The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> Circuit Simulators >> How to use spectre to measure noise of a Chopping Integrator?
https://designers-guide.org/forum/YaBB.pl?num=1386771810

Message started by prestonee on Dec 11^th, 2013, 6:23am

Title: How to use spectre to measure noise of a Chopping Integrator?
Post by prestonee on Dec 11^th, 2013, 6:23am

I have been attempting this a few various ways, each way is giving me different number results. the integrator is followed by another sc integrator so I am using noisetype=timedomain

I tried tying setting the inputs dc to cm with each having a .5 AC in opposite polarity, and make the pss period long enough to see integration in both chop phase. but this didnt seem to work. I also tried allowing it to integrate a signal and resetting the integrating cap periodically, but the frequency of the reset impacted noise results. I also was told by the local cadence rep spectre wouldnt work with chopping...but az and chopping seem like common techniques that it would be designed to handle. And lastly I was recommended by a coworker to attempt to make a .va SH and add the SH delay output to the non delayed output to basically try to cancel the low freq noise out to mimic statistical cancellation a chopper would see over time.

Are any of these the right method? Is there a tried and true procedure for handling chopping circuits?
-Pb

Title: Re: How to use spectre to measure noise of a Chopping Integrator?
Post by prestonee on Dec 12^th, 2013, 8:56am

Ok allow me ask more specific questions.

1)When performing the PAC of a SC integrator. does the output voltage of the PAC represent the linearized closed loop gain or the openloop gain?

2)In documentation it mentions that for pss that only the sample clock is being driven, such as in mixer or sc circuit, but what about when you have two clocks of different frequency, in my case the sample clock and the slower chopping clock(chopping is half freq of sampling clock). Pss forces me to state fundamental freq as the slowest freq to obtain periodicity, but I also saw references of the fundamental clock being the sample rate.

3) Question about the linearizing of the transfer function , how can i correlate the output of a PAC to the closed loop of the system? In other words if i have a closed loop gain of 5 based on feedback ratios, is there a way to manipulate the v(out) of a pac to see 5?

4) after running a pnoise on outputs, can i simply divide the n() by the v(out) waveform pf pac to get the input referred noise?

Title: Re: How to use spectre to measure noise of a Chopping Integrator?
Post by Ken Kundert on Dec 12^th, 2013, 12:06pm

You need to run a PSS analysis before running PAC. PSS analysis requires that the have a periodic steady-state solution. It sounds like your integrator is actually integrating, so that output is not periodic. To run PSS/PAC you probably want to remove the large signal input, provide feedback so that there is a stable periodic operating point, and then run PSS/PAC.

Quote:

I also was told by the local cadence rep spectre wouldn't work with chopping.

That person is wrong.

Quote:

1)When performing the PAC of a SC integrator. does the output voltage of the PAC represent the linearized closed loop gain or the openloop gain?

PAC just computes voltages and currents. The testbench determines what is measured.

Quote:

2)In documentation it mentions that for pss that only the sample clock is being driven, such as in mixer or sc circuit, but what about when you have two clocks of different frequency, in my case the sample clock and the slower chopping clock(chopping is half freq of sampling clock). Pss forces me to state fundamental freq as the slowest freq to obtain periodicity, but I also saw references of the fundamental clock being the sample rate.

There must be a periodic solution, so the chopping and sampling frequencies must be co-periodic. Use the lowest common multiple as the fundamental frequency.

Quote:

3) Question about the linearizing of the transfer function , if this is a sc integrator whose output is increasing by 100mV steps due to a given input every 1us, does the linearization mean it converts the transfer function that results in a discrete time stair step to a continuous time straight line of form out=Acl*in?

Linearization means that it is as if the input signal is very small, so small that it cannot generate a nonlinear response.

-Ken

Title: Re: How to use spectre to measure noise of a Chopping Integrator?
Post by prestonee on Dec 12^th, 2013, 12:30pm

Ok thank you.

Currently I have both inputs sources of the sample caps set to cm to allow pss converge, any offset that gets integrated is removed by the chopping integrate on the following cycle so it appears to converge ok.

