Coded Compressed Sensing with Applications to Wireless Communications
Speaker: Lorne Applebaum
Series: Final Public Orals
Location:
Engineering Quadrangle J401
Date/Time: Friday, August 10, 2012, 2:00 p.m.
- 3:30 p.m.
Abstract:
Compressed sensing is a new paradigm that exploits the sparsity of
signals to reduce the number of measurements required to recover a
representation. This is accomplished using the general notion of
inner-products as measurements, encapsulated in "measurement
matrices." In this work, we focus on designing both measurement
matrices as well as compressed sensing recovery algorithms. We
consider several measurement code designs and recovery schemes with
applications to particular systems. First, we consider chirp-coded
compressed sensing measurements which, with a jointly designed
recovery algorithm, is designed for computationally efficient
recovery. For M measurements, O(M log M) recovery is possible, a
significant speed improvement over conventional random signals and
recovery methods. Subsequently, we consider OFDM channel estimation
in the context of compressed sensing and note that measurement
matrices are restricted to the form of sub-Fourier matrices. We
provide a method to manifest suitable matrices for recovering sparse
channels by deterministically selecting a few pilot tones. Next, we
consider how the compressed sensing paradigm can be used to build
novel wireless systems. We design a muliuser detection scheme for
random access on asynchronous channels. For this system, we develop
new compressed sensing recovery theory and design a codebook suitable
for the recovery of sparse sets of active users. Finally, we design a
virtual full-duplex adhoc wireless network system using half-duplex
hardware. In the system, nodes use codes containing "listening
symbols" during which the devices sense the wireless channel. When
active nodes in the network transmit simultaneously, each node
inheirits a unique compressed sensing problem to recover the data from
its neighbors.
The work in this thesis shows how, with careful consideration of the
application, compressed sensing with coded matrices can provide great
performance improvements and novel system designs.
