doubling the sampling rate. Again,
     All four digital problems discussed above degrade the fidelity of the music waveform, at frequencies well within the audible spectrum and throughout most of that spectrum. Amazingly, these problems can be ameliorated or even eliminated by employing a higher sampling rate superfluously above the Nyquist criterion. Who'da thunk?
     If people arrogantly believe that our digital systems are perfect once they meet the supposed Nyquist criterion of a sampling rate twice the highest frequency of interest, because they think that's what guru Harry promised us, then they would have no cause to look for digital shortcomings that might need solving. These people believe that, as soon as we meet the criterion (which we have, with a 44.1 kc sampling rate for a 20 kc passband), we have achieved the promised perfection in music waveform reconstruction. So obviously there is nothing more to be gained by any further measures -- especially superfluous measures such as going beyond the Nyquist criterion to a higher sampling frequency; and especially measures basically unrelated to waveform accuracy within the 20 kc passband, such as raising the sampling rate. If we have already arrived at waveform perfection within the 20 kc passband by meeting the Nyquist criterion, then obviously we cannot go beyond waveform perfection if we do something like raising the sampling rate.
     Then it logically follows that if we do raise the sampling rate anyway (even though we don't need to), and if we thereupon hear some sonic improvements anyway, then these sonic improvements must be due to our having extended the musical passband beyond 20 kc by raising the sampling rate (since they can't possibly be due to our having improved waveform accuracy within the 20 kc passband, because that was supposedly already perfect). The obvious consequence of raising the sampling rate is to correspondingly extend the musical passband beyond 20 kc. And now their arrogant assumption also forces those people to the logical conclusion that this passband extension can be the ONLY significant benefit, since there seemingly cannot be any benefits to waveform accuracy within the 20 kc passband. And thus it comes about that these people feel impelled to start speculating wildly about our ability to hear benefits of ultrasonic passband extension: directly hearing beyond 20 kc, or hearing less of the imagined obnoxious ringing from boxcar filters, or hearing indirect effects such as amplifier IM distortion. Even such wild speculation, however, still can't account for the yet wilder truth, the sonic improvements we actually do hear from (say) doubling the sampling rate. These sonic improvements are right in the middle of the old 20 kc passband, and they include finer inner detail (with better stereo imaging), cleaner purity, more natural musicality, etc. How can one explain all that coming from extending the passband into the inaudible ultrasonic region?
     On the other hand, if we are humble instead of arrogant, and if we recognize that digital still has many imperfections in re-creating the music waveform within the 20 kc passband even when the basic Nyquist criterion is met (by sampling at 44.1 kc), then we have opened the door wide to examining real world practical problems and real solutions. We can learn why real world practical digital systems cannot meet the ideal theoretical requirements posited by Nyquist. We can study what the implications of these practical shortcomings are on degrading the accuracy of the re-created music waveform, even well within the 20 kc passband. And we can catch a glimpse of how these shortcomings might be alleviated, and waveform accuracy thereby improved, well within the 20 kc passband. Then, when we look at the possibility of raising the sampling rate, it's pretty easy to see that this offers all kinds of wondrous solutions to our practical problems within the 20 kc passband. Increasing the sample rate can dramatically benefit musical waveform accuracy well within the 20 kc passband. And this produces manifold sonic benefits which we can and do easily hear and appreciate, without needing ultrasonic hearing.
     It's time to end the misled and misleading guessing that the perceived sonic benefits of higher sampling rates imply ultrasonic hearing or evils in pre-ringing filters. There's no mystery once you understand the imperfections of digital. The sonic benefits of doubling the original recording sampling rate to twice the Nyquist requirement are in fact easily audible, and audible at middle musical frequencies, because the primary benefits have nothing to do with extending frequency response into the ultrasonic region (that's been the red herring that everyone has mistakenly followed). Instead, the primary benefits chiefly relate to improving the true resolution, the transparency and detail, the musical naturalness, the accuracy and fidelity to the original music waveform before digitization, all at frequencies well within the passband, indeed precisely in the spectral region where our hearing is most acute and discriminating.
     The Capitole 24/192 is rated as having an internal capability to handle 24 bits of resolution and a sampling rate of 192 kc, so it can handle the higher sampling rates (96 kc and 192 kc) and increased bit resolution of the new super digital media, with their improved sonics. The Capitole also has digital inputs which can accept up to 24/96 data from another transport. Of course, the transport circuitry of the Capitole itself hasn't yet been adapted to read the formats of the new super digital media, but that could easily be designed into a future model.

Update:

     There has been some speculation in the press recently that adaptive playback reconstruction filters are desirable, and that such a filter should ideally switch between a sharp boxcar shape for gentle repeating music waveforms, vs. a gentle filter shape (with less time domain ringing) for sharp music transients having lots of high frequency information. The intent here is to avoid triggering the significant time domain ringing of the boxcar filter, by turning this filter off whenever music comes along with high frequency content that would trigger this ringing.
     In our judgement, such advocacy is precisely backwards. It is precisely on sharp musical transients that the abundant ringing of a boxcar filter is most needed, in order to correctly fill in the energy gaps left by the inadequacy of the connect-the-dots model at high frequencies (this was vividly proven by our measurement of a real music waveform in IAR issue #58, page 17, figure 12). Conversely, the ringing fill in of a boxcar filter is not needed when the music waveform is smooth and repeating at lower frequencies (e.g. on a clarinet note), and here the connect-the-dots algorithm of a simpler filter would suffice -- but on such music the boxcar filter isn't triggered into ringing in the first place, so it could not possibly be "bothersome."
     In sum, the boxcar filter rings abundantly when ringing is needed to correctly reconstruct the music waveform, and it doesn't intrude with any ringing when ringing is not needed. Thus, the boxcar filter is the best filter, indeed the only correct filter, for all of the time. And there should not be any adaptive changes to it.
     Now, an adaptive filter, and any filter shapes other than the ideal boxcar, will all sound different than a boxcar filter, and there will always be some listeners who will subjectively like this different sound (especially if it is smoother and gentler). But these other filter shapes are objectively simply inaccurate and wrong.

back to Capitole review

back to table of contents