Apart from HB-QPSS of Cadence Spectre,
I've been feeling that HB-PSS of Cadence Spectre come to be enough practical level if no NPORT is included in target circuit.

But HB Analysis of Cadence Spectre is very slow when NPORT is included even if there is only one.
This is true for both convolution and rational fitting model.
I tried both tstab=0 and tstab!=0, but no improvement.

When I use HB Analysis of Cadence Spectre,
I often encounter no convergence even if there is only one NPORT in circuit.

Especially if I use large inductor such as 18nH for 2.4GHz operation frequency from the following library,
HB analysis of Cadence Spectre can't converge at all even if it is only one NPORT in circuit.
    http://www.murata.co.jp/cgi-spara/search.pl?primary=lqg15hn&lang=en
    http://www.murata.co.jp/cgi-spara/search.pl?primary=grm15&lang=en

Extraporation for high frequency might cause this convergence failure
since provided maximum frequency data is lower for larger inductor.

At least if I endure very slowness of Cadence Spectre,
I can barely get convergence when I use two or three NPORT of not so large inductor at most in circuit.

But I've never been able to get convergence of HB Analysis in Cadence Spectre when I use more than four NPORT.

When I use HB Analysis of Agilent GoldenGate,
I can get always convergence very quickly without any problem even if there are more than ten NPORT in circuit.


Is there anyone who can get convergence even if more than four NPORT are used in HB Analysis of Cadence Spectre ?
Do you have any trouble when NPORT is used in HB Analysis of Cadence Spectre ?

I tried MMSIM6.2.1, MMSIM7.0.0 and MMSIM7.0.1.
But I can not find out any improvement at all.

Although it doesn't make any sense at all that nports should slow down a harmonic balance simulation. The original benefit of HB was that every component was defined in the frequency domain. So, frequency domain n-port descriptions fit naturally to the analysis. It was always time domain techniques which choked on distributed circuit descriptions because convolution was required to evaluate distributed networks in the time domain. Weird ...

It's still necessary to come up with a DC solution, and to extrapolate for higher frequencies, even without time domain assistance (which of course does require convolution).

Reasons for non-convergence can be various - it can be due to the fact that the rest of the circuit is seeing unrealistic signals at the extrapolated harmonics, which could lead to convergence difficulties in those parts of the circuit (non-linear components do have to be evaluated in the time domain on occasion with HB analyses) which could differ between simulators.

Despite all this, there have been numerous improvements in nport in the last year - as I mentioned earlier on. I've been seeing far fewer problems with nport as a result of this - so I'd still recommend using a more recent spectre version (MMSIM72 just was released). For example, there are now options to check and or enforce passivity, and to control DC extrapolation, and so on - as well as smarts to auto-set nport settings to avoid you having to tweak too much. We recognise that this has been painful, and want to improve the robustness, and continue to do so going forward.

If you still have a problem, PLEASE contact Cadence Customer Support and allow us to try to make further improvements by providing data that illustrates the problem. There will always be specific cases (in all vendors tools) that cause trouble, and all vendors make improvements based on customer feedback and data.

Regards,

Andrew.

RFICDUDE wrote on Dec 17^th, 2009, 7:31pm:
If NPORT is used at location where it does not determine DC bias, for example NPORT is DC blocked,  
pure frequency domain HB Analysis don't have to depend on DC point of S-parameters,
although even in such case, DC components could be generated due to even order distortion.

NPORT located as DC blocked is relative common in Wireless Application, except for load inductance which also works as DC feed.

However for Wired Application, DC component of S-parameter for transmission line is very important.
Generally I set Start=0Hz and Stop=100GHz with AFS(Adaptive Frequency Sampling) as Simulation Frequency Plan in Agilent Momentum.