"I don’t think I am going to be able to provide you the clarity that you need beyond convince yourself (with simulation if needed) that you can accurately measure the rms of a tone (as “signal”) and the noise density (multiply the rms density in v_rms/root Hz by the square root of fs to get “noise”. SNR=20log()" this one.

@DanBoschen , thank you. I definetely agree your last comment. I check the noise floor for signal generator. But for previous comment, I don't understand how to say last part. Any chance of adding a snippet or image of this to your answer? Or a link will be enough for me.

@MarcusMüller I edited the question. I hope I can tell in my mind. Thank you.

Is it normal for MATLAB to give the same SINAD value for both the signal multiplied and summed by certain coefficients and the collected data? Is it because it performs an optimization in the background that gives the same result? As I said, when I correct the data, while the enob I calculated changes, that is, the SINAD changes, how does MATLAB's sinad value not change? I think he should have given different sinad values. What do you think?

@MarcusMüller Okey, thanks for your comments. I saw the conflict, sorry about that. I shared code. Another different question. Do you think my calculation or approaching is reasonably adequate without little difference SINAD value?

I think gain and offset values ​​have no meaning for SNR. But these two values ​​should definitely affect SINAD. Because linearity should completely affect distortion.

Thank you for your very detailed explanation. I try to welch method. Obviously, I analyzed "sinad" function in MATLAB. This function use kaiser windowing, power spectral density and effective resolution bandwith. You are definitely right. I am confused this point. I apply gain coefficient and offset but sinad function said the same result. How is it possible? Because I calculated the different result for SINAD. Okey, you think, Is my calculations or approaching is correct without welch's method?

@MarcusMüller The signal generator has 24 bit DAC, enob 21 bits. Just I compare ref_sin and collecting data signal. I take FFT, I find phase value. I regenerate ref_sin using this phase value. These two signal are in phase. Simple and quick.

@MarcusMüller I edited my questions. I use signal generator has very high resolution and low noise/distortion.

Great explanation, thank you.