The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl Simulators >> RF Simulators >> switched capacitor integrator noise simulation results https://designers-guide.org/forum/YaBB.pl?num=1258339986 Message started by alireza on Nov 15th, 2009, 6:53pm |
Title: switched capacitor integrator noise simulation results Post by alireza on Nov 15th, 2009, 6:53pm Hi, I have been working on noise simulations of a discrete time integrator using SpectreRF. I have used PSS/PAC/td-noise tools to find the input/output noise PSD's (power spectral densities), as well as the transfer function from the input to the output of the integrator (after the S/H). My methodology has been based on the paper: "Device noise simulation of DS modulators". However, what I get as the input noise PSD is only white noise at low frequencies, and the spectrum starts to rise as frequency gets closer to fs/2 (12.5MHz in my case.) This leads to incorrect input-referred noise power calculated by integrating the spectrum. However, if I ignore the rise of noise PSD, (assume it is white) and multiply the white PSD by the bandwidth (12.5MHz) I get the right (expected) noise power, which is approximately 2kT/C. I have attached my schematic as well as the output waveform snapshots in a single .zip file. I would very much appreciate if someone can help me figure out why I am seeing this rise of noise floor near fs/2. Thank you very much. Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 16th, 2009, 12:58am Show me time waveform of clock(25MHz) with x-axis as timeindex. Show me noise spectrum of all timeindex points. Try to set timeindex other than 0 to observe noise spectrum. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 16th, 2009, 9:35am Hi, Thank you very much for your reply. I simulated and plotted the noise spectrum at 2 other points (20ns, and 40ns). "0" and "40ns" correspond to the end of phase 2 (integration phase), while 20ns corresponds to the end of phase 1 (sampling phase). The results are approximately the same. Please find the waveforms attached. Sorry, I was not exactly sure about your first question. Did you mean the clock waveform with respect to time obtained from a transient simulation? (because if I want to plot the clk waveform when running PSS/Pnoise simulations only, Cadence doesn't plot anything.) ps: Just to provide some more detail about my test bench, I am using ideal components for this noise simulation, my opamp is ideal (VCVS, in series with an Rout), also switches are the ideal switches from analogLib, in series with a 500ohm resistor whose noise is enabled. Basically, all the noise I am simulating is the white noise from the series Ron of switches.) Thanks, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 17th, 2009, 5:08am Show me netlist portions regarding analysis statements and signal source statements. alireza wrote on Nov 16th, 2009, 9:35am:
Is there any difference ? alireza wrote on Nov 16th, 2009, 9:35am:
I request time domain waveform as result of PSS analysis. alireza wrote on Nov 16th, 2009, 9:35am:
Again show me netlist portions regarding analysis statements and signal source statements. Show me logfile. Show me input_noise spectrum, output_noise spectrum and VGain as 20*log10(). Show me "output_noise spectrum" / "input_noise spectrum" as 20*log10() and compare it with VGain of PAC result. If you compare input noise spectrums at 1Hz and 12.5MHz, they are different by 20*log10(970/660)=3.3dB. In tdnoise of Cadence Spectre Pnoise, noise PSD is overlapped around fs/2 where noise is sampled at rate of 1/fs. This overlapped noise is not physical. http://www.designers-guide.org/Forum/YaBB.pl?num=1059089369/1#1 http://www.designers-guide.org/Forum/YaBB.pl?num=1207830622 http://www.designers-guide.org/Forum/YaBB.pl?num=1218622880 http://www.designers-guide.org/Forum/YaBB.pl?num=1242797715 http://www.designers-guide.org/Forum/YaBB.pl?num=1220976977 |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 17th, 2009, 2:35pm Thanks again for your reply. I have attached the analysis and signal source parts of my netlist. I also simulated the integrator at different time index points. The results are almost the same. (please see the attached waveforms.) The waveform file also includes the PSS time-domain waveforms of my clocks p1 and p2. About comparing the gain plot (in dB) obtained from PAC versus the one obtained from: output PSD / input PSD, the plots perfectly match. (As expected, as I had originally used the PAC gain to obtain the input PSD.) About your last comment, do you think what I see in the input noise spectrum at high frequencies is an artifact caused by the way Spectre td-noise works? In other words, what I saw in my simulation results is how it should be? Thanks a lot. Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 18th, 2009, 7:12am alireza wrote on Nov 17th, 2009, 2:35pm:
alireza wrote on Nov 17th, 2009, 2:35pm:
On the other hand, VGain as result of conventional PAC(type=sources) is time averaged gain over 1/fs period where VGain don't show any periodcity regarding frequency, so VGain of PAC is not warped at all. If input noise spectrum is calculated by "warped output noise spectrum" / "non-warped VGain", such input noise spectrum is also warped around fs/2. http://www.designers-guide.org/Forum/YaBB.pl?num=1254200413/0#0 http://www.designers-guide.org/Forum/YaBB.pl?num=1221531834/0#0 alireza wrote on Nov 17th, 2009, 2:35pm:
alireza wrote on Nov 17th, 2009, 2:35pm:
|
