University of Macau

In this paper, the concept of fusion topology is proposed to tackle the development trend and control issue of hybrid parallel multi-converter applications and to achieve better performance in transient and steady-state operations. Compared to previously fixed frequency compensation strategies, data-driven finite-state machine (FSM) control can determine which voltage source converter (VSC) dominates the operation or whether two VSCs operate simultaneously rather than giving the specific target operation to each VSC alone. The results are verified by a real-time digital simulator (RTDS) with the hardware-in-the-loop (HIL) experiments. The 20kVA RTDS experimental results show that the transient time is shorter by 53.8%− 85.7%, and power loss is smaller by 28.6%− 78.3%, respectively, compared with the previous methods.

This paper proposes a centralized battery energy storage based medium-voltage multi-winding dynamic voltage compensator (DVC) for balance and unbalance operations. In this topology, the compensation voltage is added to the grid side through the transformer, and the primary side of the transformer is shunted by multiple windings to support the voltage sag on the medium-voltage distribution systems. According to the symmetrical and asymmetrical grid-side voltage sags, the positive and negative sequence mathematical models of DVC as well as their voltage-current double closed-loop control strategies are developed. As for the DC bias caused by circulating current between the primary windings of each phase of the multi-winding transformer, circulating current suppression algorithm is added to the positive and negative sequence inner current loops. Simulation and experiments results are performed to verify the effectiveness and superiority of the proposed methodology on voltage support.

Amorphous solid dispersion (ASD) is one of the most important strategies to improve the solubility and dissolution rate of poorly water-soluble drugs. As a widely used technique to prepare ASDs, hot-melt extrusion (HME) provides various benefits, including a solvent-free process, continuous manufacturing, and efficient mixing compared to solvent-based methods, such as spray drying. Energy input, consisting of thermal and specific mechanical energy, should be carefully controlled during the HME process to prevent chemical degradation and residual crystallinity. However, a conventional ASD development process uses a trial-and-error approach, which is laborious and time-consuming. In this study, we have successfully built multiple machine learning (ML) models to predict the amorphization of crystalline drug formulations and the chemical stability of subsequent ASDs prepared by the HME process. We utilized 760 formulations containing 49 active pharmaceutical ingredients (APIs) and multiple types of excipients. By evaluating the built ML models, we found that ECFP-LightGBM was the best model to predict amorphization with an accuracy of 92.8%. Furthermore, ECFP-XGBoost was the best in estimating chemical stability with an accuracy of 96.0%. In addition, the feature importance analyses based on SHapley Additive exPlanations (SHAP) and information gain (IG) revealed that several processing parameters and material attributes (i.e., drug loading, polymer ratio, drug's Extended-connectivity fingerprints (ECFP) fingerprints, and polymer's properties) are critical for achieving accurate predictions for the selected models. Moreover, important API's substructures related to amorphization and chemical stability were determined, and the results are largely consistent with the literature. In conclusion, we established the ML models to predict formation of chemically stable ASDs and identify the critical attributes during HME processing. Importantly, the developed ML methodology has the potential to facilitate the product development of ASDs manufactured by HME with a much reduced human workload.

Wireless communications aided by unmanned aerial vehicles (UAVs) have emerged as a promising technology for serving numerous terminal devices of the Internet of Things (IoT). To support such a broad range of emerging applications, millimeter-wave (mm-Wave) is an effective solution because it has the characteristics of high transmission efficiency, wide frequency band, and very low latency. Meanwhile, a UAV-assisted communication system also requires a global positioning system (GPS) reception (1.575 GHz) and ground control unit (2.4 GHz). This paper investigates a highly integrated multi-band antenna design to operate at the microwave (μ-Wave) and mm-Wave bands for robust and flexible UAV communications. Each antenna operating band utilizes the technology of a substrate-integrated waveguide (SIW) cavity-backed antenna. By strategically engineering an mm-Wave antenna into a 2.4 GHz band antenna to share the aperture, besides showing high aperture reuse efficiency with a large frequency ratio, the proposed antenna has also maintained the geometrical advantages of compactness, planar configuration, and ease of integration with other circuits. Furthermore, the proposed antenna can provide flexible radiation patterns with upper/lower-space signal coverage according to the mission demands of the UAV. Experimental results of the prototype show that the antenna can fully cover the 1.575 GHz, 2.4 GHz, and fifth-generation (5G) mm-Wave bands of n257, n258, and n261. Multiple bands and stable radiation characteristics indicate the proposed antenna's suitability for deploying a UAV-supported wireless network.

