Electrical and Computer Engineering

Permanent URI for this collection

https://hdl.handle.net/10315/35603

Browse

Now showing 1 - 20 of 52

Open Access
A Deep Reinforcement Learning-Based Caching Strategy for Internet of Things Networks with Transient Data
(2021-11-15) Nasehzadeh, Ali; Wang, Ping
Internet of Things (IoT) has been on a continuous rise in the past few years, and its potentials are now more apparent. Transient data generation and limited energy resources are the major bottlenecks of these networks. Besides, conventional quality of service measurements are still valid requirements that should be met, such as throughput, jitter, error rate, and delay or latency. An efficient caching policy can help meet the standard quality of service requirements while bypassing the specific limitations of IoT networks. Adopting deep reinforcement learning (DRL) algorithms enables us to develop an effective caching scheme without the need for any prior knowledge or contextual information such as popularity distributions or the lifetime of files. In this thesis, we propose DRL-based caching schemes that improve the cache hit rate and energy consumption of IoT networks compared to some well-known conventional caching schemes. We also propose a hierarchical caching architecture that enables parent nodes to receive the requests from multiple edge nodes and make caching decisions independently. The results of comprehensive experiments show that our proposed method outperforms the well-known conventional caching policies and an existing DRL-based method in cache hit and energy consumption rates by considerable margins. In the fifth chapter, we seek machine learning-based caching solutions for the cases in which the file popularity distribution is not static but changing over time. Taking the same system model into consideration, we propose a concept drift detection method based on clustering and cluster similarity measurements. The drift detection can trigger a process that leads the DRL agent to adapt to the new popularity distribution; we have based this process on transfer learning techniques. Transfer learning can help us leverage the existing knowledge in trained models and significantly speed up the training phase of our machine learning models.
Open Access
A Highly-Scalable DC-Coupled Direct-ADC Neural Recording Channel Architecture with Input-Adaptive Resolution
(2022-12-14) Sayedi, Sayedeh Mina; Kassiri, Hossein
This thesis presents the design, development, and characterization of a novel neural recording channel architecture with (a) quantization resolution that is adaptive to the input signal's level of activity, (b) fully-dynamic power consumption that is linearly proportional to the recording resolution, and (c) immunity to DC offset and drifts at the input. Our results demonstrate the proposed design's capability in conducting neural recording with near lossless input-adaptive data compression, leading to a significant reduction in the energy required for both recording and data transmission, hence allowing for a potential high scaling of the number of recording channels integrated on a single implanted microchip without the need to increase the power budget. The proposed channel with the implemented compression technique is implemented in a standard 130nm CMOS technology with overall power consumption of 7.6uW and active area of 92×92µm for the implemented digital-backend.
Open Access
A Mega-Hertz Micro Converter with Extended Soft Switching Operation for Photovoltaic (PV) Application
(2020-08-11) Ahmadiankalati, Samira; Lam, John Chi Wo
The increasing greenhouse effect and relative environmental pollution, along with limited fossil fuel has made it urgent to a transition towards renewable energy sources. The combined global capacity of Photovoltaic (PV) energy has increased considerably from 6.01 gigawatts (GW) to 505 GW from 2006 to 2018. A typical power configuration of a PV energy conversion system consists of a front-end DC/DC micro converter that is used to provide maximum power point tracking (MPPT), as well as to provide some step-up voltage conversion from the output of the PV solar panel. Different DC/DC PV power converters have been reported in literature. The existing DC/DC converters either require a high number of switches and magnetic components, suffer from high voltage stress over some circuit elements, or have low circuit efficiency and restricted switching frequency due to hard switching (hence large size passive components are required). In this thesis, a very high frequency DC/DC micro converter with inherent extended soft-switching operation is proposed for PV energy conversion systems. In the proposed topology, a boost-based MPPT circuit is integrated with a CL (capacitor-inductor) parallel resonant converter to form a single stage DC/DC PV micro converter. While the proposed converter has an auxiliary circuit to assist extended soft- switching operation, the inductor in the auxiliary circuit is coupled with the boost inductor so that the size and space of the overall circuit can be further reduced. A modified enhanced maximum power point tracking algorithm is also developed to work with the proposed step-up DC/DC micro converter. The theoretical analysis and the operating principles of the proposed converter will be discussed in this thesis. Simulation and experimental results on a MHz (Mega-Hertz) proof-of-concept hardware prototype are provided to highlight the performance of the proposed circuit.
