TY - GEN
T1 - An adaptive optimal-kernel time-frequency representation
AU - Jones, Douglas L.
AU - Baraniuk, Richard G.
N1 - Funding Information:
This work was supported by the National Science Foundation, Grant No. MIP 90-12747,the Joint Services Electronics Program, Grant No. N00014-90-J-1270, and an NSERC-NATO postdoctoral fellowship.
Funding Information:
Time-frequency representations are a central signal analysis tool for a wide variety of applications. A large number of bilinear time-frequency representations (TFRs) have been proposed, each differing only in the choice of a kernel function [l]. Each bilinear distribution in Cohen’s class can be interpreted as the two-dimensional Fourier transform of a weighted version of the symmetric ambiguity function (AF) A(8,r) of the signal to be analyzed. That is, if P(t,w) is a bilinear TFR, then where the weighting function @(e, r) is called the kernel. In recent years, it has become apparent that no single kernel can give adequate performance for a large class of signals; hence, there has been increasing interest in signal-dependent or adaptive time-frequency representations. Arguments for signal dependence and surveys of the related literature can be found in [2]. A number of signal-dependent time-frequency representations have been developed. Among the most promising are those based on optimality criteria, including a signal-dependent time-frequency representation based on radially This work was supported by the National Science Founda- tion, Grant No. MIP 90-12747, the Joint Services Electronics Program, Grant No. N00014-90-J-1270a,n d an NSERC-NATO postdoctoral fellowship. R. Baraniuk is on leave from Rice University, P.O. Box 1892, Houston, Texas 77251-1892, USA.
Publisher Copyright:
© 1993 IEEE
PY - 1993
Y1 - 1993
N2 - Signal-dependent time-frequency representations perform well for a much wider range of signals than any fixed-kernel distribution. However, current signal-dependent representations are generally block-oriented techniques unsuitable for on-line implementation. The time-frequency representation presented here, based on a signal-dependent radially Gaussian kernel that adapts over time, tracks signal component variations over time and supports on-line implementation for signals of arbitrary length. The method uses a short-time ambiguity function for kernel optimization and as an intermediate step in computing constant-time slices of the time-frequency representation. While somewhat more expensive than fixed-kernel representations, this technique often provides much better performance.
AB - Signal-dependent time-frequency representations perform well for a much wider range of signals than any fixed-kernel distribution. However, current signal-dependent representations are generally block-oriented techniques unsuitable for on-line implementation. The time-frequency representation presented here, based on a signal-dependent radially Gaussian kernel that adapts over time, tracks signal component variations over time and supports on-line implementation for signals of arbitrary length. The method uses a short-time ambiguity function for kernel optimization and as an intermediate step in computing constant-time slices of the time-frequency representation. While somewhat more expensive than fixed-kernel representations, this technique often provides much better performance.
UR - http://www.scopus.com/inward/record.url?scp=84941872712&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84941872712&partnerID=8YFLogxK
U2 - 10.1109/ICASSP.1993.319606
DO - 10.1109/ICASSP.1993.319606
M3 - Conference contribution
AN - SCOPUS:84941872712
T3 - ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
SP - 109
EP - 112
BT - IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 1993
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 1993 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 1993
Y2 - 27 April 1993 through 30 April 1993
ER -