Multi-label feature selection can effectively resolve the challenges of high or even ultra-high dimensionality in multi-label data. However, most existing multi-label feature selection algorithms can only handle a single data type, assume all labels are equally significant and utilize heuristic search strategies, which results in inefficient and relatively unsatisfactory classification accuracy. In view of the above shortcomings, this paper proposes a new multi-label feature selection algorithm that effectively resolves existing algorithms' issues through three innovative procedures. First, a new similarity relation metric is proposed to deal with hybrid data types effectively. Second, a label enhancement algorithm is designed to enhance and transform the logical labels into a label distribution by fully considering the analytic hierarchy process (AHP) embedded with label correlation, which can automatically identify the significance of different labels. Third, a feature weighting evaluation is redesigned in the feature selection process to obtain the optimal feature subset through feature ranking directly. Under these proposed procedures, multi-label feature selection can effectively utilize the abundant semantic information of the label significance and can significantly improve the operating accuracy and efficiency simultaneously. Comparative experiments are conducted on 20 real multi-label datasets with seven state-of-the-art multi-label feature selection algorithms. Experimental results show that the proposed multi-label feature selection algorithm in this paper is about 5–10% better than the algorithms in 80% of the compared datasets.

Distributed event-triggered secondary control in microgrids have been widely investigated to improve system efficiency. But most of them are based on consecutive triggering condition monitor, which would in turn increase the computation burden of the system. To this end, this paper presents distributed self-triggered algorithmic solutions to the frequency restoration control and active power sharing control of islanded microgrids. Different from event-triggered control schemes, in our self-triggered solutions, each distributed generator is equipped with a local algorithm that enables it to pre-compute the next triggering time instant according to the states at previous one. Our starting point is to design an triggering condition with a novel estimate error. Then, the next triggering time instant is determined by solving a quadratic equation established based on the triggering condition, rather than monitoring the triggering condition consecutively. Theoretical analysis and simulation results show that the proposed distributed self-triggered secondary controllers can highly reduce the communication and computation cost simultaneously.

Power quality disturbances (PQDs) have adverse impacts on safe operation and reliability of modern integrated power system so it is of great necessity to identify them. Existence of missing measurement data hinders accurate identification of potential PQDs and the inevitable discrepancy after data recovery vitiates the current detection methods. Besides, the related research is lacked. In this study, a novel unified framework of Wasserstein adversarial learning (WAL) is proposed on identifying PQDs with incomplete data for the first time. It consists of Wasserstein adversarial imputation (WAI) and Wasserstein adversarial domain adaptation (WADA). WAI minimizes the improved Wasserstein distance between the data distributions of observed and generated PQD parts to impute missing values. During this process, PQD characteristics can be well recovered. Then, WADA leverages the Wasserstein domain discrepancy between the feature distributions of source labeled complete and target unlabeled imputed PQDs to capture domain-invariant features. Thus, labels of target imputed PQDs can be predicted accurately. Experimental verification demonstrates that the proposed WAI and WADA outperform other typical methods with better imputation results and higher classification accuracy. Constrained Wasserstein loss empowers the proposed deep learning models with excellent convergence and gradient stability.

Additive Kernel SVM has been extensively used in many applications, including human activity detection and pedestrian detection. Since training an additive kernel SVM model is very time-consuming, which is not scalable to large-scale datasets, many efficient solutions have been developed in the past few years. However, most of the existing methods normally fail to achieve one of these three important conditions which are (1) low classification error, (2) low memory space, and (3) low training time. In order to simultaneously fulfill these three conditions, we develop the new piecewise-linear approximate measure (PLAME) for additive kernels. By incorporating PLAME with the well-known dual coordinate descent method, we theoretically show that this approach can achieve the above three conditions. Experimental results on twelve real datasets show that our approach can achieve the best trade-off between the accuracy, memory space, and training time compared with different types of state-of-the-art methods.

