Parts concerned with the process

• Gathering Data: sampling technique
• Signals Processing: depends on how good sampling technique is
• Neural Network: Extracted features from signal processing

Signal Processing:

Feature Extraction: ICA, FFT, AR

Classification: amplitude, topography, latency and frequency

Once done, feed all data into Neural Network

Refinement for Final Project Plan... # 3

This approach actually addresses the main problem of the project. Since the problem falls in the following two categories

1. Ranges/amplitude and different signal characteristics of the same action are not similar and/or do not seem to have anything in common

· We don't know what a signal means by normal testing.

2. Number of signals available is small and we have only 1 channel record.

· Cannot generalize an algorithm that would work for all of the signals

In order to over come this problem. I created a neural network for each signal set. Called XXXSignalRange where XXX = EXT, RET, PRO, SUP

To understand further, I'll clarify by an example;

EXTSignalRange

Testing phase:

Takes as a correct input all 20 EXT readings

Takes as an incorrect input all 60 non-EXT.

Training phase:

Rejects three averages as incorrect

Accepts EXT average as correct

------------------------------------

This will happen for each signal.

Thus we would have a network of neural network that predicts what each signal's features would be and what to expect and what not to…

This would give us a better classification so that we can later work with the biggest neural network.