Difference between dft and the Spectrum

The dft function simply performs a Discrete Fourier Transform and gives you a spectral output. The Spectrum tool provide other metrics as well, and eases the setup of the analysis - so it allows you to measure ENOB, SINAD, SNR, SFDR and so on. So the Spectrum tool provides a wizard to set things up.

I don't know why you think (or "people" (whoever they are) claim) that the spectrum tool is not accurate. What justification do you have for that?

Given all these questions you're asking, it seems as if you are in need of some training. I would suggest you contact Cadence Education Services (see Services->Training at the top of this page) but also look at the documentation, videos, rapid adoption kits on Cadence Online Support.

Regards,

Andrew.

Sir you are right, I can see that Spectrum is very simple and fast tool to get many parameters, may with setting they fail in some accuracy,

I have tried to run spectrum at several period of signals and make my window not start from zero rather than after one to two periods then I get more accurate result.

Actually as I am new to Cadence for sure I need training, and thank you very much you are answering a lot of my questions

If you are using the dft or spectrum tool on a continuous waveform (rather than a sampled waveform that might come out of a DAC, say), then you should set the strobeperiod parameter on the tran analysis. You should set this to coincide with the sample points that the DFT will be performing - that way, you eliminate any interpolation error that happens when the sampling occurs, and you'll see the numerical noise floor in the DFT/spectrum results drop noticeably.

For example, this is a simulation of just a voltage source generating a 1MHz 1V amplitude sine wave. I'm doing a DFT over the 1u period with 64 samples and a rectangular window. The two traces on the right side are the DFT results (in dB20) without any strobing (the red) and with strobeperiod=1u/64 in yellow. The 1MHz value matches, but the other harmonics are much higher in the non-strobed result, caused by numerical error due to the interpolation when the sampling is done.

Of course, if the system is not settled or is not periodic then you should either pick a period when it's settled (assuming it's periodic) or use a windowing function (in which case you need to include several periods of the signal to minimise the distortion effect caused by the windowing function.

Regards,

Andrew

Dear Mr. Andrew,

That was very nice explanation, I totally got your important point about  strobeperiod, I see if I don't fix the setting of it, I will see such a distortion more than the real expected one.

Thank you very much for your help

Best Regards

Dear Sir,

I used the Spectrum to find the THD and other characteristics of the output of my amplifier with sin signal. I made the Strobperiod setting you told me about. I am getting this signal shown below, the THD is looks a bit high.

I am using the default number of bin given by the window, and I am using Hanning window.

I read from Cadence help that the number of bins affect the SNR, ENOB, etc calculation. The manual also states that I can increase the number of bins up to the half of the Sample counts, however when I change it the simulator complains about the start and end frequency. for the latter setting I usually press the S tab to fix it automatically.

Do you suggest me to change the number of bins or I keep it by default better ?

One more question please Sir, there is two options of spectrum Analyses type : Signal Analyses and Noise Analyses... What are the difference between them when Signal Analyses gives me all the information about the noise ? or may be the Noise Analyses is only used when my input signal is in frequency domain ?

Thank you very much 

Please contact customer support. I'm afraid I don't have the free time over the next couple of weeks to walk you through the basics of setting up the simulation and spectrum toolbox properly (it's hard to see what you've done from the details you've provided). I don't know why the simulator would be complaining about start and end frequency because the simulator is simulating in the time domain - so there are things that don't make sense here.

Andrew.

sorry Sir if I didn't provide enough information, I would like to make my question shorter

shall I keep the number of bins as it or increase for better accuracy ?

When I have to select the Signal Analyses or the Noise Analyses ?

That its it

Thank you for your suppot 

There's all sorts of things that need to be understood when doing this kind of thing, such as if you have coherent sampling - and if your system is settled and has a periodic response (it probably does) in which case you probably don't need to have a windowing function (other than rectangular). Looking into all these issues without seeing  your setup and data and thus asking a lot of back and forth questions is something I don't have the bandwidth for (to essentially train you - given the large number of questions you are asking - in various threads). Most of your simpler questions are not necessarily simple without knowing the background of what you're doing or what you've set up.

I love to help (as I'm sure you can see by all my posts here), but if you are from a customer under a contract you should use customer support (or have training). If at an academic institution, you probably should speak to your supervisor - usually there's a route for somebody at the university to contact customer support or go via an organisation such as Europractice if they are unsure.

Maybe somebody else on the forums does have the bandwidth to help - but I already said that I don't - sorry...

Regards,

Andrew.