What's Going On Up There? Follow-up part 2
What you are seeing in these two graphs is evidence that, despite being labeled as a 24/96 recording, the Steely Dan was mixed and mastered apparently at 48kHz, hence the rolloff above 20kHz (footnote 1). It was then transferred to analog tape, which results in the plateau'd noise floor above 24kHz, which itself is rolled off above 40kHz by the dCS ADC used to redigitize the data. Yes, the digital transfer for DVD-A was made at 96kHz, but there is no original signal content above half the original sample rate of 48kHz, just analog tape hiss.
My curiosity whetted, I looked at some other tracks. Fig.5 is a spectrogram of samples from three consecutive tracks on the Warner DVD-A sampler. The first and third, from singer Luis Miguel and Emerson, Lake, and Palmer, are labeled as 96kHz recordings; the second, from Stone Temple Pilots, is labeled 48kHz. Yet there are no appreciable differences in their ultrasonic content. All cut off above 24kHz: the first two sharply, which, with the black background in the top octave, implies an original 48kHz digital master; while the blue flecks in the ELP track show that an analog master was involved (as of course it was, given the vintage of this recording).
Fig.5 Spectrogram of Warner DVD-Audio sampler, excerpts from tracks 7-9. Original DVD-A data recorded at 24 bits and 96kHz, 48kHz, and 96kHz sample rates, respectively.
The very clean-sounding "Lucky Man" from ELP doesn't roll off as sharply as the other two above 20kHz, which implies that the lack of ultrasonic bandwidth is due to the analog tape machine used to record the track back in the early 1970s. In addition, there is one vertical red line that extends to high frequencies. This is due to a splash of sibilance as Greg Lake sings the word "ssssatin," and the corresponding FFT-derived spectrum is shown in fig.6. The sibilance results in two broad energy peaks between 5kHz and 20kHz, but it can be seen that these are superimposed on a smooth rolloff of the noise floor up to the 40kHz limit of the dCS converter. The ELP track may not have had much content at all above 24kHz before it was digitized at 96kHz, but this is due to its vintage, not to the mastering.
Fig.6 FFT-derived spectrum of vocal sibilance from the word "ssssatin" from ELP's "Lucky Man" on the Warner DVD-A sampler (linear frequency scale, 0Hz-48kHz, 6dB/vertical div., left channel cyan, right channel magenta).
So, do any of these high-sample-rate recordings have the sort of ultrasonic content you'd expect from DVD-A? Yes, as can be seen from fig.7, which shows the spectrogram of a closely miked set of cymbals on the unofficial Japanese DVD-A sampler Jonathan was sent for his review. The original was recorded at 192kHz, and spectral energy can be seen to be present all the way out to 40kHz. (Ignore the vertical white line in this graph, which is the "time line"—the track was being played when I captured this screenshot, as you can see from the active meters at the bottom of the image.) The spectrum (fig.8) smoothly rolls off with increasing frequency, revealing that this is, indeed, a true high-sample-rate recording.
Fig.7 Spectrogram of "Dancing Cymbals," from an unofficial Japanese DVD-Audio sampler. Original DVD-A data recorded at 24-bit/192kHz.
Fig.8 FFT-derived spectrum of excerpt from fig.7 data (linear frequency scale, 0Hz-48kHz, 6dB/vertical div., left channel cyan, right channel magenta).
Footnote 1: If the DVD-A does sound better to you, it is perhaps because the CD had to be downsampled to 44.1kHz from the 48kHz master, which is a particularly awkward conversion, mathematically speaking.