Upsampling put to "THE TEST": R U ready 2 take it?

(WARNING!! Lengthy post ahead! Bail now if you don't have the time or the attention span! ...OK, with that out of the way, let's start...)

I recently got an usampling MSB Gold Link with a P1000 Power Base here on the 'Gon, and auditioned it against my old-school Theta DSPro Basic IIIa that does regular oversampling. I did this not because I was unhappy with the sound of my Basic, but because I decided I had to hear for myself what all the uproar about upsampling was for. The two DACs had very different sounds, especially with upsampling engaged on the Link, so to determine which one was more faithful to the data encoded on the disk, I devised a test that many of you, if you have the right equipment, may be able to try at home for yourselves - if you are interested in knowing what your DAC is really doing to the digital signal it receives.

The first thing I should say is that, after the dust settled, I decided to keep my Basic, because to me it won both the controlled test I'm going to describe, and I prefer the way it sounds. However, I'm not trying to make this post primarily into a review of the two DACs in question here. But I do want to state up front that I don't believe that it would be fair to make the Link somehow representative of all DACs on the market with upsampling. So even though I found that this feature did not improve DAC fidelity in my case, I'm hoping others will be able to try to repeat my test, maybe with other kinds of processors, and post back here with their own reports on this issue.


Let me explain what I tried to accomplish in a nutshell. I was listening to two DACs, with everything else in the system kept the same, that each gave very different presentations of the music on the disks. For some recordings, I thought one machine might sound better, but for different CDs, I would give the nod to the other. I became dissatisfied with the subjective nature of this alternating preferrence business, and thought up a way to more objectively assess what was going on. This is an audio test, not some kind of "looking the bits" thing, which I wouldn't know how or be able to do. Besides, that sort of test won't tell you anything about the decoded analog sound your processor puts out. I'm sure I'm not the first person to ever employ this test set-up, but I haven't read about anyone doing this before either.

Here's what you'll need to try this: 1) A high-quality analog source, preferrably vinyl; 2) Some kind of analog-to-digital converter (ADC) that outputs CD standard 16/44 digital, preferrably with an analog input level control, like those found in CD-R recorders; 3) Your DAC to test must have a digital input, which will rule out those found in many one-box players; 4) A control preamp with a tape loop and tape monitor function, preferrably with a full-function remote control that allows remote selection of the tape monitor switch.

Here's how to set up the test:

Take the output from your phonostage into the phono inputs on the preamp, or use the preamp's built-in phonostage if that applies.

Using the preamp's tape outputs, send the preamplified phono signal to the analog inputs of your ADC, such as the CD-R recorder I used.

On the recorder, select the analog input monitor function, and adjust the record level control referring to the level meters so that the signal level is in the proper range - not too quiet, not overloading the input.

With the ADC now encoding the analog signal from your phono into 16/44 digital, take the digital output from the recorder and send it to the digital input of your DAC that's under test.

With the DAC now decoding the digitized phono signal back to analog again, send the analog outputs from the DAC back to the tape monitor inputs of your preamp.

With a record playing, listen to the volume when the phono input is selected normally on your preamp, making any room-level adjustments needed via the preamp's volume control, and then engage the preamp's tape monitor button to switch to hearing the decoded signal coming from your DAC, in order to compare the two volumes.

Using the analog input level control on your recorder to fine tune the level attenuation coming into your ADC (and therefore out to your DAC), go back and forth between listening to the direct phono sound and the DAC sound by switching the preamp's tape monitor button, making the level adjustment at the recorder until the volumes coming through your speakers from the two sources match as closely as possible. (If you have a sound pressure level meter and an appropriate test record, you could use this to perform a final level-match check.)

With a revealing test record playing, retire to your listening chair, remote in hand, and you can now make instantaneous level-matched switches between hearing the straight feed from your phono section, and hearing the same feed as encoded and decoded in CD standard 16/44 digital by your ADC and DAC, simply by engaging the tape monitor function on the remote (leave the preamp's input selected to phono throughout).


In this test, the sound of the DAC will be contributed to by the sound of the ADC. The ADC process can be generally more transparent than the DAC process if all other things are equal (they never are) since jitter is not a factor, but a good ADC should obviously be used if at all possible. Any CD you play in your system was made in a similar way, by using an ADC to convert an analog music performance into the code on the disk. In this test, the LP stands in for the original performance - the record actually is our "absolute sound". This means that we aren't primarily concerned here with the usual question of how well the record captures a believable account of a real performance. If it's a revealing record, of course, it probably will have been recorded with a good degree of fidelity, but in this test we're not trying to decide how much what we hear sounds like what we believe it should in real life. Instead, we are only trying to ascertain to what degree the reconstituted version of the record coming from the DAC actually sounds like that record, in a no-delay level-matched comparision. I frankly think that the one-step, real-time ADC conversion process used here has got to be at least as transparent to the analog sound of the playing record as the studio-mic-feed-to-finished-CD process that occurs with the music that's on a disk you buy from the store.