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 18th, 2009, 5:56pm Hi, Thanks a lot for your comments. I just have a couple of questions. My main question is about your last comment: is there a way that I can compensate this effect (warped spectrum near fs/2) in Spectre, in order to obtain correct simulation results for the input-referred noise? if yes, would you please let me know how? The other thing is that warped noise spectrum doesn't seem to happen only for td-noise analysis. In fact the same thing happens when I run a pnoise (sources) simulation, however the spectrum in that case at some points at very high frequencies (100M, 1G) looks unrealistic, with sudden jumps in the output noise floor. (Although I am not sure, if the results of pnoise (sources) simulation make sense to evaluate the noise performance of a switched capacitor circuit.) Thank you, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 19th, 2009, 2:47am Basically I don't use Cadence Spectre mainly and intensively because I have other more more superior simulators. So you had better expect suggestions from Cadence Tool's devotee. alireza wrote on Nov 18th, 2009, 5:56pm:
Approximately you can compensate results of tdnoise by using results of pnoise(type=sources). But I think strict compensation is also possible by utilizing full functionalities of Cadence Spectre intensively. alireza wrote on Nov 18th, 2009, 5:56pm:
You mean spike like following ? http://www.designers-guide.org/Forum/YaBB.pl?num=1208334972/2#2 There's a spike in the NF curve generated from PSS/PSP. This spike's frequency is same as fudamental frequency of PSS. Due to numerical computation problem of Shooting Newton of Cadence Spectre's PSS, you often see such spikes in pnoise analysis subjected to master Shooting Newton PSS. There is another example. See http://www.designers-guide.org/Forum/YaBB.pl?num=1208334972/1#1 Compare CDS_NFssb(Green Curve) with ADS_NFssb(Magenta Curve). For low frequency less than 10kHz, CDS_NFssb is fairly large compared to ADS_NFssb. If you calculate CDS_NFssb by using HB-PSS not Shooting Newton-PSS, CDS_NFssb is consistent with ADS_NFssb completely. alireza wrote on Nov 18th, 2009, 5:56pm:
Why do you think pnoise(sources) simulation does not make sense ? What procedure do you take and what instruments do you use in actual evaluation not by "EDA Tool Play ? Consider actual situation of evaluation. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 19th, 2009, 6:26pm Hi, I have attached the input/output noise spectums that I get in my pnoise (sources) simulations with Shooting Newton PSS, with the spikes at 100MHz and 1GHz. I also tried the HB-PSS, but in that case my gain is zero! (maybe I need to change some of the simulation setup parameters to get it right?) Other than the spikes, the pnoise simulation has the same noise floor rise near fs/2. (It is about 2dB and is approximately same as td-noise.) So pnoise does not seem to be better than td-noise. The reason why I was saying that I should use td-noise (but not pnoise) for my switched capacitor circuit was based on what I read in Ken Kundert's paper where he says that for discrete time circuits, this discrete time tool (td-noise) should be used. Since it is really important for me to solve this input-referred noise problem, I'd appreciate if you could please refer me to any source of help (paper, book, etc) in order to find a solution to this problem in Cadence spectre. Alternatively, may I ask which simulator other than Cadence spectre do you recommend for this purpose which can solve my problem? Thanks a lot for your help. Regards, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 20th, 2009, 4:57am alireza wrote on Nov 19th, 2009, 6:26pm:
How do you calculate input noise spectrum ? Using direct Plot Form ? From Result Browser ? Try to simulate with more detail frequency points such as 101points/decade including 25MHz, 50MHz, 75MHz, 100MHz, 125MHz, ...., n*Fund as additional analysis points. alireza wrote on Nov 19th, 2009, 6:26pm:
alireza wrote on Nov 19th, 2009, 6:26pm:
See http://www.designers-guide.org/Forum/YaBB.pl?num=1242797715 Show me output noise spectrum for both "type=sources" and "type=timedomain" with over plot mode. alireza wrote on Nov 19th, 2009, 6:26pm:
There is nothing superior to the actual measurement using actual instruments. alireza wrote on Nov 19th, 2009, 6:26pm:
alireza wrote on Nov 19th, 2009, 6:26pm:
http://www.berkeley-da.com/prod/oview.htm |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 20th, 2009, 10:01pm pancho_hideboo wrote on Nov 20th, 2009, 4:57am:
I use the direct plot form to plot the input noise, the output noise and the gain. I also did the simulation with more points (including those specific points.) The result was that now at all those points I got those glitches in the gain plot. (please see attached) pancho_hideboo wrote on Nov 20th, 2009, 4:57am:
I think so. (I say that based on the fact that the simulation completes.) pancho_hideboo wrote on Nov 20th, 2009, 4:57am:
tdnoise and pnoise (sources) both yield the same (incorrect) spectrum for the input-referred noise. There is about 2dB rise in the noise floor near fs/2. (please see attached.) Thanks, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 21st, 2009, 12:05am alireza wrote on Nov 20th, 2009, 10:01pm:
alireza wrote on Nov 20th, 2009, 10:01pm:
Quote:
You set "hams=1" in Shooting-PSS Analysis statement. This is OK for Shooting-PSS since you set "maxacfreq=15.1G" for Shooting-PSS. But this is not OK for HB-PSS. Do you set "harms=300" in HB-PSS Analysis statement ? http://www.designers-guide.org/Forum/YaBB.pl?num=1257901096/1#1 http://www.designers-guide.org/Forum/YaBB.pl?num=1237890768 alireza wrote on Nov 20th, 2009, 10:01pm:
for both "type=sources" and "type=timedomain" with over plot mode. Not input noise spectrum. Your noise rise near fs/2 in pnoise(type=sources) is folding not alias. Folding in pnoise(type=sources) is physical. On the other hand, alias in pnoise(type=timedomain) is non-physical. Again see http://www.designers-guide.org/Forum/YaBB.pl?num=1059089369/3#3 |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 22nd, 2009, 10:04pm Hi, pancho_hideboo wrote on Nov 21st, 2009, 12:05am:
