The Digital Scene: More Theory, More Facts, More Hardware [Issue No. 12]

The Digital Scene: More Theory, More Facts, More Hardware

Harman/Kardon HD800 ; Onkyo Integra DT-770 ; Philips CD880 ; Philips CD960 (follow-up) ; Philips DAC960 ; Precision Audio D1 Analog ; Yamaha CDX-5000.

-------

We continue our ongoing, open-ended discussion of the glories and pitfalls of digital audio and check out various recent implementations, including R-DAT. We also add a new wrinkle to our test procedures.

It occurs to us that a "tree-worshiping analog druid" (latest CES epithet for the bigoted digitophobe) could have misconstrued the sweeping endorsement of current digital technology in our last issue (No. 11), qualified only by the phrase, "barring vulgar hardware and software foul-ups, of course." If we somehow left the impression that such foul ups are very rare and that smooth sailing can thus be taken for granted in the digital domain, the TWAD might have a plausible case against us. Let us hasten to point out, there fore, that today's commercial hardware (D/A converter chips, digital filter chips, etc.) and software (digital master tapes, CD's, etc.) are often quite flawed and not at all indicative of the current potential of the technology. Our feeling was that these things were rapidly improving, but now we are not so sure. Stanley P. Lipshitz and John Vanderkooy presented an excellent but somewhat upsetting paper at the March 1988 convention of the Audio Engineering Society in Paris under the self-explanatory title, "Are D/A Converters Getting Worse?" (see also the letter from Dr. Lipshitz in the " Box 392" column in this issue), and it seems that the DAC is the bad guy more often than we thought.

The DAC situation: not so good.

The two Canadians measured and analyzed 20 different CD players (17 documented in the preprint of the paper, 3 more in supplements thereto) strictly from the standpoint of digital-to-analog conversion. The players ranged from first-generation models (1982-83) to the latest-and-greatest, Philips-based as well as Japanese, moderately priced to ultrahigh-end. The results were quite dismal on the whole, although the commercial feasibility of accurate D/A con version was at the same time clearly substantiated. Only about one third of the units tested appeared to be free from objectionable conversion errors, and the majority of these dated back to 1982-85. For example, the most nearly fault less player, DAC-wise, was the JVC XL-V400, a relatively inexpensive 1985 model with a Yamaha 2-times interpolating ("oversampling") digital filter and a single time-shared Burr-Brown converter for both channels. The earliest Sony models, with analog low-pass reconstruction filters and no oversampling, tested almost as satisfactorily. One of the most dramatically flawed players was the $1700 Denon DCD-3300, introduced in 1987 and until very recently the flagship of the line, loaded with ostensibly state-of-the-art features. (It is possible that inadequate factory adjustment was the problem.) In general, the paper takes a rather dim view of "new and improved" digital circuitry for CD playback, concluding that "there is a downward trend in the accuracy of the D/A converter systems appearing in consumer CD players" and that this decline "is partially attributable to the drive for ever greater 'oversampling' ratios, without clear apparent benefits, but largely due to the use of inferior-quality D/A converter chips, even in expensive machines." It should be added, on the one hand, that the tests documenting these conclusions were quite a bit more detailed and analytical than those seen in consumer audio publications and, on the other hand, that our own tests of even more recent CD players paint a somewhat more hopeful though still not entirely reassuring picture (see the reviews below). Needless to say, D/A conversion accuracy is not the only criterion of CD player performance-the analog signal path, the power supply, mechanical design, etc., are all very important-but without accurate encoding/decoding the raison d'étre of digital sound reproduction ceases to exist, and therefore as the paper succinctly states, "it is ill-considered to skimp on performance in this one item [the DAC], especially in mid and high-priced machines." Lipshitz and Vanderkooy make a special point of straightening out the audio community on certain erroneous beliefs relating to DAC performance. To wit: test tones without added dither are totally useless, at any level significantly below 0 dB, for displaying distortion spectra or making THD-plus-noise measurements and, below -70 dB, for assessing level errors. (Offending reviewers please note.) Another thing: a correctly designed analog reconstruction filter has headroom for overshoot on transient waveforms, whereas a digital filter does not (except at the unacceptable expense of a raised noise floor); furthermore, 2-times and 4-times interpolating digital filters introduce a noise/distortion penalty as an inevitable result of the filter arithmetic (unless digitally dithered rounding is used-but it never is); therefore, the superiority of digital filters to good old-fashioned analog filters is not so clear-cut, despite the widely publicized advantages. "It is our belief," the paper states, "that, properly designed, either kind of filter can be audibly innocuous, and that neither is better per se." (Simplistic advertisers please note.) Still another tidbit: the

