Heads-up on a DAC not mentioned much here

[aljordan]

Hi,

 

I thought I would pass along some information about a DAC I am currently trying that appears to have very good sound quality and value. The DAC is the Reference 7 made by Audio-GD. It uses eight high quality R2R ladder DAC chips in a fully balanced configuration, and appears to have extremely high quality analog stages.

 

The DAC has an adjustable digital filter that can be used in either non-oversampling, or 1x, 2x, 4x, and 8x oversampling configurations, along with adjustable stopband attenuation. Power supplies are separated between digital and analog circuitry, with separate supplies per channel.

 

I decided to try this DAC out because I wanted to compare a well thought-out audiophile design with my current well designed pro-audio DAC (Prism Sound Orpheus). The Ref 7 retails for $1865.00

 

Audio-GD offers two flavors of DACs; what they call "neutral" and "musical". The Reference 7 is one from their neutral line. I bought the version without a USB input because the available USB input is not an asynchronous implementation, however the included DSP-1 digital filter module seems to deal with jitter very well. I am currently experimenting with different means of driving the DAC from my PC. The maximum internal sampling rate of the Ref 7 is 96 kHz.

 

Thus far, I had found only one other DAC that I like as much as the Prism Orpheus; the Resolution Audio Cantata. The Reference 7 uses the same DAC chips as the R.A. Cantata, but doubles the number. From what I've read on-line, the Ref 7 takes very many hours to break-in, but out of the box the sound of the Ref 7 is very smooth with no glare nor hardness. Supposedly the US importer burns them in for a few days before shipping them out, so that may have some bearing on my initial impressions.

 

While the design of the Prism Orpheus is primarily intended for studio use, I bought it because it had the best sound quality of the very many DACs I had tried out. Obviously, the Ref 7 and Orpheus are designed for very different markets, with the Orpheus having a lot more features. However, out of the box, the Ref 7 sounds as good as the Orpheus, which is quite an accomplishment given the retail price of the Ref 7. At this point the two DACs are very similar in terms of overall tonal balance, detail and smoothness. I'll give another update after I have played with the Ref 7 for a week or so and have had a chance to become more familiar with its traits in comparison to the Orpheus.

 

More info can be found at: http://www.audio-gd.com/Pro/dac/RE7/RE7EN.htm

 

Alan

 

[hiro100]

Thanks for the heads up Alan. Looks like an interesting piece and one I'd want to hear. Do you know if they offer a trial period after purchase?

 

[aljordan]

Hi,

 

It is not a true trial period, but Pacific Valve does allow a two week period, within which you can return the unit for an 11% restocking fee. If you trade in for something else within that period, the restocking fee is waived.

 

Given that you already have the ULN-8, which I haven't heard in my own system yet note that it seems to be cut from a similar grain as the Orpheus, it may turn out that the Ref 7 sounds very similar to what you already have. I'll let you know if the nature of the DAC changes over the next week or so.

 

Alan

 

[modena]

Many thanks for posting your impressions. I am considering Ref 7 + Halide Bridge BNC as an upgrade option over my HDR.

I await your update of Ref 7 - after the break in period - vs Prism, with great interest.

 

[Marcin_gps]

Alan, thanks for the great post. I've been waiting for a comparison between Audio-GD top line and Metric Halo/Orpheus DACs for a while. You didn't mention how did you connect the ref7 to PC?

 

Looking forward to your reply,

Marcin

 

[aljordan]

Hi Marcin,

 

I've been connecting the Ref 7 to the PC in one of two ways.

 

1) Using the coax S/PDIF output of an M-Audio Delta 410 PCI card. This particular card has the coax output as part of the card and hence does not use any DSUB adapter cables, and is transformer isolated. I've always had better performance feeding DACs using this card over the digital output of my Lynx Studio card.

 

2) Using the Prism Orpheus as an asynchronous firewire to S/PDIF converter, which is insane, but I own it so why not.

 

Given my experiences with the differences between the M-Audio card and the Orpheus feeding the DAC, it seems to me that the DSP-1 unit on the Ref 7 is doing a good job cleaning up the signal that is sent to it. There are subtle differences between the two ways of feeding the DAC, but those differences are not overwhelming.