Sure. Given that these (downward) spikes are natural in the S/H gain plot (due to zeros at clk harmonics), the (upward) spikes in the input-referred noise should occur too. But those spikes lead to huge input-referred noise if integrated over frequency. true? OR, do we not care about them? (Do we integrate the spectrum from 0 to inf or to fs/2?) The other related question that I had is that do we run the pnoise-sources simulation over a wider freq. range? (than fs/2), because this is what is done in the paper: "simulating switched-cap Filters with specterRF", contrary to pnoise-td which is run until fs/2. pancho_hideboo wrote on Nov 21st, 2009, 12:05am:
Yes, I had set the harmonics to 300 as well. pancho_hideboo wrote on Nov 20th, 2009, 4:57am:
I have attached the output spectrum plots both in dB and linear scale. There is a difference of about 4 dB at fs/2. (pnoise-td is higher that pnoise-sources) (pnoise-sources is in dBm, therefore the plot shows 34dB difference.) The PSS analysis was set as Shooting-PSS.. (But as I showed in the last post, when referred back to the input they both show a rise in the noise floor.) Thanks, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 22nd, 2009, 11:07pm alireza wrote on Nov 22nd, 2009, 10:04pm:
Note: n*Fund are singular points, but fs/2 is not singular point. fs/2 is nyquist point. alireza wrote on Nov 22nd, 2009, 10:04pm:
0 to inf for "Pnoise(sources)". 0 to fs/2 for "Pnoise(timedomain)". Again see http://www.designers-guide.org/Forum/YaBB.pl?num=1059089369/6#6 alireza wrote on Nov 22nd, 2009, 10:04pm:
Does iteration of HB trial reach to maximum limit without resulting in small ConvNorm ? Confirm your simulation logfile. If you have other simulators such as BDA's Analog FastSPICE, Nexxim, Aplac, ADSsim, GoldenGate, eldoRF, HSPICE-RF, SmartSpice-RF, etc, try to use them. Other simulator also have both Shooting and HB solvers. About HB solver, other simulator is far robust than Cadence Spectre. alireza wrote on Nov 22nd, 2009, 10:04pm:
alireza wrote on Nov 22nd, 2009, 10:04pm:
Upload PSF directory data for both "Shooting-PSS/PAC(sources)/Pnoise(timedomain)" and "Shooting-PSS/PAC(sources)/Pnoise(sources)". Or upload full netlist with model files and Verilog-A modules so that I can run simulation. alireza wrote on Nov 22nd, 2009, 10:04pm:
Again, n*Fund are singular points, but fs/2 is not singular point. fs/2 is nyquist point. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 24th, 2009, 10:09pm Hi, pancho_hideboo wrote on Nov 22nd, 2009, 11:07pm:
pnoise-td simulation and pnoise-sources simulation give almost identical PSDs (with slight difference near fs/2) from 0 to fs/2. And we know that in pnoise-td sim, we should integrate the spectrum from 0 to fs/2. Therefore, it makes sense that if we are using pnoise-sources for a discrete-time switched-cap circuit we integrate the output spectrum again from 0 to fs/2. Otherwise, it would yield a different result. Am I right? pancho_hideboo wrote on Nov 22nd, 2009, 11:07pm:
Is this again because of folding/aliasing issue? pancho_hideboo wrote on Nov 22nd, 2009, 11:07pm:
I have attached the full netlist of my test bench. The S/H verilog-A code used in the test bench which I had downloaded from this website is also attached. The clock generator uses ideal blocks from ahdlLib, which I assume are available to you. pancho_hideboo wrote on Nov 22nd, 2009, 11:07pm:
Does this mean that for a sampled data system we only care about the pnoise-sources spectrum from 0 to fs/2?? Thanks a lot, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 25th, 2009, 2:18am alireza wrote on Nov 24th, 2009, 10:09pm:
However from practical point of view, an integration from 0 to fs at least might be enough useful. This range is mainlobe of output noise spectrum. alireza wrote on Nov 24th, 2009, 10:09pm:
alireza wrote on Nov 24th, 2009, 10:09pm:
|
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 27th, 2009, 8:43pm Hi, pancho_hideboo wrote on Nov 25th, 2009, 2:18am:
Yes, you were right. integration beyond fs/2 does not change the result by much. I was also wondering, if you were able to simulate the netlist that I uploaded in the last post. If yes, was there any new results in the input-referred noise spectrum? Thanks, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 28th, 2009, 5:20am alireza wrote on Nov 27th, 2009, 8:43pm:
As you can see, I can get very reasonable results both for input and output noise spectrum. Show me your definition of input noise in pnoise(type=timedomain) as OCEAN Script. My definitios are followings: For type=sources Code:
For type=timedomain Code:
|
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 28th, 2009, 5:35am These are input and output noise spectrum for type=sources. |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 28th, 2009, 5:36am Hill around fs/2 in type=sources is due to folding not aliasing. |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 28th, 2009, 5:44am Attached is archive of netlist, ocean script and etc. |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 28th, 2009, 7:20am pancho_hideboo wrote on Nov 22nd, 2009, 11:07pm:
