A document by Mohaimenul Azam Khan Raiaan. Click on the document to view its contents.

This paper introduces a novel adaptive neural network-based sliding mode controller for trajectory tracking of a quadrotor UAV under unknown parameters such as inertia and mass. Unknown external disturbances and nonlinear aerodynamic forces are also considered. Due to its complex dynamics, the quadrotors need a dual controller for outer and inner loop control. For the commercial UAV, the inner loop controller, which ensures the attitude control, is usually assumed to be well developed and unmodifiable. Under this assumption, this paper focuses on the position control of a quadrotor UAV. A new position controller will be then designed and used to compute the appropriate input commands for the attitude controller to achieve the trajectory tracking task. The proposed method combines both Back-Propagation Neural Network (BPNN) scheme and sliding mode control to eliminate the effects of exogenous disturbances and model uncertainties. To illustrate the efficiency of the proposed controller, a comparative analysis is performed with different methods through experimental results.

Quantum-approximate optimization algorithm (QAOA) is promising   in  Noisy Intermediate-Scale Quantum (NISQ) computers with applications for  NP-hard combinatorial optimization problems. It is recently utilized for NP-hard maximum-likelihood (ML) detection problem with fundamental challenges of optimization, simulation and performance analysis for  n x n multiple-input multiple output (MIMO) systems with large n. QAOA is recently applied  by Farhi et al. on infinite size limit  of Sherrington-Kirkpatrick (SK) model with a cost model including only quadratic terms.   In this article, we extend application of QAOA on SK model by including also linear terms  and then realize SK modeling of massive MIMO ML detection by ensuring independence from specific problem instance and size  n while preserving computational complexity of O(16p) designed by Farhi et al. We provide both optimized and extrapolated angles for p in the set [1, 14] and signal-to-noise (SNR) < 12  dB achieving near-optimum ML performance for 25 x 25 MIMO system with p >= 4  in extensive simulations where  236500 different QAOA circuits are simulated.  We present two conjectures about the concentration properties of QAOA  and  its near-optimum performance for massive MIMO systems with large sizes covering n  < 300 promising significant performance for next generation massive MIMO systems.

Information technology organizations and companies are seeking greener alternatives to traditional terrestrial data centers to mitigate global warming and reduce carbon emissions. Currently, terrestrial data centers consume a significant amount of energy, estimated at about 1.5% of worldwide electricity use. Furthermore, the increasing demand for data-intensive applications is expected to raise energy consumption, making it crucial to consider sustainable computing paradigms. In this study, we propose a data center-enabled High Altitude Platform (HAP) system, where a flying data center supports the operation of terrestrial data centers. We conduct a detailed analytical study to assess the energy benefits and communication requirements of this approach. Our findings demonstrate that a data center-enabled HAP is more energy-efficient than a traditional terrestrial data center, owing to the naturally low temperature in the stratosphere and the ability to harvest solar energy. Adopting a data center-HAP can save up to 14% of energy requirements while overcoming the offloading outage problem and the associated delay resulting from server distribution. Our study highlights the potential of a data center-enabled HAP system as a sustainable computing solution to meet the growing energy demands and reduce carbon footprint.

The feasibility study of deep learning (DL) approaches for reliable, flexible, and high throughput wireless physical layer (PHY) has received significant interest from academia and industry. Channel estimation (CE) is one of the critical signal processing units of the receiver PHY. Recent DLbased CE approaches have outperformed state-of-the-art statistical approaches such as least-square-based CE (LS) and linear minimum mean square error-based CE (LMMSE), requiring hardware-friendly arithmetic computations than LMMSE. The first contribution is optimizing the existing state-of-the-art CE algorithms for realization on system-on-chip (SoC) via hardwaresoftware co-design and fixed point analysis, followed by performance comparison for various signal-to-noise ratios (SNR) and wireless channels. We highlight the high complexity, execution time, and power consumption of existing DL-based CE and LMMSE approaches, along with the poor performance of the LS. We develop a novel compute-efficient LS-augmented interpolated deep neural network (LSiDNN) based CE algorithm and realize it on SoC. We demonstrate substantial savings in complexity and execution time without significant degradation in functional accuracy. Specifically, the proposed LSiDNN approach offers 88-90% lower execution time and 38-85% lower resources than state-of-the-art DL-based CE. LSiDNN significantly outperforms LMMSE, and the gain improves with increased mobility. It also offers 75% lower execution time and 90-94% lower resource utilization than LMMSE

This paper investigates the implementation of a physics informed neural network (PINN) using a hardware platform and hardware description language (HDL). The main objective is to investigate how real-time physics informed neural networks (PINNs) can be utilized on edge devices for digital twin applications. To make the best use of limited resources in these edge devices, they adopt a piece-wise non-linear approximation in the design of the PINN. To validate the effectiveness of their approach, they implement the PINN on a field programmable gate array (FPGA) to solve a non-linear ordinary differential equation (ODE) related to the Reynolds equation. According to the experimental findings, the hardware-based PINN achieves an impressive 95% accuracy compared to the actual solution.