"18-bit" Yamaha bit-shifting quasi-floating-point converter system has some inherent drawbacks and is no more accurate than some of the better conventional solutions. (Please note, also, our own review below.) Department of corrections.

We were a bit sloppy in Issue No. 11 on the subject of low-level DAC errors as expressed in dB's and LSB's (least significant bits). Nobody caught us but Stanley Lipshitz (naturally); his letter and our answer (excuse?) appear on page 3 of this issue. Let us set the record straight.

Track 19 of the CBS CD-1 test disc provides a series of computer-generated 997 Hz test tones with added dither, at-70.31 dB, -80.77 dB, -90.31 dB and --100 dB. The first three of these levels represent 21 codes, 7 codes and 3 codes, respectively. Thus, in terms of LSB's, they exercise the range from - 10 to +10, -3 to +3 and -1 to +1, respectively (the extra code value in each case being 0). The -100 dB level is a special case (also synthesized with 3 codes but with many more 0 samples than -1's and +1°s), included on the test disc as a demonstration of dithered signal reproduction below the least significant bit. Now, a -6 dB error at the -90 dB level means halving the signal amplitude and is therefore a 1-LSB error; a +6 dB error at the same level means doubling the signal amplitude and is therefore a 2 LSB error. At the -80 dB level, a +3 dB error (not uncommon) results in a signal amplitude of 8.5 LSB's and is therefore a 2.5-LSB error. And so on. If we have lost you by now, all you need to remember is that respectable DAC performance at these low levels means keeping the errors to 1 LSB or less.

Having rechecked our laboratory notes on the equipment reviewed in Issue No.11, we can now correctly report that the maximum low-level DAC error found in each unit was as follows: Denon DCD-1500 II, 1.5 LSB; Denon DAP-5500, 2.5 LSB; Euphonic Technology ET650PX, 1 LSB; Philips CD960, 0.5 LSB; Tandberg TCP 3015A, 1.5 LSB. Thus, in our comments, we were too kind to Denon's DAC's and a little unfair (but just a little) to Tandberg's.

The Denon models have trimmer potentiometers on the circuit boards for making DAC linearity adjustments; we made no attempt to experiment with the factory settings, and it is possible that there was room for improvement. The Philips and Philips-based units have no such adjustments.

A test for asymmetrical transient clipping.

Most equipment reviewers tend to treat the square wave response of CD players a bit cavalierly; unless there is severe ringing or gross asymmetry, they call the wave form typical for the particular type of filter used. We have tried to be more precise by spectrum-analyzing the output of each player when reproducing the 1002.27 Hz square wave encoded at 0 dB on the CD-1. Since this test signal is computer-generated without distortion, its spectrum consists exclusively of the Fourier components of a perfect square wave, viz. the 1002.27 Hz fundamental and its odd harmonics--3006.81 Hz, 5011.35 Hz, 7015.89 Hz, etc.--at exponentially declining amplitudes. There should be no even harmonics in the output of the device under test, and indeed there are none produced by any of the Japanese CD players we have tested. To our surprise, however, we found that some (not all) Philips-based players exhibit this defect, specifically the CDB650 types and the lower-priced but more up-to-date CDB471/472 series. The higher-priced CD880, CD960 and DAC960, on the other hand, are absolutely clean in this respect.

We are reasonably certain-although we reserve the right to be wrong, this being our first stab at a rather slippery phenomenon-that what we are dealing with here is a special case of digital filter overload as mentioned above.

