A document by Hao Zhou . Click on the document to view its contents.

Fault-tolerant methods have been recognized for improving the reliability of multi-phase bidirectional DC-DC converters. Few fault-tolerant methods for current-fed dual active bridge (CF-DAB) converters have been developed because it becomes more challenging due to the asymmetrical structure between primary and secondary sides. In this paper, the behaviors of the CF-DAB converter are comprehensively analyzed when an open-circuit fault occurs in different fault scenarios. Analysis and proposed fault-tolerant method are aimed to determine the open-circuit cases which can recover full conversion power after the fault occurs. Moreover, this paper also proposed an auto-balancing control method to prevent the converter from being out of control. The proposed balancing control method can be utilized for general multi-phase current-fed converters. A 6-kW prototype is built to verify the proposed method.

Continuous Hearth Rate (HR) monitoring based on photoplethysmography (PPG) sensors is a crucial feature of almost all wrist-worn devices. However, arm movements lead to the creation of Motion Artifacts (MA), affecting the performance of PPG-based HR tracking. This problem is commonly tackled by exploiting the recorded accelerometer data to correlate them with the PPG signal and eventually clean it. Thus, automatic fusion techniques based on Deep Learning (DL) algorithms have been proposed, but they are considered too large and complex to be deployed on wearable devices. The current work presents a novel and lightweight DL architecture, PULSE, comprised of temporal convolutions and feature-level multi-head cross-attention to improve sensor fusion’s effectiveness. Moreover, we propose a relation-based knowledge distillation mechanism to pass PULSE’s knowledge to a student network that utilizes modality-wise convolutions to replace the attention module and mimic the teacher’s performance with 5x fewer parameters. The teacher and student are evaluated on the most extensive available dataset, PPG-DaLiA, with PULSE reducing the mean absolute error by 8.2% compared to the best state-of-the-art model while simultaneously reducing the inference latency by 1.6x. The student model is further compressed using post-training quantization and deployed on two microcontrollers, demonstrating its suitability for real-time execution, having a close-to-state-of-the-art MAE of 4.81 BPM (+0.40 BPM), but a 10.9x lower memory footprint of 37.9 kB, and consuming 45.9x lower energy (0.577 mJ).

In this work, for the first time, we have demonstrated the feasibility of 1T-DRAMs as synaptic elements with MLC, large dynamic range, appreciable linearity, ultra-low energy, high endurance, and large integration density. Our results may provide incentive for its experimental realization.Â

The Cognitive Agents and Interaction Lab (CAIL) at the University of Dhaka has strategically developed a focused High-Performance Computing (HPC) facility, underpinning its niche in artificial intelligence (AI) research, with a special emphasis â\euro” but not limited to â\euro” deep and reinforcement learning within the context of multi-agent systems. This technical report outlines the HPC resources at CAIL, highlighting a number of advanced GPUs that drive our research forward. The facility, supported by multiple grants, exemplifies a model for small-scale, high-impact computational research environments. We detail how these resources empower our team to achieve significant strides in AI, despite the challenges of scale, and set a precedent for innovation in the field.

Coded aperture snapshot spectral imager (CASSI) can recover three-dimensional hyperspectral images (HSIs) from two-dimensional compressive measurements. Recently, deep unfolding approaches were shown impressive reconstruction performance among various algorithms. Existing deep unfolding methods usually employ linear projection methods to guide the iterative learning process. However, the linear projections do not include trainable parameters and ignore the essential characteristics of HSI. This paper proposes a novel learning-based deep estimation-enhancement unfolding (DEEU) framework to improve the HSI reconstruction. The deep estimation-enhancement (DEE) module is used to guide the iterative learning process of the network based on the prior information of the CASSI system, and then exploits the intrinsic features of the reconstructed HSI in both spectral and spatial dimensions. In addition, a multi-prior ensemble learning module is proposed to further improve the reconstruction performance without increasing runtime. As with most of deep unfolding methods, we plug a convolutional neural network as a denoiser in each stage of the DEEU framework, which finally forms the proposed DEEU-Net. Comprehensive experiments on both simulation and real datasets demonstrate that the effectiveness of our DEEU framework, and our DEEU-Net can achieve both high reconstruction quality and speed, outperforming the state-of-the-art methods.

