Say we have an SDR receiving some digital signal. The tuner will down convert the signal to be fed to the ADC, which digitizes the signal to be further processed by software.
The signal presents a non-repetitive waveform to the ADC which can't be analyzed with a standard Fourier transform. According to this article, to analyze a non-periodic waveform, several FFT blocks must be recorded and then averaged.
My concern is in demodulating the signal. If the end result is an average of the actual signal, then how do software programs correctly decode the signal? Do the number of samples in each FFT block matter, or the number of blocks averaged? Is it possible to recreate the signal with 100% accuracy?