A metric for assessing the degree of device nonlinearity and improving experimental design

Impact of sampling domain and number of samples on the accuracy of large-signal multisine measurement-based behavioral model

“…As was shown in [8], the behavioral model extracted from 20 RF trajectories equidistantly sampled in the r-domain achieved higher accuracy of the third-order intermodulation distortion product prediction than the corresponding model based on 20 temporally-equidistant trajectories. There is however no information about the performance of both models when used to predict different types of signals.…”

“…In the nonlinearity metric piecewise uniform sampling method of [8] the RF carrier frequency cycles are selected basing on a nonlinearity metric calculated for many RF periods along the modulation envelope period. A preference is given to the RF periods associated with higher values of the nonlinearity metric, what resulted in an improved behavioral model accuracy [8]. The nonlinearity metric r is defined by (1) as the root-mean-square (RMS) of orthogonal distances between normalized trajectories b 2 ðt q ; t r Þ and the normalized reference (linear) trajectorỹ b 2 ðt q ; t ref Þ…”

“…However, the efficacy of this approach largely depended on the shape of the multisine's envelope, e.g., more dense sampling was required for signals with high crest factor [7]. Alternative sampling of the measured data was introduced in [8]. Authors showed that sampling the data equidistantly not in time-domain but in the nonlinearity metric domain resulted in improved model's accuracy.…”

“…Authors showed that sampling the data equidistantly not in time-domain but in the nonlinearity metric domain resulted in improved model's accuracy. However in [8], only limited analysis was performed in the modulation band of the RF carrier frequency.…”