Summary: | 碩士 === 國立中山大學 === 電機工程研究所 === 82 === The problem of estimating the frequency content of signals is
very important in many digital signal processing applications.
In this thesis, we are concerned with the problem of estimat-
ing and tracking the instantaneous frequency of the sinusoidal
signal together with additive white noise. Its solution has im-
portant applications in the fields of vibration measurements,
Doppler radar returns, passive sonar systems, and formant
frequency estimation of speech signals. In this thesis, a new
algorithm for IFE is developed. To do so, the forward linear
prediction filter is employed. In consequence, the modified
Trench's method along with the Bauer-Fike theorem is proposed
for solving the principal eigenvalues of the Hermitian Toepli-
tz autocorrelation matrix for instantaneous frequency estima-
tion (IFE).In fact,three kinds of eigenvalue searching schemes
can be employed in the modified Trench's method. In the new
algorithm for IFE, the modified Trench's method is first used
for solving the principal eigenvalues for initial block of
data with length N. When a new data is received,the Bauer-Fike
theorem is applied to search the new eigenvalues based on the
previous obtained eigenvalues.Such that the computational cost
can be reduced. The performance of the IFE using the presented
methodis compared with the conventional LMS adaptive method as
well as the QR based method. From the simulation results, we
found that the presented method can perform as good as the
QR based method, in terms of multiple frequencies estimation
where the frequencies are closer.But in the same situation the
conventional LMS adaptive method may not perform satisfacto-
rily. Moreover, the computational complexity of the presented
method is much less than QR based method, especially when the
presented method is implemented by the parallelized structure.
|