Yawei,

With the thd function, you specify the number of samples to take. This is converted to a power of two (I forget whether it is lower or higher, but it's the same as the dft function). Then it ends up with a waveform internally with a number of harmonics equal to half the power-of-two samples used.

That's how many harmonics are used for the THD calculation (i.e. half the number of samples, provided it is a power of two).

You also specify the frequency of the fundamental, so it doesn't necessarily have to be the first harmonic.

Regards,

Andrew.

I don't see why it should be incorrect to include all the harmonics. Why do you need more samples - remember the number of samples affects the number of harmonics - i.e. the maximum frequency. The frequency resolution is the interval you do the FFT over, so if you don't need those higher frequencies, don't do that number of samples...

Regards,

Andrew.

Yawei,

  The problem is that you are using the wrong tool to evaluate the
frequency response. If you are simulating a DAC and you want to
evaluate the effect of glitch on the response then you should use
the Fourier integral, .four, not the FFT to evaluate the frequency
response. The Fourier integral evalutes the entire waveform while
the FFT only samples the waveform. So the FFT is great for ADCs,
continuous analog --> discrete analog --> digital since the output
waveforms are sampled. However for DACs, the output waveforms
are not suited for the FFT, digital -->continuous analog. The bottom
line is that FFT is inefficient if you are sampling in small enough time
intervals to capture the glitch.  The downside of the Fourier integral
is that it does not support window functions so if you want to use
for Sigma-Delta DACs, you need to write your own window function
Verilog-A module.

BTW, there was alittle bit if discussion of this in a previous thread,

http://www.designers-guide.org/Forum/YaBB.pl?num=1113231964

                                                            Best Regards,

                                                               Sheldon

Andrew Beckett wrote on May 9^th, 2007, 9:54pm:


Andrew,

Thank you very much.

The fourier analysis is not quite the same as other simulators (see the chapter in the Designer's Guide to SPICE and SPECTRE which talks about it - first few pages can be seen at http://www.designers-guide.org/Books/dg-spice/ch5.pdf).

lteratio is a transient parameter - you need to run a transient analysis as well. So it's on the transient options form (and can be found in spectre -h tran). The default value is dependent on errpreset. It allows you to loosen the truncation error criteria, and this is particularly useful if you want to tighten tolerances and so gain accuracy without necessarily causing more timesteps to be taken. There's a good discussion of this in the above book.

Regards,

Andrew.

Hi, Yawei,

Does your cycle here mean input singal cycle or DAC clock cycle?

And secondly, I think harmonics at multiple (clock) frequencies are caused by not only sampling (when you do DFT analysis), but also the image of DAC. DAC's output spectrum (without any FFT sampling) itself has inifinity bandwith and infinity of nyquist band replica, but with sinc shaped.

driveforce


ywguo wrote on May 9^th, 2007, 12:00am:

Yawei,

  Built a simple example of the testbench that you seem
to be using to test your DAC. If the output spectrum is
showing tones similar to the tones in attached plot, then
modifying the sampling will eliminate the tones.

                                              Best Regards,

                                                 Sheldon