The antennas and metasurfaces designed for biomedical applications operate primarily in the far-field rather than the near-field. Working in the near-field is considerably more intricate in both design and theory. Furthermore, the proposed methodologies are unsuitable for seamless integration with a radar system that seeks to offer an efficient and streamlined approach to achieve heightened sensing. Our research highlights the potential of metasurfaces in redefining radar-based sensing. By integrating metasurface principles specifically designed for near-field focusing into radar design, we unlock new possibilities for high-resolution biomedical sensing with real-time capabilities.

TinyRCE is a hyperspherical classifier aimed at Continual Learning On-Tiny-Devices, a challenging task in which a Machine Learning model is required to learn from continuous streams of data while being directly installed on a (tiny) device with limited computational resources. The classifier has so far been applied to several use cases, including Human Activity Recognition, Ball Bearing Anomaly Classification, Keyword Spotting and Image Classification. The proposed work in this paper focuses on the reproducibility of TinyRCE’s experimental results already published on other papers. This to prove that all the published results are quantitatively reproducible. All the experiments have been executed on two independent computing machines to profile the impact on accuracy of the computations. As the outcomes are matching, the experimental reproducibility of TinyRCE’s accuracy over all the use cases has been positively verified.

Phase identification is the problem of determining the phase connection of loads in a power distribution system. In modern times, utility operators will generally accomplish this using smart meter data that usually requires some form of feature engineering to achieve practical phase identification using data-driven methods. Feature engineering is essential for voltage magnitude data containing noise, seasonality, and trend. Crucial components of a feature engineering pipeline are presented to perform linear denoising with Singular Value Decomposition, filtering of the denoised data to remove the seasonality and trend, and fuse multiple meter channels. We use the results of the feature engineering to perform phase label correction, a subproblem of phase identification. The authors generate a synthetic dataset from the meshed IEEE 342-Node test feeder circuit with the 2021 Electric Reliability Council of Texas load profiles to evaluate techniques. Our results show that denoising is quite effective for improving phase identification accuracy in the presence of measurement noise. We present some new insight into filtering voltage measurement data to improve accuracy, as well as eliminate the need to determine salient frequencies. We also present the application of a data channel fusion technique that is novel to the phase identification literature. This technique enhances phase identification in cases where there are both wye and delta-connected loads present.

Deep Neural Network (DNN) belongs to an important class of machine learning algorithms generally used to classify digital data in the form of image and speech recognition. The computational complexity of a DNN-based image classifier is higher than traditional fully connected (FC) feed-forward NNs. Therefore, dedicated cloud servers and Graphical Processor Units (GPU) are utilized to achieve high-speed and large-capacity computation tasks in machine vision systems. However, a growing demand exists for real-time processing of complex machine-learning tasks on embedded systems. As FC layers consume the highest fraction of computational power and memory footprint, innovating novel power-efficient and low-footprint NN architecture for embedded systems is crucial. A novel design strategy and algorithms are proposed in this article, where a power-efficient FC DNN is implemented using a pipelined and parallel Fast Fourier Transform (FFT) on a circular projection-based architecture. The footprint of the DNN is further reduced using a folded FFT network. The proposed algorithm is tested using two benchmark training set examples, the “MNIST database of handwritten digits” and the “CIFAR-10 database”. In both cases, we achieved > 90% accuracy, while the power consumption of the network is 37% less than the traditional FFT architecture-based DNNs.

This manuscript is Under Review of IEEE Transactions on Mobile Computing.

Diagnosing the type of stroke using a quantitative method of microwave imaging

Data augmentation represents an opportunity for Artificial Intelligence applications, as it aims at creating new synthetic data based on an existing baseline. In this paper, we present a new evaluation framework for Generative Adversarial Networks (GANs), a data augmentation technique, in multivariate data classification contexts. The goal is not limited to assess the performance variations obtained through GANs, but also to inspect results with explainable AI (XAI) tools, understanding how GANs work and, finally, exploiting them to discover new knowledge. To this aim, we adopt the Logic Learning Machine for performance assessment and rule extraction, and introduce a new measure of rule similarity to compare different artificial datasets. We apply the methodology on two case studies, activity recognition and physical fatigue detection, confirming that GANs can help in overcoming limitations of original datasets and lead to new discoveries.

The Internet of Bio-Nano Things (IoBNT) is a novel framework that has the potential to enable transformative applications in healthcare and nano-medicine. It consists of artificial or natural tiny devices, so-called Bio Nano Things (BNTs), that can be placed in the human body to carry out specific tasks (e.g., sensing) and are connected to the internet. However, due to their small size their computation capabilities are limited, which restricts their ability to process data and make decision directly in the human body. Thus, we address this issue and propose a novel nano-scale computing architecture that performs matrix multiplications, which is one of the most important operations in signal processing and machine learning. The computation principle is based on diffusion-based propagation between connected compartments and chemical reactions within some compartments. The weights of the matrix can be set independently through adjusting the volume of the compartments. We present a stochastic and a dynamical model of the proposed structure. The stochastic model provides an analytical solution for the input-output relation in the steady state, assuming slow reaction rates. The dynamical model provides important insights into the systems temporal dynamics. Finally, micro- and mesoscopic simulations verify the proposed approach.

