Bilinear time-frequency distributions have been widely utilized in the anal
ysis of nonstationary biomedical signals. A problem often arises where the
time-frequency components with small-amplitude values cannot be displayed c
learly. This problem results from a masking effect on these components caus
ed by the presence of high-energy slow waves and sharp patterns in the inpu
t which produce large values in the time-frequency distribution. These larg
e values often appear in the time-frequency plane as irregular patterns in
the low-frequency range (due to slow waves), and as wide-band, impulsive co
mponents at certain points in time (due to sharp patterns). In this work we
present an effective signal pre-processing method using a nonlinear operat
ion on wavelet coefficients. This method equalizes the energy of different
time-frequency components in the data so that the masking effect is greatly
reduced, while the original time-frequency features of the input signal ar
e preserved. Comparative experiments on electroencephalographic data with a
nd without using this method have shown a clear improvement in the readabil
ity and sensitivity in bilinear time-frequency distributions. (C) 2000 The
Franklin Institute. Published by Elsevier Science Ltd. All rights reserved.