Documente Academic
Documente Profesional
Documente Cultură
J A Crowe
Abstract
This paper aims to provide a relatively simple introduction to the field of wavelets and their relevance to the
pn>cessing of biomedical signals. ?he extensive bibliogmphy contains both rewmmended material on w a v e h
and papers on their application to a variety of biomedical signals for differing purposes (e.g. analysis ami
compression).
Use of the STFT allows a plot of the frequency content of a signal against time to be produced. ?Ibe
spectrogram, as it is known (and the equivalent plots produced by other time-frequency transforms) ax
valuable tools for the analysis of non-stationary signals. However, the use of a fixed length window means
that the resolution of the spectrogram, in both time and frequency, is fixed. This is often referred to as tfie
time-frequency plane being tiled by fixed size rectangular windows. The STFT is often used as the starting
point for many introductory articles on wavelets because one feature of the wavelet transform is the diilkrent
way in which it tiles the time-frequency (strictly time-scale for wavelets) plane.
The wavelet transform 0 has many similarities with the STFT but is fundamentally different in that its
basis functions (the wavelets) are not of fixed length. Rather the wavelets (literally little waves) are derived
by stretching (in time) a base wavelet to give the full wavelet basis consWng of dilated versions of the base
funclion. This repeated stretching to produce the wavelet basis obviously produces functions with decreasing
frequency (strictly scale) thus allowing the fresuencycantent of a signal to be analysed. However, the increase
in length of dilated wavelets renders them progressively less able to localise the p d s e timing of events within
the signal. This feature of the WT manifests itselfin the tiling (which is related to the time and frequency
resolution) of the time-frequency plane by prwiding good time location at high fquencies; with improved
frequency discriminaton at lower frequencies at the expense of poorer time location. Ccmsequently, the tiles at
low frequencies are shorter and wider (in frequency and time respectively)thanthe taller and narrower ones
Department of Electrical & ElectronicEngineering,University of Nottingham, Nottingham, NG7 2RD, UK.
211
at high frequencies.
The STFT's spectrogram and ifs WT equivalent, the scabgram, produce continuous the-frequency plots. An
additional, and highly significant, feature of the WT is that by using orthonormal wavelets it is possible to
effectively sample the scalogram thus producing a non-discrete transform (i.e. N input paints
produce N wavelet coefficients). "he required sampling is dyadic (it involves factors of two) which ties in
with the repeated dilation of the base wavelet by factors of two to produce the en& basis set.
One view of the discrete WT is of a transform whose coefficients detail the magnitude of the basis wavelets
(which are functions limited in both time and frequency domains) required to reconstruct the Signal under
investigation. An altemative, yet equivalent, view is that the WT analyses the frequency content of the
incoming signal via a filter bank. The link between these is that the wavelets (the basis functions of lhe
transform), are intimately linked to the filtering action of the filter-bank. Consequently, a long, low scale
wavelet gives poorer time location but better frequency location (i.e. the ccwrepmg filter is narrow) than a
short high scale wavelet with the opposite characteristics.
A &xil point to note is that unlike most transforms, whose set of basis functions are fixed t h m are a
multitude of different wavelets that can be used. "hese are usually named after their inventor (e.g. Morlet and
Daubechies). Differences between these wavelets relate to factors such as their length, smoothness and their
c o m p d i n g filtering characteristics. However, in most engineering applications it is a nidatively mall
subset that are fresuentlyused.
2/2
The above serves as a brief inmduction to the WT and its potential value as a Signal proceshg tooL Ihe
following bibliogmphy provides material which expands upon these basic ideas, and is grouped as
appropriate. Note that several of the refemus refer to whole issues dedicated to wavelets rather than
individual aIticles.
Acknowledgements
The author wishes to thankJon Hall for our many discussions on wavelets during the course of his PhD.
Bibliography
e--oms
Hlawatsch, F and Boudreaux-Bartels, GF: Linear and quadratic time-frequency Signal representationS, IEEE
Sig. F9ucMag.. Vol9, pp 21-67, Apr, 1992.
Cohen, L Time-frequency analysis, pub. prentice Hall, ISBN 0-13-594532-1,1995.
b t r O d U c t 0 ~taid!s
fioul, 0 and Vetterli, M: Wavelets and signal processing, IEEE SigPmcMag., Vol8, pp 15-38, Oct, 1991.
Bentley, PM and McDonnell, JTE: Wavelet transfoms: An introduction, IEE Elec.& Comm.Ebg.Jnl., Vol6,
Pp 175-186, AUg, 1994.
Akay, M: Wavelets in biomedical engineering, Ann. of Biomedag., Vol23, pp 531442,1995.
Kovacevic, J and Daubechies, I, Eds.: Special issue on Wavelets, Proc.IEEE, Vol84, No 4, Apr. 1996.
Books including basic theory and applications
Ruskai , MB et al.., Eds.; Wavelets and their applications, pub. Jones and Bartlett, Boston, MA, ISBN 086720-2254,1992.
Wickerhauser, M V : Adapted wavelet analysis fIDm thary to software,pub. Peters, Wdledey, MA, ISBN 156881-041-5.1994.
Combes, JM et al., Eds.:Wavelets: 'ZEme-freqUencymethods and phase space, 2nd ed pub. Springer-Verlag,
ISBN 0-387-53014-2,1990.
Chui, CK, Ed.: Wavelets: A tutorial in theory and applications, pub. Academic Press, 1992.
Other books and articles
Daubechies, I: Ten lectures on wavelets, pub. SIAM,ISBN 0-89871-274-2.1992.
Akansu,AN and Haddad, RA: Multiresolution Signal Decomposition, pub. Academic Press, ISBN 0-1204714O-X, 1992.
Vaidyanathan, PP Quadrature mirror filter banks, M-band extensions and perfect reconstruction techniques,
IEEESig.Proc.Mag., Vol4, pp 4-20,Jul. 1987.
Strang, G and Nguyen, T Wavelets and filter banks, Wellesley-Campbridge Press, Wellesley MA, 1996.
Biomedical ADFlicatiUm
-we,
JA et al.: Wavelet transform as a potential tool for ECG analysh and compression, J.Biomed.Eng.,
V0114, pp 268-272.1992.
M y , M,Guest Ed.: Special issue on time-frequency and wavelet analysis, IEEE Eng.Med.Biol.Mag., Vol
14, No 2, Mar/Apr, 1995.
Akay, M, Guest Ed.: Special issue on wavelets for Image Analysis, IEEE Eng.Med.Biol.Mag., Vol 14, No 5,
Sept/Oct, 1995.
Unser, M and Aldroubi, A: A review of wavelets in biomedical applications,k . I E E E , Vol84, pp 626-638,
Apr. 1996.
Hall, JM and Crowe, J A Wavelet packets as an appmxhation to the KLT for ECG compression,
App.Sig.Proc., In press, 1997.
2/3