i was reading about back propagation neural networks. When suddenly i realized they can be used for learning /or/ dealing with timed pattern signals.
So you could make something that puts in a missing parts of a signal Or something that raises an alert if a signal was not received. Predict the signal forward in time or delay. Or create negative signals, or different signal based on the input, etc The signals can be analog or digital, ofcourse these signals shouldnt be to fast (because the neural math is complex and needs time too perform the calculations)
I can explain a bit how it works in simple terms, a BP neural network with 2 hidden layers for example 1 input pin (digital or analog) for the network, and a timed interval in which you shift the signal pattern like this : the first signal arives at pin x.. you wait 5 sec do a new reading and put the previous reading to the next first layer node of the neural net. wait 5 sec and then again shift them to next node in the first layer. Depending on how much nodes you put in the first layer of the neural network the complexer the signal understanding can be. (and you wont need that amount of physical pins..)
Ok well al nice that things can work like that, and its quite amazing actualy such higher math. But then i wondered, what could be done with it on a arduino. *What practical use could we make of something that can react on signal paterns ? *