Open Access
A Soft Switched, Single-Switch Electrolytic Capacitor-less Step-Up Converter for Photovoltaic Energy Application
(2019-11-22) Kanathipan, Kajanan; Lam, John Chi Wo
In this thesis, a single switch, electrolytic capacitor-less quasi-resonant step-up DC/DC converter is proposed for solar energy applications. The proposed converter is an improved coupled-magnetic based topology that requires only a single switch. By operating the input inductor of the proposed converter in continuous conduction mode (CCM) the required input capacitance is reduced and hence, allows for a small sized film capacitor to be used. In addition, the proposed circuit is able to achieve a large step-up gain while minimizing the ratio between the peak switch voltage and the circuit output voltage. Two different modes of operation are presented and discussed for the proposed circuit which can achieve a very large gain and a very small peak switch voltage to circuit output voltage ratio simultaneously. A maximum power point tracking controller is also developed to work with the proposed step-up DC/DC converter through the use of variable frequency control scheme. Simulation and experimental results on a proof-of-concept, 35V/380V, 100W, 100kHz, hardware prototype are provided for both modes of operation for fixed and varying light intensities to highlight the merits and performance of the proposed converter.
Open Access
A Spherical Visually-Guided Robot
(2020-11-13) Dey, Bir Bikram; Jenkin, Michael R.
Spherical robots provide a number of advantages over their wheeled counterparts, but they also presents a number of challenges and complexities. Chief among these are issues related to locomotive strategies and sensor placement and processing given the rolling nature of the device. Here we describe Dragon Ball, a visually tele-operated spherical robot. The Dragon Ball utilizes a combination of a geared wheel to move the center of mass of the vehicle coupled with a torque wheel to change direction. Wide angled cameras mounted on the robot's horizontal axis provide a 360 view of the space around the robot and are used to simulate a traditional pan tilt zoom camera mounted on the vehicle for visual tele-operation. The resulting vehicle is well suited for deployment in contaminated environments for which vehicle remediation is a key operational requirement.
Open Access
Adaptive Reconfiguration of Protection Relays to Accommodate Distributed
(2022-08-08) Ouda, Gehad; Srikantha,Pirathayini
The rise in the installation of distributed generation systems (DGs) requires the reconfiguration of the protection system of the distribution network (DN) in place. Increased integration of DGs in the DN can cause incorrect tripping and false non-tripping of the overcurrent (OC) protection relays in the DN as well as increased possibilities of faults due to the aging infrastructure of DN. Incorrect tripping is also known as sympathetic tripping and false non-tripping is also known as blinding operation of an OC relay. As such, a reconfiguration of the adaptive protective OC relays is desired to mitigate the issues with increased DG penetration in the DN. A new algorithm is proposed in this thesis for adaptive, distributed protection relay reconfiguration for the DN with DGs. Through theoretical studies based on potential games and practical simulations, the OC relays' settings are reconfigured to adapt to the DN's changes in the system as a result of DG integration.
Open Access
An 8-Channel Bidirectional Neurostimulator IC with a Highly-Linear High-Dynamic-Range ADC-Direct Architecture for Simultaneous Recording and Stimulation
(2021-03-08) Moeinfard, Tania; Kassiri, Hossein
This thesis presents the design, implementation, and validation of an 8-channel bidirectional neurostimulator IC with a highly-linear high-dynamic-range ADC-direct architecture for simultaneous recording and stimulation. Each channel hosts a novel highly-linear high-dynamic-range recording architecture capable of amplification and quantization of brains neural signals in the presence of large differential-mode and common-mode stimulation artifacts, as well as a fully-programmable 8-bit current-mode electrical stimulator. The architecture enables the possibility of a patient-specific stimulation therapy required for the next generation of implantable closed-loop neuro-stimulators used for treatment of various neurological disorders. The proposed design adopts an ADC-direct architecture employing a dual-loop SAR-assisted continuous-time delta-sigma ADC architecture for differential-mode stimulation artifacts and offset removal. The presented channel achieves a high input impedance (1.8 G at 1 kHz), 400 mV linear input signal range, 94 dB dynamic range, and consumes 4.6 W with a signal bandwidth of 5 kHz.
