AI/ML Based RAN slicing
As part of my research internship in Samsung R&D Institute India
Worked on applying Reinforcement Learning based techniques for dynamic and efficient radio resource management in massive IoT or massive machine type communication (mMTC) in 5G
Applied learning techniques on edge for Smart Metering Internet of Things for data compression and energy efficiency. We have also implemented the proposed algorithms on commercial smart meters installed throughout the IIT Delhi campus
Estimation of transmitted signal parameters at the receiver has always been a matter of utmost importance in any wireless communication system. By using training signals , transmitter parameters can be estimated or obtained at the receiver. But the addition of redundant bits reduces spectrum efficiency. The recent trend is to estimate the parameters and synchronize the orthogonal frequency division multiplexing (OFDM) signals through blind process i.e., without the use of any training sequence. Hence , the goal of this project is to design a blind OFDM receiver which can estimate the transmitted OFDM signal parameters and synchronize the signals to demodulate it to corresponding constellation points. The parameters estimation process includes carrier frequency, signal bandwidth, useful OFDM symbol length, cyclic prefix length, OFDM symbol length, number of subcarrier, oversampling factor, symbol timing offset (STO) and carrier frequency offset (CFO). In this project we have developed a consolidated framework where we estimate all the parameters, remove the synchronization errors after estimation of STO and CFO and finally, soft constellation points are reconstructed according to the signal type. The performance of the estimator has been tested by MATLAB simulation.
In this project we explored technical aspects of using Augmented Reality in Industrial Applications. Initially existing capabilities of Vuforia (by Qualcomm) were explored by going through the sample applications. Vuforia supports most of the important and generic AR applications. Known textures were used to detect an object to track and then text or graphics was augmented on top of it. We also explored Level set method based and Kalman filter based 2D tracking algorithms in this internship where OpenCV C++ libraries were used. But again to augment in 3D we need 3D tracking. Due to unavailability of suitable open source SDK s we could not go for this. There is scope of future work which may use model based tracking which will take CAD model of all the industrial machineries as input and track them as well as augment graphics or text.
Every year the country experiences a number of railway accidents resulting in loss of human life and disruption in regular schedule. These are caused, among others, on account of invisibility of signals due to dense fog in winter months and jumping of signals by fatigued/ negligent drivers. In this project, an effort has been made to find out ways for a safer railway journey by implementing microcontroller based automated railway signaling system. The entire hardware has been tested in the laboratory using ATMEL 89C51 Microcontroller. The software has been designed based on Assembly Language.