Systems Seminar
Dimension Reduction via Subspace Mapping:
When Less is More in Signal Processing
Prof. Barry Van Veen
UW ECE Department
Abstract
Reducing dimension by mapping data into an appropriately chosen subspace is
a powerful signal processing principle for reducing computational
complexity and improving adaptive performance. These benefits usually
accrue at the expense of steady-state performance. The key issue in
managing the computational complexity and performance tradeoffs is
selection of the subspace. In general, computational complexity and
adaptive performance depend primarily on the subspace dimension, so the
subspace is chosen to minimize the steady-state performance loss. These
principles are illustrated using examples from Volterra filtering, adaptive
detection, and adaptive beamforming. The talk concludes by emphasizing
recent work in subspace mapping for space-time code division multiple
access based wireless communication systems. The subspace framework
provides an ideal vehicle for satisfying the stringent performance and
complexity requirements of anticipated high data rate mobile systems. An
appropriate subspace for signal processing of the received signal is
identified and tradeoffs between steady-state detection performance,
computational complexity, and adaptive performance are examined.
Time and Place: Wed., Feb. 24, 3:30-4:30 pm in 4610 Engr. Hall.
SYSTEMS SEMINAR WEB PAGE:
http://www.cae.wisc.edu/~gubner/seminar/