Learning Wireless Networks’ Footprints and Topologies in Shared Spectrum

Speaker: Mihir Laghate
Affiliation: Ph.D. Candidate - UCLA

Abstract:  The increasing demand for wireless connectivity and the scarcity of spectrum for exclusive use has popularized the idea of sharing spectrum between multiple communication systems. For a network to estimate its own link budget while avoiding interference from neighboring, or incumbent, networks, it is necessary to learn the spatial, spectral, and temporal usage patterns of its neighboring radios.  Instead of modifying each standard for coexistence with another, we can, as proposed by Mitola and Haykin, build learning or cognitive abilities into the radios themselves in order to achieve this coexistence irrespective of the standards used by the incumbent radios. In this work, we propose methods for such a cognitive network to cooperatively learn the spatial occupancy of incumbent radios and the topologies of the incumbent networks. In contrast to the significant body of work on spectrum sensing, this work studies the problems of detecting, distinguishing, and coexisting with multiple communicating incumbent networks rather than that of avoiding interference to a single broadcasting transmitter. Our methods are designed to make these inferences without prior knowledge of the number of incumbent radios, their locations, network topologies, transmission protocols, and channel models of the ambient wireless environment. We also do not require knowledge of the locations of the cognitive radios (CRs).

We begin by proposing algorithms to learn the footprint of each incumbent transmitter, i.e., the sets of CRs that receive signals from that incumbent transmitter. By learning the Gaussian mixture distribution of the received energy vector, we show that multiple transmitters’ footprints can be learnt irrespective of their spatial overlap and potentially anisotropic shape. Learning the footprints also enables sampling the activity of each incumbent radio. By identifying radios that transmit as a response to the transmission of another, we learn the causal links between pairs of incumbent radios, i.e., the topologies of the incumbent networks.  Thus, we can identify the potential receivers when a particular incumbent radio is transmitting. Hence, we can identify the communication links that can be established without causing interference to the incumbent receivers. Thus, our work is important for the coexistence of communicating networks with arbitrary geographical coverage such as LTE-Unlicensed and WiFi, radars and WiFi, etc.  Our inferences can also be used to improve sensing schedules, access strategies, and routing in ad hoc networks.

Biography:  Mihir Laghate received his Bachelor of Technology degree in Electrical Engineering with a minor in Computer Science from the Indian Institute of Technology Bombay in 2012. He also received his Master of Technology degree in Communication and Signal Processing from the Indian Institute of Technology Bombay in 2012. He has been pursuing his PhD at the CORES lab at UCLA since then. His research is focused on the application of machine learning algorithms to wireless systems. He received the department fellowship in 2012 and the Guru Krupa fellowship in 2013.

For more information, contact Prof. Danijela Cabric (danijela@ee.ucla.edu)

Date/Time:
Date(s) - May 17, 2017
8:00 am - 10:00 am

Location:
E-IV Tesla Room #53-125
420 Westwood Plaza - 5th Flr., Los Angeles CA 90095