Event-driven constructs such as @(cross) and @(timer) will not be triggered in HB Analysis. HB Analysis also doesn't support $abstime. You use "sh" and "nor_gate" Verilog-A modules. "sh" include @(timer) and $abstime. "nor_gate" include @(cross). @(cross) in "nor_gate" is no problem actually. But @(timer) and $abstime in "sh" are very problematic for HB Analaysis. In Agilent RFDE(ADSsim), Sample&Hold component which is effective even in HB Analysis is provided. http://edocs.soco.agilent.com/display/ads2009/SampleHoldSML+%28Sample+Hold%29 Also there is more sophisticated switch component in Agilent RFDE(ADSsim) more than Cadence Spectre "relay" component. http://edocs.soco.agilent.com/display/ads2009/SwitchV+%28Voltage+Controlled+Switch%29 And I can build various custom components which are effective in HB Analysis by using FDD(Frequency-Domain Defined Devices) http://edocs.soco.agilent.com/display/ads2009/Custom+Modeling+with+Frequency-Domain+Defined+Devices I can use these components in Cadence Spectre Syntax netlists when I use RFDE(ADSsim) as simulator. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 28th, 2009, 8:04am pancho_hideboo wrote on Nov 28th, 2009, 5:44am:
Thanks very much for your sims. Your input spectrum at different time points are (more than) slightly different. (some of the time points have more rise near fs/2) This is somewhat different that what I had. For me, they were more similar. But in any case, there is still the issue of noise folding. In your previous posts you'd mentioned there may be a solution to mitigate this problem in cadence spectre. I know you use other (more advanced) tools, but do you now have an idea of a possible solution in cadence spectre? Because this makes the integrated input-referred noise power wrong and the simulation almost useless. Thanks, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 28th, 2009, 8:10am alireza wrote on Nov 28th, 2009, 8:04am:
I showed spectrums at 0ns, 5ns, 10n, 20n, 25n, 30n. There is almost no difference between them. However If I choose time point at edge of clock signal, spectrum could be more large. alireza wrote on Nov 28th, 2009, 8:04am:
alireza wrote on Nov 22nd, 2009, 10:04pm:
alireza wrote on Nov 19th, 2009, 6:26pm:
I think your definition of input noise in tdnoise is wrong. Again show me your definition of input noise in pnoise(type=timedomain) as OCEAN Script. alireza wrote on Nov 28th, 2009, 8:04am:
Folding in pnoise(sources) is very actual. You must not confuse folding with aliasing. alireza wrote on Nov 28th, 2009, 8:04am:
The integrated input-referred noise power in pnoise(type=sources) is very correct and the simulation is very useful. Are you very familiar with IIR digital filter which is approximation as continuous time analog filter ? Frequency definition is (-infinity, +infinity) for continuous time analog filter. However this range is mapped to (-fs/2, +fs/2) for approximated IIR filter. Frequency range of tdnoise is a little similar to this. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 28th, 2009, 8:43am pancho_hideboo wrote on Nov 28th, 2009, 8:10am:
(Sorry, to clarfiy, by similar I was referring to different time points, Not sources Vs. time-domain. I meant I got similar spctra at different time points) Please see attached. This is the waveform that I had posted earlier. In my sim, the rise was about 2 dB for almost all time points (0, 5n, 10n , 15n, 20n, 25n, 30n). But in your simulations at some of the points its about 6dB while for the others it is about 2dB. (you have also plotted a waveform in linear scale, (in V/rHz), but the y-axis shows only dB scale, Does the yellow curve only correspond to the linear plot? I can see 7 plots in the figure legend, but can only distinguish three curves (colors=yellow,red, purple) on the plot. pancho_hideboo wrote on Nov 28th, 2009, 8:10am:
Based on the theory, the integrated input-referred noise of the SC integrator should be 2kT/C. With this rise in the noise spectrum it does not lead to the expected value. However, if the noise spectrum was white (as expected, since it is thermal noise of resistors) it would yield the correct result. (i.e. if I multiply the BW (12.5MHz) by the noise floor at dc I get exactly 2kT/C, otherwise I don't.) pancho_hideboo wrote on Nov 28th, 2009, 8:10am:
Sure, unfortunately, I don't have access to cadence right now, I will post it later today. |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 28th, 2009, 8:52am alireza wrote on Nov 28th, 2009, 8:43am:
Basically I don't use Direct Plot Form in Cadence ADE where there are many bugs and definition of equations are very suspicious especially for RF applications. http://www.designers-guide.org/Forum/YaBB.pl?num=1237656418/1#1 All noise spectrums are dBV/Hz unit. Again see OCEAN definition of these. Only Yellow bold line is pnoise(sources). Other six lines are all tdnoise results. See Yellow bold line in output noise spectrum. You can see only this line is from 0 to 50MHz. We can't find out any remarkable difference between these six tdnoise input noise results. Input noise results of tdnoise are all higher than pnoise(sources) by 4dB at fs/2. pancho_hideboo wrote on Nov 28th, 2009, 8:10am:
Again folding is a reality. http://www.designers-guide.org/Forum/YaBB.pl?num=1258339986/19#19 Output noise spectrum is a result of contribution from all sidebands, that is, multiple foldings. While VGain of PAC considers only zero sideband. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 28th, 2009, 6:56pm Hi, pancho_hideboo wrote on Nov 28th, 2009, 8:52am:
Here is what I have been using from Direct plot form: dB10((pow(getData("out" ?result "pnoise_td") 2) / (mag((harmonic(v("/sh" ?result "pac") '(0)) / harmonic(_drplPacVolGnExpDen("v(\"/I100/In_sh\" ?result \"pac\")" '(0) nil) '(0))))**2))) In this formula, gain is defined as: (output of the "output SH") / (output of "input SH"). However, if I use: (output of the "output SH") / (input of "input SH") I get exactly what you got. "input of input SH" is where the PAC source is connected. Therefore, the difference in the input noise spectrums was because of different definitions of gain. alireza wrote on Nov 28th, 2009, 8:43am:
