In Section 3.8, we have seen that the ordinary first-order modified Volterra model is a general purpose model both in terms of excitation signals and circuit structure; however, as a price to pay for its generality, it tends to be limited to mildly nonlinear applications. In Section 3.9, we reduced the generality of the model to make it more effective for highly nonlinear operations. The model obtained then assumes either a circuit structure with negligible long-term memory effects or a constant-envelope excitation signal. An important feature of the two models is that they require only conventional and simple frequency domain simulation or measurement techniques for the kernels extraction. In this section we will relax the constraint for conventional measurement techniques and look for a model that remains a general purpose one, but that also performs well in mildly nonlinear as well as in highly nonlinear operation.
As indicated in Section 3.5, the long-term memory mechanism mainly originates from low-frequency dispersion effects in the transistors' bias networks and electrothermal effects. These effects generate a modulation of the transistors' operating points with an index depending on the dynamics and the power of the input signal. Moreover, the low-frequency dispersion effects mentioned above can only be generated by the nonlinear behavior of the active devices. It is the nonlinear characteristics of the active devices that primarily generate the energy at beat frequencies that later on propagates through the biasing or thermal network.
TABLE OF CONTENTS 
