Department of Electrical Engineering

Indian Institute of Technology Palakkad&nbsp;(IIT Palakkad or IIT PKD) is a public autonomous&nbsp;engineering&nbsp;and&nbsp;research&nbsp;institute located in&nbsp;Palakkad,&nbsp;Kerala. It was&nbsp;one of the five new&nbsp;IITs&nbsp;proposed in the&nbsp;2014 Union budget of India. The campus was inaugurated on 3 August 2015. The institute has 94 faculty, 978 students, and 63 non-teaching staffs.

IT Palakkad currently offers four-year Bachelor of Technology (B.Tech) and two years MS programmes in 8 different engineering disciplines including&nbsp;Computer Science, Electrical, Mechanical, Civil among others.

&nbsp;

IIT Palakkad

Anti-aliased Convolutional Neural Networks (CNNs) have been proposed to overcome the shift variant nature of the CNNs. The fundamental building block of the anti-aliased CNN has been the application of Gaussian or wavelet-based smoothing before the pooling operation. However, in all these approaches, the feature maps' edges are also smoothed while suppressing high-frequency components. In this work, two novel pooling approaches are presented, namely the Laplacian-Gaussian Concatenation with Attention (LGCA) pooling and Wavelet-based Approximate-Detailed Coefficient concatenation with Attention (WADCA) that can preserve the edges in the feature maps. The results suggest that the proposed pooling approaches outperform conventional as well as blur pooling for classification, segmentation and auto-encoders. In terms of average binary classification accuracy (cats vs dogs), the proposed LGCA approach outperforms the conventional pooling and blur pooling by 4\% and 2\%, 3\% and 4\%, 3\% and 0.5\% for MobileNetv2, DenseNet121 and ResNet50 respectively. On the other hand, the proposed WADCA approach outperforms the normal pooling and blur pooling by 5\% and 3\%, 2\% and 3\%, 2\% and 0.17\% for MobileNetv2, DenseNet121 and ResNet50 respectively. It is also observed from the results that edge-preserving pooling does not have any significance in segmentation tasks possibly due to high to low-resolution translation. Meanwhile, high-resolution reconstruction has been observed for the LGCA pooling in the case of convolutional auto-encoders.

World University root node

MULTIMEDIA SYSTEMS

Edge preserved universal pooling: novel strategies for pooling in convolutional neural networks

In the recent past, with the rapid surge of COVID-19 infections, lung ultrasound has emerged as a fast and powerful diagnostic tool particularly for continuous and periodic monitoring of the lung. There have been many attempts towards severity classification, segmentation and detection of key landmarks in the lung. Leveraging the progress, an automated lung ultrasound video analysis package is presented in this work, which can provide summary of key frames in the video, flagging of the key frames with lung infection and options to automatically detect and segment the lung landmarks. The integrated package is implemented as an open-source web application and available in the link https://github.com/anitoanto/alus-package.\& COPY; 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

SOFTWAREX

SPAALUV: Software Package for Automated Analysis of Lung Ultrasound Videos

The clinical interpretability of specular interfaces in ultrasound (US) imaging is heavily influenced by the anatomical knowledge of the operator and the transmit-receive beamforming employed for imaging. The violation of the conventional diffuse assumptions by highly directive specular reflections makes their detection challenging which increases the cognitive burden of operators in tissue characterization or calibrating the beam steering/transducer to ensure their detection. However, the areas for improved operator assistance and related workflow in this regard remain largely unexplored. This work proposes a novel method called tiered subaperture (SA) directivity indexing, which employs a SA-level measurement of energy in the reflections through multiangle plane wave transmissions to automatically identify the directivity of specular reflections. The reflection directivity is vectorized and visually fed as a characterized colored image to the operator to distinguish the specular patterns in the US data. The technique is illustrated for clinical use cases such as needle tracking in guided interventional procedures, anisotropy in MSK US, and bone fracture monitoring. The proposed method improves retention, enables freehand scans, and augmentation of hand-eye coordination for the operators for tissue characterization, beamforming, and transducer calibration thus aiding quicker diagnosis by saving procedure times.