Open Access
An Efficient Machine Learning Software Architecture for Internet of Things
(2021-07-06) Chaudhary, Mahima; Litoiu, Marin
Internet of Things (IoT) software is becoming a critical infrastructure for many domains. In IoT, sensors monitor their environment and transfer readings to cloud, where Machine Learning (ML) provides insights to decision-makers. In the healthcare domain, the IoT software designers have to consider privacy, real-time performance and cost in addition to ML accuracy. We propose an architecture that decomposes the ML lifecycle into components for deployment on a two-tier cloud, edge-core. It enables IoT time-series data to be consumed by ML models on edge-core infrastructure, with pipeline elements deployed on any tier, dynamically. The architecture feasibility and ML accuracy are validated with three brain-computer interfaces (BCI) based use-cases. The contributions are two-fold: first, we propose a novel ML-IoT pipeline software architecture that encompasses essential components from data ingestion to runtime use of ML models; second, we assess the software on cognitive applications and achieve promising results in comparison to literature.
Open Access
An Encoder-Decoder Based Basecaller for Nanopore DNA Sequencing
(2019-07-02) Abbaszadegan, Mahdieh; Magierowski, Sebastian
Nanopore DNA sequencing is a method in which DNA bases are determined (basecalled) using electric current signals generated by passing DNA through nanopore sensors. The raw measured signals can be aggregated into event data presenting new bases entering the nanopore. This thesis has two contributions. First, we implemented RNN-based single- and double-strand basecallers for simulated event data to analyze the effect of signal noise. As the SNR decreased from 20 dB to 5 dB, the accuracy of the single-strand basecaller dropped 9% while the accuracy of double-strand basecaller only dropped 0.5%. Second, we implemented an end-to-end single-strand basecaller, directly processing the raw signal using an encoder-decoder model with attention instead of the CTC-style approach used in available basecallers. We achieved an accuracy of 81.9% for a viral sample and an accuracy of 90.9% for a bacterial sample. Our accuracy is comparable to state-of-the-art basecallers with a considerably smaller model.
Open Access
An Energy-Efficient Spiking CNN Implementation for Cross-Patient Epileptic Seizure Detection
(2022-03-03) Farshadfar, Parsa; Kassiri, Hossein
This research aims to develop a data-driven computationally efficient strategy for automatic cross-patient seizure detection using spatio temporal features learned from multichannel electroencephalogram (EEG) time-series data. In this approach, we utilize an algorithm that seeks to capture spectral, temporal, and spatial information in order to achieve high generalization. This algorithm's initial step is to convert EEG signals into a series of temporal and multi-spectral pictures. The produced images are then sent into a convolutional neural network (CNN) as inputs. Our convolutional neural network as a deep learning method learns a general spatially irreducible representation of a seizure to improves sensitivity, specificity, and accuracy results comparable to the state-of-the-art results. In this work, in order to avoid the inherent high computational cost of CNNs while benefiting from their superior classification performance, a neuromorphic computing strategy for seizure prediction called spiking CNN is developed from the traditional CNN method, which is motivated by the energy-efficient spiking neural networks (SNNs) of the human brain.
Open Access
An Enhanced Method for Full-Inversion-Based Ultrasound Elastography of The Liver
(2022-12-14) Mohamed Atia Aboutaleb; Ali Sadeghi-Naini
Similar to many other types of cancer, liver cancer is associated with biological changes that lead to tissue stiffening. An effective imaging technique that can be used for liver cancer detection through visualizing tissue stiffness is ultrasound elastography. In this thesis, we show the effectiveness of an enhanced method of tissue motion tracking used in quasi-static ultrasound elastography for liver cancer assessment compared to other state of the art methods. The method utilizes initial estimates of axial and lateral displacement fields obtained using conventional time delay estimation (TDE) methods in conjunction with a recently proposed strain refinement algorithm to generate enhanced versions of the axial and lateral strain images. Another primary objective of this work is to investigate the sensitivity of the proposed method to the quality of these initial displacement estimates. The proposed algorithm is founded on the tissue mechanics principles of incompressibility and strain compatibility. Tissue strain images can serve as input for full-inversion-based elasticity image reconstruction algorithm. In this work, we applied strain images generated by the proposed method in conjunction with an iterative elasticity reconstruction algorithm for full-inversion-based liver elastography. Moreover, a set of in-silico experiments were conducted to validate the assumptions used in the reconstruction technique to improve the realism of the method. Ultrasound RF data collected from a tissue-mimicking phantom and from four liver cancer patients who underwent open surgical RF thermal ablation therapy were used to evaluate the proposed method. The results showed that the proposed method produces superior results to other state of the art methods. Moreover, while there is some sensitivity to the displacement field initial estimates, overall, the proposed method is robust to the quality of the initial estimates.
