction technique is presented. In every output clock cycle, the proposed jitter reduction loop samples the periodic jitter, accumulates it, and subtracts accumulated jitter from the next output clock period. The proposed technique is applied to 10 MHz MDLL with 32 kHz reference clock, a multiplication factor of 313, shows 38% jitter reduction, and greatly improves MDLL locking even with high-supply noise. The MDLL implemented in 0.18 μm CMOS process consumes 42 μW and the core area is 0.043 mm².

Achievement of low power consumption in the field of network-on-chip (NoC) is a prominent research in recent days. Many works have attempted to improve performance in NoC using architectural and algorithmic models. The researches attempted to mitigate certain factors like power consumption, speed, complexity, scalability, and flexibility. Currently, NoC engineers incorporated external or internal sense amplifier (SA) in the architectural model of NoC. In the conventional double-tail SAs (DTSAs), more amount of energy is consumed in the heavy traffic state. Hence, many improved DTSAs like reconfigurable DTSA (R-DTSA), variable energy aware SA link for asynchronous NoC (VELAN) DTSA (V-DTSA) were proposed in the previous works and employed with the transceivers. They were evaluated on TSMC 90 nm technology with the real-time data traffics which are obtained using traffic estimators and traffic generators. An optimised logic interface for transceivers of NoC which is superior to the conventional DTSA, R-DTSA, and V-DTSA is developed. The proposed design is implemented with transceivers and evaluated on TSMC 90 nm technology for comparing the performance with the previous SAs.

Fully integrated high-voltage pulse driver with low voltage supplies and low voltage switching signals is described. Using charge redistribution for the output drive significantly reduces switching losses in the load capacitors even with the high-voltage output swing. The proposed circuit implemented with 0.18 µm HV LDMOS technology achieves 24–55% efficiency improvement for 20 k–1 MHz 29 V pulse generation with 1.2 and 3.3 V supply voltage.

A new steering mechanical structure and an improved control method of the manned self-balanced vehicle are presented in this Letter. An effective method is introduced to calculate the real-time inclination of the vehicle. Then, the integrating inclination of the vehicle could be obtained by the complementary filter and proportional derivative (PD) algorithm. In addition, the control system could be regarded as a linear superposition of three parts, including the balance control, the speed control and the steering control. Moreover, the proportional, integral and derivative (PID) parameters of the whole control system are different under different operational conditions. Furthermore, the pressure sensors placed on the left and right pedals could detect the change of the weight, which simplifies the mechanical structure and improves the flexibility of the manned steering. The experiments in the prototype are designed to prove the steering structure and the improved method.

Reinforcement Q-learning algorithm for the optimal tracking control problem with unknown dynamics and delays is proposed. Traditional reinforcement learning methods require an accurate system model, which is avoided by means of the Q-learning method. This is very meaningful in practical implementation because all or part of the model of the system is often difficult to obtain or requires an additional high cost. First, the augmented system composed of the original system and reference trajectory is constructed, then the corresponding augmented linear quadratic tracking (LQT) Bellman equation is derived. Based on this, the reinforcement Q-learning algorithm is presented at the end. To implement this method, the iteration equations are solved online by using the least squares technique.

A fast concrete crack detector based on L2 sparse representation is proposed. Specifically, via dividing the existing concrete images, many representative crack and non-crack image regions are selected for the over-complete dictionary. To suppress the noise disturbances, discrete cosine transformation is to extract the frequency-domain characteristics of these regions. For one new concrete image, it is first divided into many non-overlapping regions, and their sparse coefficients are fast computed on the established over-complete dictionary. Moreover then, a pooling operation is to extract the difference value between their sum coefficients on the crack templates and those on the non-crack ones, and easily yet effectively select the crack candidates via the sign bit of their difference values. Experiments on the practical concrete images show that the algorithm has high precision and efficiency.

Eigenvalues are intrinsic and representative values of a square matrix. They have thus been used in many image processing areas due to their important application value, but not in the image quality assessment (IQA) field. In this Letter, the authors study the correlation between local mean eigenvalues (LMEs) and perceptual quality of images, and demonstrate the applicability of LMEs in IQA. The LMEs are related to structural complexity of images. The LMEs and natural scene statistics features are utilised for a sparse dictionary learning. Experimental results conducted on promising IQA databases show their method's superiority in comparison with top-performing blind IQA metrics.

Multidimensional field embedding methods have been demonstrated to effectively characterise spectral signatures in hyperspectral images. However, high-dimensional data composed of a number of classes presents challenges to the existing embedding methods. This Letter proposes an enhanced multidimensional field embedding algorithm based on the force field formulation. The comparative performance of the proposed algorithm is evaluated in the classification and visualisation of commonly used hyperspectral images. Experimental results demonstrate its superiority over previously used field embedding techniques.

