Oscillation amplitude considerations in frequency domain experiments

Issues to consider when setting oscillation amplitudes in frequency-domain experiments are discussed, with particular reference to gradient direction estimators. Three factors which influence the size of the oscillation amplitudes for frequency-domain experiments are discussed. To ensure input feasibility and reduce higher degree term effects, the amplitudes should be set as small as possible. To reduce the noise effects, the amplitudes should be set as large as possible. It may not be possible to set oscillation amplitudes which satisfactorily address all three constraints. This situation tends to arise when experiments are run with the inputs close to a local optimum. For such situations, the priority of these constraints should be feasibility, higher degree terms, and noise effects. Thus, the oscillation amplitude should be set as small as possible so that the inputs remain feasible and higher degree term effects do not dominate the power spectrum values at linear term indicator frequencies.