Hi,

I'm trying to simulate phase noise of an injection locked oscillator in Cadence with PSS and PNOISE. The problem is when i use a periodic source for injection, PSS simulation fail and an error occurs : " 'V0' is a periodic input signal, which is inconsistent with autonomous circuits. "

So my question is:  How do we simulate the injection locked oscillator phase noise ?

Is there a way to conceal the periodic voltage source in the circuit so as not to piss the pss off ?


thanks,

Hannibal

The periodic source is for injection, it must be a source for the injection signal. The problem is when we select "Oscillator" field in PSS simulation and it fails because of that (on the one hand there is a periodic source, on the other hand there is an autonomous circuit because "Oscillator" field in PSS is selected)

@Frank Wiedmann

Ok I agree with you but how we can take into account the nonlinearity of oscillator in phase noise calculation !! based on your post, I will have only the injection signal in PSS simulation !! I'm wondering if with this configuration, PSS will consider that there is an oscillator in the circuit and thus its nonlinearities ?

Frank Wiedmann wrote on Feb 15^th, 2012, 6:00am:



That's it !! I tried your configurantion (PSS with only the injection signal) but currently I have convergence problem !!

Does anyone know the difference between shooting method and harmonic balance in PSS simulation ? Is that the shooting method is more accurate than harmonic balance ?

Thanks,

Hannibal

raja.cedt wrote on Feb 15^th, 2012, 12:40pm:

According to Wikipedia (http://en.wikipedia.org/wiki/Autonomous_circuit), an autonomous circuit is a circuit that produces a time-varying output without having a time-varying input. An injection-locked oscillator (see http://en.wikipedia.org/wiki/Injection_locking) does have a time-varying input, so it is not autonomous. This has absolutely nothing to do with the internal structure of the circuit.

I do not know exactly how cadence implemented the shooting convergence checks but I have read in the past some source code comments of Spice3 derivates with shooting implementations.

The method for non-autonomous circuits is that beginning with the starting time for the convergence check all typical non-equidistant time steps are stored. And for the next cycle of the smallest dominator frequency these time steps are recycled. If the recycled time steps lead to violation of the integration time convergence check, the shooting period starting is delayed further. If this period recycling time steps works then all vectors at all time steps are checked for convergence.

For the practical injection locked oscillator the situation could be either that the oscillator is locked after some time or not. If it does not lock because typical the injection amplitude is too low than there could be no closed period orbit in the space defined by the voltages and currents of the network.

What I have taken with me in my memory from the shooting implemenation is that there are so many parameters to tune the shooting convergence that is very difficult w/o a big number of test cases to have a robust implementation. On the other side the developers did had the strength to open access to all these numbers to the command line and provide help to the circuit designers with some kind of autotuners. That is a common practice for matrix runtime tuning to machines memory architecture today.