What should be the architecture of the neural network?

Question

Neural networks

Task: the system takes measurement waveforms of a certain environment in which it is possible to find some object (which can be scaled) against the background of noise that moves with a certain direction. Each of the waveforms can correlate quite well with the previous one. In terms of meaning, the task is very similar to the analysis of motion in an image, but for a one-dimensional case with a non-trivial signal shape.
Is it possible to build such a neural network that, regardless of the scale of the object, will determine its speed and direction of the plane? Reading Haykin's book. Neural networks, I have seen references to the use of neural networks for digital filtering and measurement systems.
Has anyone come across an example of such a neural network for measurement tasks?

Answer 1

www.pereplet.ru/nauka/Soros/pdf/9812_105.pdf
The main question of this section is what
neural networks can compute. The answer is received: in a
sense, neural networks can do everything. Another question remains open
: how to teach them this?
According to the subject, you need to use some kind of transformation (say, Fourier) to highlight this object against the background, and then use the neural network as a "window" that can recognize the necessary "correlations".

Answer 2

A neural network (NN) will not solve the subject area for you. Give her input not raw, but encoded data after preliminary filtering in the form of a normalized interval from 0.0 to 1.0.
The data must be presented as a feature vector.
I propose to read the materials on the presentation of data for the National Assembly.