In experimental sciences, time domain noises and signals are not easily analyzed simultaneously without a proper model. In this document we name “signal” a deterministic continuous time data, giving an information; a “noise” has a random nature; a “measurement” is the collected data where the signal is buried or covered with noise. Mathematically, a measured deterministic signal is almost always continuous with time variable t, and its Fourier transform also continuous; whereas the transform of a noise is not continuous nor derivable at any point on the frequency axis f. These mathematical struggles are properly handled with concepts such as power spectral density, signal-to-noise ratio, matched filtering. We present the not-well-known definition of a time domain noise by Cramér with a Stieltjes-Fourier integral; we recall the power spectral density estimator discovered by Thomson, not spread enough, except in the geophysics community. We will see that the signal-to-noise is not necessarily easy to appreciate on time domain measurements. This document will not address the noises specific to digitization. We will only consider linear time-invariant systems measuring a scalar real valued data. This document aims at improving the knowledge of hopefully useful models for students dealing with experimental devices. This document is strongly influenced by the professional practice of the scientific communities in which I participate: gravitational wave detectors, instrumentation with systems incorporating microwaves and lasers, noises in instruments, optoelectronics instrumentation design.