Constructing memory-polynomial models from frequency-dependent AM/AM and AM/PM measurements

Nonlinear behavioral models are advantageously used in system-level simulations to predict the generated distortion without the full computationally complexity of transistor- based circuit simulations. Such models are often described by two static nonlinearities called AM/AM and AM/PM-conversion. Because the modeling with purely static nonlinearities is not sufficient for wideband applications (no memory effects are considered), we introduce the concept of frequency dependent AM/AM and AM/PM surfaces, which can be easily determined from two-tone measurements by sweeping both, the input signal magnitude and the modulation frequency of the RF power amplifier (PA). With the obtained two-dimensional functions, we are able to construct a memory-polynomial model (simplified Volterra model, where the off-diagonal kernels are zero) by solving simple linear equation systems.