Implementation Efficiency Comparison Between ANN and the Cardiff Model in ADS

This paper presents CAD implementations of the Artificial Neural Network (ANN) model and the Cardiff Model (CM) in Keysight Advanced Design System (ADS). A comparative analysis of running time difference between the two behavioral models is conducted. The results indicate that the removal of certain dispensable calculations in the CM template can save 99% of time on large dataset simulations at the fundamental frequency, while maintaining a Normalized Mean Square Error (NMSE) level lower than -50 dB. The modified CM exhibits a similar runtime compared to the ANN model with selected structure. This paper also explores the impact of changing the number of the CM coefficients and hidden neurons in the ANN model on the simulation time in ADS at fundamental frequency, without considering the accuracy of models. Results reveal a linear relationship between the simulation time and the number of the hidden neurons in the ANN model whereas the simulation time of the CM is more related to the exponents for |a
2,1
|.