The first thing that comes to everyoneâ\euro™s mind, when we talk about wireless communications is â\euroœThe Radioâ\euro?. Starting from the 1800â\euro™s till date the radio has undergone several iterations of development. This paper is focused to bring out the three major development stages that are Hardware Defined Radios, Software Defined Radios & Cognitive Radios. With the introduction of cognizance and decision-making algorithms in Cognitive Radios, the scope for incorporation of Artificial Intelligence and Machine Learning has become significant. Intelligent Radios can be considered as the fourth generation of Radio in times to come, which will incorporate Artificial Intelligence and Machine Learning at the core processing unit to improve the overall capability of radio in a congested, constrained and contested electromagnetic spectrum. The Intelligent Radios would be capable of generating hypothesis based on environmental conditions on its own and take intelligent decisions as per the situation reducing the human effort. As the world is moving constantly towards information age, it is clear that future age warfare will be network and electromagnetic spectrum centric. Hence, it is very important for the military to achieve freedom operation in electromagnetic spectrum space in operational scenario. Intelligent Radios will be the weapon of future age warfare by aiding in efficient spectrum management, powerful electromagnetic warfare in active mode and easy and quick decision making leading to timely escalation and achievement of tactical, strategic, and political objectives.Â

In this paper, a circularly polarized broadband slot antenna is designed for base station applications. The antenna is built on a square-shaped FR4 epoxy substrate that measures 60mm in length and 1mm in height with a relative permittivity of 4.4. The ground plane has a recessed slot, and the feeding mechanism employs an offset microstrip line. The slot enables circular polarization and optimal impedance matching, which are enhanced further by offset feeding and the addition of stubs to create a composite-shaped patch. With an impedance bandwidth of 106% (1.28 GHz-4.16 GHz) and an axial ratio bandwidth of 69% (1.71 GHz-3.51 GHz), the CP broadband slot antenna effectively covers popular mobile communication bands such as GSM 1800, GSM 1900, 3G/LTE 2100, 3G/LTE 2300, 4G/LTE 2500, and 2.4 GHz Wi-Fi bands. The maximum gain is simulated to be 3.5 dB.Â

In this paper, we consider strategies to label mental fatigue states during a prolonged visuospatial working memory exercise. Specifically, we look to address the need to accurately forecast these states with unobtrusive measures derived from cardiac electrical activity. We formulate this challenge as a multi-step-ahead, multivariate time-series forecasting problem, where we consider the effect of annotation variations for fatigue labels on model accuracy and reliability. We found that a sequence-to-sequence (Seq2Seq) LSTM architecture was effective in identifying fatigue representations using heart rate and its variability (HR/V). Using this architecture we analyzed a heuristic-based annotation approach for fatigue that relies on changing patterns of HR/V as a signal for state changes in the human. We contrast perception-, performance- and heuristic-based labels across the forecast horizon, the input vector horizon, and sample generalizability. Our findings indicate that, under controlled conditions, HR/V can serve as a viable neurocognitive index of fatigue, however, the impact of activity, time scale, and other perturbations demand further investigation to evaluate the diagnosticity and robustness of the proposed indices.

Explicit results are obtained using simple and exact methods for the joint queue-length distribution of the M/M/c queue with an arbitrary number of non-preemptive priority levels. This work is the first to provide explicit results for the joint probability generating function and joint probability mass function for a general number of priority levels. A fixed-point iteration is developed for the stationary balance equations, which enables direct computation of the joint queue-length distribution. A multi-variate probability generating function is also derived, from which the joint probability mass function can be computed by means of a multi-dimensional fast Fourier transform method.

