The Designer's Guide Community
Forum
Welcome, Guest. Please Login or Register. Please follow the Forum guidelines.
Mar 28th, 2024, 11:47am
Pages: 1
Send Topic Print
simulating noise & transfer function of switched capacitor FIR filter (Read 7344 times)
Cardell Phillips
New Member
*
Offline



Posts: 3

simulating noise & transfer function of switched capacitor FIR filter
Jan 17th, 2009, 11:23am
 
I'm designing a switched capacitor FIR filter with a large number of taps.  I would like to simulate the noise and small-signal transfer function of the filter in spectre or spectreRF, but this circuit is fundamentally difficult or even impossible with PSS analysis.  

The problem with running this in PSS is that, with a large number of taps, the beat frequency becomes ridiculously small even though the clocks for all taps are derived from a single master clock.  For instance, say my master input clock has a period of Tck.  Then, the clock period for each tap is as follows:
tap1:  T = Tck
tap2:  T = 2*Tck
tap3:  T = 3*Tck
.
.
.
tapN: T= N*Tck

The PSS beat period for such a circuit becomes the Least Common Multiple of the period of all these taps.  So, if you had say 20 taps, the beat period would be 232,792,560*Tck - i.e. the least common multiple of 1,2,3,4,...20.  Clearly, I can't run a PSS on a 20 tap filter.  Simulating this in PSS seems to be impractical when the number of taps is greater than 8 or so.  

Is there a better solution for simulating the noise of this kind of circuit?  Transients actually run quite fast on the circuit, so I can run a 'transient noise' sim and get an idea for the noise behavior, but I can't verify that the small-signal transfer function matches my Matlab simulation.  I'm also curious how accurate are transient noise simulations when devices are hard switching.

Well I know this a long post, but it's a complex problem.  Can anybody out there point me in the right direction?  Thanks!!
Back to top
 
 
View Profile   IP Logged
pancho_hideboo
Senior Fellow
******
Offline



Posts: 1424
Real Homeless
Re: simulating noise & transfer function of switched capacitor FIR filter
Reply #1 - Jan 17th, 2009, 8:07pm
 
Your post is usage of specific EDA vendor's simulator. There is no Design issue.
Please post to "The Designer's Guide Community Forum ≫ Simulators ≫ RF Simulators".

Cardell Phillips wrote on Jan 17th, 2009, 11:23am:
The problem with running this in PSS is that, with a large number of taps, the beat frequency becomes ridiculously small even though the clocks for all taps are derived from a single master clock.  For instance, say my master input clock has a period of Tck.  Then, the clock period for each tap is as follows:
tap1:  T = Tck
tap2:  T = 2*Tck
tap3:  T = 3*Tck
.
.
.
tapN: T= N*Tck

First "beat" frequency of PSS setting UI is not beat frequency. It is a fundamental frequency not beat frequency.

Why is a period of tapN N*Tclk in your FIR ?
Your FIR fiter is decimated for averaging ? If so, a fundamental frequency has to be small.

Cardell Phillips wrote on Jan 17th, 2009, 11:23am:
The PSS beat period for such a circuit becomes the Least Common Multiple of the period of all these taps.
So, if you had say 20 taps, the beat period would be 232,792,560*Tck - i.e. the least common multiple of 1,2,3,4,...20.

No, you are misunderstanding PSS.
The PSS fundamental period for your circuit becomes lowest common multiple period of all these taps. So it is N*Tck.
The PSS fundamental frequency for your circuit becomes highest common divisor frequency of all these taps. So it is 1/(N*Tck).

In your case, fundamental frequency is 1/(20*Tclk) and fundamental period is 20*Tclk.
Back to top
 
« Last Edit: Jan 18th, 2009, 3:51pm by pancho_hideboo »  
View Profile WWW Top+Secret Top+Secret   IP Logged
Cardell Phillips
New Member
*
Offline



Posts: 3

Re: simulating noise & transfer function of switched capacitor FIR filter
Reply #2 - Jan 17th, 2009, 9:09pm
 
Thanks for the reply.  

I'll try to clafify why the period for the Nth tap is N*Tck:  The Nth tap has N storage elements.  A storage element of the Nth tap samples the input and then waits N master clock periods it is evaluated in the output.  Right after evaluation, the storage element goes back to the input for a new sample.  Hence, the total effective 'sample-evaluate' period for the Nth tap is N clock cycles.  However, the storage element is only sampling or evaluating for 1 period of the master clock.  

The problem is that each tap has a different delay.  Therefore I have all these different tones and the resulting fundamental for the collection of tones is a really low frequency.  Too low for PSS to be practical.  I'd like to figure out if another type of simulator could handle this sim - perhaps harmonic balance?  I don't know if any circuit-level simulator can handle having 20 large-signal tones.  
So... I'm thinking I'll be stuck with hand calculations and transient noise sims to verify the noise  :(

Back to top
 
 
View Profile   IP Logged
pancho_hideboo
Senior Fellow
******
Offline



Posts: 1424
Real Homeless
Re: simulating noise & transfer function of switched capacitor FIR filter
Reply #3 - Jan 17th, 2009, 9:15pm
 
I think your SC-FIR is similar to the following.
http://www.designers-guide.org/Forum/YaBB.pl?num=1218622880/4#4

Cardell Phillips wrote on Jan 17th, 2009, 9:09pm:
The problem is that each tap has a different delay.
Therefore I have all these different tones and the resulting fundamental for the collection of tones is a really low frequency.  Too low for PSS to be practical.

