EPFL doctorate Award 2016 – Ivan Dokmanić
Listening to Distances and Hearing Shapes: Inverse Problems in Room Acoustics and Beyond, EPFL thesis n°6623 (2015).
Thesis director: Prof. M. Vetterli
“For the development of the theoretical, algorithmic and experimental framework for the solution of acoustic inverse problems using tools of Euclidean distance geometry, thus opening new fields of investigations and applications.”
Bats navigate by listening to echoes, and so do pop icons: in a 2008 Hollywood blockbuster, Batman’s crew uses echoes recorded by the cellphones of unwitting citizens to visualise spaces across the wire. In reality, very much like in fiction, shapes and echoes are inextricably linked. We exploit this link to do interesting and perhaps unexpected things like hearing shapes of rooms.
Imagine an omnidirectional bat who pilots indoors by listening to echoes of its chirps, without any idea where these echoes are coming from. To divine the room’s geometry, this improbable critter must assign each echo to the right wall—that is, it must sort the echoes. We start by developing echo sorting algorithms, and continue to show how to use them to reconstruct shapes of classrooms and cathedrals, localise many microphones in unknown rooms by a finger snap, localise with a single omnidirectional sensor, and localise behind walls. Such feats are beyond the reach of traditional methods. We also show how echoes make for improved speech acquisition with “acoustic rake receivers”. The main idea is to listen behind undesired sources by listening to echoes of the desired ones. Our method—to befriend echoes—diverges from conventional approaches which struggle to do away with them.