The concern for privacy and scalability has motivated a paradigm shift to decentralized energy management methods in microgrids. The absence of a central authority brings significant challenges to promote trusted collaboration and avoid collusion. To address these issues, this paper proposes a blockchain-empowered microgrid energy management framework, which adopts a novel consensus-based algorithm with a collusion prevention mechanism. Aiming at social welfare maximization, the energy management problem is formulated into a convex and decomposable form, which can be solved in a decentralized manner. To prevent the collusion between malicious agents, we propose a random information transmission mechanism empowered by the blockchain smart contract to replace the time-invariant communication topology. The consensus-based algorithm is extended to obtain the optimal solution of the energy management problem on the random and time-varying communication topology. We theoretically proved that the proposed algorithm converges to the global optimal solution with a probability of 1, without violating the physical constraints of individual agents. The effectiveness of the proposed method was validated by multiple experiments, both within the simulation environment and on a hardware system.

Imbalanced data is a major challenge in classification tasks. Most classification algorithms tend to be biased toward the samples in the majority class but fail to classify the samples in the minority class. Recently, ensemble learning, as a promising method, has been rapidly developed in solving highly imbalanced classification. However, the design of the base classifier for the ensemble is still an open question because the optimization problem of the base classifier is gradientless. In this study, the evolutionary algorithm (EA) technique is adopted to solve a wide range of optimization design problems in highly imbalanced classification without gradient information. A novel EA-based classifier optimization design method is proposed to optimize the design of multiple base classifiers automatically for the ensemble. In particular, an EA method with a neural network (NN) as the base classifier termed NN ensemble with EA (NNEAE) is developed for highly imbalanced classification. To verify the performance of NNEAE, extensive experiments are designed for testing. Results illustrate that NNEAE outperforms other compared methods.

Secondary frequency control is one of the most effective measures to ensure the stable operation of islanded microgrids (MGs). Most research on secondary frequency regulation has only focused on realizing steady-state operation objectives, that is, frequency restoration and power sharing. However, improving the dynamic performance of secondary frequency control is of great importance, especially in synchronous distributed energy resources. These synchronous units can introduce undesired oscillation modes, which may cause the instability conditions of MGs. To improve the dynamic performance of islanded MGs, a membership-function (MF) -based control strategy is proposed. The proposed strategy can trade-off between transient frequency regulation and frequency error elimination using the MF values calculated by the time-stamped synchronized measurements of distribution-level phasor measurement units. Besides, considering the time-varying communication delays in secondary frequency control loops, an adaptive delay compensator is proposed. The weights of the proposed compensator are updated by real-time delay measurements to compensate for the phase lag of control signals. Therefore, the adverse effect of communication delays on secondary frequency control is weakened effectively. Numerical simulations on an IEEE 34-bus system and a typical 40-bus islanded MG system demonstrate the advantages of the proposed method in the secondary frequency regulation of islanded MGs.

Spectral Clustering (SC) is an effective clustering method for its excellent performance in partitioning non-linearly distributed data. On the other hand, Ensemble Clustering (EC), a different clustering technology, can promote cluster quality by ensembling the results of base clusterings. In this work, we concentrate on an EC framework that utilizes SC as the base method. Nevertheless, SC suffers from scalability due to its high computational complexity in constructing the Laplacian graph and computing the corresponding eigendecomposition. In the past decades, many efforts have been made to it. However, SC suffers from the scalability issue in processing extensive data, especially in web-scale scenarios. Additionally, EC requires multiple clustering results as the ensemble bases, which further aggravates resource consumption. To address this issue, LiteWSEC, a simple yet efficient Lightweight Framework for Web-scale Spectral Ensemble Clustering, is proposed to cluster web-scale data with limited resource requirements. It adopts the Web-scale Spectral Clustering (WSC) as the base method, which has minimal space overhead without computing overall embedding explicitly. LiteWSEC is highly flexible in the memory requirement, which is adaptive to the available resource. It can partition web-scale data (e.g., $n = 8,000~k$ ) in an resource-limited host (e.g., memory is restricted to 1 GB). Experiments on real-world, large-scale, and web-scale datasets demonstrate both the efficiency and effectiveness of LiteWSEC over state-of-the-art SC and EC methods.

