Neural networks for pattern recognition
Neural networks for pattern recognition
Learning neural networks for detection and classification of synchronous recurrent transient signals
Signal Processing - Special issue on neural networks
Applied Neural Networks For Signal Processing
Applied Neural Networks For Signal Processing
Chaos and Fractals
Digital secure communication via chaotic systems
Digital Signal Processing
| Hi-index | 0.08 |
The paper presents a concept of model-based detection of coded signals on the example of 13-element Barker code signal, embedded in additive chaos. The process of signal detection consists of two stages: approximation of chaotic dynamics and decision making. Dynamic models of chaotic signals, considered in this paper, were created in the form of linear autoregressive models as well as in the form of non-linear feedforward neural networks (of several types). The accuracy of models in one step ahead prediction of chaotic signals is satisfactory, even for chaotic signals with fast changes of their values. The error between an observed signal and its model is passed as the input to the decision-making (detection) module.When the signal received is a composite of Barker code and chaos, its dynamic properties change rapidly in the periods of Barker code appearance. Thus the error between the signal and its model becomes significant, and that allows for successful detection of Barker code. In this paper the detection module is based on a neural network; various architectures of neural net-based detectors have been proposed and tested in numerical experiments. Numerical simulations presented in this paper show good performance of detection of Barker code embedded in chaos.Robustness of such a detection scheme was also examined: the neural detectors, trained for a specific energy ratio between Barker code and chaos (SNR ratio), turned out capable detecting Barker code in a relatively wide range of SNR ratios. Also the comparison between neural detection and a detection structure, using matched filter, has been presented. Experiments have shown superiority of neural detection over detection with a matched filter, especially for low SNR ratios. It should be also noted that very simple neural network architectures were proposed as the models of signal dynamics and for the detection module.