 

If I keep the Reference 7, which I am leaning towards doing at this point, I might buy an asynchronous USB to S/PDIF converter so I can go back to using my Fit-PC Slim as the music player.

 

Alan

 

[Shibboleth]

That's one expensive Firewire converter!

 

How close in the AudioGD to the Orpheus and the Antelope Zodiak? I've been toying with auditioning all three recently . . .

 

[Marcin_gps]

Alan, if m-audio works as good as prism as a transport, it means that Reference 7 does a great job! It's a shame that it supports maximum 96kHz... If it had 24/192 capabilities, I would get one right now.

 

Cheers,

Marcin

 

[weib]

Please don't tell me you're using stock power cords for the Ref7.

 

[aljordan]

Okay, I won't tell you that I'm using stock power cords for the Ref7.

 

[weib]

The stock power cable is horrible.

If you changed to a better one, the improvements will make you keep it for good haha

 

[manisandher]

Hi Alan,

 

Thanks for bringing this DAC to our attention. It's great to see another DAC using R2R chips (and not delta-sigma, which just kill the sound to my ears). The rest of the design looks very well thought out too. For the price, it looks like a real bargain.

 

You said that the DAC has an adjustable digital filter that can be used in either non-oversampling, or 1x, 2x, 4x, and 8x oversampling configurations. This is a very interesting feature. Can you tell me how the various over-sampling rates are selected? Is the selection made automatically, depending on the input sample rate? In which case, how does one switch between OS and NOS modes? (I can't see any switches on the front faceplate of the DAC.) My interest here really stems from the thinking that at >4fs rates, over-sampling shouldn't be needed, so doing away with all filtering altogether (at these rates) may prove a worthwhile improvement in SQ.

 

All in all, a really interesting DAC!

 

Cheers,

Mani.

 

[aljordan]

Hi Mani,

 

You have to open up the cover of the unit in order to switch oversampling rates. On the DSP-1 unit inside the box, there are a series of DIP switches that can be flipped to change the following behavior:

 

- Turn PLL on or off.

- Change the stop band attenuation between -130dB, -90dB or -50dB.

- Switch between oversampling rates of 1x, 2x, 4x or 8x.

- Enable or disable 16 bit to 24 bit dithering algorithm.

- Enable oversampling (flipping this switch makes the DAC an NOS DAC).

- Change the bit rate between 24, 20, or 16 bit.

 

Alan

 

[mikechai]

1) Using the coax S/PDIF output of an M-Audio Delta 410 PCI card. This particular card has the coax output as part of the card and hence does not use any DSUB adapter cables, and is transformer isolated. I've always had better performance feeding DACs using this card over the digital output of my Lynx Studio card.

 

Interesting. Some questions.

Which Lynx card was the M-Audio compared to? In what way that the M-Audio was superior?

 

 

 

 

[aljordan]

Hi,

 

The card is a Lynx Studio 2B. The difference between it and the Delta 410, when using it to feed a DAC, are that the Lynx card adds glare and sharpness in the upper midrange, making digital sound less pleasant. This was apparent on a number of DACs I've tried over the years that have S/PDIF or AES input.

 

The Lynx has a transformer driven output as well, which allows switching between AES and S/PDIF, but if proper impedance is important in the connectors, the d-sub connector / cable system cannot maintain the proper impedance. I don't know if that is the cause of the issue though. Note that the analog outputs of the Lynx card sound quite a bit better than the M-Audio card, but the analog outputs of the Orpheus is far better than both in terms of smoothness and transparency.

 

Alan

 

[barrows]

I am curious, does Audio GD provide digital filter response curves for the different filters? and do they show the impulse response of the DAC with different filters? With all the choices, it would be nice to see exactly what their filters are doing from a technical perspective. Also, what happens if you turn off the PLL and then feed the unit a highly jittered signal (say toslink from a Mac) I would think this could cause loss of lock and data drop outs at some point. I have been to the Audio GD site, but the information on the DSP seems kind of spotty.