pancho_hideboo wrote on Nov 28th, 2009, 8:52am:
Does it mean that in reality the input-referred noise of a SC integrator (caused by white noise sources) is "not completely white"? (In the paper by R. Schreier, et al: "Design-oriented Estimation of Thermal Noise in Switched-Capacitor Circuits" in section III, they discuss that: "Due to folding the spectrum of the sampled noise signal Vnos(n) will be very nearly white." But it is not clear what "very nearly white" really means.) Another question may be: in reality on what parameters does this noise increase depend on?? For example, can it become 10dB under certain conditions? (rather than 2dB) Because in that case, a white noise assumption seems to be totally inadequate. (and/or unrealistic.) (With 2dB increase near fs/2 the integrated input-referred noise power is about 20% higher than 2kT/C.) Also, for the purpose of simulating discrete-time (SC) circuits would you recommend using pnoise (sources) (+ SH blocks at the input and the output of the circuit) instead of pnoise (time-domain)? In the paper: "Simulating SC Filters with SpectreRF", it mentions td noise analysis was built for this purpose, but it seems that the pnoise (sources) is more accurate? Am I right? Thanks a lot, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Nov 28th, 2009, 7:59pm alireza wrote on Nov 28th, 2009, 6:56pm:
You must not use Direct Plot Form in Cadence ADE blindly. alireza wrote on Nov 28th, 2009, 6:56pm:
alireza wrote on Nov 28th, 2009, 6:56pm:
There is a correlation in time domain for SC integrator naturally since you use two Sample&Hold Circuits. Note : Correlation of noises in frequency domain results in shaping of noise in time domain. alireza wrote on Nov 28th, 2009, 6:56pm:
As you know, tdnoise takes very very much time in completion. But even if you observe spectrum at clock edge by using tdnoise, you can never get any novel or valuable information so much. Also see http://www.designers-guide.org/Forum/YaBB.pl?num=1220976977/3#3 alireza wrote on Nov 28th, 2009, 6:56pm:
EDA Tool Play is no more than Tool Play. There is nothing superior to the actual measurement using actual instruments. alireza wrote on Nov 28th, 2009, 6:56pm:
Again see http://www.designers-guide.org/Forum/YaBB.pl?num=1242797715/1#1 "sources" and "timedomain" are no more than small signal analysis in frequency domain. "sources" give you Time_Averaged_Noise_Power_Density(freq), while "timedomain" give you Ensemble_Averaged_Noise_Power_Density(time,freq). But meaning or interpretation of frequency spectrum is different between these two. Again see my last comment in http://www.designers-guide.org/Forum/YaBB.pl?num=1258339986/23#23 The followings are general notes. - Use correct terminologies. - Warnigns are different from Errors. - ADS is not name of simulator. - There is no tool which name is Cadence. - Don't use Direct Plot of Cadence ADE blindly without knowing definition. - All gains in Direct Plot of Cadence ADE are "right", "true" and "practical" voltage gain. - MATLAB are different from Simulink. - Learn measurements using actual instruments. Not "EDA Tool Play |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Nov 29th, 2009, 7:35pm Thanks a lot for your help. Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Dec 6th, 2009, 6:07pm Hi, I just have a couple of more questions regarding PSS/pnoise simulations. I am right now using the pnoise(type=sources) to find the input referred noise power of the SC integrator. I am using the integrator in open loop mode (without the S/H block feeding back to the input, and I don't have any stability issues, due to finite opamp gain.) My circuit parameters are as follows: gm_opamp=1, Rout_opamp=1M, Ron_switch=600ohm, Cs=CI=1pF, R_opamp=0 (to model opamp noise) (please see the attached schematic.) I am looking at the input-referred noise spectrum, based on your OCEAN script: (mag(getData("in" ?result "pnoise-pnoise"))**2) The simulation results are fine at low frequencies (below fs). However I do have an spike occurring at fs=25MHz. (please see the attached curves.) If I want to find the total integrated input-referred noise power, I should integrate the spectrum from 0 to inf, which of course includes the singular point. Now, I am not sure, what to do with the singular point at 25MHz? Should this happen in reality? or is it again an artifact? (It only occurs at 25MHZ but not other harmonics!) (My PSS analysis engine is Shooting, as HB does not give reasonable results with the hidden-states free SH block I am using) Even if I ignore this spike, the input noise spectrum does not fall off at high frequencies, so essentially if I integrate it, it will yield a noise power which is much larger than expected. The above curve is for the noise density referred back to the input of the circuit (node:Inp), where PAC source is connected. This corresponds to the "input" of the "input SH". However, if I refer the noise back to the "output" of the "input SH" (node:In_sh), I get a different curve (second curve in the attached plot) If I integrate this curve from 0 to inf, I do not have any spikes and almost get the right input referred noise power. (10.54nV^2 for Cs=1pF). However I am not sure whether this node is the one I should refer back the noise to. Thanks a lot, Regards, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Dec 7th, 2009, 4:33am Reread all my appends from first. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Dec 7th, 2009, 8:58am Thanks for your reply. About the output noise spectrum obtained from pnoise (type=sources) sim. you had mentioned that the freq. range from 0 to fs is the main lobe of the spectrum, but in any case I should integrate it from 0 to inf: pancho_hideboo wrote on Nov 25th, 2009, 2:18am:
However for the input noise spectrum this is not the case. (as it is shown in your recent PSD as well) My specific question is: how do I find the total integrated "input" referred noise? If I integrate the spectrum from 0 to inf (or the simulation bandwidth: say 100MHz) the result will be proportional to this bandwidth, as the spectrum does not roll off at high frequencies. This does not sound correct. Am I right? Thanks, Regards, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Dec 7th, 2009, 9:02am alireza wrote on Dec 7th, 2009, 8:58am:
alireza wrote on Dec 7th, 2009, 8:58am:
alireza wrote on Dec 7th, 2009, 8:58am:
You are still misunderstanding. Consider noise spectrum from resistor, here PSD is unbounded in frequency. alireza wrote on Dec 7th, 2009, 8:58am:
Again EDA Tool Play is no more than Tool Play. There is nothing superior to the actual measurement using actual instruments. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Dec 7th, 2009, 9:01pm Thanks very much. I can see your point about the fact that the "input-referred noise" output of pnoise simulation produces infinite power, as in a resistor. But on the other hand for the switched-capacitor integrator, what is usually referred to and derived in literature is the "total integrated noise power at the input", which is effectively the white PSD*ENBW, as you mentioned. (e.g. to find the input referred SNR.) But I am not entirely sure what the ENBW of this circuit is. Based on the expected input-referred noise power, it should be fs/2. Is that correct? Regards, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Dec 8th, 2009, 1:31am alireza wrote on Dec 7th, 2009, 9:01pm:
You can determine ENBW from Integral[|gain(freq)|2, 0<=freq<+infinity]. |
Title: Re: switched capacitor integrator noise simulation results Post by pirate on Jan 7th, 2010, 6:50pm Interesting, this pnoise noise bandwith has puzzled me for quite some time as well. It infuriated quite a bit a real homeless :) appendix C4 from Joel R. Phillips and mentioned by jeffyan is clear; "Note that the spectrum of the discrete (sampled) process is periodic in frequency with period 1/T . All noise power is aliased into the Nyquist interval [-1/2T; 1/2T] (or, equivalently the interval [0; 1/T ]). Generally it is the noise spectrum which is available from the circuit simulator. To obtain the autocorrelation function or time-varying noise power, an inverse Fourier integral must be calculated" http://www.cadence.com/rl/Resources/white_papers/tdnoise.pdf mathematically you integrate the random variable on [-inf: +inf] physically you integrate the (periodic and sampled) noise on [0; 1/2T] [1/2T; 1/T] or [-1/2T; 0] is not used because the physical system (SHA, ADC else) does not work with negative frequencies. |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Jan 8th, 2010, 4:23am pirate wrote on Jan 7th, 2010, 6:50pm:
Integral range is [-1/2T,+1/2T] mathematically. pirate wrote on Jan 7th, 2010, 6:50pm:
Reason of integral range [0,+1/2T] is simple. Evaluated PSD is single sided value not both sided value. http://www.designers-guide.org/Forum/YaBB.pl?num=1059089369/1#1 |
Title: Re: switched capacitor integrator noise simulation results Post by pirate on Jan 8th, 2010, 10:38am Quote:
Sorry, by random variable I meant a non periodic one. Quote:
Yes, PSpectrumD is single sided because it is a physical concept so, herein my simplistic view of negative frequencies (there is no signal beyond or equal to 0Hz). at least periodic noise simulation is geting more and more popular and better undertsood |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Jan 8th, 2010, 10:48am pirate wrote on Jan 8th, 2010, 10:38am:
So an integral range is [-1/2T,+1/2T] mathematically. pirate wrote on Jan 7th, 2010, 6:50pm:
pirate wrote on Jan 8th, 2010, 10:38am:
Single Sided PSD has no relation with Physical System generally. For example, consider Poly-Phase filter whose spectrum must be expressed as both sided not single sided. But this is physical system. Expression as Single-Sided Spectrum is restricted to "Real Value System". On the other hand, expression as Both-Sided Spectrum can be applied for both "Real Value System" and "Complex Value System". Here negative frequencies are also physical as well as positive frequencies. Quadrature Sampling, I/Q-ADC etc. are also "Complex Value System". See http://www.designers-guide.org/Forum/YaBB.pl?num=1238705966/3#3 pirate wrote on Jan 8th, 2010, 10:38am:
I don't think noise analysis like "tdnoise of Cadence Spectre" is popular. And I don't know actual instruments which can evaluate noise as same manner as "tdnoise of Cadence Spectre". |
Title: Re: switched capacitor integrator noise simulation results Post by pirate on Jan 8th, 2010, 1:20pm Quote:
sorry, I meant mathematical in the core sense, tdnoise of Cadence Spectre is physics if not circuits to me. A non periodic probability density ranging [-inf; +inf] = 1 or 100%. At the circuit level that gives me a hint on the noise bandwidth [0; +1/2T] for a cyclostationary process. Interesting your polyphase filter example, I have no experience with them. I have only done periodic noise, aka pnoise, (bda, cadence) simulations on low pass circuits where the density belong to the Real domain so I never considered the negative side of the spectrum. Thanks for clarifying the real and complex domain. :wq |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Jan 9th, 2010, 2:16am pirate wrote on Jan 8th, 2010, 10:38am:
pirate wrote on Jan 8th, 2010, 1:20pm:
Noise bandwidth of [0, +1/2T] has no relation to cyclostationary process. Even in "Pnoise(type=sources)", noises are treated as cyclostationary. pirate wrote on Jan 8th, 2010, 1:20pm:
