A/D Converter Listening - part one

by Bob Hodas & Paul Stubblebine

Oh Lordy, thank you for digital, the perfect medium. We keep seeing these ads for the perfect sound of the new DAT recorders, just as perfect as CD. Yeah, sure.

This article was supposed to be a DAT recorder review, but the DAT didn't sound that good to us. So we listened to just about every DAT on the market and couldn't find one that sounded like the source material. We came to the conclusion that we should look into more sophisticated converter technology for our answer and so here we go.

A listening was set up with the five outboard converters currently in production. Listening was conducted at Rocket Lab mastering studio (see Dec. 89 article) in San Francisco on Meyer HD-1 speakers. Our source material was analog tape recordings, one pop, one spoken voice, one choir, and one orchestral. The first three recordings were 15ips Dolby SR and the last was 30 ips with no noise reduction. Recordings were made through the converters at 44.1kHz into a Studer-Dyaxis and played back through an Apogee D/A 1000 converter. We started off with a more expensive D/A "audiophile" converter but determined through careful comparisons that the Apogee better represented reality. The converters were always compared directly with the source material, not with each other. The idea was to find out which ones most closely reproduced the analog source. Let's look at the converters we used.

The Singular Solutions A/D 64X is certainly the most unique converter we tested. It was designed to interface with the NeXT computer but also works as a stand alone unit. It is also unique in that it contains a pair of mic preamps in addition to stereo line level inputs. Conversion technology is Delta-Sigma claiming 16 bit operation.

The 1.88" x 15" x 7.5" box is a tabletop model weighing 7 lbs (now also available in rack mount version). The front panel contains two continuously variable mic preamp gain pots, matching line gain pots, an input select switch with LED indicators for analog or digital inputs, left and right overload LEDs and a power LED.

The rear panel contains the AC cord receptacle with accessible fuse cover, two female XLR mic input connectors (pin 3 hot) and low cut filter switch, a phantom power switch, two female XLR fixed line level input connectors with +4dBm/-10dBm selector switch, and two female RCA variable line level input connectors. These RCAs are tied into the front panel gain pots. The DB25 connecter interfaces with the NeXT computer or a Loopback Adapter. This adapter for stand alone operation contains a switch to select 44.1kHz or 48kHz sampling frequency. There is also a switch position for external clock (slaving several units together) with RCA clock in/out connectors. The digital I/O is curious. It contains an RCA input connector and a Female!!! XLR output connector. This is the first time I've ever seen a female line level output connector but I guess these guys are into computers and don't know much about audio standards. Also, referring back to the Digital I/O, you will notice that the input is SPDIF while the output is AES/EBU. Retail price is $1,295 ($95 extra for rack mount option).

The Yamaha AD2X is a 1U x 11.5" rackmount unit weighing almost 10 lbs. The front panel contains a power switch, four LEDs indicating external clock, 44.1kHz or 48kHz sampling frequency, and emphasis. There is a 12 segment LED input level meter, and two stepped input level pots. Conversion technology is Delta-Sigma claiming 19 bit operation.

The rear panel contains an attached power cord, two female XLR +4dB inputs connectors (pin 2 hot), emphasis on/off switch, an XLR male connector for AES/EBU digital out, DIN connector for Yamaha digital out, RCA connector with on/off switch for SPDIF out, BNC connector and in/out/off switch for word clock, and a switch to select 44.1kHz or 48kHz sampling frequency. Retail price is $1,695.

The DCS 900A is a 1U x 15.25" rackmount unit weighing 15.4 lbs. The uncluttered front panel contains only five LEDs. These indicate overload, 44.1kHz or 48kHz sampling frequency, and master/slave operating mode. Conversion technology uses 128 times oversampling claiming up to 24 bit operation.

