The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> Circuit Simulators >> Spectre Fourier Analyzer vs. FFT
https://designers-guide.org/forum/YaBB.pl?num=1147369975

Message started by schehrazi on May 11th, 2006, 10:52am

Title: Spectre Fourier Analyzer vs. FFT
Post by schehrazi on May 11th, 2006, 10:52am

Hi,

Spectre documents and Ken's book clearly say that the spectre fourier analyzer is more accurate than common FFT. Besides, it does not have aliasing problem. If so, I was wondering why so many people still use FFT and there is no thread about Spectre fourier analyzer on this forum??? Does it have a problem, or people are just not aware of its benefits?

Besides, I could not find out how to use Spectre fourier analyzer by looking at the Spectre user's guide. I think using "fourier" or "fourier2ch" parts which are in analogLib will automatically invoke fourier analyzer but how? Does anybody know a good document which explains this clearly?

Thank you

-schehrazi

Title: Re: Spectre Fourier Analyzer vs. FFT
Post by Ken Kundert on May 11th, 2006, 11:42am

The best source of information about Spectre's built-in Fourier analyzer is the Spice & Spectre book (http://www.designers-guide.org/Books/dg-spice/index.html). It is available either by using the fourier or fourier2ch components (the output is simply printed as ascii text in the log file) or by using the built-in Fourier analysis capabilities of SpectreRF (in PSS).

The Spectre Fourier analysis is not a fast transform as is the FFT, thus it is expensive if many frequency components are needed.

And it is not really right to say that the Spectre Fourier analysis is more accurate than the FFT. Really they are used in different situations. Generally the FFT is used on sampled signals and the Spectre Fourier analysis is used for continuous signals. So the Spectre fourier analysis might be used to measure THD of a continuous-time amplifier or filter, while the FFT would be used to compute the distortion of sampled circuits such as switched-capacitor filters, sample and holds, DACs, ADCs, etc. With Spectre, the FFT is generally used in conjunction with transient strobing in order to eliminate interpolation errors.

-Ken

Title: Re: Spectre Fourier Analyzer vs. FFT
Post by schehrazi on May 11th, 2006, 12:06pm

Thank you so much, your explanation clarified everything for me.

-schehrazi

Title: Re: Spectre Fourier Analyzer vs. FFT
Post by sheldon on May 12th, 2006, 7:56pm

Schehrazi,

  Some  additional comments, there are simulation challenges
that the Fourier Analyzer is uniquely suited to analyze. For
example, when characterizing circuits DACs, Spectre's Fourier
Analyzer is better. In the process of sampling you lose
information about waveform. So while it is difficult to analyze
the effect of glitch impulse on the re-constructed waveform
using an FFT, it is simple to perform this analysis with the
Fourier Analyzer.

  Some comments on run time:

1) The analysis time is a function of the number of frequency
    components and the number of transient time points.

2) For a typical for a Pipeline ADC, equivalent to a 256-point
   FFT , the overhead is not a big issue.  The overall run time
   is on the order of the time to run the FFT.

3) For a Sigma-Delta Modulator, the overhead can be significant,
    ~10-100x (or more) the time to run the FFT. Simulating
   a Sigma-Delta requires many time points, for example 8192
   points, and high accuracy. So even though the number of
   frequency components is low, the znalysis time is long.

                                                            Best Regards,

                                                               Sheldon

Title: Re: Spectre Fourier Analyzer vs. FFT
Post by schehrazi on May 12th, 2006, 9:56pm

Thank you so much Sheldon.

Title: Re: Spectre Fourier Analyzer vs. FFT
Post by Ken Kundert on May 12th, 2006, 11:21pm

Another situation where Spectre's internal Fourier analysis capability is essential is when analyzing pulse-width or pulse position modulated signals. The discrete sampling nature of the FFT cannot accurately resolve the transitions, but Spectre's Fourier analysis does not sample the waveforms, it includes the information from every time point computed by Spectre.

-Ken

The Designer's Guide Community Forum » Powered by YaBB 2.2.2!
YaBB © 2000-2008. All Rights Reserved.