Optical flow estimation has bottlenecks such as large displacement and motion blur. In this Letter, the authors propose a geodesic-based probability propagation (GeoFlow) method combining the global geodesic with local spatial similarity to build a non-local superpixel graph. To achieve efficient belief propagation, a probabilistic framework for optimising the Markov random field (MRF) objective is proposed. In this way, the limitation of local propagation can be tackled in the global image level, and the probabilistic framework reduces computational complexity in the optimisation. In experiments, their method showed promising performance by improving the results on two public large displacement benchmark datasets.

A method for horizon detection in maritime scenes using a scene parsing network is proposed. First, each pixel from an input image is segmented into corresponding semantic categories using a scene parsing network, which relies on a deep neural network. Then, the boundary information related to the horizon and the sea is extracted. Scene segmentation allows the proposed method to identify the horizon, regardless of whether the boundary between the sea and sky is smooth or blurry, or whether the image contains many line elements like the horizon. Moreover, least squares and median filtering are iteratively used to retrieve an accurate estimation of the horizon line. Experimental results demonstrate the superior accuracy of the proposed method to identify the horizon when compared to state-of-the-art methods.

During a radiation treatment, the images of kilovoltage digital reconstructed radiograph (KV-DRR) and megavoltage digital radiograph (MV-DR) are registered to guide the therapy. Such registration is difficult since the images belong to different modalities. To reduce the difficulty, a fractal convolutional network is developed to map MV-DR images into the modality of KV-DRRs. The key idea is to split a hourglass-shape network into multiple similar networks at reduced scale, yielding a fractal topology that is self-similar at multiple scales. This division allows to predict images of unprecedented high resolution at low graphics processing unit memory usage. Experiments demonstrate that perceptual plausible and numerical accurate results are achieved out-competing recent alternative architectures.

An analytical model is described to estimate the performance of the carrier sense multiple access with collision avoidance (CSMA/CA) protocol during the simultaneous transmit/receive (STR) operation of the upcoming IEEE 802.11ax amendment for WLANs. Furthermore, based on that model, an enhancement to the operation of CSMA/CA to increase its performance during the STR mode is proposed.

Systematic Luby transform (LT) codes with grey-mapped QAM constellations over additive white Gaussian noise (AWGN) channels are investigated. For high-order modulations, it can be learnt that the traditional Gaussian approximation method developed for binary PSK (BPSK) is not accurate any more. To improve BER performance of systematic LT codes with grey-mapped QAM, therefore, presents a novel coded modulation scheme, which only extracts Gaussian alike log-likelihood ratios for decoding, based on the modified QAM constellation. More importantly, the proposed scheme allows to optimise degree distributions for QAM-modulated systematic LT codes as easily as for BPSK. Simulation results show that the degree distribution related to the proposed coded modulation scheme can provide outstanding BER performance for systematic LT codes over the AWGN channel.

An ultra-compact pulse compressor for generating ultrawideband short pulses is proposed and tested. The proposed pulse compressor consists of a thin copper-coated dielectric laminate to form an ‘ultra-compact’ reverberant cavity, where a time-reversal-based input pulse is compressed into an ultrawideband short pulse. The proposed pulse compressor is validated via numerical simulation and measurement, and results show that a short pulse is generated with a peak gain of >10 dB.

Field emission current is widely used to measure various characteristics of nanoscale. In this Letter, a cantilever of a carbon nanotube is focused and the amplification of the current is discussed. Increasing the voltage which induces the current gives the amplification; however, this method results in the large power consumption. This Letter provides the alternative solution with the use of the surrounding noise. Exploiting the non-linear behaviour of the current and introducing a signal processing of time-averaging, the amplification is observed. The theoretical discussion provides the concrete theory of the amplification: essentially such an effect is obtained by the modulation of the probability density of the noisy voltage. The numerical results demonstrate the effectiveness of the method.

Piezoelectric nanogenerators (PENGs) based on micro-fibre of polyvinylidene fluoride (micro-fibre polyvinylidene fluoride (PVDF)). Flexible nanogenerator was developed by addition nanoparticle composition. This PENG fabricated using micro-fibre PVDF/BaTiO₃, which was deposited on paper interdigitate substrate by wire electrode free surface electrospinning method. The device from micro-fibre PVDF performed open-circuit voltage and short-circuit current of 13 mV and 380 nA recently. The device from composite material addition value of open-circuit voltage and short-circuit current of 528 mV and 661 nA recently.

Dispersion in the multimode fibres (MMF) limits the transmission distance of the modern data communication networks. The authors show that the use of single-mode (SM) vertical cavity surface emitting lasers (VCSELs) in combination with the receiver module of a commercially available transceiver can increase the transmission distance to several kilometres. They report 25.78 Gbit/s error-free transmission over 1800 m of multimode fibre with 850 nm and 1400 m with 910 nm SM VCSELs. Data transmission up to 2600 and 2400 m can be achieved with BER below 3.8×10⁻³ forward error correction limit accordingly. The use of multiple wavelengths demonstrates the opportunity to increase the capacity of MMF links with shortwave wavelength division multiplexing.