Radar is widely used in aviation, meteorology and military fields, and radar pulse signal detection has become an indispensable and essential function of cognitive radio systems as well as electronic warfare systems. In this paper, we propose a deep learning based radar signal detection method. Firstly, we propose a detection method based on raw in-phase and quadrature (IQ) input, which utilizes a convolutional neural network (CNN) to automatically learn the features of radar pulse signals and noises, so as to accomplish the detection task. To further reduce the computational complexity, we also propose a hybrid detection method that combines compressed sensing (CS) and deep learning, which reduces the length of signal by compressed downsampling, and then feeds the compressed signal to the CNN for detection. Extensive simulation results show that our proposed IQ-based method outperforms the traditional short-time Fourier transform method as well as three existing deep learning-based detection methods in terms of probability of detection. Furthermore, our proposed IQ-CS-based method is able to achieve satisfactory detection performance with significantly reduced computational complexity.

In the grand tradition of Sherlock Holmes, the master of deduction, an investigative journey unfolds into the enigmatic world where the steadfast realm of German taxation meets the cryptic and ever-evolving landscape of blockchain technology. Much like the fog-laden streets of Victorian London, the intertwining of these two domains presents a puzzle wrapped in a mystery, shrouded in an enigma. Â The tale begins in the orderly corridors of German tax offices, where meticulous records and an unyielding adherence to regulation reign supreme. Into this world of precision and predictability, enters the maverick of modern finance: blockchain. With its promise of decentralization, transparency, and security, blockchain stands at the threshold, poised to revolutionize the very foundations of tax collection and administration. Â Yet, as Holmes himself would attest, the introduction of such a disruptive force into a system as established and rigid as the German tax regime is akin to introducing a wild card into a well-ordered deck. The potential for both unparalleled efficiency and unforeseen chaos exists in equal measure. Will blockchain technology prove to be the trusty Watson to Germanyâ\euro™s tax system, a loyal ally in the pursuit of order and justice? Or will it be akin to the elusive Moriarty, a force of disruption and uncertainty? Â In this scholarly endeavor, the labyrinth of regulations is traversed, potential harmonies and discords are explored, and deductions are made whether this alliance between blockchain and the German tax system is a match of mutual benefit or a partnership doomed to complexity and contention. Â Standing at the crossroads of tradition and innovation, donning the proverbial deerstalker, and arming with the magnifying glass of scrutiny, this intellectual pursuit embarks. For in the words of the illustrious Holmes, “The game is afoot!” Â

Christiaan Huygens, in the 1600s, discovered the synchronization of coupled pendulums while considering a specific closed system. This paper furthers Huygensâ\euro™s ambitions by considering the effects of synchronization time based on changes in string length in coupled simple pendulums designed on a moving platform. Two simple pendulums were connected through the medium of a wooden board which was then placed on cylindrical cans. String length and synchronization time seemed to display an inverse relationship based on trends of raw data. Explanations for other behaviors such as brief stops in motion and anti-phase versus in-phase synchronization are explained using various laws of Classical Mechanics and are modeled with linear approximations. The effect of synchronization arises from the medium between the pendulums and the various dampenings of the system. The findings presented generally show that synchronization can be optimized which is useful in various fields of study like the medical field where many diseases are caused by the synchronization of neurons. Finally, the equations of motion and energy are modeled with Lagrangian physics and Mathematica software. Possible extensions, like creating a model similar to the Kuramoto Model, and other applications of the problem are discussed.

A co-design approach is proposed in this communication for an efficient design of frequency scanning coupled-patch arrays (CPAs) with wide scanning range and high scanning rate. It combines characteristic mode analysis (CMA) and coupling matrix analysis in the design, where the former is employed for radiation pattern synthesis while the latter tunes the feeding reflection and mode excitation based on a coupled-resonator network model. With this systematic approach, it becomes efficient to control both beam scanning properties and feeding reflection. 3rd-order and 6th-order design examples are provided to validate the proposed method. The 6th-order CPA achieves a wide scanning range of 73◦ (from −45◦ to 28◦ across broadside) and a high scanning rate of 26:1◦/1%. Both scanning range and scanning rate are much higher than other reported CPAs.

