[bhobba]

What I am going to mention is way out of left field so I would not be surprised if no one is really interested.  It is simply something that has piqued my interest enough to mention it.

A Swiss company, Anagram Technologies  http://www.anagramtech.com/ has created a software based up sampler running on a shark processor that is supposed to reduce jitter to unbelievably low levels. The detail are discussed here -
http://gearslutz.com/board/showthread.php3?s=&threadid=1531
It has produced a DAC using this processor that it sells to OEM manufactures.  

Players that use it such as the Audio Aero Capitole (and costing $8,500 it better be good) have received rave reviews comparing its sound to the best vinyl.  My curiosity piqued I looked around to see if any kits were available offering this in say a combination DAC preamp. It seems a number of people are interested but Anagram will not sell evaluation boards to individuals - only to companies.

To me this seems such a shame for what looks like a ground breaking product.  I wonder if there would be any support to have this board included as an option in the GK-1?  I suspect the resulting product would sell for much less than 8,500 and probably be better.

As I said it is just an idea I wanted to get out there.

Thanks
Bill

[dawkimi]

Bhobba,

This DAC technology has also interested me.  I believe one of the most cost effective ways to sample this technology would be to purchase a Camelot Uther DAC.  I believe they make two versions using the Anagram DAC chipsets (Mk IV and Mk V).  I've seen an upgraded Mk IV listed for around $2,000 on Audiogon.  Good luck.

Mike

[bhobba]

Hi Dawkimi

Thanks for the reply

Yep there are undoubtedly a number of ways you can get it.  The point is to find a reasonable way to get it without paying too much.  Here in Australia that dac your talking about would equate to $3000 Australian plus taxes etc.  Even then a quick internet search reveled to go down that path would require purchasing an old unit and having it upgraded.  I think new ones with it already installed cost about $4000 US or about $6000 Australian.  We are now talking real serious money - to the point where you may as well go and buy an Audio Aero for $8500 Australian and be done with it. I suspect an OEM like Aspen could incorporate it into their product like the GK-1 for a MUCH more realistic price - the product is supplied as a stand alone DAC that would seem drop dead easy to integrate.  From what I have read it is a drop dead product whose only competitors are the DCS DACS and up samplers at 2 to 3 times the price again.

As I said it was just an idea that intrigued me enough to post.

[Tinker]

What I am going to mention is way out of left field so I would not be surprised if no one is really interested.  It is simply something that has piqued my interest enough to mention it.

To me this seems such a shame for what looks like a ground breaking product. I wonder if there would be any support to have this board included as an option in the GK-1? I suspect the resulting product would sell for much less than 8,500 and probably be better.

I looked into this technology a while back and I was certanly very interested too.

Hugh has asked me to make a response to this. So here it is.

This is old news to most of you, but some have asked me off list for a technical discussion, so here it is.

With a typical DAC interface using SPDIF or AES/EBU receives a signal (audio data) and an IMPLIED clock. Various things (long list) make this data quite accurate but the implied clock quite messy - jitter. So the idea is to introduce a new clock which is much cleaner.

The traditional approach to use a PLL, or Phase-Locked Loop. This a like a new clock which locks on to the incoming clock matching it's speed, but ignoring short term irregularities (ie jitter) producing a smoother less jittered output. However, there is a limit to this "smoothing" before synchronisation with the incoming clock is lost, and that means lost bits, which means audio glitches. Some very complicated math and a little soldering later we discover that in practical terms we can get only a few dB jitter reduction at the low end of the audio band, maybe tens of dB by 10kHz and up to 100dB out at the MHz range. Physics usually steps in, though, particularly thermal noise and grounding issues make such juicy figures hard to get.

Really what you want to do is break the dependence between the input and output clocks. One way to do this is asynchronous resampling. In this scheme you are essentially putting a waveform (some music) in one end of a device and pulling it out the other AT A DIFFERENT SAMPLING RATE. Such devices were initially built to solve equipment interfacing problems, eg playing a CD (Fs=44,100Hz) into a digital broadcast system (Fs=32,000), or matching multiple digital devices supposedly running at the same speed but not quite precisely matching owing to manufacturing tolerance etc. With ASRC DACs what we are doing is resampling a dirty input to a clean output which differs by only a fraction of a percent.