I mean results of "Pnoise(type=timedomain)" are virtual world where an ideal impulse sampling is assumed which is not realized physically, while results of "Pnoise(type=sources)" are real world. Again I don't know actual instruments which can evaluate noise as same manner as "tdnoise of Cadence Spectre". pirate wrote on Jan 8th, 2010, 1:20pm:
Pnoise of BDA's Analog FastSPICE is a type of "Pnoise(type=sources)" not "Pnoise(type=timedomain)". Don't you confuse post processing theory of Transient Noise Analysis of BDA's Analog FastSPICE with Pnoise ? http://www.designers-guide.org/Forum/YaBB.pl?num=1259714206/2#2 Again I don't think noise analysis like "tdnoise of Cadence Spectre" is popular. |
Title: Re: switched capacitor integrator noise simulation results Post by Andrew Beckett on Jan 16th, 2010, 9:30am pancho_hideboo wrote on Jan 9th, 2010, 2:16am:
The main thing I see tdnoise used for these days is for investigation of how the noise varies throughout the period, particularly in switching circuits. Not so much for actual measurements, but to aid understanding of the cause of the noise. So it does get used - not sure how you class "popular" though ;) In the past it used to be used more for measuring the noise at a transition, and converting that into jitter (using the slew rate at the transition) - again for switching circuits. Now that you can do this directly with the jitter mode of pnoise (actually pmjitter) there's less need to use tdnoise itself (although pmjitter is still using tdnoise underneath). I would say that the jitter mode is quite "popular" though... Regards, Andrew. |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Jan 16th, 2010, 10:33am pancho_hideboo wrote on Jan 9th, 2010, 2:16am:
although both "sources" and "timedomain" are no more than small signal analysis in frequency domain. There are many many many people who have much confusion about interpretations of this periodic noise PSD in this forum. This is an original topics of this thread, "Pnoise(type=timedomain)" v.s. "Pnoise(type=sources)". BTW, see BDA's Analog FastSPICE RF http://www.berkeley-da.com/news/news_pr/news01_pr_2009_12_08.html http://www.berkeley-da.com/prod/datasheets/Berkeley_BDA_AFS_RF_WP.pdf BDA claims like followings. AFS RF provides full-spectrum periodic analysis (AFS RF pnoise) that performs device noise computations directly in the time domain and then translates the results to the frequency domain. Using a unique computational approach, AFS RF provides the equivalent of an infinite number of harmonics or sidebands on every run. AFS RF also provides the industry’s only full-spectrum oscillator device noise analysis(oscnoise). Phase noise is a large-signal effect, while amplitude noise is a small-signal effect. AFS RF uses a proprietary non-approximate stochastic nonlinear engine for transistor level phase noise analysis and full-spectrum pnoise analysis for amplitude analysis. Cadence Shooting-PSS also doesn't require setting of maximum harmonics, but setting of maximum sideband is required in Cadence Pnoise. This is true for tdnoise. See http://www.designers-guide.org/Forum/YaBB.pl?num=1257901096/1#1 On the other hand, BDA's AFS RF doesn't require setting of maximum sideband even in pnoise analysis. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Feb 18th, 2010, 9:26pm Hi, alireza wrote on Nov 28th, 2009, 8:43am:
pancho_hideboo wrote on Nov 28th, 2009, 8:52am:
I totally see your point regarding the fact that pnoise (sources) simulation takes folding effects into account, which is quite natural and physical, but PAC only considers zero-side band. However, I cannot accept that 2kT/C which is a well-known estimate for the input-referred noise of a switched-capacitor integrator "ignores" noise folding! Otherwise, design methodology used for design of SC delta-sigma modulators (and other SC filters) would all be in question! I am quite sure that 2kT/C estimate does take noise folding into account. (Noise folding is the basic principle of noise analysis for SC circuits.) Regards, Alireza |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Feb 19th, 2010, 1:57am alireza wrote on Feb 18th, 2010, 9:26pm:
Did you evaluate ENBW ? http://www.designers-guide.org/Forum/YaBB.pl?num=1258339986/34#34 |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Feb 19th, 2010, 8:49pm pancho_hideboo wrote on Feb 19th, 2010, 1:57am:
The reason why I mentioned this was that I believe the 2kT/C estimate for the total input-referred noise power is very true, and includes everything physical which happens in the real world. (i.e. measurement.) Therefore, this value should be treated as a reference for the total discrete-time input-referred noise power. (In fact, 2kT/C noise power is caused by the switches noise, i.e. when the opamp noise is negligible compared to switches. In the literature, there are also derivations of the total noise power when the opamp noise is dominant.) For the discussion here, we are only analysing/simulating switches noise. Based on what you mentioned the rise in the spectrum of the input-referred noise in pnoise (sources) sim is due to the fact that output noise spectrum is folded, but the gain transfer function only considers zero side band. Therefore, the input spectrum has the effects of noise folding uncancelled. Folded noise only manifests itself at high frequencies, since the integrator gain is low. However, when we look at the system as a discrete-time system folding should not be taken into account, as in discrete-time domain there is no freq. component beyond fs/2. So, the discrete-time input-referred noise power should be calculated based on the low-frequency portion of the psd obtained from spectre multiplied by fs/2, where the effect of folding is negligible due to high integrator gain. This way the simulated DT input-referred noise power is almost exactly 2kT/C, and matches very well with the theory. Otherwise, if someone only integrates the input psd obtained from pnoise (sources) from 0 to fs/2, the result is 10-20dB (depending on opamp parameters, etc) larger than the 2kT/C value which is not acceptable. Regards. |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Feb 20th, 2010, 7:18am alireza wrote on Feb 19th, 2010, 8:49pm:
And still you can not evaluate ENBW as maximum frequency for integration. Simply consider single noise observation point for multiple noise inputs. Here multiple noise inputs mean actual noise sources inside circuits and folding noises of each actual noise sources. Then consider meaning of equivalent input noise for one defined input. There are two switches which generate noise in SC integrator at least. Contribution to output from these noises are correlated in time domain. Again see http://www.designers-guide.org/Forum/YaBB.pl?num=1258339986/27#27 [Note] Output noise and input noise are both k*T/C for simple continuous time RC LPF, although input noise PSD is perfectly flat until infinite frequency. In this case there is only one noise source and there is no folding noise. alireza wrote on Feb 19th, 2010, 8:49pm:
Output noise and output noise PSD are both absolutely physical observables. But both input noise and input noise PSD are not. alireza wrote on Feb 19th, 2010, 8:49pm:
I think Discrete-time system which you mean is an ideal impulse sampling system whose transfer function described by H(z=ejωTs). Even by an ideal impulse sampling, noise folding effects are included naturally. See last comments in http://www.designers-guide.org/Forum/YaBB.pl?num=1258339986/23#23 I think you are misunderstanding my answer in the following. http://www.designers-guide.org/Forum/YaBB.pl?num=1260505047/1#1 Any gain given by PAC is not H(z=ejωTs). I don't care at all whatever interpretation or understanding you take. So take any interpretation or understanding which you think most preferable. |
Title: Re: switched capacitor integrator noise simulation results Post by alireza on Feb 21st, 2010, 10:18pm pancho_hideboo wrote on Feb 20th, 2010, 7:18am:
The fact that you are suggesting to use ENBW, implies that the input noise to the system should be "white", Otherwise it makes no sense to use ENBW for a colored input noise! In this case, the input noise spectrum from spectre is NOT white! How do you use ENBW for it? pancho_hideboo wrote on Feb 20th, 2010, 7:18am:
So far you have provided different theoretical reasons for the rise in the input noise spectrum, and one was correlation between the noise sources. But this does not make sense. In a SC integrator there are 4 switches and their noise sources are all uncorrelated. I have confirmed this statement with simulations. In simulations the noise of the switches was enabled one at a time, and still the same colored spectrum can be observed for the input-referred noise. In fact, it is only the switch noise of the integration phase which causes the colored psd. The psd due to sampling phase switch is white! Please see the attached plot. This plot has been obtained using pnoise (sources) simulation, but the result is plotted up to fs/2. Because the singular points at fs harmonics, do not allow the rise near fs/2 be observed. Given that with only one noise source this phenomenon still happens proves that this explanation is wrong! alireza wrote on Feb 19th, 2010, 8:49pm:
I am not sure if you are familiar with design of delta-sigma modulators. In DS Mods, there is the concept of STF (Signal Transfer Function). In many DS Mods STF=1; meaning that when you measure the output SNR you have effectively measured the input SNR, hence the input noise. BUT I prefer not to be so picky about details, and look at the main problem instead. what I meant in my previous post was that 2kT/C is correct, accurate and the simulation is accurate only if it can match to it. (I can provide references, if there is any doubt in accurateness of this statement.) Given this, and the fact that with the rise in the input psd, simulation results is 10-20dB higher than reality, how would you justify the simulation result? IF you think the integration bandwidth is not fs/2 then what is it? I think if you can help me in answering this question, the problem is solved easier. This has been my question since post #1! 1) pancho_hideboo wrote on Dec 11th, 2009, 4:14am:
2) pancho_hideboo wrote on Feb 20th, 2010, 7:18am:
What do you mean it is not H(z)? Aren't 1) and 2) in contradiction? The gain of the output zero-side band/ input zero side band IS EXACTLY H(z). I have again confirmed this by plotting the PAC gain, importing it to MATLAB and comparing with |H(z)| given by MATLAB. |
Title: Re: switched capacitor integrator noise simulation results Post by pancho_hideboo on Feb 22nd, 2010, 12:21am I think you can't uderstand differences between ideal circuit assuming ideal impulse sampling and actual circuits. Actual SC integrator is no more than continuous time domain circuits if seeing from Cadence Spectre. This is true even if you don't use Sample&Hold circuit for output. Also this is true for any SPICE type simulator including Cadence Spectre. And this is true for real physical actions of SC integrator. alireza wrote on Feb 21st, 2010, 10:18pm:
alireza wrote on Feb 21st, 2010, 10:18pm:
Only output noise [Vrms2] and gain are real observable. But both input noise [Vrms2] and input noise PSD [Vrms2/Hz] are not. Approximately, input noise[Vrms2] is "output noise"/gain(freq=0)2 for LPF application. |
The Designer's Guide Community Forum » Powered by YaBB 2.2.2! YaBB © 2000-2008. All Rights Reserved. |