This test is pretty good for listening to the sound of your DAC vs. the phono feed as a way of assessing the accuracy of the DAC, but it is even better for comparing the sounds of two DACs, or of comparing the sound of an upsampling-switchable DAC with it's upsampling turned on vs. turned off. This is because, in addition to the sound of the ADC making its contribution to any sonic deviation from the straight phono feed, there are also two extra runs of interconnect to account for: the run from the preamp's tape outs to the ADC, and the run from the DAC's analog outs back to the preamp's tape ins. Even if you had a perfectly transparent ADC and DAC combo, the extra wire runs would mean that the sound from the loop out would never be the same as from the phono itself. So expect some degradation just from this factor, and use the best cables you can for the test. (This reality is ameliorated somewhat by the fact that one of these cable runs, from the DAC to the preamp, will always be there in normal use anyway - you couldn't listen to the DAC without it, so you might as well just consider it to be part of the DAC's sound for practical purposes. The same is true of your digital interconnect, and any outboard jitter-reduction boxes you regularly employ; use them in this test if you use them for normal listening.) The reason, of course, that the test is better for comparing at least two conditions of digital-to-analog conversion rather than one, is that these other unavoidable factors (the ADC and cables) will always be held as a constant, allowing you to focus in on just the isolated differences between the two (or more) DAC variables under test. With only one, non-adjustable DAC to test, you'll never really know which deviations you hear are attributable to the ADC or the wires, and which to the DAC itself.


This was the system I auditioned with:

Modified Technics SL-1200 turntable
Benz-Micro Glider M2 cartridge
Camelot Technologies Lancelot phonostage

Innersound remote preamp

HHB BurnIt CDR-830 CD-R recorder

Monarchy Audio DIP 24/96 jitter-reduction box

Conrad-Johnson MV-55 power amp

Thiel CS2.2 speakers

All Cardas Cross analog interconnects (single-ended RCA)
All Cardas Lightning digital interconnects (single-ended RCA)
Cardas Cross speaker cables
Synergistic Research Master A/C Coupler on preamp

Audio Power Power Wedge Ultra 116 (balanced, filtered AC for all digital front end components and preamp)

(The digital components were used with their stock power cords, as I don't own multiple identical aftermarket cords.)

All right, let's get to the listening test results. Like I said, this is not supposed to be as much of a review of the MSB Gold Link or Theta DSPro Basic IIIa DACs as it is a report on my test set-up, an assessment of upsampling as implemented in the Link, and an encouragement for some of you all to try this experiment too and report back here. So I'm just going to jump right to my conclusions and skip all the listening impressions, if that's OK with you.

First, just a little background: Both of these units sport dual-differential DAC chips per channel for balanced digital operation (lesser Links do not), with multi-bit PCM architechture, 8X oversampling, and op-amp output buffering. The Theta implements output filtering in the digital domain on DSP chips incorporating proprietary software, whereas the MSB boasts of 'minimal' filtering. The Gold Link comes with an installed upsampling board that can be set to 96KHz or 132.3KHz frequencies internally via a removable jumper, and an exterior switch to turn upsampling off. Build quality seemed roughly comparable, nods going each way. The Basic originally cost about $2,700 - about a grand above the Gold Link with its Power Base - but used today the Theta sells for slightly less than the MSB combo used.

I esentially did two kinds of listening, and reached a few significant conclusions. I'll start with the second kind first. The last comparitive listening I did was with the phono bypass test descibed above. I did this part for a few hours, after I had already spent many more hours auditioning the units conventionally (all auditioning was done over several days). The controlled tests basically confirmed for me a handful of things. First, my Theta was the more accurate DAC of the two, and not by a little. Doing the bypass test with the Basic in the loop proved to be remarkably transparent. I really wouldn't want to have to put money on telling you whether I was listening to the direct phono feed or the DAC if you blindfolded me. What I did hear being lost could've reasonably been attributed to the cabling alone, that's how small it was. This made me glad about my CD-R recorder selection, too. Just a slight thinning-out of tone color, a bit less transient jump, and a small addition of background texture were all that was there to let me know I was listening to the processed feed. This also led me to conclude that a lot of what I've always assumed was just a part of the Theta's sound - a little coolness and dryness despite its virtures - is actually on the CDs I play, because I couldn't hear those things in this test. The sound wasn't any cooler or dryer than the records, just a tiny bit paler.

The second finding from this part of the testing concerns the Gold Link and its upsampling choices. As I implied already above, the Link could not do as good a job of disappearing in the test set-up as the Basic did. What is most interesting to me, however, is that it showed its best with the upsampling defeated, meaning that this sought-after option, which has been the main source of all those subjective raves, is actually taking the unit's performance *further* away from reality. However pleasing this DAC's upsampling effect might be on some CDs to some people, it's going to be very hard to argue that a process which takes the encoded sound of an LP record and decodes it so that it sounds even less like that record than when it's switched off, is somehow going to make all your CDs more 'analog-like', as has been claimed.