What information is available about the ANAGRAM system tells us that it is a very clever variation on the asynchronous sample rate conversion (ASRC) concept. In ASRC schemes a series of digital filters are used to construct a new output waveform which is synchronised to a very clean clock in the DAC. This breaks the association between input and output timing systems and effectively gets rid of most of the dominant jitter modes, although it must be emphasised that this can only ever be as good as the new master clock. The new master clock is in fact a theoretical limit reached should everything else be perfect. During the development of the DAKSA we considered the AD1896 and the AD SHARC 21061 system. In fact the SHARC is still being considered for other projects. The 1896 is the latest in a line of high quality standalone ASRCs by AD which uses a 64 tap FIR polyphase filter to do the business and was a strong contender for the DAKSA (about the third thing we tried). Of course a DSP like the 21061 virtually and number of taps can be used, what is more 32-bit floating point math can be used on this system. This means much more ideal (and nicer sounding) filters can be realised and other adaptive functions can be implemented using the DSP's computing power. These "other functions" are part of the IP of ANAGRAM and adaptive signal processing is a whole other discussion.

At this point I should also state my usual legal disclaimer, I make no claims about the quality of other manufacturer's products. I mention names only as exemplars of particular approaches that interested readers can go and look up.


I am not ruling out using these chips or similar in future products, however, there are at least two reasons Aspen did not pursue an ASRC approach. Firstly, state of the art devices like the 1896 or a 21061-based system would have ruled out a kit. The core of these issues were simply assembly problems, and although Aspen could turn these out, fully-built retail units are for a later stage in the AKSA business plan. Although it is well known that Hugh does custom builds on a commission basis, every AKSA product must be available in kit form. AKSA is lucky to have so many talented builders as clients and friends, but despite this, in principle all AKSA kits must be buildable by the average hobbyist with no special equipment. A second factor is cost. The idea was to be able to offer a complete DAC for under US$1000, well under $1000 if possible (and so far it's looking good!). A DSP-based system would blow the AKSA budget and then some. It wouldn't cost $8k, but it would cost a good fraction of that.

Those who saw my talk in Sydney will also know I have some personal objections to the limtations of CERTAIN (but not all) ASRC implementations, and for this reason we have pursued a different approach in the last few months to see if we can do better. The iteration of the DAKSA currently under test works on a very different concept. There is no resampling. The bits from the CD are passed straight through from the input chip to the anti-aliasing and DAC sections. As with the ASRC approach there is a new very clean clock which drives the output, but instead of interpolating the input signal the bits are buffered in a memory unit and the speed of the clock is very subtly altered based on an analysis of the incoming signal. This uses a microprocessor with custom software. This is *like* a PLL in that we are tracking the incoming clock, however, because the memory element means we don't have to react to clock irregularities quickly (ie bits will not be lost) we can attenuate jitter enourmously right down to the bottom of the audio band and come close to the theoretical limits of a crystal oscillator.


There are many other elements to a good DAC as well, and these are at least as important as the problem of jitter and in many ways more fundamental, for instance the actual DAC itself. The DAKSA has a carefully chosen DAC chip with excellent linearity and low-level resolution specs, a topology that puts the output clock within about an inch of the DAC, plus huge work on the analogue parts of the circuit - particularly power supply - which reek of the Hugh R Dean AKSA style which has made the 55 and 100 amps so musical and such a huge bang for the buck. In true AKSA style the design is robust and uses components which are of high quality, but easily available.


This was meant to be a short post! But, we are really beginning to get into some very detailed areas of design, engineering and philosophy. I hope this sparks some further discussion. I am always pleased to see this kind of independent research and market savvy, and this is exactly why AKSA prizes it's customers so much. I am more than happy to discuss these issues and incorporate suggestions into product development if they are compatible with the AKSA philosophy.

T.

[bhobba]

First thanks for the very detailed and informative post.

