Features

Regarding the necessary word depth required for a Linear PCM (LPCM) system to perfectly preserve the original waveform: Bob Stuart was adamant that, to capture everything the human ear can hear with all kinds of music under all playback conditions with a linear PCM system, you need 20-bit quantization. For those who then ask why we are going for 24-bit data words in DVD-Audio when 20 bits are sufficient, remember that computers store data in 8-bit bytes. For a practical 20-bit system, therefore, you might as well use 24-bit data, 24 being the nearest integer multiple of 8. And as is evident in the screen shots of the spectrograms, the computer's hard drive actually stores the 24-bit data as 32-bit words (two 16-bit bytes), padding the second byte with eight zeroes to the right of the LSB.

However, it can also been seen from these spectrograms that when there is ultrasonic energy present in the recordings, it is always at a low level. This means that to use conventional long-word-length LPCM to encode the wide-band audio signal will inherently be inefficient. Meridian's Lossless Packing (MLP), which has been adopted as part of the DVD-Audio standard, takes advantage of this inefficiency to squeeze the data into a smaller space without losing any information.

The next question involves the second of Michael Riggs's points: that humans cannot hear much above 20kHz, and then only when they are young. But as reader Dr. Charles King pointed out in the August issue's "Letters" (p.17), research exists that indicates a change in measured brain activity when listeners are presented with full-bandwidth music compared with music that is bandwidth-limited to the CD's 22kHz cutoff frequency (footnote 4).

But that brings me to my final and somewhat puzzling spectrogram. I have written before in these pages about how privileged I felt at being asked by Canadian pianist Robert Silverman and Stereophile recording artist to record him in a complete Beethoven Sonata cycle (released in October on the Canadian OrpheumMasters label. I captured the sound of Bob's Bösendorfer at 88.2kHz sampling and a 24-bit word length, using four DPA (formerly B&K) capacitor microphones. The mikes use a ½" diaphragm and extend to above 30kHz in response. Nevertheless, as can be seen in fig.6, the piano sound has almost no content above 15kHz or so. In theory at least, a CD sample-rate version of these data should sound as good as the 88.2kHz original.

Fig.6 Spectrogram of Robert Silverman's recording of the Allegro assai from Beethoven's Piano Sonata 3 in C, Op.2 No.3. Master data recorded at 24 bits and 88.2kHz sample rate.

Yet when I downconverted the data to 44.1kHz to prepare the CD masters, I was disconcerted to hear in level-matched comparisons that the high-sample-rate original was better, in that it sounded more natural and, paradoxically, had a cleaner, better-defined bass register. I used a dCS 972 digital-to-digital converter for the downsampling, and, as shown in Stereophile's February 1999 review of this unit, the 972 is measurably transparent in this divide-by-two downconversion, other than discarding the signal content above 22.05kHz.

Something other than pure frequency response must be going on. Michael Story of dCS has conjectured that, at the higher sample rate, the time-smearing introduced by the ubiquitous low-pass anti-aliasing and reconstruction low-pass filters is shortened, hence reduced in audibility (see "A Suggested Explanation For (Some Of The) Audible Differences Between High Sample Rate and Conventional Sample Rate Audio Material," available as a downloadable PDF file.) Empirical evidence that reducing the timing of the pre- and post-echo is audibly beneficial comes from the sonic improvements from a redoubling of the sample rate to 192kHz that has been reported by respected engineers like Tony Faulkner. There is almost nothing in the spectra of musical instruments that will benefit from a 192 or 176.4kHz sample rate, yet Tony is so convinced of the sonic benefit of the 4x rate that he prefers to record at 176.4kHz with 16-bit word lengths rather than at 96kHz with 24-bit word lengths. (Both require identical amounts of data storage on a Tascam MDM recorder.)

Something's happening here but we don't know what is, do we, Mister Jones?

---

Footnote 4 T. Oohashi et al, "High-Frequency Sound Above the Audible Range Affects Brain Electric Activity," AES Preprint No.3207 (91st Convention, New York), available from the Audio Engineering Society; "Inaudible High-Frequency Sounds Affect Brain Activity: Hypersonic Effect," Oohashi et al, Journal of Neurophysiology 2000 83: 3548-3558.).