Designer’s Guide Community

venki_vlsi Junior Member Offline Posts: 29 warangal	Re: Fourier analysis and DFT compare Reply #15 - May 30^th, 2007, 5:43pm hi driveforce and seldon. I got result what i want. I too added normharm in CDF and extracted the result. Thanks Regards venkats
Back to top	s.venkat_nitw IP Logged

ywguo

Community Fellow

Offline

Posts: 943
Shanghai, PRC

Re: Fourier analysis and DFT compare
Reply #16 - May 31^st, 2007, 7:01pm

Hello,

The fourier component must have voltage input. Am I right? Is it feasible to do fourier analysis for a current?

I tried the following method in a spectre netlist, but it reported fatal error.

FOUR0 (I0:OUT_QN I0:OUT_QP) fourier fund=122.0703125K harms=4096 normharm=9

The error information in output log is shown below.

Error found by spectre during circuit read-in.
"input.scs" 2249: No master specified for instance 'FOUR0'
"input.scs" 2249: Unexpected signal or attribute name "I0:OUT_QN". Expected end of file or end of line.

Thanks
Yawei

Back to top

IP Logged

venki_vlsi Junior Member Offline Posts: 29 warangal	Re: Fourier analysis and DFT compare Reply #17 - May 31^st, 2007, 10:56pm hi, I didn't try with current node as input. Please keep posting messages if u got any new ideas on dac (measuring mismatch, modelling ..etc)and i too will do the same. regards venkats
Back to top	s.venkat_nitw IP Logged

RobertZ

Community Member

Offline

the art of analog

Posts: 55
San Diego, CA

Re: Fourier analysis and DFT compare
Reply #18 - Jun 4^th, 2007, 12:17pm

hi, yawei,

Check this out.
insert a vdc = 0 into the branch you want to measure.
use a cvvs, and put fourier at cvvs output. This doesn't load your circuti at all.

regards,
Robert

ywguo wrote on May 31^st, 2007, 7:01pm:

Hello,

The fourier component must have voltage input. Am I right? Is it feasible to do fourier analysis for a current?

I tried the following method in a spectre netlist, but it reported fatal error.

FOUR0 (I0:OUT_QN I0:OUT_QP) fourier fund=122.0703125K harms=4096 normharm=9

The error information in output log is shown below.

Error found by spectre during circuit read-in.
"input.scs" 2249: No master specified for instance 'FOUR0'
"input.scs" 2249: Unexpected signal or attribute name "I0:OUT_QN". Expected end of file or end of line.

Thanks
Yawei

Back to top

Thanks,
Robert

IP Logged

Ken Kundert Global Moderator Offline Posts: 2386 Silicon Valley	Re: Fourier analysis and DFT compare Reply #19 - Jun 4^th, 2007, 2:12pm The fourier analysis component will directly analyze a current waveform if you use the term or refterm parameters. See spectre -h fourier for more information. -Ken
Back to top	IP Logged

ywguo Community Fellow Offline Posts: 943 Shanghai, PRC	Re: Fourier analysis and DFT compare Reply #20 - Jun 4^th, 2007, 7:41pm Hi, Rorbert, Quote: Check this out. insert a vdc = 0 into the branch you want to measure. use a cvvs, and put fourier at cvvs output. This doesn't load your circuti at all. Good idea. I will try your method. Thank you. Yawei
Back to top	IP Logged

ywguo

Community Fellow

Offline

Posts: 943
Shanghai, PRC

Re: Fourier analysis and DFT compare
Reply #21 - Jun 4^th, 2007, 7:53pm

Ken,

Ken Kundert wrote on Jun 4^th, 2007, 2:12pm:

The fourier analysis component will directly analyze a current waveform if you use the term or refterm parameters. See spectre -h fourier for more information.

-Ken

The help on fourier is not very clear, which is shown below.
Instance Definition
Name [p] [n] [pr] [nr] ModelName parameter=value ...
Name [p] [n] [pr] [nr] fourier parameter=value ...
The signal between terminals p and n is the test or numerator signal. The signal between
terminals pr and nr is the reference or denominator signal. Fourier analysis is performed on
terminal currents by specifying the term or refterm parameters. If both term and p or n
are specified, then the terminal current becomes the numerator and the node voltages
become the denominator. By mixing voltages and currents, it is possible to compute large
signal immittances.

Does the instance definition look like the following?
Name fourier fund=122.0703125K harms=4096 normharm=9 term=I0:OUTP term=I0:OUTN,
where I0 is a current-steering DAC, OUTP and OUTN are plus/minus current outputs. However, I doubt that the parameter term supports differential current signals.