Many biological processes rely on endogenous electric fields (EFs), including tissue regeneration, cell development, wound healing and cancer metastasis. Mimicking these biological EFs by applying external direct current stimulation (DCS) is therefore the key to many new therapeutic strategies. During DCS, the charge transfer from electrode to tissue relies on a combination of reversible and irreversible electrochemical processes, which may generate toxic or bio-altering substances, including metal ions and reactive oxygen species (ROS). Poly(3,4-ethylenedioxythophene) (PEDOT) based electrodes are emerging as suitable candidates for DCS to improve the biocompatibility compared to metals. This work addresses whether PEDOT electrodes can be tailored to favor reversible biocompatible charge injection. To this end, different PEDOT formulations and their respective back electrodes are studied using cyclic voltammetry, chronopotentiometry, and direct measurements of H2O2 and O2. This combination of electrochemical methods sheds light on the time dynamics of reversible and irreversible charge transfer and the relationship between capacitance and ROS generation. The results presented here show that although all electrode materials investigated generate ROS, the onset of ROS can be delayed by increasing the electrode’s capacitance, which has implications for future bioelectronic devices that allow longer reversibly driven pulse durations during DCS.

The purpose of this project is to mitigate and solve issues regarding health care services such as rehabilitation and specific health guidance while alleviating spatio-temporal, economic, and cognitive constraints by establishing remote technology foundation. There are four themes in this project titled “Multimodal XR-AI (XR powered by AI) platform development for tele-rehabilitation and the reciprocal care coupling with health guidance.” In theme #1, we have been developing MR3 (Multi-Modal Mixed Reality for Remote Rehab) devices consisting of Wear and Mannequin for supporting detailed assessments of customers’ physical functions and tactile interaction respectively. The central issue of theme #2 is to support intrinsic motivation for rehabilitation and exercise training through XR technologies as in virtual co-embodiment and hand redirection. In addition, we have also been investigating how to deal with one-to-many (a small number of providers) and zero-to-many (no providers) situations. Theme #3 has aimed on establishing AI technology foundation for creating, monitoring progress of, and updating tele-rehab programs mainly for the upper limb. Systems for always-on monitoring during daily life and work developed in theme #4 is expected to be served as a common foundation for various tele-healthcare services.

Atrial fibrillation (AF) is a common cardiac arrhythmia causing severe complications if left untreated. Due to its sporadic nature, early detection often requires longitudinal ambulatory electrocardiogram (ECG) screening. Recently, deep learning (DL) has gained prominence in analysing long-term ECG and automating AF detection. However, like any medical classification problem, obtaining diverse labelled ECG data for DL model training is expensive and time-consuming. This paper proposes a semi-supervised learning (SSL) based AF classification model employing a variational auto-encoder (VAE). It leverages varying amounts of labelled and unlabelled ECG data to optimise the AF detection performance on ambulatory ECG. As ambulatory contexts under free-living conditions influence ECG recordings, we incorporate context via accelerometry data and experiment with its influence on model performance. Our proposed SSL model was trained on ECG data from 72,003 unique patients and can classify between sinus rhythms, AF, and other arrhythmias. Experimental results on our unseen test dataset and the publicly available CACHET-CADB dataset clearly demonstrate the model’s generalisability, achieving an accuracy of over 91% with just 20% of the training set being labelled. With extensive experiments, our study exhibits the ability of SSL to improve AF detection from ambulatory ECG using small amounts of labelled data.

Elliptic curve scalar multiplication (ECSM) stands as a crucial sub-block in elliptic curve cryptography, which represents the most widely-used pre-quantum public key cryptography. Hardware constructions of cryptographic systems utilizing ECSM have been subject to permanent or transient errors. In cryptographic systems, it is important to validate the correctness of the underlying computation performed on hardware or software to identify such errors. In this paper, we present new fault detection schemes in window method scalar multiplication, which, to the best of our knowledge, has not been previously investigated. Our approach involves introducing refined algorithms and implementations that can effectively counter both permanent and transient errors. We assess this by simulating a fault model, ensuring that the evaluations conducted reflect the obtained results. As a result, we achieve a significantly extensive coverage of errors. Lastly, we benchmark our proposed error detection scheme on ARMv8 and FPGA to demonstrate the implementation and resource overhead. On Cortex-A72 processors, we maintain a clock cycle overhead of under 3%. Additionally, when implementing our error detection method on different FPGAs including Zynq Ultrascale+, Artix-7, and Kintex Ultrascale+, we achieve comparable throughput while introducing a mere 2% increase in area compared to the original hardware implementations.