the upper end of the musical spectrum that this upper end of the musical spectrum was still too noisy. Thus, the noise shaping averaging of Bitstream, while it helped music's mid and bass frequencies very effectively, was inadequate for helping music's treble frequencies enough. Note that Bitstream upsampled the 44.1 kHz coming off the CD by a whopping 256 times, all the way up to 11,300 kHz. So there was a wide margin of high frequency spectrum above 44.1 kHz into which to dump the extra noise. Even with this wide margin of upsampling and oversampling, Bitstream still could not deliver good enough noise reduction, good enough improvement in effective bit resolution, just below 20,000 Hz for music's trebles.
If Bitstream couldn't do an effective job for music's trebles with a whopping 256 times oversampling, how then can the Purcell hope to do an effective job with a puny 4 times oversampling, which is the most that it can deliver from a 16/44 CD? This question brings us to the third tool required for the Purcell to be effective at noise shaping averaging.
A clue to the answer may be found in Sony-Philips' new DSD digital system. The DSD system oversamples music at merely 1/4 the rate of Bitstream, 64 times oversampling instead of 256 times. The DSD system also has problems with music's trebles, but it still handles the mid and lower frequencies of music very well, as well as Bitstream (indeed even better). How could this be? The answer is that DSD uses a much steeper filter slope than Bitstream. Bitstream, in common with other previous noise shaping systems, used just a second order filter slope. The classic engineering dictum has been that a second order noise shaping filter is the steepest you can have, since any steeper filters tend to be unstable and can overload easily. But then Vimak engineered a 3rd order noise shaping filter for its D/A processor, which successfully gave improved resolution and quieter background. And Sony-Philips' DSD now uses an even steeper 5th order noise shaping filter. The DSD system has new, different problems with treble distortion (described in our previous articles), and these distortions might (or might not) be related to instability or overload in this 5th order noise shaping. But, at least for music's mid and low frequencies, the very steep 5th order noise shaping of DSD is very effective at improving effective bit resolution and reducing noise and distortion, in spite of the fact that the oversampling rate is much lower than Bitstream, much closer to the musical spectrum with less margin.
If there is less margin between the oversampling rate and the musical spectrum, then there's less space to dump the extra noise. And therefore the transition zone, between the dumping space and the musical spectrum you want to preserve (indeed enhance) has to be made narrower and more sharply defined, which requires a steeper filter. This consideration is similar to that which pertains in speaker crossover design. If a speaker driver misbehaves at a frequency far away from its intended passband, then a gentle (low order) crossover filter suffices, since there is a wide spectral margin between the music you want and the unwanted misbehavior. But if a speaker driver misbehaves at a frequency close to its intended passband, then there is less spectral margin, and so the speaker system designer needs to use a steeper (higher order) crossover filter, to establish a sharper barrier between the music you want and the unwanted misbehavior. Likewise, if there is less spectral margin in a noise shaping digital system, between the music you want and the spectral dumping ground for the unwanted extra noise, because the upsampling or oversampling is not a high enough multiple of the original sampling rate, then a steep, high order filter is required to accomplish the noise shaping averaging.
A steep filter has the further advantage that it can accomplish even better noise reduction, and even greater improvement in effective bit resolution, for the heart of the passband (music's mid and low frequencies). Any noise shaping averaging system generally provides progressively improving benefits at progressively lower music frequencies, since there are more samples to average together per cycle of music signal. With a steeper filter slope, this progressive improvement can be dramatically enhanced.
The best oversampling margin that the Purcell can possibly provide, even from the slowest likely input program source (44.1 kHz CD) to its fastest possible output (192 kHz), is a paltry 4 times, which pales in comparison to the 64 times of DSD and the 256 times of Bitstream. So, if the Purcell is to be able to do much good in its noise shaping averaging, it had better employ a mighty steep filter. And it does. The Purcell is capable of doing 9th order noise shaping averaging. This very steep filter is the third tool that the Purcell needs, in order to be effective at noise shaping averaging. Incidentally, performing the calculations to implement this very steep filter takes a hell of a lot of computing power, and that internal computing power is primarily why the Purcell costs a not inconsiderable $5000.
But it's this computing power, and the steep 9th order filter slope it provides, that allows the Purcell to achieve meaningful sonic benefits, even when working on a slim 4 times oversampling margin. We verified this when we checked the Purcell's sonic performance using its less steep 3rd order filter option. The 3rd order filter, even when used in conjunction with the Purcell's maximum output upsampling settings of 24 bits and 192 kHz, scarcely provided any sonic improvement over the input signal, coming from a 16/44 CD. Note that, if the Purcell upsampler's sonic improvement were chiefly due to upsampling per se (from 44.1 to 192 kHz), or to increased number of bits per se (from 16 to 24), or both, then we would still have heard a major sonic improvement (over the 16/44 input signal) when we tried the 24/192 output settings with 3rd order noise shaping averaging. But instead we heard virtually no improvement. That's why this experiment proved that the Purcell chiefly achieves its sonic benefits by being a noise shaping averager, not an upsampler per se. This experiment also proved that the expense of the Purcell, while high, is justified in terms of the computing power it buys, since less expensive 3rd order noise shaping averaging didn't cut the mustard. Also, as noted above, if the Purcell were merely an upsampler, then its expense couldn't be justified at all, since virtually all CD players already internally upsample to about 192 kHz or even higher, and they already internally increase bit resolution to 24 bits as part of their digital filter calculations.
