Main Info

Recent Work:

• Wireless Communications:  Worked on developing many next-generation wireless technologies such as turbo equalization for vehicular channel (this work resulted in a chip); improving WiFi using cellular technology; advanced channel codes for next-generation cellular technology. Our work on the Design of Low-Density Parity-Check Codes for 5G New Radio can be found in the IEEE Communications Magazine: Key Technologies for 5G New Radio, March 2018. Most of our LDPC design features have been adopted in the 5G standards.

• Reed-Muller Codes: We show that Reed-Muller codes achieve capacity on the binary erasure channel. Reed-Muller codes are one of the oldest codes and are found in almost all digital devices. For long they have been conjectured to achieve the Shannon capacity. In this work we combine ingredients from modern coding theory and thresholding of monotone functions to show that Reed-Muller codes are capacity-achieving on the binary erasure channel.

• Coding for Parallel Data Buses in Chips: We explore joint error-correction coding (ECC) and crosstalk-avoidance coding (CAC) for data communication across data buses in chips. Traditional approaches of ECC followed by CAC or vice-versa do not work and our joint encoding scheme identifies wires in the bus which are not constrained and uses them to embed the parity-bits (ECC). Analysis turns out to be non-trivial since the factor graph for the joint scheme consists of both linear and non-linear factor nodes.

• Spatial Coupling:  We show that Spatially Coupled Ensembles Universally Achieve Capacity under Belief Propagation. We describe the fundamental mechanism which explains why "convolutional-like" or "spatially coupled" codes perform so well. In essence, the spatial coupling of the individual code structure has the effect of increasing the belief-propagation (BP) threshold of the new ensemble to its maximum possible value, namely the maximum-a-posteriori (MAP) threshold of the underlying ensemble. For this reason we call this phenomenon threshold saturation. We gave a tutorial for Spatially Coupled Codes at the International Symposium on Information Theory (ISIT) 2013. Slides are available here.

• Indoor Positioning: We developed an indoor positioning system on a mobile device which fuses the information available from all the sensors such as RF signals (WiFi), inertial sensors (accelerometer, gyro) and magnetic sensors using a bayesian inference and learning engine. The algorithm uses a particle filter to jointly infer the position of the mobile device and learn the parameters of the models used for measuring the signals. We achieved centimeter-level median accuracy. 

Research Summary:

• The broad area of my research is in information science. Most of my work has been in the design and analysis of algorithms for inference and learning in graphical models appearing in computer science, communications, machine learning and signal processing. I have also applied methods from statistical physics to analyze problems in computer and communication sciences. 

• I am fascinated by the resurgence of neural networks via deep learning. Currently, I am interested in understanding adversarial robustness of neural networks, training neural networks with smaller number of examples without sacrificing accuracy and in understanding the generalization power of deep networks.