The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl Design >> Mixed-Signal Design >> FFT simulation of SNR for Sigma delta Modulator https://designers-guide.org/forum/YaBB.pl?num=1150350571 Message started by hk2004 on Jun 14th, 2006, 10:49pm |
Title: FFT simulation of SNR for Sigma delta Modulator Post by hk2004 on Jun 14th, 2006, 10:49pm Dear, I am facing a stranger problem as following. In Cadence , I run transient analysis for model based SDM circuit and strobe the output. My setting is like this: quantizer clock frequency: 64M or period 15.625ns skipstart: 319.5ns = 312.5+7ns. the 312.5 is 20 cycle for skipping the inital data. The quantizer is rising edge triggered. So 7ns is used to settle. The problem is like this: if I store 8192 data , the FFT results shows the SNR is 48dB. But if I store 4096 data, the FFT results shows the SNR is 79dB that is expected. FFT operation is performed by matlab, and hann window is used. Does anyone help on this? Thanks a million. |
Title: Re: FFT simulation of SNR for Sigma delta Modulato Post by sheldon on Jun 15th, 2006, 9:52pm HK, It is a little difficult to analyze the problem from the available data. One experiment that may be useful would be to run the 4096 sample FFT using different data sets. In particular, FFT A: 319.5n to 64.3195u FFT B: 64.3195n to 128.3195n If the FFT results are not consistent, then 20 cycles may not be long enough to settle the start-up transients. One other point, you say that you allowed 7ns for settling the transient before performing the FFT. Shouldn't you allow half a period + 7ns, (15.625n/2) +7ns, or 14.8125ns? Best Regards, Sheldon |
Title: Re: FFT simulation of SNR for Sigma delta Modulato Post by ee05s030 on Jul 26th, 2006, 10:46pm how do u get a decreased SNR when u take a 8192pt FFT? do u see the noise shaping? is the noise floor increased? if noise shaping is maintained u might be doing some scaling factor mistakes. check that. |
Title: Re: FFT simulation of SNR for Sigma delta Modulato Post by sheldon on Aug 2nd, 2006, 2:37am ee05s030, If there are long internal time constants inside the circuit, then the effect of settling the time constants will degrade the SNR. A settling time issue looks like a 1/f noise plot, the FFT noise floor decreases with increasing frequency. The request was to perform to FFTs. If settling time was an issue then the SNR from the two FFTs would be different. The SNR of the first FFT would be lower than the SNR of the second FFT. Settling issues can occur for many reasons, for example, the common-mode feedback loop might have a long settling time. Best Regards, Sheldon |
Title: Re: FFT simulation of SNR for Sigma delta Modulato Post by Jerry Kwon on Sep 8th, 2006, 11:38pm Hello HK2004! I read your problem so I'd like to suggest something. First of all, sometimes there are faults with transient simulation results because of step problem. Thus, I think, more point sizes make FFT's SNR reduce. If you can get text output, how about you perform following as: Using Matlab-> command box -> (typing) psd(output,length(output),frequency) and then you can see the exact specrum you want. Good luck! Best, Jerry Kwon In Cadence , I run transient analysis for model based SDM circuit and strobe the output. My setting is like this: quantizer clock frequency: 64M or period 15.625ns skipstart: 319.5ns = 312.5+7ns. the 312.5 is 20 cycle for skipping the inital data. The quantizer is rising edge triggered. So 7ns is used to settle. The problem is like this: if I store 8192 data , the FFT results shows the SNR is 48dB. But if I store 4096 data, the FFT results shows the SNR is 79dB that is expected. FFT operation is performed by matlab, and hann window is used. Does anyone help on this? Thanks a million. |
Title: Re: FFT simulation of SNR for Sigma delta Modulato Post by filipe on Mar 9th, 2007, 1:42pm a good way to calculate the snr, and plotting your psd is using a script from a toolbox about delta sigma modulators, with matlab. bye |
Title: Re: FFT simulation of SNR for Sigma delta Modulato Post by sheldon on Mar 9th, 2007, 4:56pm Jerry, Sorry did not see your post. Your results are certainly contrary to expectation. A couple of questions about the FFT issue. 1) What is the stop time for the simulation for the 4096 point FFT and for the 8192 point FFT? --> The stop time for the 8192 point FFT is 2x longer than for the 4096 point FFT 64.3125us [4096] or 128.3125us [8192] 2) What input frequencies do you use for each simulation? Adjusting the input frequency for for each simulation may help. 7/4096 --> 13/8192 3) Also, I have found that using the zvcvs, ideal S/H, can give better results than using the strobe function. Just my experience maybe but adjusting the input frequenices based on the number of points seems to improve the results. Again really sorry for the slow response time. Best Regards, Art Schaldenbrand |
The Designer's Guide Community Forum » Powered by YaBB 2.2.2! YaBB © 2000-2008. All Rights Reserved. |