How to submit input data to the neural network?

Question

Python

I don’t quite understand how to apply input signals to the neural network to analyze analog data from sensors. For example, there is a 16-bit ADC, which gives a range of values ​​from 0 to 65535. How to apply this signal to the input of the neural network? Convert this value to a binary number and represent each digit by a neuron? Or just take an array with a dimension of 65535 and fill it in the following way: for example, the value 200 = set the first 200 digits in the array to one, and the rest to zeros and feed the array as input.
The first option has fewer neurons, but will take longer to learn. The second one is larger. Maybe that's not how it's done at all?

Answer 1

The simplest thing is to convert to a fractional form and divide by the maximum. Then the range of values ​​will be from 0 to 1. Sometimes [-1, 1] is preferred, you can translate into this range as well.

Answer 2

A 16-bit audio signal is not suitable for input to the HC. It needs to be translated into frequency response and even beaten into some atomic readings like sounds or letters or musical notes. How to beat is a topic for a separate topic and should be studied separately. In your case, you cannot consider the neural network as a meat grinder where they throw whatever they get. Information must be discrete and meaningful. That is, if you train the NN with a teacher, then you must present the frequency response of the note DO to the first octave to the network, for example, as a training signal and give a hint that this is a note DO. If you feed white noise to the input, the network will not learn anything.