Department of Chemistry, Cambridge University, Cambridge, U.K.

Radiofrequency pulses specifically tailored to generate a particular frequency profile are increasingly used in high-resolution NMR spectroscopy and magnetic resonance imaging. The search for these 'designer' pulses makes use of an entire family of numerical techniques, ranging from the Fourier transform approximation through the standard gradient descent methods, linearization of the Bloch equations, average Hamiltonian theory, simulated annealing, and evolutionary methods. The authors propose the use of neural networks to address this important problem; this technique has already been successfully used in NMR for spectral recognition problems.