PSS doesn't converge with nport component

[ezbentley]

I need to run PSS on a mixer circuit which includes a SAW filter, whose s4p file was imported using the nport component. However, no matter what convergence trick I do, the circuit just doesn't converge. The circuit without SAW can easily converge. With SAW, I dont get any error, but it always either reaches the maximum iteration or stops at some point and doesn't proceed.
I already loosen parameters such as reltol, steadyratio, tstab, etc.
Has anyone run PSS simulation on a circuit involving nport component? Any help will be greatly appreciated.

 

[Mazz]

PSS is a time based simulation, while your s4p file is frequency based. the simulator has to convert into a time based model to perform the simulation. This could be the problem.
2 suggestions:
1) use a freq based simulator, if available, such as harmonic balance (ADS or EldoRF)
2) make an RLC network that reproduce the inband SAW characteristics (insertion loss, input and output imedance) and use it to simulate.
I hope it will help
Mazz

 

[BigBoss]

If the s-aparameters are not very well extrapolated or defined, PSS may not converge.Instead, you may try transient and then DFT to obtain very close ( often more accurate ) result.

PSS doesn't like s-parameters but ADS loves them..

 

[ezbentley]

Thanks for the comment from BigBoss and Mazz.

My nport is an IF SAW filter, whose s4p ranges from 25~50MHz with 20dB loss. Transient shows 60dB loss and a huge 2GHz component at the output of the SAW. I don't know if this transient analysis is believable.

Also, I cannot see the passband ripple using transient.

I love ADS, but I have to use spectre at my work. Seems to me spectre isn't a convenient RF simulator at all.

 

[BigBoss]

Have you ever done s-parameter simulation of your circuit?? What they gave you???
Normally, insertion loss of a SAW filer may be around 35-50dB( it's true!!) at that frequency.And if it has a mismatch at its input-output ports, this attenuation can be higher.
Be careful !! Insertion loss may be normal..
Therefore a buffer amplifier is used before to drive SAW filters to compansate this loss..

 

[ezbentley]

I have done both SP and AC analysis on the saw under matched condition, and the results agree with the spec(20dB loss). I found through transient that the output of the saw shows 60dB loss, but suddenly jumps to only 20dB loss after about 1us in simulation. I guess it's an extrapolation issue. So I tried to set tstab to 2us in PSS, the output result is still inconsistent with transient or AC simulation, but I will keep "guessing" ways to make PSS work.

PSS sucks. ADS rocks.

 

[Mazz]

ezbentley
don't penalize too much spectre. Is today most probably the most used RF IC simulator.
Take care how you measure insertion loss using transient. As you are looking at Spar, your insertion loss is a power loss, so measure output POWER/input POWER also with transient.
Anyway, if you're not interested in out of band performance of the SAW, but only at in band components, follow the suggestion to make a passive equivalent network (also equivalent in term of input and output impedances) and do your simulations.
To be sure that, in the real application, you will not have bad "surprises" follow the data sheet of your SAW, and design the proper source impedance for the SAW (that is the output inpedance of the buffer, eventually with external matching network) and the correct load impedance (that is your next stage (VGA?) input impedance.
I hope it can help.
Mazz

 

[BigBoss]

Some primitive versions of Spectre had some convergence difficulties at that time..
Update your simulator..

 

[ezbentley]

Mazz,

Thanks for your advice.

Let me clarify something. AC sim shows voltage loss of 20dB, which agrees with the spec. If I apply a sine wave in the passband, the output of the saw would show a sine wave of the same frequency with 60dB loss in TRAN simulation. The weird thing is, the output wave amplitude would suddenly increase after about 1.2us in TRAN time. After that, the output amplitude would correspond to 20dB loss, which is the correct behavior. I guess this weird behavior is a result of inadequate extrapolation?

So then I set tstab to 3us to run PSS, and it finally converged. But the output amplitude of the saw still shows 60dB loss.

Those are my observations.

Any comment will be appreciated. Thanks again.

 

[Mazz]

Passband response, my friend. Reflect about the meaning of group delay and you'll have the solution.
I bet that GD is around 1.2 us.
It is normal that the filter shows a transient response of that shape. So it is normal that PSS convarge to a correct value after some us. I think your model is correct.
I hope it can help.
Mazz

 

[ezbentley]

Thanks a lot Mazz for pointing me to the right direction! The time domain delay on the spec sheet is about 1.2us, and TRAN simulation confirms this behavior.

However, even though PSS converges after 3us, the voltage amplitude from PSS still differs from the measurement from TRAN. I am suspecting that PSS is using the initial time domain response instead of the final steady state response, but I will need to look into that to be sure. Also, even in time domain, the steady state amplitude would differ depending on the order of the converted time domain model. Interestingly, a higher order model(n=110) does not give correct response.(n=110 gives more loss than specified) I used n=50 and the loss is as specified.(20dB)