Related Experiment Videos
A complex-valued nonlinear neural adaptive filter with a gradient adaptive amplitude of the activation function
Andrew I Hanna1, Danilo P Mandic
1School of Information Systems, University of East Anglia, Norwich, NR4 7TJ Norfolk, UK. aih@sys.uea.ac.uk
Summary
A new complex-valued adaptive amplitude nonlinear gradient descent algorithm enhances neural network filtering for complex signals. This method outperforms standard algorithms in predicting dynamic, nonlinear data.
Area of Science:
- Signal Processing
- Machine Learning
- Artificial Neural Networks
Background:
- Nonlinear adaptive filters are crucial for processing complex signals.
- Existing complex-valued gradient descent algorithms face challenges with highly dynamic data.
Purpose of the Study:
- To introduce a novel complex-valued adaptive amplitude nonlinear gradient descent (CAANGD) learning algorithm.
- To improve the performance of nonlinear neural adaptive filters for complex-valued signals.
Main Methods:
- Development of a CAANGD algorithm modifying the activation function's amplitude.
- Implementation of a finite impulse response (FIR) nonlinear neural adaptive filter.
- Simulation using complex-valued, colored, and nonlinear input signals.
Main Results:
- The CAANGD algorithm demonstrated superior performance compared to the standard complex-valued nonlinear gradient descent (CNGD) algorithm.
- Effective prediction of complex-valued colored and nonlinear signals was achieved.
- The adaptive amplitude feature proved beneficial for signals with rich dynamical behavior.
Conclusions:
- The proposed CAANGD algorithm offers significant advantages for adaptive filtering of complex dynamic signals.
- This advancement in neural adaptive filters enhances signal prediction accuracy.
- CAANGD is a promising approach for applications involving complex-valued data analysis.