The inevitable overshoot of the full-scale square wave constitutes a worst-case scenario (very rare in music) resulting in filter clipping, but that alone would not generate even harmonics. Only some kind of asymmetrical condition, such as a DC offset, would cause the even harmonics to appear. Now, every one of the anomalous Philips players uses the older SAA7220P/A digital filter chip, and every one of the unaffected units uses the newer SAA7220P/B chip. The A chip has a negative offset, intended or not, whereas the B chip operates symmetrically. It is interesting that the oscilloscope display of the square wave gives little or no hint of the A chip's asymmetrical clipping; the spectrum analyzer, on the other hand, makes it loud and clear.

The amplitude of the even harmonics is of the order of 3% of the odd harmonics preceding them (i.e., 30 dB down, typically), and that is not a negligible distortion figure. We admit the possibility that the test is a mere exercise, since musical transients punching through the 0 dB ceiling will be rarely encountered, but we must still downgrade audio equipment that spontaneously outputs at -30 dB what is not inputted. What still remains to be explained is why the 100 Hz square wave encoded at -10 dB on the Denon Audio Technical CD (38C39-7147, Track 71) also causes even harmonic generation in the same players, but only higher up in the kilohertz region, not in the hundreds-that could not be a transient overload level, even with a sizable negative offset, now could it? Let Philips worry about that one...

Incidentally, the SAA7220P/B filter is the mandatory complement to the TDA1541A converter with the S1 marking in the higher-priced Philips models. (Quite confusingly, the A suffix in this last case stands for selected quality and little whirs and clicks of automatic threading, watching the control panel take charge, handling the smoothly functioning controls, etc., gives considerable aesthetic satisfaction.

On the lab bench, we found every measurable specification of the DT-7700 fully met or exceeded. As there is no DAT equivalent of the CBS CD-1 test disc, we had no way to measure playback-only performance characteristics.

(Digital-to-digital copies of CD's cannot be made on un-modified DAT decks, and an analog-to-analog copy of the CD-1 is not an accurate digital reference.) We can report, however, that the record/playback frequency response is ruler flat over the entire audio range, much flatter than we have seen in any professional open-reel analog recorder, and that wow and flutter are literally nonexistent-the expanded spectral display of the recorded test tone shows no sidebands whatsoever, not even tiny blips. In general, the analog output of the recorder appears to be identical to its line input, except of course on square waves, which show the typical sampled and quantized profile, in this case overshooting just a little at the leading but not the trailing edge-the early CD player look, probably due to an analog low-pass filter at the input. No even harmonics, though, not even at 0 dB. We did not get as much as a block diagram, let alone a schematic, with the machine, so our cause-and effect insights are limited. All we know is that the DAT standard specifies 16-bit quantization with 48-kHz sampling (although 44.1 kHz and 32 kHz are also provided, but only for playback); for encoding the Onkyo uses a 2-times interpolating digital filter with independent left- and right channel A/D converters, for decoding a 4-times interpolating digital filter and independent left- and right-channel DAC's. The digital input and output jacks-by the way and apropos of nothing-offer the choice of either coaxial or optical connection.

Our listening tests also gave complete satisfaction.

We recorded pink noise on the cassette and tried to hear differences as we switched back and forth between the generator and the DT-7700's line output (a more critical test in many ways than when the signal is music). No difference was audible between the source and the recording. That is total accuracy, and nobody can ask for more than that, We also compared the dmp and Sheffield Lab demo DAT's with the CD versions of some of the same tracks. Such a comparison is ever so slightly booby-trapped because of (1) possible sampling-frequency conversions from the master to the CD or DAT in some cases but not others and (2) the difficulty of setting equal levels without a prerecorded 0 dB reference standard for DAT. Nevertheless, we managed to adjust the levels accurately enough by ear to yield at least reasonably credible results, to wit: exceedingly small and possibly nonexistent differences. The lack of rigor in this particular test permitted us, however, to indulge in just a bit of subjectivity, so that we ended up with a vague feeling that some of the dmp tracks were marginally better sounding on DAT than on CD. Well, Tom Jung at dmp records on the Mitsubishi X-80 digital master recorder with 48-kHz sampling, and his DAT's are therefore digital-to digital duplicates without conversion, whereas his CD's are converted at 44.1 kHz... who knows? In any event, we think that DAT is the bee's knees and the cat's meow, that the Onkyo Integra DT-7700 is a lovely example of the breed, and that the politicking provincial protectionists who are keeping this important technological development off the market are nothing but picayune pea-brained Luddites who probably beat their wives and change their underwear infrequently. Do we make ourselves clear? Philips CD880 N.A.P. Consumer Electronics Corp., I-40 & Straw Plains Pike, Knoxville, TN 37914-1810. CD880 compact disc player, $749.00. Tested sample on loan from manufacturer.

