Abstract
A very common problem in signal processing is parameter estimation
of exponentially damped sinusoids from a finite subset of noisy observations.
When the signal is contaminated with colored noise of unknown power
spectral density, a cumulant-based approach provides an appropriate
solution to this problem. We propose a new class of estimator, namely,
a covariance-type estimator, which reduces the deterministic errors
associated with imperfect estimation of higher order correlations
from finite-data length. This estimator allows a higher order correlation
sequence to be modeled as a damped exponential model in certain slices
of the moments plane. This result shows a useful link with well-known
linear-prediction-based methods, such as the minimum-norm principal-eigenvector
method of Kumaresan and Tufts (1982), which can be subsequently applied
to extracting frequencies and damping coefficients from the 1-D correlation
sequence. This paper discusses the slices allowed in the moments
plane, the uses and limitations of this estimator using multiple
realizations, and a single record in a noisy environment. Monte Carlo
simulations applied to standard examples are also performed, and
the results are compared with the KT method and the standard biased-estimator-based
approach. The comparison shows the effectiveness of the proposed
estimator in terms of bias and mean-square error when the signals
are contaminated with additive Gaussian noise and a single data record
with short data length is available
Users
Please
log in to take part in the discussion (add own reviews or comments).