With upsampling defeated, it was still a lot easier to hear whether the Link was in the circuit or not than with the Basic, but I don't want to go overboard with this impression. We're still talking about maybe a 6% change in the sound now, but I would classify the Basic as being around 2% - and that's including the contributions (and subtractions) attributable to the ADC and cables. The Link was more intrinsically colored and opaque, and more greatly reduced dynamics, extension, and transparency to detail. In particular, it altered tonal color noticably, in a way similar to a classic transformer-coupled tube amp of yesteryear. It rolled the highs and a little bit of the lows, thickened the upper bass and low mids, and gave the midrange a dollop of yellowy-colored warmth. It also pinched the soundstage from the sides and back, and laterally splayed images somewhat.

Switching in the upsampling definitely added more murkiness to the sonic picture - veils were being lowered! The treble was even more rolled this way, but now the deep bass also was pronouncedly lightweight, despite the increasingly thick upper bass. The midrange coloration mentioned above increased in intensity, while dynamic shadings got glossed over. 96KHz was better than 132KHz in these regards - it became sort of a progression from best showing to worst as I moved from regular 16/44 to 24/96 to 24/132, but I have to caution you not to assume that this means that higher is necessarily not better in this regard, since MSB in the manual admits that 132KHz 'overdrives' their merely 96KHz-capable DAC chips. But as it is implemented in the Link, oversampling in this test took the processed sound a few more percents farther away from the reference with each increase in frequency. I do have to admit here that, despite these objective results, more than a couple of times I found myself greatly enjoying the Link's renditions of different CDs in my preliminary auditioning, and some of those most with 132KHz upsampling engaged. However, even before running the bypass tests, I strongly suspected that some pretty hefty accuracy and detail babies were being thrown out with the digital bathwater, so to speak. It's no wonder that the Link's upsampling seemed to be fairly effective at banishing digital texture artifacts, along with such things as harsh-sounding instruments or sterile-sounding CDs, because it was covering up a lot of stuff - with a shovel and an axe - that I guess some folks would rather not know about. But there is a real price to be paid for that, which should not go without being fully acknowledged.

The other main finding I got out of this exercise also involves the upsampling function, and was first run across during my preliminary auditioning before I got to the controlled testing later on. Again, I want to stress that this complaint is not to be taken as an indictment of upsampling as it is implemented in other products which I have not heard, only the Link. But even in reviews I've read of this product, what I'm reporting now has never come up. After auditioning for many days, I decided I had to pick a favorite between 96KHz and 132KHz upsampling, if for no other reason than you can't easily reselect the other option after the unit is installed, but only when the case is opened up. 132k sounded smoothest and most different from my reference, while 96k seemed to preserve a little more of the sound I was used to. I had mostly been comparing either one or the other upsampling mode against my Basic, but now I wanted to compare the two modes against each other, so I completely removed the top cover from the Link in order to have quick access to the jumper on the upsampling board. Focusing on which upsampling rate gave the best overall results the most often, I began to notice a bothersome trend.

When I changed the upsampling frequency and played a song over again, I found that the harmonic emphasis had shifted. This was going on above and beyond any of the more obvious changes or colorations I've already covered. It was probably harder to notice at first, because upsampling in general rolled-off the highs where the overtones live, but just listening to the two upsampled processing choices without interruption from non-upsampled sound brought the phenomenon to the fore. Especially on material with artificially-enhanced harmonic content, such as chorused electric guitar, it sounded as if each setting of the jumper was selecting a different filter for the overtones on the CD. Once identified by the ear, this was hard to ignore, even on acoustic material. The effect was very dependent on the exact source being processed, but seemed disturbingly 'mathematical' in nature. When upsampling was turned off, the selective emphases disappeared, and the harmonics regained a more neutral perspective. (If you listen carefully for this in a revealing system with sensitve test material, it still won't hit you over the head right away, but I do want to see if anybody can confirm this finding. To those who have a Link with upsampling and want to try this, it's easier to remove the top cover completely out of the way if you first unplug the front panel display.)


I know that even if everybody did these comparisions and heard what I heard, they still wouldn't all agree with my conclusions, because a lot of folks don't care about whether or not the sound coming out of their system is 'accurate' or not - they take the emminently practical position that they just want it to be pleasing. After all, we are listening for entertainment. For this type of listener, even if an arbitrary 'correction' to the sound on a CD just reminds them superficially of a type of sound they'd rather hear, that's fine with them. And I'm not here to criticize their choice, though I can't listen happily that way. But when certain prominent writers who shall go unnamed preach from the bully pulpit about gear containing "magic bullets" to cure something they label "CD sound", don't automatically take them as prophets spreading the only true gospel. Try this for yourself if you can, and let us know what you find out.
I agree completely with your findings!!!!!!!!!! We have found the same thing here using even much more expensive upsamplers. Good job!
Whewwwhhh! I finally ended reading but with curiosity and from top to bottom THANKS!

I knew you would stand towards "Former Heavy Weight Champs".
I believe that you was someway wrong comparing MSB Gold with Theta DSPro Basic III. You've mention a difference in built quality as you should've mentioned a cheap board with cheap parts inside that unit despite the name "Gold".

I have an assumptions that I want to share and to comment your test. I'm about to wash brains of my dealer and drive him to setup the same as you've greatly suggested.

