Software analyzes complex, rapidly changing signals
Fourier transforms have long been used to study the frequency distribution of time-series data. But the Fourier transform can fail when frequencies change rapidly over a short time period (see "How Fourier and wavelet techniques compare," Vision Systems Design, Sept. 1996, p. 54).
Another method of signal analysis, the Wigner distribution, overcomes this problem but can introduce undesirable effects. At the Johns Hopkins University Applied Physics Laboratory (Baltimore, MD), Amir Najami has used the Wigner distribution to detect signals with rapidly changing content. To do so, he used IDL, a signal and image-processing package from Research Systems (Boulder, CO), to develop custom kernel functions to attenuate the nonlinear effects caused by the Wigner distribution. Najami runs IDL on a 753 workstation from Hewlett-Packard. To test the new kernels, Najmi used experimental ocean-current data and digitized recordings of whale songs. Each species of whale produces different signals that form a sound fingerprint. These signals provided Najmi with complex time-series data with rapidly varying frequency content. Processing the collected data using the Wigner distribution and kernal functions within IDL proved the usefulness of the new algorithm.
"IDL has opened up a method of rapid prototyping that will have implications beyond whale song," says Najmi. "Medical studies into the effects of drugs on patients` EEGs are one possibility," he says. ©