Open Access
An Exploratory Study on the Platforms of Sharing Reusable Machine Learning Models
(2021-03-08) Xiu, Minke; Jiang, ZhenMing "Jack"
Recent advances in Artificial Intelligence, especially in Machine Learning (ML), have brought applications previously considered as science fiction (e.g., virtual personal assistants and autonomous cars) into the reach of millions of everyday users. Since modern ML technologies like deep learning require considerable technical expertise and resource to build custom models, reusing existing models trained by experts has become essential. Currently the ML models are shared, distributed, or retailed on multiple ML model platforms which can be divided into two categories based on their usage patterns: (1) ML model stores whose models can be deployed and served with the help of cloud infrastructure, and (2) ML package repositories whose models are free but need to be deployed and used (e.g., embedded into users applications as a software component) manually. We conducted an exploratory study on the above two categories of ML model platforms: ML model stores and ML package repositories. We analyzed the structure and the contents of the ML models platforms, as well as functionalities provided by the package managers. The research subjects were three general purpose ML model stores (AWS marketplace, ModelDepot, and Wolfram neural net repository) and two popular ML package repositories (TensorFlow Hub and PyTorch Hub). When studying the structure of ML model platforms and functionalities of package managers, we compared them against their counterparts from traditional software development: ML model stores vs. mobile app stores (e.g., Google Play and Apple App Store), and ML package repositories vs. programming language package repositories (e.g., npm, PyPI, and CRAN). Through our study, we identified special software engineering practices and challenges for sharing, distributing, and retailing ML models. The implications from this thesis will be helpful for stakeholders to make the ML model platforms better serve the users (i.e., software engineers, data scientists and researchers).
Open Access
API Knowledge Guided Test Generation for Machine Learning Libraries
(2022-12-14) Narayanan, Arunkaleeshwaran; Wang, Song
This thesis proposes MUTester to generate test cases for APIs of machine learning libraries by leveraging the API constraints mined from the corresponding API documentation and the API usage patterns mined from code fragments in Stack Overflow (SO). First, we propose a set of 18 linguistic rules for mining API constraints from the API documents. Then, we use the frequent itemset mining technique to mine the API usage patterns from a large corpus of machine learning API related code fragments collected from SO. Finally, we use the above two types of API knowledge to guide the test generation of existing test generators, for machine learning libraries. To evaluate the performance of MUTester, we first collected 2,889 APIs from five widely used machine learning libraries (i.e., Scikit-learn, Pandas, Numpy, Scipy, and PyTorch),then for each API, we further extract their API knowledge, i.e., API constraints and API usage patterns. Given an API, MUTester combines its API knowledge with existing test generators (e.g., search-based test generator PyEvosuite and random test generator PyRandoop) to generate test cases to test the API. Results of our experiment show that MUTester can significantly improve the corresponding test generation methods. And the improvement in code coverage ranges from 18.0% to 41.9% on average.In addition, it also reduced 21% of invalid tests generated by the existing test generators.
Open Access
Assisted Target Detection in Airborne Search and Rescue
(2020-11-13) Taheri-Shirazi, Maryam; Elder, James H.
