A maximum likelihood type nil-estimates of the time-invariant amplitude and
time-varying frequency are developed for observations of a complex-valued
harmonic contaminated by impulse random errors having an unknown heavy-tail
ed error distribution. The M-estimates are different from the standard shor
t-time periodogram by a nonquadratic loss function minimized in order to ob
tain the estimates. A choice of this loss function on the base of the Huber
's minimax robust statistics enables the M-estimates to be robust with resp
ect to the error distribution. Asymptotic analytical results are given for
the variance and bias of the estimates; The l(1)-periodogram is introduced
as a special class of the robust periodogram where the loss function is a s
um of absolute values of real and imaginary parts of residuals. The simulat
ion demonstrates an effectiveness of the proposed robust estimates. (C) 199
9 Elsevier Science B.V. All rights reserved.