Integrated sensing and communication (ISAC), which enables the joint radar sensing and data communications, shows its great potential in many intelligent applications. In this paper, we investigate the unmanned aerial vehicle (UAV) aided ISAC with mobile edge computing (MEC), where the ISAC device deployed on the UAV senses multiple targets with the sensing scheduling and offloads the radar sensing data to the edge-server to train a machine learning model for target recognition. The radar estimation information rate is utilized to measure the radar sensing performance. We aim to minimize a system-wise cost that includes both the UAV’s energy consumption and the data collecting time, while satisfying the requirements on both the model training error and the radar sensing performance. We formulate a joint optimization problem of the sensing scheduling, the number of time-slots, the sensing power, the communication power, and the UAV trajectory. Despite the strict non-convexity of the formulated problem, we propose an efficient algorithm for solving it. Our algorithm jointly leverages the vertical decomposition that exploits the layered structure of the formulated problem and the horizontal decomposition that utilizes the block coordinate descent (BCD) method. Numerical results are presented to validate the effectiveness of our proposed algorithms and show the performance gain of our proposed scheme.

A noise-shaping successive approximation register (NS-SAR) ADC combines the merits of the $\Delta $ - $\Sigma $ and SAR ADC, transforming it into an emerging ADC architecture to reach high resolution with good power efficiency. The single-channel NS-SAR with high resolution, however, suffers from bandwidth (BW) limitations. The time-interleaved (TI) NS-SAR mitigates the speed bottleneck but faces challenges in obtaining high resolution and BW simultaneously due to the lack of a sharp noise transfer function (NTF). This article presents a calibration-free two-channel TI-NS-SAR with an aggressive second-order NTF for high resolution. Based on a one-time error feedback (FB) at midway, we propose a second-order error-feedforward (FF) to enhance the noise-shaping (NS) effect further meanwhile avoiding the excessive NTF peaking and dynamic range (DR) loss. A dynamic residue amplifier shared between two channels lowers the offset, which reduces the redundant bit to only one bit, thus improving the efficiency of SAR conversion. Fabricated in a 28 nm CMOS with 1 V supply, the prototype achieves 73.2 dB-signal-to-noise-and-distortion-ratio (SNDR) over 30 MHz-BW when operating at 330 MHz. It consumes 3.07 mW and exhibits a Schreier FoM (FoMs) of 173.1 dB.

This paper presents a cross-coupled CMOS rectifier for ambient RF energy harvesting, which utilizes an advanced topology amalgamation technique to significantly improve the rectifier’s power conversion efficiency (PCE) and power dynamic range (PDR), also known as the high-PCE range. Our novel technique involves adaptively self-biasing the rectifying PMOS in the last stage of the proposed rectifier, which deactivates the cross-coupled counterpart and enables the subsequent diode-based rectifier circuitry to operate efficiently during high-power operation and extends the high-PCE range by mitigating the leakage current. Fabricated in a 65-nm CMOS process, our proposed rectifier scores a wide PDR of 21 dB at 900 MHz and 15 dB at 1.8 GHz with a peak PCE of 79.77% and 51.3%, respectively, under a 100-kΩ load. Moreover, we perform a Monte Carlo simulation to showcase the impact of transistor variations. Our proposed rectifier, which offers the widest PDR compared to recently-reported designs operating at a similar frequency, is highly adaptable to the varying RF environment, enabling efficient and reliable energy harvesting for Internet of Things devices.

This brief presents a compact and power-efficient full ring-oscillator (RO)-based cascaded fractional-N PLL. The proposed cascaded PLL consists of a RO-based DLL and type-II PLL as the first and second stages, respectively. The first stage serves as a frequency multiplier that increases the operating frequency of the delta-sigma modulator (DSM) in the second stage, thereby suppressing its quantization noise. A burst-mode sampling (BMS) scheme is introduced to improve the phase noise (PN) of the frequency multiplier and achieves a PN multiplication factor removal. Implemented in a 28nm CMOS technology, the PLL prototype occupies a 0.016 mm2 active area, achieving a 686 fs integrated rms jitter from 10KHz to 40MHz at a 4 GHz output frequency; while consuming 10.21mW with -233.6 dB FoMjitter. The measured fractional and reference spurs are -59.8 dBc and -54.5 dBc, respectively.