In general terms I can assume that for reading an upsampled or any digital information there have to be certainly devices either switching transistors or opamps that are capable to keep up the sampling speed(please do not compare it to the sine waves in this case). There are micro-transistors and there are micro-opamps in the DAC chips that combine inside a number of other micro-elements with certain re-locking switching freequency. And it's essencial for the DAC chip to capture the exact form of the input signal.

Now let's continue to talk about up- and over-sampling. On one side it reduces a probability of error but the distortion is INCREASING with larger number of samples due to imperfection of DAC mechanizm (mainly chip) to keep up the presice signal form and speed as well. Thus we loose linearity and colour the original signal. This can be reflected on any audiable freequency spectrum and also depended on the digital software mastering.

Despite this reality there are still a possibility to create a fast-acting DAC chips with every responce manually tested through the oscilloscope... What price of the unit or even the chip we're now talking about?

In one-box player case DAC chips can be carefuly matched to the responce delay and signal form imperfections to mate the transport near-perfect. In outboard DAC the DAC chips have to be perfect by themselves.

So in reality there is a finite number(that realy makes a sence) for up- and over-sampling despite the theory of Mr Nyquist. There are not to many variables to define this number or maybe... Maybe it's already been defined more than 20 years ago and we'll come back again to the same conclusion?
We have used much more expensive DAC's that upsample and have come up with basically the same conclusions. Maybe not as dramatic difference as Zaikesman using the MSB but the same basic result all the same. A general smearing of information which if you have used a older DAC or one which was bright due to brick wall filters can come as a relief as it generally is not as harsh. But at what cost. We have found a really good DAC which doesn't upsample depending on how it is implemented will have more air, transparency and will be more involving. This has come from a dozen or more shootouts in which customers were unanimous in their thoughts. I am not against upsampling. I hope it really someday makes as huge a difference as it looks like it should on paper. It Just doesn't do it yet to the ear. It was great marketing though in a time when cd player sales were very flat.
While I have no doubt the basic findings would be similar, we are taking the reviewer at his word when exclaiming "everything was the same throughout". This is a bit hard to swallow. Tweaking has a greater impact on sound then many/most realize. The accuracy of the test would require identical shelving, isolation (above and below), power cords, interconnects AND each power cord interconnect plugged into the same locations. With that said, that too could skew the results, as different chassis require different isolation devices, as will different internal parts require different interconnects to extract their magic. Does each have identical manufacturer RCA connections? These alone could skew the comparison. Sorry, I don't by the paradigm. However...

Even with all that accounted for, I would not be surprised that the Basic was the sonic equal, or preference to, the MSB. However, this should not imply that all upsamplers therefore sound the same. The Northstar 192 DAC sounds best when used with their transport, as you can use I2S to take the signal directly from the optical pick-up avoiding a bunch of nasty jitter. Better still, the GamuT CD-1 gives 2 times DSD up/oversampling in a one-box player, eliminating at least two variables. I am sure their are other examples.

I applaud your common sense approach for comparative listening. It seems that perhaps we have a little bit of processing going on.

My conclusion is, regardless of source material, that recording, engineering and mixing are the key elements in audio bliss. Give me a well crafted LP or CD and I never complain, nor do I want to rush out and buy another box, cable or power cord. Most recordings however, SUCK.

Happy listening,
Ohh!! that GamuT CD-1... Excellent machine!
...but still it's one-box with perfectly matched transport and DAC i believe and it's still not a turntable...
Zaikesman, I think it is great that you took the time and effort to do this test and report it to us. Even though I do not use an outboard DAC, I found your test to be very interesting. While some may argue your findings, I am happy to have at least one good source of information on this subject, where effort was made to be as accurate as possible in listening comparisons. Kudos to you for your test, and we are all benefitted by your work. Thanks for adding this to our information bank.
Yes the Gamut is very good. As over and upsamplers go it is definately one of the better ones out there.
I should also add a couple of notes. First, by not keeping the MSB Gold Link, I am passing up the opportunity to move to a DAC that accepts data formats other than Red Book 16/44 CD, such as DVD/DAD audio or DVD video. I don't own any of these now, and don't know too much about that stuff, but maybe I'll want a DAC that can handle those in the future (if there's still such a thing as a digital output!) Second, I haven't yet sold the Gold Link. I trust that my article won't make it too much harder for me to do so, but if I advertise it on A-Gon, I don't want to get the reputation of someone who first sells a piece, then goes on the forum and puts it down. I laid out the bucks for this unit, and I've put my money where my keyboard is, so to speak. Third, the Link is supposed to work much better with its AES/EBU XLR input, so I have a cable on the way right now to try this with both machines, and will post an update. (BTW, Tellig was wrong about balanced input engaging the 'MSB Network' proprietary input option - you need to get a $600 mod done to your transport by MSB first.)