The rear panel contains female XLR analog line inputs (pin 2 hot), input gain trims, AES/EBU XLR in/out connectors, SPDIF RCA connector, a switch to select 44.1kHz or 48kHz sampling frequency, DB-9 remote control connector (several units can be slaved together), and a switch to select 16/18/24 bit operation. The four BNC connectors for SDIF-2 are for channel 1/2 out only and word clock in/out. An AC power receptacle and a power switch round out the panel. Retail price is $10,000. By the way, for $10,000 they don't give you a power cord, you have to buy your own.

The Pygmy Computer Systems AD-1 is a 1U x 13" unit weighing 7.5 lbs. The simple front panel hosts a power switch, two gain pots, a phase switch with dual LEDs indicating normal/inverted phase, and another switch with dual LEDs indicating 44.1kHz or 48kHz sampling frequency. We were disappointed that there was no overload indicator. Conversion technology is Delta-Sigma claiming 16 bit operation.

The real panel houses the AC receptacle with accessible fuse, two female XLR analog line inputs (pin 2 hot), and a male XLR AES/EBU digital output. SPDIF is supported by a switch inside the AD-1 but you must build an XLR/RCA adapter. Three BNC connectors support SDIF-2 word sync out, and left/right digital audio. Word sync in is not supported. Five BNC connectors support two or more AD-1s operating together for multichannel use. These are system clock, two for external clock, and two for word sync in. SPDIF operation is available with an internal switch and building an adaptor for the AES connector. Retail price is $3,995.

The BTS BAC-3000 is another unique unit in that it contains two separate stereo pairs of A/D as well as D/A converters. This unit was designed to run very long lines in broadcast situations. It is a 1U x 16" unit and weighs 16.3 lbs. The front panel contains two sets of identical controls. Backlit buttons select left/right/stereo operation for each pair. Each channel has three LEDs sensing level at -25/0/+14dBu but there is no overload indicator. Although the BTS has no variable input pots, by removing the top you can access jumpers that allow you to set input sensitivity in four increments (-10, 0, +4, +8dBu). Conversion technology uses 128 times oversampling claiming 20 bit operation.

The rear panel contains a barrier strip for wiring analog line level inputs and outputs. There are XLR ins and out for each AES/EBU pair and a sync input. A DB-9 connector for serial control is provided as well as ground jumpers (remember this is designed for broadcast). A power receptacle and accessible fuse panel are also present. Retail price is $6,000.

As described above, all but the BTS have some type of variable level trim on the input. The input stage implementation is quite different on these units and on two, undesirable results could occur if you're not really careful! On both the Singular Solutions and the Yamaha you can overload the input stage without seeing it on the overload indicator. The overload indicators only show DAC overload. This is poor design in our book and dangerous for the operator. In order to avoid this potential problem, the Singular Solutions should always be operated with the input pots at 1 o'clock or above. The Yamaha pots should be set to 3 o'clock or above.

Regarding bit rate, we obviously got a truncation to 16 bits in the Dyaxis but decided that this represents a real world situation for most people, especially those recording to DAT. Those of you who use a Mitsubishi X-86 can take advantage of those converters offering extra bits while recording but it still comes down to the 16 bit CD in the end. This doesn't mean the extra bits are just marketing bits though. For instance, the DCS can operate at either 16, 18, or 24 bits. Through listening, we determined that even with truncation, 24 bit operation was superior and so we did all recording in the 24 bit mode.

Measurements were done using the new Audio Precision system after the listening test to see if there was any correlation to what we heard. That will be discussed later as the most important question is "How did they sound?". For the most part, these comments will be an amalgamation of both our impressions. If we had distinct differences, those will be noted. Converter reviews are arranged in ascending order based on list price.

The first piece we listened to was "Waiting Song", a current folk-pop release by Barbara Higbie. Instrumentation was piano, vocal, electric bass, drums, congas, percussion, and acoustic guitar.

Singular Solutions: Top has a distinct edge, a bit harsh, tonal balance is shifted, lo-mid richness reduced, imaging precise, lots of low bottom.

Yamaha: Treble edge, bite on top, vocal tone change, lost warmth, mid-bass resonances, where's the beef, no reverb definition, imaging good.

