(Scholarly Paper)
Simon G. Wood, Jr.
MS EE Candidate, ECE
Department
Advisor: Dr. Peter Paris
Thursday August 22, 2002,
1:00 P.M.
Source
separation had its early start in the area of independent component analysis
(ICA) where it is desirable to recover a set of "source signals" from
the observation of several mixtures of these signals. Blind source separation
is the problem of recovering all of the source signals without using prior
information about the channel and/or training sequence. In the context of
statistical array processing blind source separation translates to not having
any information about the array manifold vector. In the context of
communication systems blind source separation translates to not having any a
priori knowledge of multipath channel distortion or waveform training
sequences.
In
recent years techniques to exploit the finite alphabet, constant modulus,
cyclostationarity and other properties of digitally transmitted waveforms have
arisen. This paper addresses the problem of detecting and demodulating direct
sequence spread spectrum (DSSS) signals without knowledge or use of the
pseudo-noise spreading sequences. An overview and performance analysis of an
expectation maximization (EM) technique that exploits the finite alphabet
property is presented. The complexity and performance along with subspace
mismatch analysis of the EM-based technique is also presented.
Source
separation has become a hot research topic in such application areas as
wireless communications (capacity improvement), speech processing (separating
multiple voice conversations), medical diagnosis (separation of a mother's EKG
from that of the fetus's EKG) and remote sensing (low probability of intercept
communications).