As to Tweak1's questions, yes they were on the same shelf in the rack, sometimes separately, sometimes stacked with both seeing time on top and bottom. No, I did no footers or anything special underneath. No, I didn't try different interconnects than what I listed above - I don't have anything better than these, though. The DACs were both fed from two different sources, and in each case my impressions were consistent. Although I appreciate Tweak1's concerns, frankly what I heard went to a degree of difference that is beyond tweaking - we're talking about fundamental character here. I would not expect this to change a lot, but don't rule out some synergy in my system that would be changed for others. And I make no attempt to extrapolate my findings to other products, although I do have a new respect for those companies, like Theta, that aren't jumping on the bandwagon. I had previously assumed Theta were just distracted by HT, and that other companies simply had too much tied up in their current technology, or rolled out product too slowly. And it's obviously not as simple as a case of them merely being Luddites. JC's point about marketing in hard times for CD is very astute, I think. Any takers on trying the test yet?
Thanks for the work Z-man. I will certainly use your information when looking into upgrading my DAC (I understand you have one for sale;)). I tend to fall for these new products without doing this type of comparison, but unfortunately I cannot afford to have 2 of the same type of gear in my system at once to do side-by-sides. I wanted to like SACD and DVD-A just as I wanted to like Upsampling, but as Lugnut pointed out, just give me a well engineered cd and I'm happy. Maybe the industry should stop throwing these new wonder tweaks/formats at us, and spend time and $$$ getting the recording industry up to snuff on how to mix a cd consistently.
Zaikesman " a degree of difference that is beyond tweaking - we're talking about fundamental character here. "

That is a really interesting concept. Would make a good thread topic: "What actually can be expected from tweaks?"

Would power cords, interconnects, speaker wire, vibration dampening, be considered "tweaks"? Where would one draw the line between tweak and upgrade?
Z, thanks for posting your results here. I know that conducting such a test is very time consuming let alone writing a "novel" explaining the whole scenario. Your effort is much appreciated.

I would have to say that i believe your results 100%. Given the fact that the motivation behind your testing was strictly for your own benefit, there is no logical reason to believe that you would want to lie to or fool yourself by making the tests anything less than "fair" in your own eyes. As such, i'm sure that you did the best possible in those terms.

With that in mind, my limited experience with upsampling is somewhat different than what you stated. I don't think it is so much "upsampling" as a whole that instigated the lesser quality performance that you witnessed, but rather how upsampling was implimented and the parts quality & circuitry that they chose to do it with.

As a case in point, there are many paths that one can take when going from Point A to Point B. All of them might be called "highways", but some will have a lot of bumps along the way, making the trip less comfortable and more fatiguing. Others may have more traffic and be quite congested and slow, which would increase frustration. A select few may provide excellent travel times with beautiful scenery and minimal road hassle and congestion. Obviously, this is the path that most would choose to take. That is, if they were aware of such a path. The end result of all of those paths are the same i.e. you got from Point A to Point B, yet the way that you got there was done in a very different manner. The same can be said for circuit designs in any electronic based products. There are more than a few methods and designs that claim to do the same thing, yet they all do them very differently.

I think that most of what people here in any cd player or DAC ( upsampling or not ) are the power supply, passive components and filtering used. I am of the belief that use of a quality power supply with a lot of filtering and regulation is of utmost importance. Since transformers and digital gear produce both RF hash and magnetic fields, one has to take steps to minimize the potential for problems in both of those areas. This means paying special attention to shielding and / or how wiring is laid out on the board. After all, a wire connected to and running through a magnetic field can become a radiating antenna. Use of ribbon wire inside any digital component can only lead to "cross contamination" in various parts of the circuitry. The wires in such a design run parallel and so close to each other that the inductance involved pretty much makes them an easy target. As such, point to point wiring becomes mandatory. Even if one were to do that, you could still run into problems if steps were not taken to minimize the potential for that wire to pick up RFI / EMI. This means individually shielded wires or wires using non-standard geometries become mandatory.

I also think that different brands & types of resistors, capacitors and diodes have various sonic signatures. Knowing how to blend the various sonic signatures of various brands in each part of the circuit to achieve good tonal balance, proper harmonic structure and good transparency is a very fine art. This is known as "voicing" a component. Being able to do this and keep parts count to a minimum can result in a very short and excellent sounding signal path. However, even a short signal path with minimal parts count can suffer from impedance mismatches, microphonics and RFI. This means more damping and shielding with special attention to how things are laid out on the circuit board.

I also think that filtering on the analogue output is detrimental to performance. Besides creating phase shifts, filters of any type create impedance mismatches and reflected energy back to the source. Since digital is basically an RF signal, these "internal reflections" are technically called VSWR ( voltage standing wave ratio ) and create loading irregularities within the circuit. This can result in loss of coherence, liquidity, increased grain and an overall lack of transparency. Getting rid of, or minimizing, the filtering on the output of the DAC makes a HUGE difference in my opinion. Upsampling and using high quality passive components in a well thought out design only adds more icing onto the cake.

The most "organic" presentation of digital recordings that i've ever heard has come courtesy of equipment that has extremely little to no output filtering in the DAC. As far as i know, there are only a very few products that make use of this design strategy and philosophy with none of them available in SACD or DVD-A players.