No, you are still misunderstanding.
There is no effect to a decision of a fundamental frequency even though each tap has a different delay.
So even though there are 20 large signal tones, a fundamental frequency of PSS is 1/(20*Tclk) not 1/(232,792,560*Tclk).

Show me portions of your netlist regarding signal sources, analysis statements, option statements
and related parameter statements where a fundamental frequency of PSS is 1/(232,792,560*Tclk).


Cardell Phillips wrote on Jan 17th, 2009, 9:09pm:
I'd like to figure out if another type of simulator could handle this sim - perhaps harmonic balance?

About a fundamental frequency, a situation is same even in HB analysis. And HB analyisis is not suitable for switched capacitor circuits.

Cardell Phillips wrote on Jan 17th, 2009, 9:09pm:
I don't know if any circuit-level simulator can handle having 20 large-signal tones.
 
I think any circuit-level simulator can handle having 20 large-signal tone if consumption memory is permissible.
If there is no common divisor frequency between 20 large-signal tones, you have to invoke QPSS analysis. But in this case, there are too many fundamentals, so it is not practical.

In your case, all frequencies are multiple of 1/(20*Tclk), so you can't use QPSS.
PSS with fundamental freq=1/(20*Tclk) is enough useful for your case.
Back to top
 
« Last Edit: Jan 18th, 2009, 8:14am by pancho_hideboo »  
View Profile WWW Top+Secret Top+Secret   IP Logged
Cardell Phillips
New Member
*
Offline



Posts: 3

Re: simulating noise & transfer function of switched capacitor FIR filter
Reply #4 - Jan 18th, 2009, 12:49pm
 
Thanks again for the response.  I may have been unclear in my description...let me clarify.

The frequencies for each tap are not a multiple of 1/(20*Tck).  That is the reason this is a difficult simulation problem.  

For instance, the clocks for the 19th tap have a frequency of 1/(19*Tck).   1/19 is not a multiple of 1/20.  Also, 1/18 and 1/17 are not multiples of 1/20.

Imagine the simple case of a 4-tap FIR.  The first tap has a delay of Tck and a frequency of 1/Tck = fck.  The 2nd, 3rd, and 4th taps have sample periods and frequencies as shown:

tap1   T1 = 1*Tck   f1 = fck
tap2   T2 = 2*Tck   f2 = fck/2
tap3   T3 = 3*Tck   f3 = fck/3
tap4   T4 = 4*Tck   f4 = fck/4

The fundamental frequency for this circuit would actually be fck/12, not fck/4.  This is because the least common multiple of 1,2,3 and 4 is 12.  
Back to top
 
 
View Profile   IP Logged
pancho_hideboo
Senior Fellow
******
Offline



Posts: 1424
Real Homeless
Re: simulating noise & transfer function of switched capacitor FIR filter
Reply #5 - Jan 18th, 2009, 3:55pm
 
I understand your situation.

In your case, QPSS is also not suitable.

You should use Envelope Analysis with fudamental frequency of 1/(20*Tclk).

Envelope Analysis of Agilent ADSsim and GoldenGate support small signal noise analysis.
I recommend you to use RFDE2008U2 or GoldenGate-4.3.0. GoldenGate-4.3.0 is more recommended since convergence is very robust.
But GoldenGate-4.3.0 might not be suitable for your purpose.
So you might have to use RFDE2008U2 not GoldenGate-4.3.0 since we can't evaluate Time_Averaged_Noise_Power_Density(freq).
http://eesof.tm.agilent.com/products/goldengate_main.html#whats_new

By using these simulators, you can observe small signal noise Noise_Power_Density or Noise_Power around any harmonics you specify.

RFDE2008U2 calculates Time_Averaged_Noise_Power_Density(freq).
GoldenGate-4.3.0 calculates Ensemble_Averaged_Noise_Power(time).
So you can observe time behavior of small signal Noise_Power by using GoldenGate-4.3.0.
This is a little similar to Pnoise/TDnoise Analysis of Cadence Spectre where Ensemble_Averaged_Noise_Power_Density(time,freq) is evaluated.

About monte carlo type noise analysis similar to transient noise analysis, it is also available in Envelope Analysis of RFDE2008U2.
But it is not available in Envelope Analysis of GoldenGate-4.3.0.

Although you can evaluate small signal transfer function by using Envelope Analysis of RFDE2008U2, you can not evaluate it by using Envelope Analysis of GoldenGate-4.3.0.
Envelope Analysis of GoldenGate-4.3.0 doesn't support small signal analysis although it supports small signal noise analysis.
But you might be able to evaluate a transfer function by using two tones Envelope Analysis of GoldenGate-4.3.0 where one is N*1/(20*Tclk), the other is small input signal tone.

I don't know whether Envelope Analysis of Cadence Spectre supports small signal noise analysis although it can calculate AC charactersitics at specific envelope time.
http://www.cadence.com/Community/blogs/rf/archive/2009/01/14/tip-of-the-week-mms...
Back to top
 
« Last Edit: Jan 19th, 2009, 8:07am by pancho_hideboo »  
View Profile WWW Top+Secret Top+Secret   IP Logged
Pages: 1
Send Topic Print
Copyright 2002-2024 Designer’s Guide Consulting, Inc. Designer’s Guide® is a registered trademark of Designer’s Guide Consulting, Inc. All rights reserved. Send comments or questions to editor@designers-guide.org. Consider submitting a paper or model.