Pygmy: No tonal skew, good dynamic impact, good bass & good bass pitch definition, good imaging, missing just a bit of warmth, slight loss of life.

BTS: Low bass undefined, lacking lo-mid richness, felt masked, hard to hear into reverb, reverb mushed together, imaging less precise, treble a little clangy.

DCS: lack of deep bass effects impact, good imaging, shimmer and life of source reduced a bit, loss of sibilance present on tape, slight reduction of low level detail and differentiation.

Score:
5. Yamaha
4. BTS
3. Singular Solutions
2. DCS
1. Pygmy

The next recorded piece was a mono talking head reading copy from a magazine. There was some background room noise that shut off partway through the recording that added some dimension to the piece.

Singular Solutions: Hard treble edge, warmth a bit reduced, sibilant splash.

Yamaha: No sparkle, low level background noise missing, pops less dynamic, warmth gone, not live.

Pygmy: Warmth reduced slightly, treble exaggerated slightly, good pop reproduction, slightly nasal.

BTS: Tonal change, nasal, not live, warmth reduced, very high treble gone.

DCS: Top rolled off a little, presence a little reduced, warmth slightly boosted, tone good, good pop.
Score:
4. Yamaha
3. BTS
2. Singular Solutions
1. DCS/Pygmy (tie)

Our next piece was the Slavyanka Millenium Chorus of fifty male and fifty female voices recorded at the Stanford Memorial Chappel.

Singular Solutions: Grainy, voices not distinguished, too blended.

Yamaha: Unpleasant, certain intervals in voices cause distortion or beating, crowd noise and clapping less real - more like spurious noise, air in hall gone, tone change.

Pygmy: Very detailed, a little bright, lacks a bit of richness & warmth.

BTS: Image good, tonal change, some beating, top air gone, a little smeared, some detail missing, bright noise on Forte section.

DCS: A little warmer but pleasing, very top slightly rolled off.

Score:
5. Yamaha
4. BTS
3. Singular Solutions
2. Pygmy? (see final evaluation)
1. DCS? (see final evaluation)

The final piece was a Mozart recorded at the Oklahoma Mozart Festival.

Singular Solutions: As above.

Yamaha: As above.

Pygmy: good detail, slight lightening of orchestral weight.

BTS: As above.

DCS: Slight reduction of image size, warmer, slight lightening of orchestral weight, slight reduction of transparency.

Score:
5. Yamaha
4. BTS
3. Singular Solutions
2. Pygmy
1. DCS

Did the measurements show us anything significant that could point out this cult of personality? We can give you a positive maybe. Some of the things that would support our listening results were very slight, leading us to believe that the industry still needs to some homework on what is significant when measuring digital. We do not have space to publish all the graphs but will print a few of the most significant findings.

Frequency response: Almost everybody was extremely flat out to 20kHz and be out by 22kHz so that doesn't seem to be much of a factor. The only possibly significant corelation to what we heard is that the DCS was .5dB down at 20kHz and the roll starts around 7kHz (fig. 1). Hardly significant but this could account for the smmothness that the DCS exhibited throughout the tests. Also, Yamaha appears to drop off steeply just before 20kHz (fig. 2).

THD+N vs Frequency: This was a best case measurement. The DCS was clearly superior followed by Pygmy, then BTS, with Yamaha and Singular Solutions almost identicle.

THD Spectrum: Once again the DCS proves superior, followed by Pygmy, Singular Solutions, BTS and Yamaha. Measurements we taken at -6/-20/-80dB below full scale. One significant characteristic showed up in the Yamaha which may contribute to its sound performance. There is at least a 20dB change in the noise floor when the input signal passes a threshold between -18dBFS and -20dBFS (fig. 3 & 4). We could not measure how fast this change ramped up or down. Although Yamaha feels it would improve measurements with low level inputs, the transition could be disconcerting to the listener. Could this be why air and background noise got lost on the Yamaha recordings?

IM Distortion: Nothing significant showed up in these measurements.