A zero voltage switching (ZVS) detection method, which relies on the half-bridge output voltage slopes during the MOSFET turn-on and turn-off transient processes, is proposed for the wireless power transfer (WPT) systems in this Letter. A voltage slope detection circuit and its active control method are also proposed to convert the nanosecond transient signal to a steady voltage level, which can be handled by the microcontroller easily. By having the information of the turn-on and turn-off voltage slopes, the state of ZVS can be evaluated to optimise the switching losses. A 500 kHz, three-coil resonant WPT system is built in order to evaluate the performance of the proposed detection method. Experimental results show that the ZVS state can be detected effectively. Also, the turn-off current can be controlled to a constant value to optimise the switching losses under the load and parameter variations.

We report a method of recovering degraded lead-acid batteries using an on–off constant current charge and short–large discharge pulse method. When the increases in inner impedance are within ∼20% of the initial impedance value, their system will permit discharge times to recover to a level approximately matching their initial time values. In one experiment, when the discharge time of a <5-year-old lead-acid battery used for engine starting had degraded to about 50% of its initial discharge capacity, the authors found that 80% of the initial state discharge time could be recovered and maintained via their method. In a separate experiment, a deep-cycle battery was forcibly degraded by 12 h of depletion at a constant current of 10 A discharged at a terminal voltage of 10.2 V. The results of that experiment show that use of their on–off constant current charge method for the deep-cycle battery (recovered at 80% initial discharge time) quadrupled the post-recovery discharge time in comparison with a battery recovered via constant voltage charge. Taken together, these results indicate that use of their proposed on–off constant current charge system can successfully recover both degraded engine start and deep-cycle lead-acid batteries.

This Letter proposes a multi-objective self-synchronised virtual synchronous generator (VSG) strategy in unbalanced power grid. A second-order generalised integrator is applied to the traditional VSG strategy to deal with the double-frequency ripple components. Therefore, the precision of the self-synchronisation process can be confirmed, and the VSG can follow the power references of both DC and double-frequency ripple components. On this basis, the power compensation strategy is combined with the VSG strategy to achieve different control objectives. Comparative experimental studies of the traditional VSG and proposed VSG with different targets are conducted to verify the effectiveness of the proposed strategy.

A novel coherent high-speed manoeuvering target detection method based on the parameterised autocorrelation function and three-dimensional integration is presented. The parameterised autocorrelation function realises the searching parameters replacement and reduces the noise correlation without the signal energy loss. Consequently, compared to two representative detection methods, the proposed method can strike a balance between the computational cost and detection performance without the blind speed sidelobe.

Ultra-thin InAlN/GaN heterostructure field-effect transistors (HFETs) having high maximum oscillation frequency (f _max) are fabricated by scaling lateral dimensions. A 3 nm GaN cap layer is adopted to reduce the electron density and suppress the short-channel effects. Non-alloyed regrown n⁺-GaN ohmic contacts with total ohmic resistance (R _tot) of 0.13 Ω.mm is also introduced into the device, in which the virtual source-to-drain distance is 600 nm. T-shaped gate with 40 nm length is formed in the centre of the source-to-drain region by self-aligned e-beam lithography. The peak extrinsic transconductance (g _m) reaches 956 mS/mm. Most of all, a high f _max of 405 GHz is obtained, which is the highest value among the reported InAlN/GaN HFETs. These obtained results mean that the InAlN/GaN HFETs having reliability should be still suitable for G-band (140–220 GHz) power-amplifier application with further optimisation.

Some blind humans have developed the ability to perceive their silent surrounding by using echolocation based on tongue clicks. Past research has also shown that blind echolocators can use information gained from multiple echoic ‘views’, provided through head movements, to successfully identify 2D shapes. Here, echo features that might be used by blind humans to discriminate 2D shapes are investigated. Echoes from four shapes are collected with a custom-built acoustic radar and various features are extracted. By piecing together individual features across the measurement plane, it is found that total power and spectral centroid are two salient features for shape discrimination from multiple echoic views.

In this letter, the authors formulate the energy-efficient user association and cell sleeping optimisation problem which applies to the multi-tier ultradense network scenarios as a complex 0–1 integer programming and propose an alternating iterative method to solve the problem via two low-complexity sub-problems. The simulation results show that the proposed strategy can steadily reduce overall energy consumption compared with the existing strategies.

Media-based modulation (MBM) concept is exploited to increase physical layer (PHY) security in the presence of an illegitimate listener. A precoding design is developed to increase the mutual secrecy rate of MBM, and the secrecy mutual information expressions are derived. The provided analyses and numerical results have also shown that MBM is a perfect candidate for PHY security by providing the maximum achievable secrecy mutual information, even in the presence of channel estimation errors.