Device for spectral analysis.

摘要

The signals coming from sources (10) are converted into digital form (14) and undergo a Fourier transformation in order to obtain their frequency transforms. The processing circuit (20) is configured to compute the periodogram of a signal, the cross-periodogram of two signals and to estimate a cross-spectral density matrix. Means (24) carry out optimal smoothing of the Log-periodogram or of the cross-periodogram of two whitened signals by computing an optimal weighting window. The estimated auto-spectra and cross-spectra are obtained after smoothing. By employing a recursive algorithm comprising estimates of auto-spectra and of cross-spectra, a set of signals can be decorrelated and whitened, leading to estimation of the cross-spectral matrix. The invention finds a particular application in the processing of electroencephalographic signals, in particular for brain mapping, the locating of sources of brain activity and the search for relationships between these latter. <IMAGE>