Improved block truncation coding using Hopfield neural network - Electronics Letters

We believe this is the first report that demonstrates that the inclusion of physically realistic intrinsic PM-HEMT noise parameters into an accurate PM-HEMT equivalent circuit predicts not only the noise figure, but also the optimum noise impedance and noise resistance, giving excellent agreement with experiment over a wide frequency range, when fitted at only one frequency. Previous efforts to compare noise models with experiments have generally concentrated on just calculating the minimum noise figure at spot frequencies,’ using equivalent circuits which are too simple to be able to simulate a realistic PMHEMT. For example, in the original paper by Pucel,’ the gate-drain capacitance and output conductance and capacitance were not taken into account. The results of this work suggest that it should be possible to go directly from a physical noise model (which requires input layer thicknesses, doping densities, mobilities, saturation velocities etc.) to accurate predictions of noise figure, optimum noise impedance, and noise resistance, provided that a reasonably accurate equivalent circuit model of the PM-HEMT is known. The reason for not using such an approach here is that the numerous parameters entering the physical model, and their uncertainties, are such that, a t present, it is not possible to calculate P, R, and especially C, to the accuracy required to fit to experiment. Nevertheless, the results of this work suggest that such an approach is feasible in principle. (The reason why the correlation coefficient C needs to be known to high accuracy is that it appears in the noise expressions as the factor (1 C’)), and because C is likely to be in the range 0.74.95, small differences in C can result in large differences in the predicted noise figure, optimum impedance, etc.). In conclusion, we have demonstrated the feasibility of going from a physical noise model, via a realistic equivalent circuit (including gate-drain capacitance), to predicted values of noise figure, optimum noise impedance, and also noise resistance. In addition, we have shown that this approach is capable of giving an excellent fit to experimental measurements of the four noise parameters over a wide frequency range. It is also possible to investigate, using the same intrinsic noise parameters, the effect of the parasitics on the noise performance, and so use the noise model in a CAD role to predict the optimum PM-HEMT layout for low noise performance.