I have only recently started posting so did not know that AKSA had already started work on its own DAC.  I fully agree with the following:

'A second factor is cost. The idea was to be able to offer a complete DAC for under US$1000, well under $1000 if possible (and so far it's looking good!). A DSP-based system would blow the AKSA budget and then some. It wouldn't cost $8k, but it would cost a good fraction of that.'

Yes - It is doubtful a DIY product costing more than $1000 would be a commercial success.  I estimate use of the Anagram DAC would make the cost, between 2 to 3K (based on the fact the module looks like it sells to OEM manufactures for about $1000 US or $1500 Australian).  We certainly have no definite evidence it so much better than other approaches to justify that kind of price difference.

Having said that I do not think construction by the average hobbyist would be a problem - the DAC looks as though it comes prebuilt by Anagram so it would be simply a matter of adding a power supply case etc.  The issue would seem purely one of cost.

However there are a couple of things I would like to mention.

Tinker:
'This breaks the association between input and output timing systems and effectively gets rid of most of the dominant jitter modes, although it must be emphasized that this can only ever be as good as the new master clock.'

Information from the Manley site indicates that the upsampler produces jitter free output by taking into account jitter at the source, in the clock of the processor itself, and jitter of the clock at the DAC (if one is used).  They claimed to have tortured the clocks in all sorts of ways and was unable to produce any jitter related effects.  Indeed for the SLAM product Manley claim using more stable jitter free 'super clocks' are simply not necessary and probably a waste of time.  Thus the technology would seem one of algorithmically reducing the effects of jitter so that clock accuracy is no longer an issue.  

Of course that begs the question of if such is really necessary and could not be achieved by cheaper methods.  Your method for example looks fine.

Tinker:
The DAKSA has a carefully chosen DAC chip with excellent linearity and low-level resolution specs, a topology that puts the output clock within about an inch of the DAC, plus huge work on the analogue parts of the circuit - particularly power supply - which reek of the Hugh R Dean AKSA style which has made the 55 and 100 amps so musical and such a huge bang for the buck.

One thing that I have always found a bit strange about dacs is the use of a final output stage at all.  I have always thought paralleling a few dacs on the output and simply sinking the current output through a resistor would be better than having an output stage and the associated problems.  I seem to recall reading somewhere that doing so also increases linearity.  My background is in IT and I have always found the Keep It Simple Stupid (KISS) principle to be the best.  Any large out of band artifacts could be ripped out by say a simple passive second or third order filter at say 200 - 300 khz.  Since the frequency 'image' as a result of the DAC conversion is mirrored by the conversion process (ie 21kz produces 21kz and 23kz etc) I have always suspected the conversion artifacts would not cause audible problems.  I seem to recall Pioneer did something similar to this and found not only did it not cause audible problems people preferred it.  I don't know if you use this approach or not - it is just an idea.

Tinker:
Those who saw my talk in Sydney will also know I have some personal objections to the limitations of CERTAIN (but not all) ASRC implementations, and for this reason we have pursued a different approach in the last few months to see if we can do better.

As you probably have guessed I like thinking about technical issues.  Would it be possible to get a hold a transcript of something from that talk?

What would be very interesting indeed would be to compare the sound of the AKSA DAC with others such as the Anagram.

Again thanks for the detailed reply
Bill

[bhobba]

First thanks for the very detailed and informative post.

I have only recently started posting so did not know that AKSA had already started work on its own DAC.  I fully agree with the following:

'A second factor is cost. The idea was to be able to offer a complete DAC for under US$1000, well under $1000 if possible (and so far it's looking good!). A DSP-based system would blow the AKSA budget and then some. It wouldn't cost $8k, but it would cost a good fraction of that.'

Yes - It is doubtful a DIY product costing more than $1000 would be a commercial success.  I estimate use of the Anagram DAC would make the cost, between 2 to 3K (based on the fact the module looks like it sells to OEM manufactures for about $1000 US or $1500 Australian).  We certainly have no definite evidence it so much better than other approaches to justify that kind of price difference.