What do you get for $200 less than the CD960? A somewhat cheapened though commercially more up-to-date player which, in our opinion, is less attractive by a wider margin than $200. We feel that if you can afford the tourist class ticket here you can also afford the first-class one.

The CD880 offers the same top-of-the-line Philips digital chip set (see above) as the CD960 and DAC960 flag ships, although we are inclined to suspect that the selection process includes some further fine-tuned grading, so that the low-level DAC linearity in this case is third best by a hair, but really just a hair, and possibly not even in all samples. Maximum error is still of the order of 0.5 LSB, and who can object to that? What we like less about the CD880 is the decidedly sluggish disc drawer, for one thing. It opens by means of a button right on the drawer, just like the CD960 but much more slowly; to close it, however, you have to nudge the front-no button available-and hope the drawer responds, an eventuality on which the odds are about fifty-fifty. Somebody's idea of simplified ergonomics, no doubt. The front-panel display and controls are also up dated-or shall we say Japanized?--with additional bells and whistles; the trouble is that when you want to punch in 2-1 for track 21 and linger a little bit over the 2, track 2 starts to play automatically. The damn thing takes charge instead of letting the user have the last word; Sony and Denon do these things better. On top of it our CD-1 test disc, which allows perfectly normal track search and skip ping on all other players known to us, made the CD880 flash the error sign when certain tracks were punched in, although the same tracks were played normally in a continuous sequence. One other minor annoyance is that the up and down buttons for index are on the remote control unit only.

The latter is a new and very nice design, but we like to be able to operate all functions from the front panel if we so desire and resent the assumption that most users just pork out on the couch and never get up again.

In sound quality the CD880 falls very much into the same excellent but not necessarily unexceptionable category as the CD960 and DAC960; more about that below in our review of the latter. If the CD880 were the only CD player available in the company's Philips-branded line, we would have to recommend it enthusiastically on the basis of its superior chip set and digital performance, if nothing else; as it is, we feel a little blah about it.

Philips CD960 (follow-up) N.A.P. Consumer Electronics Corp., I-40 & Straw Plains Pike, Knoxville, TN 37914-1810. CD960 compact disc player, $949.00. Tested sample on loan from manufacturer.

It turned out that the "improved" version we had been promised and had finally received was not a reengineered design but merely a unit double-checked to incorporate the selected chip set discussed above. Not all CD960's did, at least initially-our first one, as it happened, yes; a second one we looked at, no; this latest one, yes; Len Feldman's sample as reviewed in the June 1988 issue of Audio, obviously no, judging from his measurements; his sample of the DAC960 reviewed in the same issue, again no. By now we are reasonably certain that things have settled down and nothing but the primo chips are used in the three top-of-the line Philips models.

This time we obtained absolutely symmetrical square waves (without even-order harmonics of course), and the low-level linearity was, if anything, a smidgen better than in our first sample, with a maximum error of less than 0.5 LSB. Very impressive performance, very beautifully built machine, four separate power supplies, outstanding control and display facilities with very good human engineering, still the old-style (CDB650-type) remote control though, and no volume-controlled main output-as if it mattered with a maximum of only 2 volts out. (That CD convention was established, we are absolutely convinced, as affirmative action to assure jobs for preamplifiers even where there is no phono source.) The sound of the CD960 appears to be identical to that of the topologically extremely similar DAC960; further comments in our review of the latter below.

Philips DAC960

N.A.P. Consumer Electronics Corp., I-40 & Straw Plains Pike, Knoxville, TN 37914-1810. DAC960 DIA converter unit, $949.00. Tested sample on loan from manufacturer.

