Soham Naha

M.Tech Project Phase-I

• Title: Optimal Precoder Feedback for MIMO Wireless Systems
• Guide: Prof. Kumar Appaiah, EE Dept., IIT Bombay
• Duration: 8 months
• Objective: To find efficient optimization techniques using flag manifolds for precoder quantization
• Completed Work:
  • Reviewed existing techniques for efficient feedback methods for MIMO systems and the use of Unitary, Grassmannian, Stiefel and Flag Manifold in precoder parameterization and codebook generation
  • Simulated ADPCM based scalar parameterization of precoder matrices of MIMO systems
• Future Work:
  • Use of manifold based techniques to construct scalable codebooks for precoder tracking
  • Effective parameterization of precoder updates for minimal feedback while not sacrificing rate

M.Tech Seminar

• Title: Effect of General Channel Aging on the DownLink Performance of Massive MIMO Systems
• Guide: Prof. Kumar Appaiah, EE Dept., IIT Bombay
• Duration: 5 months.
• Objective: Investigate the joint impact of user mobility and phase noise in Massive MIMO systems.
• Simulated and analyzed the joint-effect of user-mobility and phase-noise due to antennas in the downlink achievable sum-rate of Massive-MIMO Systems by a joint-model, for low and high relative velocities
• Simulated achievable rates for MISO, SIMO with MRC and VBLAST receiver using ITPP library in C++

R&D Project (EE692)

• Title: Improving Energy-Based Out-of-Distribution (OOD) Detection
• Guide: Prof. Amit Sethi, EE Dept., IIT Bombay
• Duration: 6 months
• Implemented energy-based OOD detection techniques using energy scores on the CIFAR10 dataset
• Experimented with noisy labels to analyze the robustness of the proposed energy score for OOD detection
• Reference Paper: here

R&D Project (EE691)

• Title: Python and SQL based UI for ZBTT project of Indian Railways
• Guide: Prof. Kumar Appaiah, EE Dept., IIT Bombay
• Duration: 7 months.
• Objective: Create UI for efficient analysis and correction of Zero-Based Time-Table (ZBTT) simulator output
• Accelerated error detection, analysis and correction speed of the IITB-simulator output, by creating a Dash and SQLite based UI, which was handed over to the Indian Railways as an add-on to the original deliverables
• Streamlined different components of the project using Git version control and GNUMake
• Integrated simulator output (10,000+) lines with SATSANG database, used by Indian Railways

Automatic Speech Recognition

• Title: Speech to Sign-Language(with emotions) for the Hearing-Impaired
• Guide: Prof. Preeti Jyothi, CSE Dept., IIT Bombay
• Duration: 5 months
• Trained a Convolutional Neural Network on RAVDESS audio samples to detect emotion from speech
• Used a Conformer-based pre-trained model from ESPNET-model zoo to convert Speech to English text
• Created a streamlit based WebApp, to record audio and display the corresponding predicted text and emotion

Information Theory and Coding

• Title: Degraded Broadcast Channel (DBC) and Superposition Coding
• Guide: Prof. Bikas Kr. Dey, EE Dept., IIT Bombay
• Duration: 5 months.
• Reviewed capacity bounds of discrete memoryless channels with focus on degraded broadcast channels
• Studied codebook generation techniques using Superposition Coding to attain achievable rate for DBC

Advanced Machine Learning

• Title: Evolution of Human Pose Estimation
• Guide: Prof. Amit Sethi, EE Dept., IIT Bombay
• Duration: 4 months
• Reviewed the state of the art method of human pose estimation which propagates known person locations forward and backward in time and searches for poses in those regions.(combined detection and tracking)
• Studied multi-timescale trajectory prediction for long term abnormal human activity detection

• Title: Sentiment Analysis
• Guide: Prof. Amit Sethi, EE Dept., IIT Bombay
• Duration: 4 months
• Trained various neural networks LSTM, GRU, Bi-LSTM, and Bi-GRU on IMDB dataset to study their performance on changing the number of units and layers and achieved an accuracy of 87% using Bi-GRU
• Studied the effects of using different learning-rate schedulers like Cosine Annealing, Exponential and Step

• Title: Medical Image segmentation
• Guide: Prof. Amit Sethi, EE Dept., IIT Bombay
• Duration: 4 months
• Objective: Segmentation of the right ventricle from cardiac MRI using UNet Based Deep Learning models
• Performed data augmentation on RVSC-MICCAI 2012 dataset using color jitter, random horizontal flip, and random vertical flip to include defects that might occur in real image acquisitions
• Experimented with various loss functions like BCE loss, Focal Loss, BCE+Dice loss, BCE+Dice+InverseDice loss, Switching loss and obtained an accuracy of 72% with BCE+Dice+InverseDice loss

