Noise reduction based on local linear representation using artificial neural networks

A new noise reduction algorithm is presented which cleans noise contaminated chaotic sample data by applying artificial neural nets. The algorithm is attractive in real time applications including communication systems that exploit chaotic signalling. In the last decade chaotic coding and chaotic-based communication have been proposed based on discrete maps or based on continuos dynamical systems mostly implemented in form of electric circuits. A major drawback of the proposed chaotic coding strategies is their poor performance with respect to signal reconstruction in the presence of noise. Several investigations have been made on noise reduction but in an iterative manner and so are not applicable in real time applications. The novel proposed algorithm achieves a SNR gain of 5.9 dB independent from the actual noise level in only one step.