RESUMO
Nuclear quadrupole resonance is a spectroscopic technique that is dependent on the measurement conditions and one of its challenges is to find the optimal excitation sequence parameters that ensure the highest signal-to-noise ratio. This is typically performed empirically and this paper proposes a different approach by using black-box optimization techniques. This work explores several optimization algorithms in a black-box environment, where no model is required for the spectrometer and measurement conditions. The techniques can be applied to any excitation sequence, they can adapt to the measurement conditions and enable the automatic and simultaneous optimization of multiple parameters. Single and multi-parameter optimization experiments are performed on sodium nitrite and the algorithms are shown to converge to similar optimal values with small standard deviations. The use of an optimized excitation sequence is shown to improve the signal-to-noise ratio by approx. 8 dB compared to the initial sequence. Further research directions are discussed and include the study of different acquisition functions and the use of reinforcement learning algorithms.