We've seen how and why the Purcell benefits the sound of 16/44 CDs, especially when set to its maximum output of 24/192. How does the Purcell at this setting, upsampling a 16/44 CD, compare with 24/96 DVDs played straight without the Purcell? If you gullibly swallow the pitch that this is all a numbers game, you would think that the 24/192 upsampling by the Purcell would sound better than a 24/96 recording played without the Purcell. After all, 192 represents higher horsepower than 96, right? But in fact the 24/96 DVD sounds better than the 24/192 upsampling. The 24/96 DVD (of the same Chesky recording) sounded more transparent, open, airy, fast, and extended, with more overall sparkle and life. The Purcell might be able to take a sow's ear and make a silk purse, but it has limits, and can't make a gold purse. Specifically, noise shaping averaging can detect musical trends amidst noise and coarse resolution, and can therefore make genuine improvements in effective bit resolution, but its improved version of 16/44 is still not as good as being able to capture the full genuine resolution in the first place, on a PCM system such as 24/96.
Remember too that the 24 bit output of the Purcell merely represents increased word length, a carrier to hold any additional effective bit resolution that might be generated by the Purcell's noise shaping averaging calculations. There's no promise that all 24 bits of this longer output word length can actually be filled with meaningful musical detail detected by the statistical trend calculations of the computer -- in contrast to the 24/96 DVD, which does contain genuine 24 bit resolution for all musical details. Likewise, the Purcell's 192 kHz output sample rate does not represent music information up to 96 kHz, nor even up to 48 kHz as is contained on the 24/96 DVD recording; the music bandwidth is still limited to 20 kHz. The Purcell's 192 kHz output sample rate instead merely represents slots to hold improved resolution data, derived by a computer calculation that needs to run at a faster clock rate in order to provide the desired noise shaping averaging improvements.
If genuine 24/96 DVD sounds better than the Purcell upsampling 16/44 CD to 24/192, then the forthcoming genuine 24/192 recordings from the new WG-4 audio DVD standard will upstage the Purcell even further. So, for future recordings, your best bet is to invest in a new WG-4 DVD player and the WG-4 software. For future re-issues of current CD recordings made off analog masters, the WG-4 software at a genuine 24/192 should likewise sound better than the Purcell upsampling your present 16/44 CDs.
Thus, the primary value of the Purcell is in getting the most out of your present investment in 16/44 CDs that will not be reissued. The Purcell will also be useful in getting the most out of CDs which come from digital masters no better than 16/48; even if these are later reissued on 24/192 DVD, the upsampler used to make the new reissue software will likely be the Purcell, so you might as well use the Purcell on your present 16/44 CD and save yourself the cost of buying this reissue software.
The Purcell might also be of some benefit in playing today's 24/96 DVDs, made to the ad hoc standard using video DVD. There's not much upsampling to be done (96 to 192 kHz), and no place to put increased bit resolution (24 bits input stays 24 bits for output). But there might still be some benefit to the limited noise shaping averaging that the Purcell can do, with its computations working within the limits of merely 2 times oversampling. And, even though there's not an output word length longer than 24 bits to hold any additional musical detail beyond the 24 bits input, there would at least be twice as many 24 bit sample words being fed to your D/A processor. The D/A processor does its own averaging with its reconstruction filter, so that any different detail information put into every second 24 bit word by the Purcell could still beneficially affect the final result heard from your D/A convertor.
There's still the question of cost. The computational power in the Purcell makes it fairly priced at $5000. But, to take full advantage of its noise shaping averaging improvement of effective bit resolution, you should run it at its full 24/192 output, and that means you have to use a D/A convertor capable of accepting a 24/192 input. The only such D/A convertor available today, the dcs Elgar, costs another $12,000 (in the future, other less expensive convertors may well be made with this input capability). So your tab today already sits at $17,000. On top of that, you should employ a sophisticated CD transport, with excellent control of noise, jitter, and vibration, so that the advantage of the Purcell/Elgar's improved bit resolution is not squandered in resolution lost to the distortion and smearing caused by an inferior transport. Most such luxury CD transports cost about $13,000. So now your tab for the trio has risen to $30,000. All this to improve the sound of your present library of 16/44 CDs. Granted, it's important to enhance your enjoyment of your present music library and protect your investment in your present music library. But just how much is your investment in your present CD library worth?
