Fast complex adaptive spline neural networks for digital signal processing

We study a complex-domain artificial neural networks, called the adaptive spline neural network, defined in the complex domain, which is able to adapt its activation functions by varying the control points of a Catmull-Rom cubic spline. This kind of neural network can be implemented as a very simple structure which is able to improve the generalization capabilities using few training epochs. Due to its low architectural complexity the network can be used to cope with several nonlinear DSP problems at high sampling rate. In particular, we investigate the application of this new neural network model to the adaptive channel equalization problem. The goal is to design a receiver which compensates the high power amplifier nonlinearities in digital radio links and performs the symbols extraction from the received data (demodulation process), when a 16-QAM is used.