 

[aljordan]

I don't think that the DSP-1 unit allows you to choose between the types of filters that effect pre or post ringing, such as the Ayre CD players allow. I think that you could write the manufacturer to find out information about the type of filter used, but from what I understand English is a tough language for him.

 

I have not yet tried bypassing the PLL. I'll do that at some point this weekend and let you know the results. One owner of a different Audio-GD model said he liked the PLL off better, although at the expense of soundstage expansiveness. I've only tried using the DAC in two modes thus far; in the manner it was shipped from the factory and in NOS mode. While I've heard some NOS DACs that I've thought sounded musical, I did not like the Reference 7 in NOS mode.

 

Alan

 

[CKKeung]

The Audio-gd Ref7 mentioned looks good but the coming NFB-1 & NFB-7 are even more attractive because they use the Sabre32 ES9018 DAC chip & support up to 32/192!

The NFB-7 seems to have a similar internal construction as the Ref7.

http://www.audio-gd.com/Pro/dac/NFB7/NFB7EN.htm

http://www.audio-gd.com/Pro/dac/NFB-1/NFB1EN.htm

 

BTW, anyone have tried their Digital Interface USB-to-coaxial converter that has an optional external power-supply?

http://www.audio-gd.com/Pro/dac/USBface/Digital1EN.htm

 

[barrows]

for higher sample rates is nice, but I suspect one the main reasons why people are interested in the current designs from Audio GD is particularly because they use the 1704 DAC chips. These chips are the last of R2R multibit DACs, and there are those who believe R2R DACs offer better sound than current designs (which are almost all Delta Sigma).

Of course the ESS chips are different from both R2R DACs and common Delta Sigma designs, and some think they are better.

It really is difficult to sort through the differences in sound among DAC chip designs though, as it is impossible to compare the sound of different chips in the same implementation, unless you are a manufacturer trying to evaluate different chip designs and can apply them into similar circuit topographies.

It will be interesting to see how the Audio GD ESS DACs turn out...

 

[PeterSt]

Hi Barrows,

 

It is the other way around :

DAC chips don't make sound (quality) as such; it is the implentation (configurations) they do or they do not allow (in more or less easy fahsion for the designer) that creates the sound.

 

Peter

 

[aljordan]

Thanks for the link. I had read about the NFB 7 a little while back. The analog circuitry appears very similar to the Reference 7. Looks like he doesn't need the DSP-1 because he is using the DSP features on-board the Sabre 32. That and the need for less DAC chips looks like he can get it out the door for $400.00 less than the Ref 7.

 

I've now owned a couple of different well designed resister ladder DACs. There is a similarity between them that are different from various delta sigma DACs I've owned. The delta sigma DACs seem to slightly homogenize timing and very low level details, such that in comparison to the ladder DACs, acoustic music has a slightly more sampled sound and a slight loss of richness. In this regard, the Reference 7 is very good in resolving low level details. I'd love to have a chance to hear how a well implemented Sabre 32 DAC performs in this regard.

 

Alan

 

[barrows]

I am not quite in total agreement here, but I understand what you are saying.

I do believe that different DAC conversion topologies sound different (disregarding the effects of implementations, digital filters, etc-just the converter section of the chip)

I suppose you are aware that the current TI BB delta sigma chips: like 1796,1792, 1794 all allow the designer to use only the D/A part of the chip, and bypass the internal oversampling/digital filter(s) as long as the correct data rate is sent to the D/A section (in other words, the oversampling and digital filtering is done off the chip). As an example, Ayre uses this approach in their QB-9 USB DAC and CD players-they use proprietary oversampler/filters running on a FPGA, and then send this data directly to the converter section of the TI DAC chip, bypassing all on chip processing.

I believe this means that you could, for instance, use these delta sigma designs with your DAC design implementation instead of the 1704, please correct me if I am wrong considering your design (does it use a data rate not compatible with the 1796,1792,1794 chips?). I am interested in your work, and specifically your "predictive arc" approach to oversampling: I wonder if sometime you might consider licensing "predictive arc" as an oversampling approach to be used within a DAC, rather than in the computer-I see no reason why this approach could not work, given the processing power of many DSP chips these days.