Having said that I do not think construction by the average hobbyist would be a problem - the DAC looks as though it comes prebuilt by Anagram so it would be simply a matter of adding a power supply case etc.  The issue would seem purely one of cost.

However there are a couple of things I would like to mention.

Tinker:
'This breaks the association between input and output timing systems and effectively gets rid of most of the dominant jitter modes, although it must be emphasized that this can only ever be as good as the new master clock.'

Information from the Manley site indicates that the upsampler produces jitter free output by taking into account jitter at the source, in the clock of the processor itself, and jitter of the clock at the DAC (if one is used).  They claimed to have tortured the clocks in all sorts of ways and was unable to produce any jitter related effects.  Indeed for the SLAM product Manley claim using more stable jitter free 'super clocks' are simply not necessary and probably a waste of time.  Thus the technology would seem one of algorithmically reducing the effects of jitter so that clock accuracy is no longer an issue.  

Of course that begs the question of if such is really necessary and could not be achieved by cheaper methods.  Your method for example looks fine.

Tinker:
The DAKSA has a carefully chosen DAC chip with excellent linearity and low-level resolution specs, a topology that puts the output clock within about an inch of the DAC, plus huge work on the analogue parts of the circuit - particularly power supply - which reek of the Hugh R Dean AKSA style which has made the 55 and 100 amps so musical and such a huge bang for the buck.

One thing that I have always found a bit strange about dacs is the use of a final output stage at all.  I have always thought paralleling a few dacs on the output and simply sinking the current output through a resistor would be better than having an output stage and the associated problems.  I seem to recall reading somewhere that doing so also increases linearity.  My background is in IT and I have always found the Keep It Simple Stupid (KISS) principle to be the best.  Any large out of band artifacts could be ripped out by say a simple passive second or third order filter at say 200 - 300 khz.  Since the frequency 'image' as a result of the DAC conversion is mirrored by the conversion process (ie 21kz produces 21kz and 23kz etc) I have always suspected the conversion artifacts would not cause audible problems.  I seem to recall Pioneer did something similar to this and found not only did it not cause audible problems people preferred it.  I don't know if you use this approach or not - it is just an idea.

Tinker:
Those who saw my talk in Sydney will also know I have some personal objections to the limitations of CERTAIN (but not all) ASRC implementations, and for this reason we have pursued a different approach in the last few months to see if we can do better.

As you probably have guessed I like thinking about technical issues.  Would it be possible to get a hold a transcript of something from that talk?

What would be very interesting indeed would be to compare the sound of the AKSA DAC with others such as the Anagram.

Again thanks for the detailed reply
Bill

[bhobba]

Sorry guys - something went wrong and it got posted twice.

Sorry
Bill

[bhobba]

Just to let people know I have looked into stuff already posted about the DAKSA that has answered some of the questions I had.

I now understand the idea of no oversampling has already been investigated and found to have sonic problems.  Got it.

I also see that considerable effort has gone into designing the best output stage possible with a very low output impedance, much lower than could be achieved by my idea of paralleling dacs then simply sinking the current into a resistor - 4 Dacs sunk into say a 500 ohm resistor and then fed into a passive anti-alaising filter would have a much higher output impedance.  The advantage of no output stage must be balanced against this and the damage the stage will do to the signal.  I can see where it would favor a well designed output stage.

I certainly see the logic behind the choice of filter at 350kz.

The only question that leaves I suppose is why a valve output stage was chosen.  I have nothing against valves at all and understand they add that indefinable 'something'.  However I have also read where their use tends to 'add up' so that if you were to use the DAKSA with say the GK-1 with it valve output then the sound may have a bit too much of a good thing.  Just a thought to find what others think.

Bill

[AKSA]

Bill,

You got it!  There is no tube at the output of the DAKSA;  it is envisaged that this component would be used with the GK1, and too much of a good thing is indeed an issue we've investigated.

And, of course, only a CF will give us the Zout we need,  but certainly not down to 3R, the Zout of the output stage we are using.  The idea is to minimize the sonic influence of any interconnect used.......