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT

A Novel Technique for Estimation and Detection of Specular Reflections Leveraging High Frame Rate Ultrasound Imaging

In the recent years, deep learning techniques have received a great attention in a wide range of practical problems, including the event detection, localization, classification, and prediction. In this paper, we present a real-time bidirectional long-short term memory (BiLSTM) based human presence sensing using the WiFi channel state information (CSI) feature for non-intrusive occupancy detection systems in indoor environments by selecting best hyper-parameters. In this study, we created real-time Wi-Fi enabled Raspberry Pi computing platform to collect CSI datasets for training and testing purposes by considering different room sizes, fixed objects and different activities. On the wide variety of CSI datasets, evaluation results showed that the proposed magnitude CSI-BiLSTM based human presence detection (HPD) method had an average sensitivity (SE) of 92.23 %, specificity (SP) of 91.18 % and accuracy (ACC) of 91.68% for the 20% test CSI segments taken from the same training dataset and had an average SE of 85.71 %, SP of 91.59% and ACC of 88.67% on the unseen CSI dataset collected under different room conditions. In real-time Raspberry Pi testing environment, the proposed method had an average SE of 99.35%, SP of 94.39 % and ACC of 97.00 %. The proposed magnitude CSI-BiLSTM based HPD method had an average computational time of 1.37 ms with a model size of 509.67 kB. © 2023 IEEE.

2nd International Conference on Microwave, Optical and Communication Engineering, ICMOCE 2023

Bidirectional LSTM based Sequence Classification Approach for Human Presence Detection using Channel State Information

Photoplethysmogram (PPG) is a noninvasive technique to estimate heart rate and respiration rate with advantages over electrocardiogram (ECG) in terms of ease of sensing. In the realm of energy-efficient health monitoring, compressive sensing (CS) is gaining prominence, particularly for signals that exhibit sparsity. This study explores to find the best sparse recovery algorithm for a fixed sparsifying matrix including impulse (I), discrete cosine (C), discrete sine (S) [I C S] and deterministic binary block diagonal (DBBD) sensing matrix for the reconstruction of the CS PPG signal, with a focus on quality metrics and potential very large scale integration (VLSI) implementations. At a compression ratio (CR) of 3, the orthogonal matching pursuit (OMP) algorithm, DBBD and [I C S] combination outperform other recovery algorithms, achieving an average percentage root mean square difference (PRD) of 1.86%, maximum absolute error (MaxAE) of 0.0976 and signal-to-noise ratio (SNR) of 36.07 dB. However, as CR increases, the approximate message passing (AMP) algorithm, DBBD and [I C S] combination often surpass OMP, achieving an average PRD value of 9.79% at a CR of 5. Additionally, AMP is computationally simpler than OMP for lower CRs. Although OMP is recommended for its accuracy across various CRs, AMP often matches or even outperforms it at higher CRs with the fixed combination of DBBD and [I C S] in compressed PPG signal recovery. The results emphasize the importance of finding suitable sparse recovery algorithms for specific CRs, sparsifying matrices, sensing matrices and the trade-offs between compression, distortion and complexity. © 2023 IEEE.

ICSIMA 2023 - 9th IEEE International Conference on Smart Instrumentation, Measurement and Applications

Performance Evaluation of Sparse Recovery Algorithms for Efficient Signal Reconstruction in Compressed PPG Sensing Node

By considering limited resource-constraints of medical devices and advanced deep learning networks, in this paper, we explore a lightweight convolutional neural network (CNN) based AFibri event detector by finding suitable hyperparameters and activation function with best trade-off between the detection accuracy and model size (or computational time). This study presents extensive evaluation results of CNN-AFibri event detection methods that are obtained for different combination of model parameters: number of convolutional layers (CLs of 3, 4, and 5), number of filters (8, 16, 32, 64 and 128), activation functions (including the rectified linear unit (ReLU), leakyReLU (LReLU), exponential linear unit (ELU)), and kernel sizes (<inline-formula> <tex-math notation="LaTeX">$3 \times 1~$</tex-math> </inline-formula>, <inline-formula> <tex-math notation="LaTeX">$ 4 \times 1$</tex-math> </inline-formula>). In addition to different CNN-AFibri models, we validate their performances under different ECG segment duration of 5, 10 and 30 seconds. On the standard databases and unseen databases, the CNN-AFibri model with the CLs of 5, ELU function and kernel size of <inline-formula> <tex-math notation="LaTeX">$ 4 \times 1$</tex-math> </inline-formula> had a highest accuracy of 99.97% (specificity of 99.98% and sensitivity of 99.95%) for 5 second ECG segments as compared to the performances of 54 CNN-AFibri models reported in this paper and other existing deep learning based methods on the same validation databases. Real-time implementation of the best CNN based method with a model size of 3.14 Megabyte is demonstrated using the Raspberry Pi computing platform with Broadcom BCM2711, 1.5 GHz Cortex-A72 quad-core CPU with 8 GB RAM. Results demonstrated that the average processing times are less than 3 ms and 11 ms for processing 5 s and 30 s ECG segments, respectively with an accuracy reduction of less than 1% as compared to the same model tested on the personal computer with Intel(R) Xeon(R) W-2133 3.60 GHz Processor with 6 core and 128 GB RAM. IEEE