This is not a "digital audio preamplifier" package a la Denon DAP-5500, as there are no analog inputs, not even for line-level sources. The DAC960 is strictly a digital-to analog converter of great versatility, designed to be inserted between a digitized signal source (coaxial or optical, CD, DAT or satellite tuner) and an analog input (preamp, power amp or what have you). It uses the same selected and top graded Philips chip set (TDA1541A-S1 and SAA7220P/B) as the top-of-the-line Philips CD players; we even suspect that the absolutely best chips may be reserved for this one model because the low-level linearity we measured was close to perfection, with a maximum error of 0.25 LSB in the better channel and only infinitesimally more in the other. All other parameters of digital performance were of the same order of excellence. The unit is probably the most sumptuously built of all the Philips’es, with three separate power supplies incorporating three individual transformers, balanced/fixed/variable outputs, an extra 6 dB of gain avail able at the volume-controlled output (no preamp needed), beautiful control panel with automatic indication of sampling frequency (48, 44.1 or 32 kHz), and other goodies too numerous to mention. Since all the quality, both digital and analog, is built into this one box, the CD player you plug into it does not have to be particularly sophisticated; a good disc drive, drawer, laser and error-correction chip are the main requirements. Even so, to eliminate any possibility of a compromise, we decided that the CD960 plugged into the optical input of the DAC960 would be our arbitrary reference-the most elaborately engineered hookup available to us-for at least one of our full-fledged double-blind listening comparisons. The variable output option on the DAC960 made it easy to match volume levels accurately.

The CD player we pitted against the "double 960" was the Precision-Audio-modified Sony CDP-910, which had impressed us equally in casual listening. The details of that ABX test are given in our Precision Audio review below; suffice it to state here that there was a minuscule audible difference and that our very hesitant vote at the end was not in favor of the Philips combination. It was strictly a choice between analog output stages, and the high-feedback integrated op amp of Philips, although unquestionably excellent in sound, was judged to lack the special qualities of the other player's custom analog board with discrete components. It was close, though, very close.

Precision Audio D1 Analog

Precision Audio, 223-47 65th Avenue, Bayside, Long Island, NY 11364. D1 Analog section replacement for Magnavox, Philips and Sony CD players, $450.00. Tested samples (Magnavox CDB472 and Sony CDP-910 mods) on loan from manufacturer.

Precision Audio is a tiny company with an uncompromising attitude toward audio circuit design. David Rich, their R and D man, is a highly articulate young man who teaches electrical engineering to college students, designs VLSI chips (i.e., integrated circuits of great complexity) as his main professional activity and yet believes that integrated op amps should not be used in the audio signal path, least of all in the analog section of a high-quality CD player. The company has published two position papers on this subject, "Application Note #1: Ten Reasons Not to Use Integrated Operational Amplifiers in Audio Applications" and "Application Note #2: The Transimpedance Amplifier & Its Use in CD Players," the latter a technical explanation of the circuit concept used in the analog replacement board under review here. These short but highly informative papers are available directly from Precision Audio.

The D1 Analog board, which must be installed by its makers, incorporates two discrete, class A, low-feedback, high-slew-rate and very fast current-to-voltage amplifiers in the proprietary configuration called transimpedance amplifier, plus all the associated analog circuitry. The DAC of the CD player produces current, not voltage, and this cur rent signal is not smooth and continuous but full of high frequency glitches at the transition points of the staircase waveform. Converting this input into a line-level voltage output is not at all the same problem as amplifying a small voltage signal with a stage of gain; the solution is of necessity somewhat different, and the transimpedance amplifier appears to be a very good solution.