In plain English, i've heard better "red book" digital than i have heard come from newer highly touted "high tech" methods of recording such as SACD and DVD-A. Bare in mind that this is not to say that these formats do not have potential. I think that they do. Just as is the problem with redbook, my thoughts are that the prime limitations are based in the recording studios and disc making process and how the design of such players is implimented for home use. I think that we've finally gotten to the point that we've got a good handle on how to make "good" digital discs and playback systems using a low sampling frequency. Things should only get better as we increase the sampling rate ( as SACD & DVD-A does ) and fully incorporate all of what we already know into future playback equipment designs and recording techniques / equipment. Obviously, this will take some time as it has taken us almost 20 years to get this far refining a much simpler design.

Luckily, some engineers are smart and they will carry over and incorporate what they already know into future designs. This means we should have much more rapid advancement of the SACD / DVD format than what we originally had with redbook, but that is only if the manufacturers and designers choose to apply themselves and produce that level of product.

Since none of these products exist today, or what i should say is, IF these products exist today, i am not aware of them nor do i think that i can afford them. As such, i've settled on a redbook system that takes advantage of all of the aforementioned design goals. While it is not perfect, it is the best that i've been able to assemble with the means and resources that i have. It just so happens to be based on an upsampling DAC. While i've not ruled out any other platforms of reproduction, i have heard enough to know that what i have is a very solid performer and that i have good reason to be happy with it.

As such, all i ask is that people keep an open mind about the ENTIRE "digital subject" and look at things objectively and on a product by product basis. Otherwise, we are bound to run into classifying all digital products into 1X sampling / oversampling / upsampling / SACD / DVD-A, etc... categories. This will not help us in any way just as it hinders us in the all too common "tube vs SS" debates that seem to run rampant. I think that the bottom line is "musicality with good accuracy" and any / all of these methods are capable of doing so. That is, IF the designers put their mind to it and we thoughtfully assemble the correct components. It is up to us to tell the desingers what we want and are willing to pay for. Hopefully, they will listen and try to give us those things, regardless of the method used. Sean
Zaikesman, I applaud your efforts and thank you for sharing your results. I want to suggest a simpler and possibly more accurate test setup. Rather than listening directly to the two D/As using your preamp as a switcher, record directly from the D/As to your CD-R and make a compilation disc. Some of the tracks on the "test CD" will be taken from the analog outputs of the Theta, others taken from the MSB, and to make things more challenging, transfer some tracks digitally to the CD-R without going thru either of the D/As. Obviously, you will have to match signal levels. You now will be able to compare the effects of the D/As to each other and to an "uncoverted" signal. Furthermore, if you set your CD player to random play, the test will be blind. Sit down, listen, take some notes. You can then confirm your impressions by going back and doing a "sighted" listening.

FWIW, I've never understood exactly what upsampling is or how it differs from oversampling or sample rate conversion. I think there's an element of marketing hype involved.
Sean, thanks for the thoughful response. First thing I would like to know from you is, are you set up to try and replicate this test at some point? If you have the needed ancillary gear, and your DAC has an upsampling defeat switch, I'd love to see what you would come up with here.

61, let me address your proposal first: the problem with this is that to listen to such a CD-R, you need to put the signal back through one of the DACs again, or another DAC - you cannot listen to an 'unconverted' signal. So your test would have the effect of comparing a signal that's been digitized and converted twice against one that's been so processed once, which I'm afraid wouldn't be very useful. Remember, in the set-up I describe above, you are not hearing the two DACs [or two DAC modes] pitted against each other, but rather separately compared one at a time directly against the analog feed they are processing the digitized version of. Also, the random component you mention couldn't make your test 'blind', because you would have no way of comparitively evaluating what you were listening to unless you noted the track numbers as they came up, in which case the test is no longer blind (it is generally not possible, for the same underlying reason, to construct a 'blind' test run all by yourself - someone has to be able to take note of which choice the listener is evaluating at any given point in the test, even if that person doesn't know which variable that selection actually represents, in the case of a 'double-blind' test, which requires yet a third person).

Now, if in my set-up proposed above, I had had two identical digital outputs on my ADC, then a friend could have helped me to do a blind ABX test using both DACs and the direct analog feed at the same time, since my preamp does have two tape monitor inputs. As it is, my CD-R recorder has one S/PDIF output on RCA and one Toslink optical, which would have confounded such a test. My Theta does happen to have its own digital record loop on RCA, so I could send a feed to the MSB, but that would introduce an extra digital cable running to the Link. While running the test as a blind ABX would be interesting and valid to do, if you are able to try the procedure I detail above in your own system, I think you will wind up agreeing with me that it is very useful and illuminating in and of itself.

Getting back to your points, Sean, I can't argue with your caveats, and indeed do believe that I essentially acknowledge the main thrust of your comments in my posts: that this test can only evaluate the particular components under review, and should not be construed as an overall indictment of the upsampling technique. It was a provocative result, however, precisely because the other factors affecting the sound that you bring up (parts quality, filtering scheme, power supply, etc.) stayed the same when I compared the different modes (no upsampling, 96KHz upsampling, 132KHz upsampling) present within the Gold Link itself. But as you correctly say, implementation is everything. The observed finding that the Link's output most closely resembled its input when upsampling was defeated is what prompts me to urge others, some with other kinds of upsampling machines, to try and run this test themselves if they can.

