The issue is that the body is not rigid and it is not firmly attached to the bike. A spring-mass model for the human body has been developed which shows how vibrations are attenuated between body parts (no snickering from the crowd: that wasn't modeled). The result of this is, the higher the frequency, the less the fraction of the body mass which is contribute to the "inertia" term in the spring-mass system. The mass ends up being frequency-dependent. In other words, it's complicated.
So we come back to Champoux's result:
From this table, you'd expect frequencies above around 20 Hz would not be effectively transmitted from the road to the bike, although the bike contains a resonances at 34 and 51 Hz which might be a factor, depending on the nature of the impact and where the vibrations are measured.
As I already noted, Champoux rode his bike on a treadmill on which he'd attached a bump which he hit every 0.9 seconds or so. He moved accelerometers around the bike to search out the modes. Unfortunately he showed time-domain data only for the stem. I'm not so interested in vibrations at the stem, which can clearly flex, since the point of this whole exercise was to model what's happening at the pedal spindles.
Hastings, Blair, and Culligan @ MIT and Pober at UMass did a similar experiment in 2004, with Kevin Monahan (then US criterium champion) riding each of three Cervelo bikes set up with equivalent wheels and with essentially equivalent geometry, also on a treadmill with a bump. The bike hit the bump once per second. An accelerometer was attached to the base of the seatpost to measure vibrations. Here's a plot of a typical response:
It stands to reason the acceleration spectrum at the pedal spindles would be similar. They saved me the work of doing a Fourier transform of their data. Here's a plot they made of the power spectrum (proportional to acceleration squared) for the three bikes, "CF" = carbon fiber, "AL" = aluminum, and "S" = steel:
Note the broad peak in the vicinity of 55 to 60 Hz. This corresponds nicely with Champoux's mode identified at 51.4 Hz: frame torsion and lateral motion of the fork. Since the fork is generally more flexible than the frame (as demonstrated in Tour Magazine tests), the fork will have a large influence on the frequency of this mode. Since the same fork model was used on each of the Cervelo test bikes, it is understandable that the spectra of the three bikes near this frequency are similar.
As an aside, the authors were particularly interested in how the frame materials affected the spectra. The carbon fiber dampen vibrations in the 30 Hz range, and Al has higher vibration magnitudes near the 55-60 Hz peak.
Anyway, it's clear the effect of vibration modes within the bike-body system are important. The vibration "noise" I constructed intuitively doesn't quite cut the mustard.
An interesting issue here is these vibration components show amplitudes exceeding 1 g. That means the seatpost, at least, experiences negative g's. So does the bike go flying? Well, peak displacement at a given frequency equals peak acceleration at that frequency divided by (2πf)². So at one g (9.8 m/sec²) at 50 Hz, the peak displacement is 0.1 mm. So the bike doesn't go flying: we're talking minor vibrations substantially less than the inflated tire thickness.
Anyway, I'm changing my noise model to better represent the importance of this oscillation mode near 55 Hz.