ICSE 2008 Proc. 2008, Johor Bahru, Malaysia HEMT Transistor Noise Modeling using Generalized Radial Basis Function Mohsen Hayati , Ali Shamkhani , Abbas Rezaei , Majid Seifi Electrical Engineering Department Faculty of Engineering, Razi University Tagh-E-Bostan, Kermansshah-67149, Iran 0098918812041(Phone), 00988324274542(Fax), mohsen_hayati@yahoo.com , ali.shamkhani@yahoo.com, arezaei818@yahoo.com , majidsfy@gmail.com Abstract: In this paper, one important architecture of neural networks named a generalized radial basis function (GRBF) is applied in order to model HEMT Transistor Noise Parameters dependence on bias conditions such as dc drain-to-source voltage, dc drain-to-source current, frequency and S- parameters that can accurately predict transistor noise parameters in a wide frequency ranges for all bias points from the operating range including transistor S- parameters. Keywords: Generalized Radial Basis Function, HEMT Transistor, S-Parameters. I. INTRODUCTION Accurate and reliable noise models of microwave transistors are necessary for analyses and design of microwave active circuits that are parts of modern communication systems, where it is very important to keep the noise at a low level. Model development is basically an optimization process and can be time-consuming. Furthermore, measured signal and noise data for each new operating point are necessary for model development, which could take much effort and time, since these measurements require complex equipment and procedures [1, 2]. In many of these cases, neural modeling could be a good alternative to the classical modeling. Neural models are simpler and retain the similar accuracy. They require less time for providing response, therefore, application of neural model can make simulation and optimization processes less time-consuming, shifting much computation from on-line optimization to off-line training. Neural networks have been applied in modeling of either active devices or passive components, in microwave circuit analysis and design, etc. It has been proposed in microwave MESFET and HEMT transistor signal and noise performance modeling [3- 5]. In this paper, a Generalized Radial Basis Function (GRBF) network for HEMT transistor noise modeling is proposed. This network receives bias such as dc drain-to-source voltage, dc drain-to-source current, frequency and S- parameters as inputs and produces transistor noise parameters at its output. Therefore, bias conditions and frequency are inputs and minimum noise figures, magnitude of optimum reflection coefficient, angle of optimum reflection coefficient and normalized equivalent noise resistance are outputs of the neural network. A simplified overview of proposed ANN model is shown in Fig. 1. GRBF Model min F n R opt Γ opt ∠Γ 11 s ∠ 11 s 12 s ∠ 12 s 21 s ∠ 21 s 22 s ∠ 22 s ds v d i f Fig. 1 A simplified overview of ANN model. The GRBF network is a generalization of the RBF network, which allows to different variances for each dimension of the input spaces by replacing the radial Gaussian kernels with elliptical basis functions. The 475