Certainly not every DAC chip will work in every design implementation, and a given DAC design should be optomised to suit the DAC chip chosen-these are all reasons why I tried to point out that it is hard to evaluate the different chips alone, and decide whether one prefers delta sigma, R2R, or other variations on these approaches.

 

[PeterSt]

Hi Barrows - Thanks for a constructive post ...

 

Of course in all you are saying you are right, but the "problem" is ... which chip has the same spec as the other, while allowing that same configuration behind - or in front of it ? So, you could say that this (different specs) is the exact reason why "chips" sound different, but I don't think it will work out like that.

 

So, suppose two chips will allow for the same process in front of it, and behind it. Those two chips will have a "sound" allright, but only because of their specs. Thus, if one has a better THD figure over the other, I estimate that one will "sound" better. But usually it doesn't work out that easily, and the PCM1704 might be the best example. So, the U-K version of it is nice specs (0.0008% THD+N by heart), but creating an I/V that meats its requirements of seeing no resistance makes it a quite impossible chip to deal with. But, it can be done in "some" optimal way. Now, applying the same I/V to any other chip which isn't so crazy about this requirements would be stupid. Why ? well, because it can be done better with that other chip, and "better" means a 100 more choices to choose from. It's only an example ...

 

I wonder if sometime you might consider licensing "predictive arc" as an oversampling approach to be used within a DAC, rather than in the computer-I see no reason why this approach could not work, given the processing power of many DSP chips these days.

 

Yes, you said this before, and I didn't answer to it;

Of course this can be done, but there is so much more to it that IMO it would be useless. It's the total concept from start to finish which makes it work, or maybe better : which makes it judgeable. So for example, for the DAC subject to this thread it wouldn't work at all, no matter it's also using 8x PCM1704U-K. But indeed, I guess any FPGA will be fast enough to process what is needed here.

At least at this moment I'd say it is also too soon to go that direction, because -although a year old now I think- it is still only a first version and can be improved. And now the conveniency of having it off board comes off handy, because it's so easy to change the software, test a bit (analyser), change again, etc. This will not be the case when everything is on-board, also incorporating the various means of filtering which theoretically (but also practically at this moment) can occur.

 

I think I said it before ... I like modularity, which is a bit different from an engineer's thinking, who generally gets his kicks out of another more small footprint with everything in there (but which sometimes is needed for better performance, I'll admit).

 

Back to the configurations again, apart from the specs which will be more or less different per chip, it really is about the so important possibilities. But, maybe first you have to be as crazy as I am, at not accepting *any* "processing" means which can be avoided. And notice that "processing" for me already can be a resistor;

 

I hope it's not getting too much off topic, but once you're into thinking like that (and everybody will agree that all electronics added will debit negatively to the sound) you can get mad of selecting that chip which will depict your possibilities. By itself for me this was easy, because I just wanted the R2R 24 bit in the first place, so there's just nothing else than the 1704. But just as well it creates all the problems. So, I'll try to give an example, which is just at the far end of the design ...

 

Once the whole concept is running, and, say, "no" resistor is in there to degrade sound, you know the merits of it, and it's ultimate. *Then* people start to ask about a volume control, which btw had been in my mind forever, but completely depends on the design again. For example, if that one resistor would have been in there already, you just as well can make it a variable one, and there's your volume control (assuming a variable VC would be as good as the fixed resistor !!). But if it's not, because you "could do it without resistor" with the particular design, you created yourself the problem of now having to add one.

I can tell you, it took me a whole month to create something which a. can't be measured at all (which doesn't tell all, because we measure at certain frequencies only, while an infinite number of them are there) and b. isn't audible at all (that tells all). The only penalty there is, is that the output impedance rises from 33R to 100R, BUT which is good enough assuming that there's a main amp behind it, and no pre-amp. So you see, here again is the "whole concept" thing, because why to use a VC when there would be a pre-amp containing it. So sometimes it is a matter of "reasoning" too. Or finding the logic.

And all it takes for those who still want to use a pre-amp is creating the additional output without the VC in it, them having the 33R impedance again.

