I wanted to offer some very general comments about getting accurate results with the spectre nport component...  

Good s-parameter data is key to nport model stability and accuracy.
Common data problems that I see:
-No DC data available due to hardware limitation
-Not enough bandwidth due to hardware limitation
-Not enough frequency resolution
-Passivity not enforced in the data
-Noisy data caused by poor calibration and DUT contacts
Data interpolation is always needed for frequency-domain sampling
Data extrapolation is needed when DC data is not available, or when extra data bandwidth is needed. Extrapolation can be highly unreliable

So, how do you tell if your s-parameter data is reasonable???
-Does it have a dc point?  (or a reasonable amount of data close to dc?)
The DC point is essential, because without it Spectre cannot do the inverse Fourier
transform. If the DC point is not given, Spectre does extrapolation based on the first
two data points, assuming constant magnitude, and linear phase. This simple extrapolation
works well in some cases, but in general it is better to provide correct DC point. And of
course, if you just add an arbitrary DC point, you cannot expect accurate simulation results.
The DC point should be consistent with the rest of the data

-Are there enough data points to accurately model the DUT?  Use the Smith Chart to verify your data.
Is the S-parameter data (when plotted on the Smith Chart) smooth, rotating in a clockwise direction?  Or is it very choppy and piecewise-linear looking? Sufficient frequency resolution is needed to ensure the frequency-domain data is smooth, and can be safely interpolated

-When you are running your simulation, are you getting any warning messages about "passivity",
"risky extrapolation", etc. ?  Pay attention to any warning messages.

-If you are using the Spectre sp analysis to produce S-parameter data, always produce 3 times the bandwidth data needed.  I would generate s-parameters to 3X the highest fundamental frequency needed.  


A couple of notes on Linear vs Spline vs Rational interpolation:
Linear and spline
When the input data points are sufficient, linear and spline interpolation produce
comparable model accuracy. However, when the input data points are scarce,
linear interpolation is preferred to bound jumps between data points.

Rational interpolation
The rational model is not as robust and accurate as the convolution based model, particularly when the input data is insufficient, and the DUT is a multi-port system (with n> 4).
Use interpolation = rational only for analyses where interpolation = spline or linear is not supported.

For the nport option interpolation method = rational, a state-space model is synthesized to match the frequency domain data, and that model is simulated in the time domain along with the rest of the circuit. This approach can be problematic with non-passive data, and occasionally with elements that are high impedance at high frequency, such as inductors.

Best regards,

Tawna