Digital Image Processing

• Title: Flash No-Flash Photography
• Guide: Prof. Suyash P.Awate, CSE Dept., IIT Bombay
• Duration: 4 months.
• Implemented denoising and detail transfer to merge the ambient qualities of the no-flash image with the high-frequency flash detail, using cross-bilateral filtering
• Performed white-balancing to change the color tone of ambient images, continuous flash to interactively adjust flash intensity, and red-eye removal to repair artifacts in the flash image

Speech Processing

• Title: LPC Analysis of Sythesized vowels and Natural Speech
• Guide: Prof. Preeti Rao, EE Dept., IIT Bombay
• Duration: 4 months.
• Synthesized phonemes by implementing source filter model to replicate the human glottal source
• Performed Linear Predictive Analysis of natural and synthetic speech which involved LP coefficient estimation, constructing LP magnitude spectrum and sound reconstruction

Introduction to Machine Learning

• Title: Attrition Classification
• Guide: Prof. Amit Sethi, EE Dept., IIT Bombay
• Duration: 5 months.
• Achieved accuracy of 89.3% by training a SVM classifier on Kaggle IBM Attrition dataset
• Extracted relevant and less correlated features and applied One-Hot Encoding for features with multi-classes

• Title: Support Vector Machine
• Guide: Prof. Amit Sethi, EE Dept., IIT Bombay
• Duration: 5 months.
• Implemented modified SVM algorithm in the paper Primal Estimated sub-GrAdientSOlver using NumPy
• Used the SVM classifier on linear data and kernelized-SVM on non-linear data

Communication Networks

• Title: Routing Information Protocol (RIP) using C
• Guide: Prof. Sharayu Moharir, EE Dept., IIT Bombay
• Duration: 5 months.
• Objective: To implement RIP using socket programming (in Linux)
• Implemented RIP (Distributed Bellman Ford Algorithm) using C socket programming that read a given network topology and generated the cost matrix for the shortest paths between the nodes

Summer Research Intern

• Title: Development of Image Processing Technique to Reduce the Effect of Blurring
• Guide: Dr. Ashok Sharma (Senior Project Research Scientist, NCPRE) and Dr.Prabir Kanti Basu (Senior Project Executive Officer(Scientist), NCPRE)
• Duration: 2 months.
• Work done:
  • Use CCD cameras to capture photoluminescence images of Si-wafers.
  • Use techniques of edge spread function and laplacian filtering on the raw images.
  • Enhance the raw images, removing the filtering effect of the camera, for proper visualization of defects in the Wafers.
  • Removal of the camera effects, enhanced the quality measurement of the cell.

B.Tech Project

• Title: Implementation of Subtractive Clustering on Functional Magnetic Resonance Images
• Guide: Dr.Bandana Barman (Assistant Professor, Dept. of ECE, Kalyani Government Engineering College)
• Duration: 8 months.
• Work done:
  • Use Python packages like nilearn and nibabel to visualize fmri data.
  • Code Subtractive Clustering in Python
  • Use subtractive clustering to cluster regions of FMRI images, based on activities.

Summer Research Intern

The internship led to my exposure to industrial solar cell manufacturing process and characterization methods.
My work primarily focussed on:

• Generating correct spatial maps of Photoluminescence (PL)
• Extracting correct information on electronic quality and enhancement of Si wafers and Solar Cell images
• Clear identification of defects and true distribution of minority carrier lifetime of these Cells and Wafers.

Advanced Telecommunication Vocational Training

In this training period of 1 month, we gained insights into the aspects of modern communication systems like GSM, GPRS, 3G architecture, Routing Protocols, MPLS VPN,etc.

Digital Marketing Intern

The internship lasted for 2 months. Kahaniya.com began a new initiative for budding story-writers to publish and showcase their work, with the winner able to meet Mr. Shashi Tharoor himself. My work in this was to help spread the great initiative via Digital Media.

Brand Ambassador

Worked as a Social Media Marketing intern for 1 month. The work included Digital Marketing and preparation of Report of activities.

Data Analyst

• Duration: -
• Responsible for assisting Insight Ed-Board for creating articles by analyzing and visualizing survey data
• Scraped, collected, pre-processed, visualized and analysed data from various department websites
• Helped publish independent articles on Insight social media handles and dynamic blogs on ML using streamlit

Teaching Assisstant

• Duration: 3 months.
• Conducted tutorial sessions on basic coding in Python and introductory machine learning algorithms using libraries like scikit-learn for a diverse class consisting of industry professionals from various domains

System Administration

• Duration: -
• Responsible for configuring, maintaining and securing e-mail, backups, LDAP and computational servers of EE
• Managed e-mail communication, handled shortlisting and coordinated interviews for EE PG Admissions

Institute Interview Coordinator

• Duration: 2 months.
• Coordinated with a team of 250+ members for interviews of 1600+ students.
• Assisted in conducting Pre-Placement Talks, Placement Tests and Interviews for 15+ firms