 

To keep it in this context, for some chips of the same brand, the difference could be a digital volume on-chip. So, the whole problem would be solved by that. But will it ? will it be as good as my own in-software digital volume ? of course not. Not when you're me and know what it does and what can be wrong hence will be wrong.

 

And then to think that the whole digital volume thing is 100% related to the filtering, knowing that you can't use both without a penalty somewhere.

 

Ok, I tried to give one very small example in the area people may understand, because it is a thing "we" touch. But there are so many many more deep down. I think I referred to it before, but there's no more "hot" subject than I/V's on the (DIY) internet, and apart of the 100 designs which can exist for one chip, there are so many differently working chips. And mind you, the I/V creates the sound (largely), and at doing it wrong (for the chip concerned) you'll have distortion which may not be audible as such, but which *will* create the sound.

 

I dare say that no matter the one chip may have a 10 times better distortion figure (which is like going from 0.0008% to 0.008%) this doesn't matter at all to sound, because what's happening around it degrades much much more. Of course, this is exactly why you want to hear a chip's real merits, but I claim that the task of creating an apples and apples situation for two chips is an undoable task. Indeed, if you make it all more course and don't take into account all the things you can do to optimze, it is not difficult to have the apples and apples situation, but *then* you will again be listening to that course sound.

For me this is easy : take any OS chip and whatever you'll do further, you'll hear it's OS. This is the other way around, because of the chip being "course". But never mind this as a subject (*again*); I just try to point out the example.

 

All 'n all, if you have two different chips with the same specs on everything for their internal working, all they'll do is converting a digital number to an output voltage, and both will do that the same. Suppose a PCM63 would have the same specs as the PCM1704, next you'll meet the "outside" requirements, and you'll see that taming the PCM63 is like taming a dog, while the 1704 is like taming a shark which is near to impossible (with the objective of maintaining the chip's specs at the end of your interlink).

So, wrong chip and give in to delta-sigma ?

 

What a "hobby" :-)

Peter

 

 

[barrows]

Agreed. Especially in that the I/V/Analog stage of a DAC is very imortant to overall sound (perhaps the most important factor, also, power supplies), and that evaluating different DAC chip topologies is difficult, as implementations must differ to suit the chip.

But I still wonder about the differences amongst DAC chips... ESS' papers about the 9018 are quite interesting in pointing out the level of inter-transient noise present with typical delta sigma chips, which they claim are many times lower with the ESS 9018, and I suspect, also lower with R2R chips. Also, I often wonder how the very high frequency noise level of delta sigma interacts with the following analog stages effects the overall sound.

RE your Predictive Arc developments, I totally understand why you want to keep them in software, but still suggest that you consider, some time in the future, the possibility of licensing to DAC manufacturers. I do not know if you care, or are interested in the possibility of making some money for yourself, but if your developments are even close to as revolutionary as you seem to believe them to be, they could be shared with a far wider range of audiophiles through distribution in hardware (DACs). I also wonder about the true viability of very high sample rates via SPDIF/AES-it would seem to me that sending 384 kHz (and higher) over SPDIF is putting a big degree of faith in the SPDIF transmitter and receiver circuits (many of which seem to struggle with much lower data rates).

In any case, thanks for the discourse here, it is always interesting to your hear about your work.

 

[aljordan]

The other day I brought the Audio-GD Reference 7 over to a buddy's house who has been listening to the Resolution Audio Cantata for the past month or so. I left the Ref 7 there for him to play with for a bit. The DACs are similar to the degree that they both use the same PCM1704 ladder DACs, but the Reference 7 doubles the quantity in a fully balanced circuit path. He used the asynchronous USB input on the Cantata and fed the Reference 7 with an S/PDIF signal from the Prism Orpheus.

 

After spending a night with the Ref 7, my buddy ordered one. Both DACs sound very good and have a similar flavor of sound, but the Reference 7 further resolves low level details. The differences were quite apparent with acoustic bass material, but the Reference 7 also presented layers of sound-stage depth compared to a flatter, more forward presentation of the Cantata.

 

Alan