I ask about PAC because it shows my output voltage at around 3000 with input of 1v, at first I thought this was showing openloop, but since it is an integrator that isnt being reset, maybe the pac results are of the integrate function(transfer function of a integrator should be infinite at dc right, 1/s)

And if I am trying to do input referred noise, can i take the result of pnoise and divide it by the pac waveform?

using the amp in an integrate configuration is messing with me a bit :-?

Title: Re: How to use spectre to measure noise of a Chopping Integrator?
Post by Ken Kundert on Dec 12^th, 2013, 12:37pm

Please remember that we cannot see your circuit or your testbench, so it is difficult to say with any confidence what you are doing. It sounds like you are running the integrator open loop, which is probably wrong.

PNoise can compute the input referred noise directly. No need for you to combine the PAC results with the PNoise results to get input referred noise. Just specify the input source to the PNoise analysis and it will work.

-Ken

Title: Re: How to use spectre to measure noise of a Chopping Integrator?
Post by prestonee on Dec 12^th, 2013, 1:01pm

I apologize for not showing something graphical. This is a typical sc integrator with chopping. phase1 is sampling input caps with respect to cm. phase2 connects the input caps to the integrator with input being tied to cm.

are you suggesting i have some type of vcvs feeding the output back to the input? i was thinking of doing this if i had convergence issues.

I cant refer my noise tot he very input because of the voltage sources driving the lines. I am interested in having the noise referred to the very input to verify it wont wash my small input signal being generated by the preceding circuit.

Title: Re: How to use spectre to measure noise of a Chopping Integrator?
Post by Ken Kundert on Dec 14^th, 2013, 7:48am

To get meaningful results, the testbench should try to mimic the way the circuit is intended to be used. Since integrators have very large gains, one tends to use them in one of two ways:

They are embedded in a feedback loop (ex. in active filters).
They are active for a fixed period of time and then reset.

To use SpectreRF, you need a periodic solution. That is generally easy in the first case, you include the feedback in the testbench and then run PSS with the clocks applied and the input amplitude set to zero. In the second case, you would add reset circuitry to your testbench and allow the integrator to operate for a fixed number of cycles (the PSS fundamental would then be set to the reset frequency).

After the PSS analysis you would run the PNoise analysis. If in your overall system the output of the chopper will be sampled, then you should probably use a sampled noise analysis. This is because only the noise at the sample point is of interest, noise at other time is ignored by the overall system because of the sampling. If instead the chopper is driving a continuous time filter then you should use 'sources' ('sources' is a non-descriptive name, a more descriptive name would be 'time-average noise').

I don't understand your comment on the input referred noise. I don't know what the 'very input' is. If there is additional circuitry between the true input and the input modeled in your testbench, then you can get the true input referred noise by adding that additional circuitry, or at least a model of that circuitry, to your testbench.

-Ken

Title: Re: How to use spectre to measure noise of a Chopping Integrator?
Post by prestonee on Dec 16^th, 2013, 6:53am

The integrator is part of a sigma delta modulator so while it does have feedback in the system its not periodic feedback.

The input thing. I have 2 voltage sources that drive the input( voltage sources are connected to the sample caps during the sample phase. Ideally I would like to refer the noise to these nodes but because of the ideal voltage sources whos ac=0 the noise referred to these nets = 0, at least that is why I assume its =0 when i set noise referred to v(inp,inn).

Thank you again for replying to me in this thread.
-Pb

Title: Re: How to use spectre to measure noise of a Chopping Integrator?
Post by prestonee on Dec 18^th, 2013, 6:26am

Sorry I messed up with the input , no issue there.

as far as adding feedback, As long as the system works without using reset, the noise numbers shouldnt be affected correct? example having a system feedback that swaps the inputs every three samples vs every 5 samples. the noise/xfer function should be the same since the integrating cap is never losing charge due to a discontinuous reset.

Title: Re: How to use spectre to measure noise of a Chopping Integrator?
Post by Ken Kundert on Dec 18^th, 2013, 8:05am

I recommend that you read http://www.designers-guide.org/Analysis/delta-sigma.pdf.

-Ken