If you don't want to spend the $30,000 this trio might cost, then you can get close to this sound with a well engineered integrated package for far less money, such as the Oracle CD player at $8950 total. Note that the Oracle also employs noise shaping averaging to improve effective bit resolution, albeit with a 1 bit sigma delta system. The Oracle also has the advantage of being engineered as an integrated package, which sets out from the start to reduce vibration, noise, and jitter. Thus it provides a much cleaner, quieter, higher resolution music signal than other 1 bit sigma delta systems, even to the point where it can be competitive with the super PCM hybrid system offered by the Purcell/Elgar combination.
There's another way of looking at your budgeting choices. The best value in a luxury CD transport worthy of the Purcell/Elgar's capabilities is surely the Oracle CD transport, at $6950. Once you have allotted the $6950 to a worthy CD transport, then you have a choice between the $2000 Oracle D/A convertor and the $17,000 Purcell/Elgar combination. Both do a superb job of getting the best sound out of your 16/44 CD library. Both are worthwhile choices for upgrading your present CD playback system. Only you can decide which your budget can afford.
What about the relevance of the Purcell to recordings in the near future, which will include 24/192 audio DVDs made to the forthcoming WG-4 standard? Today's Purcell, with its maximum output of 24 bits at 192 kHz, obviously won't be able to provide much improvement to a digital data stream that is already encoded at 24/192. But the concept taught by today's Purcell will live on, and the usefulness of this concept will expand.
The significance of the Purcell goes far beyond today, and far beyond the sonic improvement of yesterday's 16/44 CDs. The Purcell heralds a new kind of digital system, which we call Hybrid Super PCM. This is an important third alternative to WG-4 PCM and DSD sigma delta for tomorrow's music. That's why we have discussed the workings of the Purcell in such depth.
The forthcoming WG-4 PCM standard, encoding music at 24/192, sounds superb, especially for music's trebles, but we still hear a touch of the canned glaze that afflicts conventional PCM. The forthcoming DSD/SACD digital system, with its noise shaping averaging, has a more naturally musical, open, airy sound for music's middle and lower frequencies, but has tragic flaws of distortion in handling music's trebles. Each of these two digital systems has complementary strengths and weaknesses, and the strength of one clearly shows up the weakness of the other.
Why can't we have a single digital system that combines the strengths of these two digital systems and overcomes their weaknesses? Why should we compromise? Why can't we have the best of both worlds?
We can. Hybrid Super PCM can give us the best of both worlds, the treble accuracy of WG-4 PCM plus the open, airy musical naturalness of DSD. Hybrid Super PCM would piggyback onto the WG-4 PCM standard, essentially running the PCM digital signal from a WG-4 audio-video DVD transport through a processor like the Purcell, which would apply noise shaping averaging processing. The competing DSD/SACD digital system might as well die away, since the Hybrid Super PCM system can provide all its advantages and more
We can envision dcs making a Purcell 2 box, similar to Purcell but with higher sampling rate and bit resolution options, to provide the headroom needed to improve 24/192 PCM. Also, dcs might integrate the Purcell 2's computational processing into an Elgar 2 D/A convertor (to keep path lengths short for this higher bandwidth digital signal). Eventually, as costs continue to drop for digital chips and computational power, we should even see some caring specialty audio companies making a complete player, which incorporates the noise shaping averaging processing on board, to give you Hybrid Super PCM sound from all your 24/192 PCM audio DVD recordings. When that happy day comes, we will look back upon today's Purcell as the milestone that inaugurated Hybrid Super PCM as a third digital system. And we will thank the Purcell for teaching us that even high rate PCM such as 24/192 is still not quite perfect sound forever, that there is something even better within reach.
A number of new D/A convertors and CD players are now appearing that incorporate upsampling. Once again, specmanship reigns; an upsampler promising upsampling to 24/192 must be better than one offering upsampling to only 24/96, and upsampler A at 24/192 must be as good as any other upsampler B at 24/192. Of course, you know better, having read the above article. You know that the sonic benefits of upsampling are achieved not by the upsampling per se, and thus a fortiori not by upsampling per se to a higher specmanship number. You know that the sonic benefits are instead achieved by the high power averaging algorithms, and that the upsampling is merely a tool, a prerequisite for getting these high power averaging algorithms to work. These high power averaging algorithms are so complex that it is very unlikely that two competing D/A convertors or CD players will use the same algorithm (unless they both use the same chip or module programmed with the same software). And therefore it is very unlikely that competing units will sound the same, or will realize the same sonic benefits from upsampling and high power averaging, even if perchance they both upsample to the same 24/192 spec in order to bring their high power averaging algorithms into play. In short, the 24/192 spec means virtually nothing in predicting the sonic benefits of upsampling in competing units. As always, listening carefully and analytically to both units is the only guide to assessing how well they are enhancing and delivering your music. See also the background analysis article on the Audio Aero CD player, for further discussion of this topic.
back to table of contents