Modeling of complex-valued wiener systems using B-spline neural network.

Area of Science:

• Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Complex-valued Wiener systems are crucial in modeling nonlinear systems, particularly in communication engineering.
• Accurate modeling of the nonlinear static function within these systems is challenging.

Purpose of the Study:

• To introduce a new complex-valued B-spline neural network (CVBSNN) for modeling complex-valued Wiener systems.
• To represent the complex-valued nonlinear static function using tensor products of univariate B-spline neural networks.

Main Methods:

• The complex-valued nonlinear static function is modeled using tensor products of two univariate B-spline neural networks, utilizing the real and imaginary parts of the input.
• Parameter estimation is performed using the Gauss-Newton algorithm, incorporating the De Boor algorithm for B-spline curve and derivative recursion.
• A least squares method is employed for initial parameter estimation.

Main Results:

• The proposed CVBSNN effectively models complex-valued Wiener systems using observational input/output data.
• The method demonstrates high accuracy in representing the nonlinear static function.
• Successful application to a nonlinear high-power amplifier model in communication systems validates the approach.

Conclusions:

• The developed complex-valued B-spline neural network provides an effective tool for modeling complex-valued Wiener systems.
• The tensor product approach for nonlinear function representation is efficient and accurate.
• The method shows significant potential for applications in communication systems and other fields involving complex-valued nonlinear dynamics.