Obviously, the only final conclusion that can be drawn from my own experience is that I prefer my reference, but I declined to get fully into the reasons why for this, because I didn't feel like turning my article into a more subjective personal take on what I think of these two machines' sounds when playing music. I will say, however, that doing this test and observing the more objective results generated under controlled conditions, did make it basically impossible for me to then go back and listen to the Link playing music and just simply enjoy its particular presentation of the program, when I knew in that in fact it was editorializing the sound more than my reference - however pleasing that presentation might have been with any given disk. JCAudio above has much more experience in this area than a lot of us, and holds a broader opinion of what he's heard concerning upsampling as a catagory. Again, the more folks who attempt this test set-up (how about you, JC?), the more general conclusions we may either be able to come to, or to know to be wary of.

Years ago, long before the advent of upsampling products in the marketplace, I wondered why the same basic process that enabled players to carry out error correction could not be employed to interpolate additional 'smoothing' data points in between the samples contained on the CD. I thought, if a player can generate a reasonable facsimile for a missing sample, or even several in a row, than why shouldn't that technique be used to double or triple the sampling frequency? I couldn't believe that this wouldn't have occured to digital engineers however, so I simply assumed that there must have been some inherent flaw in my reasoning that I was just too ignorant to understand.

Upsampling certainly seems to be the realization of my daydream scheme from back then, so I consider myself to be more than open to the possibility of its unqualified success. And of course, that is precisely why I went ahead and laid out the cash to hear some fruits of the technique for myself, even though you could never accuse me of being an early adopter. I was quite prepared to sell my Basic and keep the Link even if I heard merely comparable sound, because of the Link's more up-to-date input capabilities.

Maybe in the future I will try out a different implementation of the upsampling technique, but I'm not as curious anymore - not because of my experience with the Link, but because of the results my Basic returned. Since my test revealed it to be quite transparent to the source, I consider myself satisfied, for the time being, standing pat with the DAC whose sound I was pleased with when I got it, and continue to be. I'll just listen to it knowing that, if I hear something I'd rather not, it's probably the disk and not my DAC that's to blame.
I will probably have a device that allows a "straight wire bypass" test to be performed on various devices within the next two weeks. This will allow me to insert any analogue based product ( that lets DAC's out ) and compare them to running a straight wire. It can be used for doing blind ABX testing or simply comparing product A to B to see which sounds "better" or "different" ( depending on point of view )or if the difference between them is discernable at all. I went and looked at it today and will probably end up purchasing it when all is said and one. Mind you, it is nothing that i need but i think it could result in a helluva lot of fun and learning : ) Sean
May I join the cohorts of those who thank Zaikesman for the thorough experiment and thoughtful reporting of results.
The link below will connect you to an article on exactly this topic from hi-fi news and record review. It is the very least.......nice experiment btw.
Hello again. It has been over a year since I originally posted this article, and in it's time it received over 900 views, the most of any thread I've started on Audiogon. To my mild surprise, I have yet to discover any other accounts about anyone else, say a reviewer, constructing a test similar to the one I describe, but I continue to maintain that the basic procedure I outlined should be an objectively valuable method for assessing DAC performance - one that ought to be copied IMO.

My method enables a feasible (provided you have the necessary ancillary gear listed above), common-sense, in-home way to comparision-test - instantaneously in real-time - the sonics of one DAC against a near-absolute reference of the signal it’s decoding, or even better, to compare the relative performances of two or more DAC’s (or of one DAC with two or more selectable operational settings available, such as filter slopes, upsampling rates, etc.) in a manner which allows a positive determination to be made of which option remains truest to the sound of the reference feed, and to identify whatever deviations from reality are occurring with much more certainty than through conventional subjective auditioning alone. Then an audiophile can pick his or her poison, be it literal accuracy or euphonic massaging, without feeling forced to simply wonder in some degree which of two or more ostensibly good-sounding presentations is more correct and which is less correct, in addition to helping aid in quantifying what audible system shortcomings should be rightly attributed to DAC performance in isolation.

[BTW, I am happy to report that I was able to successfully sell my MSB Gold Link/P1000 on Audiogon for what I had invested in them, even after the appearance of this thread. Today I am still using my long-discontinued Theta, and have not tried any other snazzier, 'new and improved' digital sources in my system during the interim, and have no immediate plans to either, be that to my credit or shame.]

My only disappointment with the response to this thread has been that no one has attempted to duplicate the described test set-up and report their findings back here. As this was one of my primary goals upon writing the article in the first place, it is also my main motivation for posting anew today: there are many newer members on Audiogon since the last thread activity, and I wanted to bring this article back up from the depths of the archive in hopes of freshly stirring the pot (might not work too well, but what the heck).

In that spirit, I also intend to re-perform this test set-up again myself, even though at present I can only do it with my one (non-upsampling) reference DAC and no other competition. I want to do this a second time mostly because there have been several upgrades throughout my system during the meantime that should make the test more revealing, and I am curious whether I will still come to the conclusion that there isn't a hell of a lot of meaningful difference to be heard between my straight analog feed and the same feed that's been run through my reference ADC and DAC conversion and reconversion. (Those system changes will be detailed when I post my listening results.)

