To see my recent research, go here. On this page is a collection of various projects (research and otherwise) from my B.Tech. at IITB.
Research
The Doppler effect is well-known - relative speed causes a shift in frequency observed. An application of this is the micro-Doppler - used to detect rotating and vibrating bodies (such as the blades of a helicopter or the swinging arms of a walking person), which give an oscillatory frequency variation. Here we attempted to study these and their detection from radar signals using compressed sensing.
Studied the detection of micro-Doppler parameters from RADAR signals using Inverse Radon Transform and L-statistics
Devised an algorithm to express the demodulated signal in terms of Bessel functions of the first kind and use this to extract the micro-Doppler components by filtering out appropriate elements [NCC 2022 paper]
Obtained some preliminary results in extending a Finite Rate of Innovation sampling framework to micro-Doppler detection, using Variational Mode Decomposition [SPCOM 2022 paper]
Presented with an Undergraduate Research Award (URA01) for our work
It is well-established in information theory that feedback does not increase the capacity region for single-user discrete memoryless channels, but this is not the case when there are multiple users involved. Our goal was to characterize the capacity region for Broadcast Channels with two users, with potentially noisy feedback from both receivers.
Derived an upper bound for the erasure probability in the feedback from the strong receiver that can possibly provide any improvement in binary erasure broadcast channels
Attempted to extend a similar analysis to output-quantized Gaussian broadcast channels
Low-density parity-check (LDPC) codes have achieved great fame as capacity-approaching codes, but coupling them together in a convolutional manner has been shown to be even better. We studied this improvement in detail, and looked at an application of this code over fading channels.
Performed literature review and simulations to understand why these perform better than conventional LDPC codes
Studied the performance improvement achievable using interleaving over fading channels, subject to a latency constraint
Analysed the performance of low-complexity, reduced-latency windowed decoding
Power amplifiers have inherent non-linearities (underlying physics), which leads to a spreading of the signal spectrum (mulitplication in time-domain, convolution in frequency-domain). This is especially problematic when dealing with allocated bandwidths, as is done in all of wireless communication. The effect of these non-linearities can be removed by Digital Pre-Distortion, for which we first model the non-linearity mathematically, then apply an 'inverse'.
Conducted literature review and implemented the latest models on MATLAB, obtaining considerable improvement over those currently in use
Devised an efficient "peeling" algorithm (inspiration) to make the model implementable on an FPGA, hence ready for use in a real product
Given a Pre-Placement Offer to join as a full-time Systems Engineer following graduation
Others
This was a survey of recent works in compressing graphical data, motivated by the ubiquity of graphs as modern data structures. Using Delgosha and Anatharam, 2020 as a starting point, I read related papers that look at variants: universal lossless compression, using sparse and heavy-tailed graphs and summarized them.
Traditional multi-armed bandit settings seek to "learn" an environment by performing actions and obtaining corresponding rewards. In distributed learning settings, the action-performing and reward-collecting is done by resource-constrained, lightweight agents which communicate with a central server that does the "learning". We studied Hanna et al., 2022, which provides algorithms and bounds on the regret and amount of communication needed for such a setting. This project continued even after my graduation from IITB, leading to this research project.
Devices that are used in IoT applications are resource-constrained, so they cannot emply high complexity crytpography schemes to secure their data. This motivates the study of "lightweight" techniques. We suggest a highly preliminary idea for such a purpose, combining two previously known schemes.
A typical compressed sensing task is to recover a high-dimension, sparse vector from noisy measurements. Whether knowing something about the sparsity pattern (such as the probability of each index being zero) can improve compressed sensing algorithms is an interesting question, one investigated by Scarlett et al., 2013. We study the same and provide preliminary attempts at improving the proposed algorithms.
Imagine yourself at a bank. You enter, get yourself a token from a dispenser, and wait until your number shows up on a screen, directing you to an unoccupied counter. This is exactly what this project aims to replicate by using an 8051 microcontroller on a Pt-51 board, interfaced with a 16x2 LCD and a keyboard using UART. There are four counters. Typically, the display shows the token number being serviced at each counter. Customers take tokens by pressing t on the keyboard, at which moment the display shows the token number allotted (for 2 seconds). The tellers at the counters notify their availability by flicking a switch on the Pt-51 (one switch per counter), at which point the next token is allotted (on a first-come-first-serve basis).
Solar-powered street lights are enviroment-friendly, but require extra protection for the battery due to the source being inconsistent in its output. Frequent undercharge and overcharge severely diminish the life of the battery, undoing all of its benefits. Hence a Battery Management System is called for, which is what we built here - designing a protection circuit for the battery connected to an LED load and a Printed Circuit Board (PCB) to make it ready for use as a real-world product.
"Applications of Digital Signal Processing for Atmanirbhar Bharat (Independent India)" was the broad topic of the project - we picked Audio Watermarking as a prime example of the use of technology in a growing country to encourage individual creativity by protecting property rights.
Putting together everything we learnt in the course, we used the TI OPA344 as a reference to design an op-amp. Given target specifications of unity gain bandwidth, phase margin, slew rate, common mode voltage range, and open-loop small-signal low-frequency gain, we calculated the circuit component values. The designed circuit was simulated using ngspice to verify that the target requirements were met.
Here I made software applications to aid in larger research projects. This was my first real experience with actually writing large (or any, for that matter) quantities of code in Python, C++ and Javascript (basically any language that isn't MATLAB).
This was my first association with any kind of research --- I was involved in the quality improvement of the doctoral thesis of Dr. Peeyush Sahay, who successfully defended his thesis in 2020. I also helped edit and complete associated publications (this and this) by going through comments by reviewers and making corrections as needed, which led to a mention in the acknowledgements section. It was interesting to see first-hand the amount of effort that went into a successful publication.
This is often used in quizzes to determine the fastest person to press a buzzer. With four push-button switches representing four participants, each with an LED attached and connected to a common buzzer, the circuit is designed to break when any of the switched is pressed, turning the corresponding LED on and blaring the buzzer by using IC 555 as a bistable vibrator. It also displays the time on 7-segment displays, using an IC 555 timer to count the seconds, stopping when a switch is pressed.