Finding and rescuing people from downed aircraft is challenging in many parts of the world, including Canada. Because the Canadian military still relies on the naked eye to conduct searches, airborne search and rescue could benefit greatly from advanced sensor systems. Partial automation of target detection could alleviate operator workload and potentially improve rescue efforts. One of the obstacles to developing such a system has been the lack of a large, realistic, and ground-truthed search and rescue (SAR) dataset. I used a new dataset for airborne SAR collected in 2014 by the National Research Council Flight Research Laboratory (NRC-FRL) and labeled approximately 40,000 frames, to extract roughly 20,000 negative and 20,000 positive images. Then I tested three ATD methods on this dataset in order to develop more advanced assisted target detection algorithms for thermal infrared (IR) images.
Open Access
Data Rate Enhancement in Ultrasonic Data Telemetry Links
(2022-08-08) Abbasi Shakooh, Ali; Sodagar, Amir
This research aims at introducing a novel technique for the enhancement of data rate in wireless data transfer through ultrasonic telemetry links. Referred to as INtervened-Timing, Enhanced-Rate Interval Modulation (INTERIM), the proposed technique is an isochronous pulse-based modulation technique, which embeds the telemetered symbol in the time interval between two consecutive ultrasonic tone bursts of different frequencies. To realize this idea, two pairs of ultrasonic transducers with different central frequencies and non-overlapping frequency responses are used. Each transducer pair oversees the transmission and receipt of one of the tone bursts. This is while using the same transducers in the same operational conditions (i.e., with the same excitatory-inhibitory pulse (EIP) excitations, same link medium, and at same distance), the INTERIM approach offered a maximum data rate of 500 kbps and 800 kbps at bit error rate (BER) of 1.4% and 6% respectively.
Open Access
Deep Reinforcement Learning based Energy-Efficient Multi-UAV Data Collection for IoT Networks
(2022-03-03) Khodaparast, Seyed Saeed; Wang, Ping; Nguyen, Uyen Trang
Unmanned aerial vehicles (UAVs) are regarded as an emerging technology, which can be effectively utilized to perform the data collection tasks in the Internet of Things (IoT) networks. However, both the UAVs and the sensors in these networks are energy-limited devices, which necessitates an energy-efficient data collection procedure to ensure the network lifetime. In this thesis, we propose a multi-UAV-assisted network, where the UAVs fly to the ground sensors and control the sensor's transmit power during the data collection time. Our goal is to minimize the total energy consumption of the UAVs and the sensors, which is needed to accomplish the data collection mission. We formulate this problem into three sub-problems of single UAV navigation, sensor power control, and multi-UAV scheduling, and model each part as a finite-horizon Markov Decision Process (MDP). We deploy deep reinforcement learning (DRL)-based frameworks to solve each part. Specifically, we use the deep deterministic policy gradient (DDPG) method to generate the best trajectory for the UAVs in an obstacle-constrained environment, given its starting position and the target sensor. We also deploy DDPG to control the sensor's transmit power during data collection. To schedule activity plans for each UAV to visit the sensors, we propose a multi-agent deep Q-learning (DQL) approach by taking the energy consumption of the UAVs on each path into account. Our simulations show that the UAVs can find a safe and optimal path for each of their trips. Continuous power control of the sensors achieves better performance than the fixed power and fixed rate approaches in terms of the sensor's energy consumption and the data collection completion time. In addition, compared to the two commonly used baselines, our scheduling framework achieves better and near-optimal results in the simulated scenario.
Open Access
Deep Unsupervised Learning for Network Resource Allocation Problems with Convex and Non-Convex Constraints
(2023-03-28) Alizadeh, Mehrazin; Tabassum, Hina
Deep neural networks (DNNs) are currently emerging as a potential solution to solve NP-hard wireless resource allocation problems. However, in the presence of intricate constraints, e.g., users' quality-of-service (QoS) constraints or base station quota, guaranteeing constraint satisfaction becomes a fundamental challenge. In this thesis, I propose a novel unsupervised learning framework to solve the classical power control and user assignment problem in a multi-user interference channel, where the objective is to maximize the network sum-rate with QoS, power budget, and base station quota constraints. The proposed method utilizes a differentiable projection function, defined both implicitly and explicitly, to project the output of the DNN to the feasible set of the problem. Extensive simulations depict that the proposed DNN solutions not only improve the achievable data rate, but also achieve zero constraint violation probability, compared to the existing DNNs, and also outperform the optimization-based benchmarks in computation time.