Oversampling permits use of very high frequency filters, with relatively gentle, non-damaging slopes.  The Sallen Key filter on the DAKSA was designed by Fred Dieckmann, a hugely talented Texan engineer who frequently posts to DIYaudio.  He also is behind the folded cascode of the I/V conversion, though there's a lot of my input there as well.  Both Ben and I did the pcb layout, which is approaching epic, Rembrandtian proportions even as we speak.......

I believe Ben has covered most of the salient points of digital design;  our approach is a further refinement (using unique software written in assembler for speed) of the Levinson Type 36 DAC of a few years back.  This was a very, very well reviewed FIFO DAC.

I believe that our combination of software, FIFO, I-V conversion and active SE output stage with Sallen Key filtering will be one of the best in the world, and probably about the best kitset DAC in the market.  That's the goal, anyway, and right now the hardware is much cheaper than it was five years ago and it can actually be realized as a DIY, kitchen table project.  I'm pretty enthused about this, though I admit the time and money expended has become something of a trial!

Since there are such huge legacy collections of Red Book CD out there, and since the medium has been demonstrated again and again to offer audiophile qualities regardless of the new high speed, mega-sampling formats now available, and since our target is guys of technical savvy and usually 30 years and older, it seems like a sensible commercial approach to offer this at this time.  This may not be true in a couple of years, of course, but right now it makes good sense, particularly as Sonic Frontiers and AudioNote (AFAIK) have withdrawn their DACs from the market.

Thank you for your interest;  I hope you find this as fascinating as we do!

Cheers,

Hugh

[bhobba]

Hi Hugh

Thanks for your excellent post.  

This originally started simply as a query about the use the Agaram Dacs but evolved into a discussion on the design of the DAKSA.

After thinking about the design I see and understand the issues you grappled with.  I think the solutions are well thought out and innovative.

With regard to a DAC that just works with standard Redbook CD's and not the newer formats like SACD and DVD audio I believe that is the best choice.  Review after review shows that the very best implementations of standard CD does not suffer sonically in comparison to the new formats and often is judged superior.  Blind tests with the very best equipment simply does not prove the new formats are sonically superior.  The very best standard CD implementations are approaching the holey grail of vinyl.

Combine this with the fact that the CD player as a source component is on the way out - it will almost certainly be replaced by networked music/ video servers in the future - then I think it is a wise design choice.  The newer formats have got themselves in bind here - the distribution of audio looks set for a shakeup.  It will be replaced by cheap internet download - file sharing technology makes it inevitable.  Sony etc will fight to the death to stop it - but they will loose.  The internet is a force that cannot be stopped.  No standards exist for distributing SACD or DVD audio electronically.  With their being no discernable improvement on the very best equipment then I believe they may slowly die.

It will be very interesting to compare the sonics with other products such as the Audio Aero Capitole.

Bill

[Tinker]

Hi Bhobba,
   just thought I'd drop this one in quickly. I have a reply for you with regard to some of your earlier questions, however, I have been too busy to write it up. I will address some of these issues and continue our discussion later today!

Cheers,
   T.

[Tinker]

Information from the Manley site indicates that the upsampler produces jitter free output by taking into account jitter at the source, in the clock of the processor itself, and jitter of the clock at the DAC (if one is used). They claimed to have tortured the clocks in all sorts of ways and was unable to produce any jitter related effects. Indeed for the SLAM product Manley claim using more stable jitter free 'super clocks' are simply not necessary and probably a waste of time. Thus the technology would seem one of algorithmically reducing the effects of jitter so that clock accuracy is no longer an issue.

I have only read Anagram's old whitepapers, I will have to browse thorugh the Manly site to respond to this. What I have rad so far does not make their exact modifications clear to me beyond the ASRC interface.
The main Anagram innovation is "integral time estimation". A far as one can tell this adds a time dimension to the filtering routine (now the interpolation is a 2-dimensional adaptive process) which increases teh fidelity of the resampled signal.Anagram themselves still recommend a high accuracy clock placed as close as possible to the output.

One thing that I have always found a bit strange about dacs is the use of a final output stage at all. I have always thought paralleling a few dacs on the output and simply sinking the current output through a resistor would be better than having an output stage and the associated problems.

