Two of the most important parameters for accuracy are:
(1) maxharms
(2) oversample
maxharms:
The accuracy of HB is greatly dependent on the number of harmonics chosen. As a general rule, you should simulate with with maxharms [5 3] and (if memory allows) re-simulate with maxharms [7 3] to see if the answer changes (or not). If the answer does not change, you can be confident about your results.
For circuits with dividers, the number of harmonics chosen should equal 5 times the divide ratio. This is regardless of whether you are using the shooting or HB engine.
examples:
You are simulating with PSS and (say) the L0 input is 400M. There is a divide by 4 in your circuit, so the PSS beat frequency is 100M.
You will need 5x4= 20 harmonics.
For QPSS, if you have a divide by 4 with these frequencies: Mixer LO=400M, RF=100M+1M, you'll need [20 3] for your maxharms. Also, you'll need to set the freqdivide parameter in the qpss Choosing Analyses form in ADE to 4.
oversample:
The harmonic balance algorithm contains the Fourier Transformation. If oversample=1 (which is the default) and the harmonics number is K, then there will be (2K+1) samples in one period of waveform. If the oversample=S, there are (2K+1)*S samples. Increasing the number of oversamples will result in more sample points in the time-domain, and thus the HB spectrum and subsequent noise calculation will be more accurate.
Unlike the harmonic number (maxharms), oversample only affects the number of samples in the time-domain, and won't increase the number of unknowns (which is determined by the equation and harmonic number). Thus increasing oversample is an effective way to improve accuracy without using too much memory as increasing harmonic number.
An application where one would want to use oversample is on the LO of a mixer, because it is a strongly nonlinear signal in need of more samples in time-domain for an accurate Fourier Transformation.
For more information, similar tips, and design topics, please visit sourcelink.cadence.com.