IEEE Transactions on Circuits and Systems I: Regular Papers

AFibri-Net: A Lightweight Convolution Neural Network Based Atrial Fibrillation Detector

Conventional adaptive filters, which assume Gaussian distribution for signal and noise, exhibit significant performance degradation when operating in non-Gaussian environments. Recently proposed fractional-order adaptive filters (FoAFs) address this concern by assuming that the signal and noise are symmetric α-stable random processes. However, the literature does not include any VLSI architectures for these algorithms. Toward that end, this article develops hardware-efficient architecture for fractional-order correntropy adaptive filter (FoCAF). We first reformulate the FoCAF for its efficient real-time VLSI implementation and then demonstrate that these reformulations cause negligible performance degradation under the 16-bit fixed-point implementation.

IEEE Transactions on Very Large Scale Integration (VLSI) Systems

Novel VLSI Architecture for Fractional-Order Correntropy Adaptive Filtering Algorithm

This study involves the determination of the impact of mutual coupling between antenna elements on the performance of a dual-polarized, wide-angle scanning, phased array antenna for weather radar applications. Weather radars require dual linearly polarized antennas with low cross-polarization, and a narrow beam phased array for wide scanning angle. For this simulation-based study, a microstrip dual linearly polarized 2x20 phased array antenna operating at S-band (2.65 to 3.0 GHz) is designed. This antenna has been designed to have a cross-polarization level less than -45 dB at both polarizations and for the scan angle range of -55o to 55o, which is better than most of the existing dual-pol phased array antennas. This antenna has been used to analyze the impact of mutual coupling on cross-polarization, beam width, and antenna gain at various scan angles. Mutual coupling is studied in terms of antenna active element pattern and the corresponding cross-polarization value as well as the active reflection coefficient and impedance values for inter-element spacings of 0.4λ and 0.5λ. It has been found in this study that cross-polarization levels of the whole array (at various scan angles) are affected significantly because of mutual coupling between elements.

Fulltext

AEM Journal

Study on Impact of Mutual Coupling on Performance of Dual Polarized Phased Array Antenna

We propose an energy-based framework to analyze the statics and dynamics of a ferroelectric negative capacitance-hybrid microelectromechanical system (MEMS) actuator. A mapping function that relates the charge on the ferroelectric to the displacement of the movable electrode is used to obtain the Hamiltonian of the hybrid actuator in terms of displacement. We then use graphical energy-displacement and phase portrait plots to analyze static pull-in, dynamic pull-in, and pull-out phenomena of the hybrid actuator. Using these, we illustrate the low-voltage operation of the hybrid actuator to static and step inputs, when compared to the standalone MEMS actuator. The results obtained are in agreement with the analytical predictions and numerical simulations. The proposed framework enables straightforward inclusion of adhesion between the contacting surfaces, modeled using van der Waals forces. We show that the pull-in voltage is not affected, while the pull-out voltage is reduced due to adhesion. The proposed framework provides a physics-based tool to design and analyze negative capacitance-based low-voltage MEMS actuators.

IEEE Transactions on Electron Devices

Analysis of Ferroelectric Negative Capacitance-Hybrid MEMS Actuator Using Energy-Displacement Landscape

Electrostatic MEMS actuators suffer from an instability called pull-in, wherein the movable electrode snaps onto the fixed electrode beyond a certain applied voltage. Thus, the entire allowed range is not utilized for stable operation. We propose pull-in free, low-voltage operation by using a cubic non-linear spring with a ferroelectric negative-capacitance hybrid MEMS actuator. We use a physics-based framework based on the energy landscape to illustrate the stability improvement. This framework uses energy-displacement and voltage-displacement plots for analysis. We predict that the actuator can operate in three distinct modes: (a) monostable, (b) bistable and (c) always-stable, based on the value of the cubic spring constant. We also estimate the threshold values of the cubic spring constant that demarcate the three modes of operation. By proper design of the cubic spring constant, we predict pull-in free, low-voltage operation of the hybrid actuator, as compared to the standalone MEMS actuator. The results obtained are in agreement with the numerical simulations. This work will aid in the design of electrostatic MEMS actuators for low-voltage applications without pull-in instability.