Great question! You can choose to add transient noise to the signal to which you apply the FFT. That will result in a much higher noise floor, depending on the statistical properties you choose in the setting. Right now, the floor is at around -300 dB, which implies that the noise is only limited by the accuracy of floating-point numbers (i.e. there is no discernable noise). For more info on how to determine such a noise floor's behavior as a function of temperature and system parameters, have a look at this: en.wikipedia.org/wiki/Minimum_detectable_signal.
Hi Thomas, great video. I was just wondering at 13:12 the THD is 0% even though Harmonics are present. Any idea how to get an actual value of THD in this case? Thanks
Hi Jerome! Really good point, the real THD definitely isn't 0% here! I don't have access to the simulation from here, but I expect that you might have to indicate what the fundamental frequency is somewhere. Otherwise, I'd expect that Cadence can't distinguish between the signal and its harmonics (esp. in the case of intermodulation, where the number of harmonics increases significantly). Hope this may help as a pointer for getting closer to that real THD-value.
@@ThomasBooij I see, very interesting. I think I managed to get a value for Total Harmonic Distortion by using the "THD" function in the Cadence calculator. This function asks for the fundamental frequency and returns a percentage for THD. Thanks Thomas
Hello Thomas Thanks for this tutorial I wonder if you could help me with this. I've designed a 10 bit 1v ref fully differential SAR ADC. what I want to do now is to calculate SNR, ENOB and other factors. Here I can not understand if these values such number of samples and Fs you're talking about is just for FFT function requirements or the exact idea can be applied for ADCs too. I mean should I apply a sine wave and sampling frequency for 1M hertz and then all values will be calculated by software automatically? If the sampling frequency is to be 1 Mega hertz so fin has to be only 1K hertz so that 1024 specific samples will be produced.
Hi Mohamad, the rules laid out in the video are meant for getting a meaningful result from the FFT. The sampling rate of the FFT and that of the ADC are, thus, not the same. I am therefore also not completely certain as to whether the contraint you list is correct, but from intuition I would expect the Fs of the FFT to be highter than that of the ADC. Otherwise, you are taking too few samples in the FFT to get a realistic idea of the signal's spectrum. Hope this helps!
@@vincentf502 Indeed, after thinking about it, you're absolutely correct. As long as M is relatively prime with 2^N this will work, which is indeed the case for M = 77. When M is a prime number itself, we have the immediate guarantee that there can be no common divisors. In other words, while there are many other values of M for which this will work, choosing a prime number guarantees that there will be no common divisors.