We investigate AM/PM distortion models and compare them with baseband (BB) Volterra models. We show that the AM/PM model can be considered a special case of memoryless baseband Volterra models, and that adding memory can improve modeling accuracy by allowing the simulation of more complex nonlinearities. We report models of an LNA, a downconversion mixer, an upconversion mixer, and one class-AB Power Amplifier. All circuits are simulated using the 45nm STMicroelectronics CMOS process with Virtuoso, while the PA, with discrete devices, is simulated using ADS. Adding memory improves performance at the expense of increased numerical complexity: this makes real-time simulation and real-time calibration more expensive, so that there is a trade-off between complexity and accuracy or linearity (after calibration). Foreground calibration’s techniques only require the real-time computation of the correction (inverse) system’s output, whereas background calibration also requires the real-time estimation of the model coefficients, so the relevant complexity is that which is required during correction.