The D1 modification is compatible only with DAC's having low glitch currents on bit transitions and therefore requiring no sample-and-hold circuit after the current-to voltage converter stage. Right now, that leaves the Philips TDA1541 pretty much in a class by itself, although others are expected to be forthcoming. We first tested the Magnavox CDB472 ($229.00 in the stock version) with the D1 board added; this is one of the Philips-based players with the asymmetrical transient clipping problem discussed above, and it also has a linearity error of 1.75 LSB at the lowest level in one channel (only 1 LSB in the other). Even so, we found the sound to be very pleasing on a purely subjective basis without a double-blind listening test; furthermore, the player is quite impressive in mechanical operation and ergonomics for such a cheap machine (not so cheap, of course, once you add the D1). Precision Audio is a little bit down on this model for the reasons mentioned and also because of inadequate power supply regulation, which they try to improve in their mod. Unfortunately, the TDA1541-equipped choices are somewhat limited. Their best shot so far has been the Sony CDP-910 (discontinued but still available here and there in the upper $300's), which is a marvelous CD player digitally, even though its 1541 has no suffixes of any kind (a selected chip predating the current designations?). We measured mind-blowing low level linearity in our D1-modified sample, equal to that of the DAC960 or possibly even a hair better-if such a thing is possible. Mechanically, too, and in ease of operation, control facilities, display, etc., the CDP-910 is close to the top, regardless of price. Great value. The bad news: its successor, the CDP-950 ($450.00) is without the +15-volt supply rail, which was removed-and the whole design watered down-to reduce the manufacturing cost and hold some sort of price in the face of the dollar-yen exchange rate. Precision Audio wants to have nothing to do with this model. The Sony CDP-507ESD ($600.00), on the other hand, is quite similar to the CDP-910 but uses an 8-times interpolating system which requires time-multiplexing two TDA1541 chips. Since the chip was not designed to operate in this mode, Precision Audio is hesitant to do a D1 mod on the unit. The best immediate solution? The Philips CD960, 36 most probably, but then we are talking big bucks. Get up to-date information from the company before you make a move, in case you are interested.

As already summarized above, we were sufficiently impressed by the sound of the Sony CDP-910 with the D1 Analog board to run for our ABX comparator and set up a double-blind listening comparison at matched levels with the Philips CD960 plugged into the optical input of the Philips DAC960 ($1898.00 the package and until then our more or less arbitrary reference). We must repeat once again that our sporadic ABX tests with just a few highly motivated participants cannot possibly have the statistical authority of a massive test program with dozens of panelists and hundreds of comparison trials. On the other hand, we firmly believe that if one listener in a hundred can provably hear a difference, then the difference is real-even if 99 others are unable to hear it. Our conclusion in this case was that the modified Sony is indeed distinguishable in sound from the Philips combination; the difference, however, is very small and not easy to zero in on. One panelist did it with a confidence level of 98.9%, and another was not far behind, but everybody agreed it was a tough one. The main difference, after A and B were identified, was felt to be an "etched" or super defined quality in the sound of the Philips combination as against a slightly softer, spatially more convincing and somehow more musical quality in the Precision Audio mod. It seemed to make sense, as excessive etching is considered one of the possible artifacts of high-feedback integrated op amps. Remember, though- we are nitpicking between two outstanding CD playback systems.

Even so, the Sony CDP-910 with the Precision Audio D1 Analog modification became our new reference.

Yamaha CDX-5000 Yamaha Electronics Corporation, U.S.A., 6660 Orangethorpe Avenue, Buena Park, CA 90620. CDX-5000 compact disc player, $2200.00. Tested sample on loan from owner.

A friend who had paid only $1000 for a brand-new CDX-5000 brought it to our lab for just a few hours to find out how it performed and why it was being discounted from $2200. It is certainly a massive, luxurious, beautifully built unit. We measured some seemingly negligible DAC nonlinearities at the middle as well as the lowest levels; it must be noted, however, that an error as small as 0.5 LSB measured at the output of this bit-shifting quasi-floating-point "18 bit" player means a 2 LSB error in the DAC itself because of the 12 dB attenuation resulting from a 2-bit shift in the binary point-and some of the errors were bigger than that.

In a quick ABX test at carefully matched levels (the CDX.-5000 has a digital volume control), your Editor was able to distinguish the Yamaha from the Precision-Audio modified Sony CDP-910 with a confidence level of 92.7%, the owner with only 61.3%. Very inconclusive, but we still prefer the Precision Audio for the reasons stated above.

-------

[adapted from TAC, Issue No. 12]

---------

Also see:

Reference Audio Hi-Fi Systems: Some Tentative Recommendations: Reference A; Reference B

Further Thoughts on Double-Blind Listening Comparisons at Matched Levels

Various audio and high-fidelity magazines