A wavelet based neural network procedure for the field reconstruction of NMR main magnets

A novel procedure to identify and correct the field inhomogeneities in Nuclear Magnetic Resonance (NMR) devices is proposed. Instead of local field measurements, the procedure makes only use of the information supplied by the standard output (absorption spectrum) of the device to determine the correction field to be supplied to the magnet. The reconstruction of the actual field map starting from the spectrum is carried out by means of a Neural Network (NN) approach. The inputs of NN are, in turn, the moments of the spectrum or a pattern of features extracted by a wavelet pre-processor. The accuracy achieved in correcting the field is acceptable and the computational burden of the operation phases appears to be smaller with respect to other methods. The nature and shape of the wavelets are particularly suitable for treating the asborption spectra signal. The procedure promises to be a possible alternative tool to the frequently needed routine calibrations of the magnetic field in Nuclear Magnetic Resonance devices.