| The main goal of this thesis is to implement noise cancellation as a pattern recognition problem by using a radial basis function network because the network can converge much faster than other standard feed-forward networks such as Backpropagation network. The input vectors are segments of a known signal with noise. The noise can be normally distributed random numbers or real background noise recorded in a room and tested with the sound card through a desktop computer and the output vectors are the same segments of the known signal. After training the neural network, the network will be working as a noise filter. After the training phase, the radial basis network will be tested with segments of the known signal contaminated with another noise. The convergence of the radial basis networks will also be evaluated with the CPU time, and the signal-to-noise ratio at the output. |