So I look forward to playing with the test procedure once again, and especially to any and all additions to this thread from the forum contributors. I know it takes a bit of time and effort to run this procedure (not to mention read through this thread! :-) , and that not everyone has the needed gear or system configuration to do so even if the inclination is there, but I'm hoping that some of you will be able and inspired to to run this test for yourselves, and write about it for all of us. (Anyone interested who has further questions about how to do this test for themselves is encouraged to post them here or email me directly for clarification.) Thanks for reading, Z.

I have read your long post very, very quickly but I think that I have the gist of it.

Sometime back (say 1 to 1.5 months ago) some other A'gon members & I had a long discussion on upsampling & oversampling. Much EE stuff was discussed in those 35 or 37 posts. Guess what......that thread has been deleted!!! Somebody started a "24/192.." thread recently & I wanted to point him this upsampling/oversampling thread but I discovered that it was deleted!!! I'm quite pissed!

We cut thru the chase in that thread & discussed what upsampling & oversampling do & why the sound of the music was different in each case. It laid bare the basic fundamentals of both techniques. Anyway, it *appears* that A'gon & the sponsors didn't like this very much!

Anyway, with upsampling I'm convinced that one is listening to the digital reconstruction filter (that follows the upsampling event). Each manuf. has their own proprietary technique for doing this. It seems that very few of these implementations are actually liked by the end-user while most of them just add distortions to the music signal.

It is no wonder that you like an oversampling design. This too requires a filter after the oversampling event but since no new data points are being estimated between data points sampled off the CD, the filter is less in the critical path. Hence the reproduced sound is more true to the recorded signal.

Anyway, looks like you have gone thru some pain to discover this yourself. I'll read your post in details & *might* have some more comments to make.
Bombaywalla: I can't pretend to possess the technical expertise to comment authoritatively on your post (though I'm not at all sure you are on the right track when you opine that "very few of these [upsampling] implementations are actually liked by the end-user", and I don't really follow where you talk about a filter being "less in the critical path").

I will say that I tried my best to stress, in all of my posts above, that my findings should in no way be taken as some sort of overall indictment of all so-called upsampling techniques or implementations. To the contrary, as best I can understand it (hard definitions in this area seem tough to come by), the theory behind upsampling appears to make a lot of sense on its surface.

But getting any takers to try to running my controlled auditioning test and shed some more light on the topic is proving a tall order. When it comes to testing upsampling effects in particular, part of this is predictable: there aren't too many DAC models out there featuring selectable/defeatable switched upsampling, and lots of the newer models incorporating upsampling technology are all-in-one players, not stand-alone boxes with digital inputs.

I know I prominently mentioned upsampling in the title of this thread, and that was deliberate. I thought it would attract the most interest to my article, and investigating upsampling's effects in one of the DAC's I was testing was a big part of what I was trying to accomplish when I first hit upon the described test set-up.

But I think the wider implication for employing my test design lies in simply comparing one or more DAC's against each other indirectly by making serial comparisions against a known constant reference feed. Believe me, I would love it if someone with another DAC model featuring switchable upsampling were to run my test and post their results, but I'm no longer crossing my fingers on that wish.
Zaikesman, I just stumbled upon this thread via current thread called 'Up and Over sampling Is one superior to other' by Blindjim.

Have you done any updates/auditions with current batch of reference UP and oversampling DACs. Has your conclusion changed?
No, but I hasten to state again that I never reached, or ever tried to reach, any sweeping conclusions about upsampling, however that term may be defined. My limited experience couldn't support that. I was mostly hoping for other listeners with other DACs to try out this test themselves and post their results here, but as far as I know none have repeated it. I still think it's a valid and objective way to assess the accuracy of a DAC. I did repeat this test once with my longtime DAC (the Theta described in the header), mostly because I'd made several other improvements around my system and wanted to see if my impression of that DAC's perfomance in a bypass test remained as high (basically it did). Where this test falls short of being as informative as I might wish for is that it doesn't tell anything about how a DAC will interface with a transport reading a disk, in terms of jitter rejection.
One of the things I found very interesting in your test, Alex, was your finding that the Theta sounded practically indistiguishable from the bypass test. From that, I take away that digital may not be the culprit we tend o think of it as (re: preferring the sound of vinyl) but that the problem may lie somewhere in the creation and/or extraction from the compact disc itself. I would appreciate comments on my reasoning.

This interests me because it suggests A-D conversion of our analog sources, for purposes of room correction or of supporting systems with digitally-controlled crossovers, may not be the unacceptable heresy it appears to be at first blush.
Dan: I agree, that's the conclusion I came to as well. The bigger problem with CD may not be resolution so much as getting the data from the disk to the converter in real-time with accuracy. It's not that there aren't losses going from analog to 16/44 digital and back again in real-time without an intervening optical-read disk -- there are and I can hear them better now than I could at the time I wrote this thread -- but they don't really seem to be the defects that audiophiles usually complain about when comparing vinyl to CD, either in kind or magnitude.