No. Spice will generate the same IMD from multi-tones, but the lower
level products are easily lost in numerical noise, and are a pain to
extract with FFT.
Actually, the numerical noise point is good. The error mechanism in
harmonic balance is mostly through having too few harmonics. The
differential equations are solved in the frequency domain, where d/dt
becomes multiplication by Laplace 's' -- no difference approximation
error as in transient analysis. And harmonic balance is a uniform
approximation method, unlike transient analysis which can accumulate
error with each timepoint. For circuits where the waveforms are
represented well by a reasonable number of harmonics, accuracy is good
and the noise floor is orders of magnitude lower than transient
analysis. So, as you say, small mixing products are less likely to
get lost in the numerical noise.
HB is best for time-varying circuits with one principle
large signal and any number of smaller signals.
Hmmm. This is true for the periodic steady-state methods, which can't
easily handle multiple large signals -- to be efficient, they really
want one large signal, then they approximate the small signals as
perturbations on a time-varying operating point. But multi-tone
harmonic balance can handle multiple large signals -- wouldn't be very
useful in RF if it couldn't. And the frequencies of these signals
need not be related. This is a killer for time-domain methods, since
the period required must be one for which all signals are periodic,
which could be _very_ long. In this case, harmonic balance can be
_much_ faster than transient analysis.
Transient analysis and periodic steady-state methods have an advantage
in that they don't care (much) about how horribly nonlinear the
circuit is. Harmonic balance does care, in that more harmonics are
required to (a) get convergence, and (b) represent the waveforms well.
And more harmonics will slow the HB simulation. So there are
certainly appropriate places to apply both...