The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> System Simulators >> MATLAB DFT vs Cadence DFT
https://designers-guide.org/forum/YaBB.pl?num=1351060146

Message started by Sharath Raju on Oct 23rd, 2012, 11:29pm

Title: MATLAB DFT vs Cadence DFT
Post by Sharath Raju on Oct 23rd, 2012, 11:29pm

Please bear with me since the post is a bit long. I had to run a couple of simulations before getting here.

Experiment:
A sine wave of unit amplitude and frequency 100 Hz is sampled at a frequency of 2048 Hz for a period of 0.5s yielding 1024 samples. A 1024-point FFT is taken and the power spectrum is calculated from the FFT. The above experiment is performed separately in MATLAB and Cadence. The next step is to compare the power spectrum plots from Cadence and MATLAB and see if they match.

Analysis/Result:
Figure 1 shows a plot of the power spectrum obtained from MATLAB. It also shows a snapshot of the signal in time domain. Bear in mind that this is simply a partial snapshot since the sampled signal is actually 0.5s long. The peak occurs at 100 Hz as expected and has a magnitude of 20.97 dBm. The noise floor is at -300 dBm.

Link to Figure1:
https://www.dropbox.com/s/bywnv0h958yovk3/Figure1_Power_Spectrum_MATLAB.png

The above experiment was repeated in Cadence by using vsin from analogLib to generate the sinusoidal signal and setting the length of simulation time to be 0.5s, and the time step to be 488.28125 us (=1/2048 Hz). Figure 2 shows a plot of the power spectrum from Cadence. While the peak is still at 100 Hz, it has a magnitude of 29.9 dBm, a difference of 9 dBm, and the noise floor is -125 dBm. I cannot understand the reason for this difference.

Link to Figure 2:

https://www.dropbox.com/s/2vrtjtjs5rp6my3/Figure2_Power_Spectrum_Cadence.png

Since the plots do not match, I imported the data generated by vsin as a .csv file into MATLAB and plotted the FFT (Figure 3). The MATLAB plot continues to show a peak of 20.97 dBm and the noise floor is lower than Cadence.

Link to Figure 3: https://www.dropbox.com/s/eehbywfcs2i5ldz/Figure3_Power_Spectrum_imported_into_matlab.png

Could someone comment why the MATLAB DFT and Cadence DFT give different results even though I feed the same data to both the functions ?

On a minor note, could someone tell how to post images inline, or at least to attach multiple images in a single post ??? :(

Title: Re: MATLAB DFT vs Cadence DFT
Post by Ken Kundert on Oct 24th, 2012, 4:57pm

1 Vp corresponds to 30dBm? Is your reference impedance 500mΩ?

Title: Re: MATLAB DFT vs Cadence DFT
Post by Sharath Raju on Oct 24th, 2012, 6:26pm

Hi Ken,

Thanks for the reply. The load impedance in my testbench is 1 kilo-ohm. I changed it to 50 ohm, bwut didn't see any difference in the power spectrum. (The peak tone continues to have a magnitude of 30 dBm and the noise floor is approximately -125 dBm).

Snapshot of testbench: https://www.dropbox.com/s/lqkaieyf2kc152o/Power_Spectrum_Cadence_Testbench.png

Thanks,
Sharath

Title: Re: MATLAB DFT vs Cadence DFT
Post by Ken Kundert on Oct 24th, 2012, 11:28pm

1 Vp with a 1kΩ reference resistance is -3 dBm. So your first step is to understand why both approaches are giving you the wrong power at the fundamental frequency.

-Ken

Title: Re: MATLAB DFT vs Cadence DFT
Post by rfidea on Oct 25th, 2012, 9:42am

What is causing your dft plots to be in dBm? You are talking about volts in the text. How is the conversion done?

Title: Re: MATLAB DFT vs Cadence DFT
Post by Sharath Raju on Oct 25th, 2012, 11:00am

The conversion to dBm is done as follows:
Power in dBm = 20*log10(abs(Xk)) + 30
where,
Xk = FFT of x
x = voltage input.

The factor 30 causes the power to be in dBm. Otherwise, it would be in dB.

Thanks

Title: Re: MATLAB DFT vs Cadence DFT
Post by Ken Kundert on Oct 25th, 2012, 11:54am

dBm is decibels relative to 1mW. To convert voltage to dBm you need to know the load resistance, which is generally referred to as the reference resistance. The formula is

dBm = 30 + 10 log(Vp2/(2R))

where the voltage is assumed to be sinusoidal with a peak value of Vp. R is the reference resistance.

I don't know how you are calculating dBm in Matlab (because even with your equation 1Vp does not correspond to 20 dBm), but I am pretty sure that ADE would be using the formula I gave above, but you would have to specify a value for the reference resistance.

-Ken

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