This work exploits, for the first time, the well-proximity effect to develop a sub-0.5 V voltage reference with a high power-supply rejection ratio (PSRR) and a compact area. The layout-dependent effect (LDE) is deemed to affect the matching and characteristics of analog circuits in the deep-submicron CMOS process. Here we explore the LDE effect in designing analog circuits, by exemplifying it with a CMOS voltage reference. Validated in 65-nm CMOS, the voltage reference occupies 8,400 μm2 and outputs a reference of 107.2 mV at a VDD of 0.4 V, with a power of 56.7 nW. The temperature coefficient is 79.4 ppm/∘C across.20 to 80 ∘C (average of 12 samples) after a two-point batch trimming and scores a high PSRR of.66.5 dB. The standard deviation of 12 chips is 2.6 mV, evincing the robustness of the voltage reference exploiting the LDE.

This paper presents a hybrid boost DC-DC converter for driving piezoelectric actuators that overcomes the frequency limitation of traditional boost converters and the 3-level, double-step-down (DSD) topology when dealing with high conversion ratios (VCR). The proposed converter combines the benefits of conventional hybrid Fibonacci and Dickson switchedcapacitor (SC) converters to achieve high VCR with fewer capacitors and lower-voltage-rating switches. Two inductors are used to reduce DC resistance (DCR) loss and achieve charge and current balance. The converter operates as a 14× boost circuit with a duty ratio (D) as high as 0.5. With the SC stages reducing the voltage stress of inductors, only two small inductors are required to complete voltage conversion, reducing the inductor current ripple. This work was fabricated in the 180nm 1P6M BCD process. It can deliver a maximum load current of 2mA to an output voltage of 40-70V with a standard battery input voltage of 2.5-5V. The measurement results show a 54.8% peak efficiency and 2.312mW/mm3 power density, which are higher than the commercial products and the prior arts.

This work presents a 14-bit 500 MS/s single-channel pipelined-successive-approximation-register (SAR) analog-to-digital converter (ADC) with an adaptively biased floating inverter amplifier (AB-FIA) as the residue amplifier (RA) and a hybrid reference ripple mitigation (H-RRM) technique to relax the power and area burden on the reference stabilization. Leveraging the adaptively biased architecture in the last stage FIA, the speed and open-loop gain of the proposed two-stage FIA are enhanced compared with the conventional cascode counterpart. Besides, the impact of the reference error on the pipelined-SAR conversion accuracy is alleviated by hybridizing the improved reference ripple cancellation (RRC), reference ripple neutralization (RRN), and reference buffer (RBUF). The improved RRC removes the potential noise coupled from the floating capacitor to counter the decision error during the sub-SAR conversion in the first stage. Meanwhile, the RRN facilitates a rapid reference recovery. These acts constitute the H-RRM, which assists a high-speed and high-resolution pipelined-SAR process with a relaxed integrated reference RBUF with low-power and compact area. The prototype ADC was fabricated in a 28 nm CMOS process; it consumes 6.34 mW total power at 500 MS/s, including 2.4 mW dynamic power of RBUF. It occupies an active area of 0.018 mm $^{2}$ , which the ADC core area of 0.0168 mm $^{2}$ and the area of RBUF with a 2.3 pF decoupling capacitor is 0.00105 mm $^{2}$ . The measured signal to noise and distortion ratio (SNDR) and spurious free dynamic range (SFDR) are 64.2 dB and 80.55 dB with a Nyquist input, respectively, leading to a 170.2 dB Schreier figure-of-merit (FoM) (FoM $_{\mathrm{S}}$ ) and 9.6 fJ/conversion-step Walden FoM (FoM $_{\mathrm{W}}$ ).

This brief presents a switched-capacitor network (SCN)-based bandgap voltage reference (BGR) with a leakage current injection technique for curvature correction, improving performance in terms of temperature coefficient (TC). A deep N-well NMOS transistor biased with a complementary to absolute temperature (CTAT) voltage generates a leakage current with a concave upward curvature. Subsequently, we inject this current into the SCN network during the holding state for curvature correction. The injected leakage current changes the sampled CTAT voltage, achieving TC compensation without consuming too much additional power. The proposed BGR, fabricated in 65nm CMOS, occupies an active area of 0.0442 mm2. Measurements from 5 chips show that the achieved reference voltage is 432.4 mV under a 0.5 V supply. The average TC is 22.5ppm/∘C over a temperature range of -40∘C to 120∘C, significantly improving on TC while the power consumption is only 29 nW, which is comparable to previous SCN BGRs. This validates the effectiveness of the proposed leakage current injection technique for curvature correction.