Open Access
Design and Analysis of Non-Orthogonal Multiple Access Techniques for Terahertz Networks
(2022-08-08) Melhem, Sadeq Bani; Tabassum, Hina
Fueled by the emergence of machine-type communications in a variety of wireless applications, the provisioning of massive connectivity becomes instrumental. On the other hand, accommodating trillions of devices within the extremely congested and limited sub-6GHz spectrum is becoming challenging. In this context, shifting to higher frequency terahertz (THz) communication is under consideration to obtain the data rates in the order of hundreds of gigabits per second (Gbps). Also, non-orthogonal multiple access schemes are becoming popular to support multiple users in the same frequency and time resource block, while leveraging on efficient interference cancellation mechanisms. In this thesis, I develop a comprehensive mathematical framework to analyze the performance of emerging non-orthogonal channel access schemes, such as non-orthogonal multiple access (NOMA) and rate-splitting multiple access (RSMA), in THz networks. In the first part of the thesis, I develop a statistical framework to analyze the performance of NOMA in the downlink of a single-carrier and multi-carrier THz network considering Nakagami-m fading and molecular absorption noise. In this context, I first develop a novel user pairing scheme which ensures the performance gains of NOMA over orthogonal multiple access (OMA) for each individual user in the NOMA pair and adapts according to the THz molecular absorption. Then, I characterize novel outage probability expressions considering a single-carrier and multi-carrier THz-NOMA network in the presence of various user pairing schemes, Nakagami-m channel fading, and molecular absorption noise. Specifically, I propose a moment-generating-function (MGF) based approach to analyze the outage probability of users in a multi-carrier THz network. For negligible thermal noise, I provide simplified single-integral expression to compute the outage in a multi-carrier network. Numerical results demonstrate the efficiency of the proposed user-pairing scheme compared to the existing benchmarks and validate the accuracy of the derived expressions. Finally, in the second part of the thesis, I extend the developed framework to analyze the performance of RSMA in the downlink transmission of Sub-6 GHz and THz networks.
Open Access
Digital Sun Sensor Design for Nanosatellite Applications
(2020-08-11) Bolshakov, Konstantin; Lee, Regina S. K.
In this research a novel, semi-custom design of a Sun sensor, based on the orthogonal photodiode array design approach, applicable for nanosatellites and other small spacecraft is proposed. A common and well-known strategy of application of a geometrical aperture mask on the light detectors is improved upon and utilised in a non-conventional fashion in the design presented in this thesis. The main characteristic of this design, that is investigated in this work, is the inclusion of a chirped pattern of slits, while using a digital readout of the photodiode arrays. This pattern is expected to allow greater angle detection accuracy while digital photodiode readout decreases complexity and power consumption. The array based design approach is chosen due to lower power requirements, mass savings and simpler readout interface and signal processing in comparison to the matrix based approach. The Sun sensor design presented maximises the accuracy, while keeping the cost, development and implementation time and complexity to minimum. The design will be demonstrated on DESCENT mission CubeSat as a part of the Moth-Eye Anti-Reflective solar cell coating payload. The characterisation of the non-calibrated manufactured part shows that the sensor is capable of 1 degree accuracy, which is expected to improve to under 0.5 degrees with calibration and sensor data processing.
Open Access
Distance Protection Challenges of Converter-Interfaced Renewable Energy Sources
(2018-11-21) Banaiemoqadamfariman, Amin; Hooshyar, Ali
Full-scale converter-interfaced renewable energy sources (CIRESs) can cause misoperation of distance relays installed in their vicinity. Such failure stems from the different fault behavior of CIRESs compared to synchronous generators (SGs), based on which existing relays have been developed. Several measures have been devised to improve the performance of distance protection by modifying existing relays. This thesis proposes a new approach to tackle this problem. The prime objective of this method is to mimic certain features of SGs' fault current while the constraints of a converter, such as its limited fault current magnitude, are satisfied. As a result, correct operation of distance relays close to CIRESs is ensured regardless of the fault characteristics, including its type, resistance, and location. Some salient features of the proposed method are its simplicity, compatibility with off-the-shelf relays, independence from the voltage and power rating of the CIRES, using only local measurements and being cost-effective.