There are a number of engineering considerations. Current drive dacs (which usually have the best linearity) can be paralleled up, but then any mismatch between them at all results in glitches. If you consider that the the typical output is aobut +/-1.5mA for a full range 16-bit signal the mismatch doesn't have ot be much for this to be nasty. The old resistor to ground trick is OK, but does have problems. DACs generally have inbuilt protection diodes which limit the voltage swing (in one direction) to about 0.6 of a volt. This jsut ain't enough and even this maximum doesn't really allow useful choices of resistors. Obviously paralleling outputs doesn't get around this.

If you can find that reference to parallel up DACs I would like to see it. I do not mean this in a nasty way, I would genuinely like to learn about how this is done. Usually such practices decrease linearity and create little spikes in the ouptut. However, multistage DACs are a differnet animal altogther.

Any large out of band artifacts could be ripped out by say a simple passive second or third order filter at say 200 - 300 khz. Since the frequency 'image' as a result of the DAC conversion is mirrored by the conversion process (ie 21kz produces 21kz and 23kz etc)

I see you have made a new note on this, buit I'll post a bit anyway. Nearly all D/A systems have a anti-iamging filter. I don't have a sonic preference for passive or active, except that passive filters generally present impedence problems: either they are jsut too high to drive  a preamp or involve inductive elements which get REAL nasty at high frequencies presenting impedence problesm to the DAC.
The audibility of out-of-band noise is a subject of debate. Humans can't hear it in pure tone tests, however, out of band noise has been shown to cause intermodulation noise in many SS amps, which is one reason a lot of amp have input filters (yuk!)
If you read about a DAC without a filter, it's probably a 'bitstream' (TM?) or DSD/SACD which outputs it's pulse-train through an integrator. Which in truth is a filter. These kind of filter (sampling rates in teh MEGAHERTZ) can contrive to be first order, critically damped and hence don't ring. However, some single-bit techologies play with other tradeoffs, one of which is a shaped noisefloor and problems with idle tones caused by feedback.

I could go on and on. And will if encouraged.    No DAC to date is perfect. Dare I also suggest that the DAKSA will not perfect, it will IMHO be very, very good. I certainly intend it to be the last CD DAC I ever have to buy. The DIY format allows us to put in features typical of far more epensive DACs, and as well as incorporating some novel circuits into the design we have chosen the engineering tradeoffs with scrupulous attention to sound quality based on our listening experiences with other DACs.

Clearly you are one smart guy and have read a lot about the subject. I look forward to disucssing these issues further and learning from your own travels.

Cheers,
     T.

[bhobba]

Tinker:
'I have only read Anagram's old whitepapers, I will have to browse thorugh the Manly site to respond to this. What I have rad so far does not make their exact modifications clear to me beyond the ASRC interface. The main Anagram innovation is "integral time estimation". A far as one can tell this adds a time dimension to the filtering routine (now the interpolation is a 2-dimensional adaptive process) which increases teh fidelity of the resampled signal. Anagram themselves still recommend a high accuracy clock placed as close as possible to the output.'

I would say a direct reading of what Anagram has to say is of greater relevance than my reading between the lines of a comment made by a company that uses it.  

Tinker:
'There are a number of engineering considerations. Current drive dacs (which usually have the best linearity) can be paralleled up, but then any mismatch between them at all results in glitches. If you consider that the the typical output is aobut +/-1.5mA for a full range 16-bit signal the mismatch doesn't have ot be much for this to be nasty. The old resistor to ground trick is OK, but does have problems. DACs generally have inbuilt protection diodes which limit the voltage swing (in one direction) to about 0.6 of a volt. This jsut ain't enough and even this maximum doesn't really allow useful choices of resistors. Obviously paralleling outputs doesn't get around this.'

As you probably noticed in a later posting I came around to the idea, providing one has an output stage that does negligible damage, then this is the preferred approach.

Tinker:
If you can find that reference to parallel up DACs I would like to see it. I do not mean this in a nasty way, I would genuinely like to learn about how this is done. Usually such practices decrease linearity and create little spikes in the ouptut. However, multistage DACs are a differnet animal altogther.