Hate speech targets different social groups such as race and gender, and poses a significant threat to social harmony. Researchers are increasingly motivated to devise efficient techniques to improve automatic hate speech detection on social media platforms. However, current models are usually evaluated without considering hate speech targets and fail when the targets are unseen in the training data. In this study, we examine target (domain) shifts of hate speech and propose, \emph{Tad}, an adaptation framework for neural models that adopts domain-aware networks to improve cross-domain hate speech detection. Tad features a hate knowledge lexicon infusion network, a domain-specific network, and a weighting network. We demonstrate that incorporating Tad improves the performance of leading neural models in hate speech detection when tested on unseen domains. Specifically, Tad yields improvements of up to 8$.1\%$ and an average of $2.4\%$ in macro F1-scores. Moreover, we identify data quality and quantity as vital factors to address performance gaps between models tested on seen and unseen domains. Our results reveal that excessive knowledge infusion may result in a decrease in performance such as for \textsl{Religion}. In addition, we find trade-offs in cross-domain hate speech detection. For example, weighted loss for heavily imbalanced data generally improves performance.

We propose a solar sensor-based smart farm system to provide high monitoring quality while preserving sensor energy in the presence of adversarial attacks. Since solar sensors are attached to cows to monitor their health under varying weather conditions, ensuring that the system provides energy-adaptive, high-quality monitoring services is critical. Further, the smart farm system should be robust against diverse adversarial attacks that will disrupt its monitoring quality. We use deep reinforcement learning (DRL) to identify the optimal policy for maximizing monitoring quality and prolonging the systemâ\euro™s lifetime while maintaining sufficient energy. We introduce transfer learning (TR) into the DRL process to achieve fast learning by DRL without experiencing a cold start problem. In addition, we develop an uncertainty-aware anomaly data detection method to filter out deceptive data caused by adversarial attacks. Via extensive comparative performance analysis conducted in our experiments based on real datasets, we demonstrate the superior performance of TL-based DRL strategies over other competitive counterparts regarding system lifetime, monitoring quality, learning convergence time, and energy consumption.

Most single image super resolution (SISR) methods are developed on synthetic low resolution (LR) and  high resolution (HR) image pairs, which are simulated by a  predetermined degradation operation, such as bicubic  downsampling. However, these methods only learn the  inverse process of the predetermined operation, which fails to super resolve the real-world LR images, whose true  formulation deviates from the predetermined operation. To  address this, we propose a novel SR framework named  hardware-aware super-resolution (HASR) network that first extracts hardware information, particularly the camera degradation information. The LR images are then super resolved by integrating the extracted information. To  evaluate the performance of HASR network, we build a dataset named Real-Micron from real-world micron-scale  patterns. The paired LR and HR images are captured by  changing the objectives and registered using a developed  registration algorithm. Transfer learning is implemented  during the training of Real-Micron dataset due to the lack  of amount of data. Experiments demonstrate that by  integrating the degradation information, our proposed  network achieves state-of-the-art performance for the blind  SR task on both synthetic and real-world datasets. Impact Statementâ\euro” The proposed HASR method has  significant impact on various areas, such as enhancing the  accurate inspection of manufactured products for quality  control and enhancing the resolution of medical images to  enable more accurate diagnosis and healthcare. Current SR solutions neglect the uniqueness of each imaging system,  hence cannot produce accurate HR images across the  different systems. Taking advantage of the known hardware information, HASR can differentiate low?resolution images across different imaging systems and  produce HR images that are closer to the real-world  scenario. Given sufficient training images, the proposed  HASR method can overcome the physical optical limitation  and generate higher quality images. The proposed method  improves the overall performance by about 0.2 dB and 0.5  dB on the synthetic and the real-world datasets,  respectively. Â

This paper addresses the distributed optimal coordination problem for a network of heterogeneous agents. To the best of our knowledge, few works where the control design requires no information regarding agents dynamics can be found in the literature. Compared to these works in which agents are only allowed to be of relative degree one, this paper extends the results for networks that are mixtures of agents with different relative degrees. The main goal is that all agents' outputs globally converge to the minimizer of the global cost function. We introduce a novel distributed two-layer control policy to do so. The top layer searches for the minimizer and generates reference signals for the bottom layer. The top layer is identical for all agents. The bottom layer provides each agent with an adaptive controller, enabling the agent to track its associated reference signal. Local cost functions are assumed to be strictly convex and have smooth gradients. The proposed control policy is fully distributed, meaning agents only depend on their own and their neighbors' information to reach a consensus on the global minimizer. Numerical simulations validate the theoretical results.