Results (
Thai) 1:
[Copy]Copied!
For this simulation each input signal is sampled an arbitrarily chosen 64 times over a single period. Each sampled value is then a separate input to the network. As with the alphabetic character example, a separate output is provided for each signal (square, triangular, and sine). However, the signals with noise are represented by inputs, which are no longer two-valued (binary or bipolar). Thus, our network has 64 multi-valued inputs and 3 bipolar outputs. Thus, a square wave input gives us the outputs 1, -1, -1; a triangular wave results in -1, 1, -1; a sine wave produces -1, -1, 1. A MATLAB routine was written to train the network and then allow testing for arbitrary levels of noise. Portions of the output from several runs of the routine are shown below. The number of epochs was first increased until learning was complete. Then, the trained network was tested with decreasing signal to noise ratios, to see where the noise becomes too large to enable proper classification. Outputs from sample runs are shown below.
Being translated, please wait..