I came across this from a number of sources eg page 5 of
http://www.mlssa.com/pdf/Upsampling-theory-rev-2.pdf

'One way to reduce differential non-linearity is to average the outputs of two or more DACs connected in parallel. Indeed, several CD players use this technique to improve sound quality. The idea is that the differential non-linearity errors in each individual DAC will tend to be random and thus tend to average out when several DACs are connected in parallel.'

The circuit where I have seen it done is:
http://www.tnt-audio.com/clinica/convertus3_e.html

As you can see all the designer of this DAC did was parallel the outputs of 4 dacs to a resistor paralleled with a capacitor to provide some high frequency roll off then a second order passive filter for further ripping out of frequencies above about 200khz.  I have since changed my mind about the approach believing it is now possible to design a proper output stage having negligible sonic degradation and allowing a much lower output impedance.  

Tinker:
Nearly all D/A systems have a anti-iamging filter. I don't have a sonic preference for passive or active, except that passive filters generally present impedence problems: either they are jsut too high to drive a preamp or involve inductive elements which get REAL nasty at high frequencies presenting impedence problesm to the DAC. The audibility of out-of-band noise is a subject of debate. Humans can't hear it in pure tone tests, however, out of band noise has been shown to cause intermodulation noise in many SS amps, which is one reason a lot of amp have input filters (yuk!) If you read about a DAC without a filter, it's probably a 'bitstream' (TM?) or DSD/SACD which outputs it's pulse-train through an integrator. Which in truth is a filter. These kind of filter (sampling rates in teh MEGAHERTZ) can contrive to be first order, critically damped and hence don't ring. However, some single-bit techologies play with other tradeoffs, one of which is a shaped noisefloor and problems with idle tones caused by feedback.

The DAC I linked to above does not use it and is part of its design.  I personally think it must cause problems not the least of which is the fact the amp doing more work than it really should have to leading to unnecessary intermodulation distortion.  What I was referring to is that some people seem to think it sounds better with this noise.  It is this age old question of well it may not be a reflection of the original but I think it is better based it sound to my ears better - not always the best augment IMHO - but not something you can dismiss out of hand either.  The link I gave regarding upsampling conjectures the presence of uncorrelated high frequency information is the reason upsampling sounds better.

Tinker:
I could go on and on. And will if encouraged.  No DAC to date is perfect. Dare I also suggest that the DAKSA will not perfect, it will IMHO be very, very good.

Of course not.  All that can reasonably be asked is that it is based on well chosen design principles - which your design meets very well.  

Tinker:
I certainly intend it to be the last CD DAC I ever have to buy. The DIY format allows us to put in features typical of far more epensive DACs, and as well as incorporating some novel circuits into the design we have chosen the engineering tradeoffs with scrupulous attention to sound quality based on our listening experiences with other DACs.

I think that beyond having chosen a reasonable principles to base the design on you can really do no more.  The final answer will come in the listening and how it compares to other top of the line implementations such as the anagram.  My gut feeling is you have chosen well the principle the DAC will be based on and that any sonic improvement other dacs may have will be marginal.  But only time will tell.

Tinker:
Clearly you are one smart guy and have read a lot about the subject.

Blush Blush.  I just enjoy thinking about technical issues and this piqued my curiosity.  As an aside I often post to sci.physics.realtivty and the brain power of some guys there bring you down to earth pretty fast.

Tinker:
I look forward to disucssing these issues further and learning from your own travels.

OK the DAC I currently have is the well reviewed MSB link with upsampling.  Before that is was the Cambridge Audio dac.  To my ears (and my friends as well) the MSB DAC was more detailed but much more bass heavy - to the point where you think it must have some artificial bass added.  I had a particularly bass heavy system with an Evocotor subwoofer from Axis LS 88's so it was particularly revealing in this department.  It would not have surprised me if MSB did dome tinkering there - but it is also possible it was an artifact of the upsampling.  At present I do not have the space of such a big system and am looking to one more suited to my current abode.