Thank you very much.
Yawei

Back to top

IP Logged

ywguo

Community Fellow

Offline

Posts: 943
Shanghai, PRC

Re: Fourier analysis and DFT compare
Reply #22 - Jun 11^th, 2007, 1:02am

Hi, Ken,

The parameter term and refterm don't support differetial current. So I do fourier integral using the method proposed by Robert, which put CCVS sources to convert currents to voltages. I remember that you have mentioned that equally spaced time steps reduce the error of fourier. However, I cannot find that post again.

To reduce the interpolation error, I set the max step as 1 ps, a little smaller than the minimum step in the last output log. The simulation is running, much slower than before. Is it necessary? Or is it a good solution?

Thanks
Yawei

Back to top

« Last Edit: Jun 11^th, 2007, 2:37am by ywguo »

IP Logged

boe

Community Fellow

Offline

Posts: 615

Re: Fourier analysis and DFT compare
Reply #23 - Jun 11^th, 2007, 9:31am

ywguo wrote on Jun 11^th, 2007, 1:02am:

Hi, Ken,
I remember that you have mentioned that equally spaced time steps reduce the error of fourier. However, I cannot find that post again.
To reduce the interpolation error, I set the max step as 1 ps, a little smaller than the minimum step in the last output log. The simulation is running, much slower than before. Is it necessary? Or is it a good solution?
Thanks
Yawei

The Fourier transform uses equally spaced samples to calculate the harmonics; if you do not have a point at the required time, the simulator does linear interpolation and this may be too inaccurate (especially for low-distortion systems).
If you limit the time step to 1 ps, you will be very slow, calculating at least 8M points for your 8 us simulation time, which is much more than you are likely to need...
I use the strobeperiod option of Spectre (set to 1/fsample) whenever I have to do a Fourier analysis to get samples that are correctly spaced for the Fourier analysis and run the Fourier transform as post-processing job over my data.
OK, then you need to calculate all the parameters for yourself, but you also know what you get.

BOE

Back to top

IP Logged

ywguo

Community Fellow

Offline

Posts: 943
Shanghai, PRC

Re: Fourier analysis and DFT compare
Reply #24 - Jun 11^th, 2007, 7:22pm

Hi, BOE,

Quote:

I use the strobeperiod option of Spectre (set to 1/fsample) whenever I have to do a Fourier analysis to get samples that are correctly spaced for the Fourier analysis and run the Fourier transform as post-processing job over my data.

Do you mean that you run a DFT function in the calculator?

In spectre user guide, it reads that You can also dramatically improve the accuracy of external Fast Fourier Transform (FFT) routines. I think that means the parameter strobeperiod improves the accuracy of DFT function.

However, I am runing a transient simulation with a FOUR component from analogLib, which is a fourier integral immediately after the transient simulation.

Now the FOUR (fourier integral) result is much larger than that of DFT function at high frequency. I want to know whether the FOUR result have large error due to intepolation error. Then I run a simulation with maxstep = 1ps. Now the simulation speed is unacceptable slow. Sad

Thank you.
Yawei

Back to top

IP Logged

boe

Community Fellow

Offline

Posts: 615

Re: Fourier analysis and DFT compare
Reply #25 - Jun 12^th, 2007, 3:59am

Hi, Yawei,
ywguo wrote on Jun 11^th, 2007, 7:22pm:

Do you mean that you run a DFT function in the calculator?

Yes.

ywguo wrote on Jun 11^th, 2007, 7:22pm:

I think that means the parameter strobeperiod improves the accuracy of DFT function.

The DFT is sensitive to the accuracy of the samples used - so you use strobeperiod to force the simulator to produce exactly the time points you need and therefore eliminating the interpolation error.

ywguo wrote on Jun 11^th, 2007, 7:22pm:

Now the FOUR (fourier integral) result is much larger than that of DFT function at high frequency. I want to know whether the FOUR result have large error due to intepolation error. Then I run a simulation with maxstep = 1ps. Now the simulation speed is unacceptable slow.

Spectre uses a quadratic numerical integration algorithm to calculate the Fourier integrals for the requested frequency points (cf. Spectre User Guide). To get reasonable results for high frequencies, you will probably need small time steps. Unfortunately, Cadence tells you hardly anything about the algorithm (which is why I prefer the DFT method), but I suggest a minimum of 10 or 20 points per period (of your harmonic), i.e. maxstep = 0.05/f ... 0.1/f.

Did you try the suggestion of the Spectre User Guide ("If you are concerned about accuracy, perform an additional Fourier transform on a pure sinusoid generated by an independent source")? If you want to check harmonics, you might want to use a triangle wave (with you fundamental frequency).
BOE

Back to top

IP Logged