Alias Spectrum: We put in a 24kHz tone and took a look at what came out. The Pygmy and Sonic Solutions both had some full spectrum activity about 120dB down (which could just be noise) with a 20kHz spike down 95dB and 101dB respectively. A good showing. The BTS spectrum was way down in the noise too but its its 20kHz spike was only 65dB below signal. The DCS spectrum was out of sight as well but its 20kHz spike was only down 30dB and it had additional spikes at 6kHz, 9kHz, and 16kHz all down at least 90 dB though (fig. 8). We found this curious as the DCS had measured so well in other tests and sounded so good. Maybe these spikes from out of hearing range artifacts aren't significant unless the spectrum is gross? An example of this would be the Yamaha measurement (fig. 9). The full spectrum reaction was significantly higher than the other units, only about 94dB down from 20-800Hz. From 800Hz on up there are a plethora of high level spikes, many only 60dB down (well within hearing range). The worst spikes were 800Hz (50dB), 7kHz (50dB), 8.5kHz (50dB), and 16kHz (35dB).

Alias Level: A tone was swept from 22kHz to 200kHz while we looked at the output. Only the BTS and the Yamaha showed abnormal and higher than average responses. Look at the difference between the normal Singlular Solutions and the others (fig. 5,6,7).

These last two tests show a specific problem of designing with Delta Sigma. Out of band artifacts can cause beating within audible frequencies. This was apparent in our listening tests as both the BTS and Yamaha developed strange harmonics or beatings especially audible on the complex waveform of the Slavyanka vocal chorus. We feel that we did see some corelation between listening and measurement here. The other tests are hard to quantify.

This was first and formost a listening test though. Our results did form a pattern that gave us a fairly unified conclusion.

Overall it was very clear in the lower ranking:
5. Yamaha
4. BTS
3. Singular Solutions

The upper ranking was a little less clear in some cases. On the pop piece we definitely liked the Pygmy better, you could dance to it. On the Mozart the DCS had the edge, you could dance to it, in your seat of course. On the voice and Slavyanka it was hard to make a choice, a question of warmth (DCS) vs. detail (Pygmy).

We can say without qualification that even the units of the lower ranking make some improvement, however small, over the sound of DAT. The question is how much is a small improvement worth? In the lower ranking, the Singular Solutions provided the most noticeable improvement in quality and if that is worth the price then by all means buy it.

But the real improvment comes with the higher ranking units. We believe that one of these is necessary if you want to more closely represent reality in your recordings.

The significant thing was that nobody hit the nail on the head as an exact duplication of the source. Each unit had its own personality. We could see some people choosing a converter personality for a certain type of music just as one would make a microphone selection. If we had to make a decision it would be the Pygmy for for pop/rock and the DCS for classical and possibly jazz. We also felt that the Pygmy and DCS were significantly better than those units of the lower ranking.

There are now several more converters coming on to the market. At this time we know of future products by Apogee, Drake, Quantec, and Vacuum Tube Logic. In our ongoing search for digital perfection (an oxymoron?) we will test these and give you a report as soon as possible.

SIDEBAR - The question of digital transfers:

When this project was completed, we decided to store our musical selections from the Dyaxis disk to 1630 for comparison to units in future listenings. This was a digital transfer of course, just ones and zeros. Lo and behold, something that wasn't supposed to happen occured. When we transfered the 1630 back into the Dyaxis, we found that our signal was somewhat degraded. We were now only two digital generations down. Some of the life and air was gone and the stereo soundstage had collapsed a bit. This was universal for all of the converter recordings.

One of the advantages we are supposed to have with digital is the ability to make multiple dubs with no noticable signal loss. Wrong! In our exploration of this problem, we have found that we are not the first to run into the problem. Recording notables such as Bob Clearmountain, Doug Sax, Bernie Grundman, and George Massenburg have all made statements regarding this issue. This appears to be an issue that equipment manufacturers should start to address and mixing engineers need to be aware of...