I found the Cambridge audio quite good.  Not as detailed as the MSB but better balanced.

From my perspective what I would like to see is exactly how it sounds compared to other DAC's such as the MSB, Orpheus (that uses the Anagram upsampler) etc.  The design principles of the DAKSA are well enough chosen that all you can do is build a prototype and compare.

Greatly appreciative of the detailed response
Bill

[EchiDna]

Bill, Just a minor request...

please use the quote function... your posts are a bit tough to read as to who is writing what!


Interesting discusion

[Tinker]

One way to reduce differential non-linearity is to average the outputs of two or more DACs connected in parallel. Indeed, several CD players use this technique to improve sound quality. The idea is that the differential non-linearity errors in each individual DAC will tend to be random and thus tend to average out when several DACs are connected in parallel.

OK. Now I understand what you mean. I thought you were reffering to a way of getting mroe drive. Paralleling device for drive is often done with logic circuits when no other optionis available.

Incidentally, there are a number of DAC chips which have an internal mechanism for achieving this, for example the state of the art AD (designed by Burr-Brown, but bought out by AD!) PCM1702/1074 which internally has two interleaved DACs to minimise differential non-linearity at MSB changes. Preety cool.

The paper you linked seems to cover many of the issues I have considered extremely important, and I commend this paper to anyone else out there in AKSAland who wants a succint overview of upsampling. I am not sure about the concept of "time smearing." Yes CD has limited tmeproal resolution, but still the signal is preserved when band-limited. Can of worms ahead there... Anyway, for my money you can't do without an anti-aliasing filter and filters active or passive are going to have a sonic impact. Oversampling is a great way of dealing with filter ringing and phase shifting, moving the worst muck outside the audioband. Recent debates in pro-audio have actually suggested that carefully filtered 44.1kHz audio sounds every bit as good as 96kHz. The way in which this was done suggests very strongly that ringing in anti-imaging filters is the culprit.

I was interested to hear your comparison of other DACs and would like to go into it further later.


Cheers,
    T.

[bhobba]

Hu EchiDna

Bill, Just a minor request...

please use the quote function... your posts are a bit tough to read as to who is writing what!

Just testing the function.  

Thanks for the tip
Bill

[bhobba]

Hi Tinker

Incidentally, there are a number of DAC chips which have an internal mechanism for achieving this, for example the state of the art AD (designed by Burr-Brown, but bought out by AD!) PCM1702/1074 which internally has two interleaved DACs to minimise differential non-linearity at MSB changes. Preety cool.

Sure is - and probably one reason why you chose the PCM1702 for the DAKSA.  Just out of interest why did you prefer the units 20 bit accuracy over the 1074's 24 bit accuracy?

The paper you linked seems to cover many of the issues I have considered extremely important, and I commend this paper to anyone else out there in AKSAland who wants a succint overview of upsampling. I am not sure about the concept of "time smearing."

I think it is a good paper too.

Oversampling is a great way of dealing with filter ringing and phase shifting, moving the worst muck outside the audioband. Recent debates in pro-audio have actually suggested that carefully filtered 44.1kHz audio sounds every bit as good as 96kHz. The way in which this was done suggests very strongly that ringing in anti-imaging filters is the culprit.

Would it be possible to get a rundown of the oversampling digital filter the DAKSA will use?  Hugh mentioned the analogue stage had a filter at 350kz.  This would indicate 16x oversampling.  Is that correct?  Do you envisage your approach to improve on the ringing performance  

Thanks
Bill

[Raj]

Hi Hugh and Ben,

Is there any chance whatsoever that after the dac is released, a tube output stage will be developed? Maybe not as an upgrade but an option for those people who prefer all tube gear, or don't mind too much of a good thing?

Thanks
Raja

[Tinker]

Hi Tinker

Sure is - and probably one reason why you chose the PCM1702 for the DAKSA. Just out of interest why did you prefer the units 20 bit accuracy over the 1074's 24 bit accuracy?

Call me a nitpicker (you're a nitpicker Ben) but I think "resolution" is a better description than "accuracy."  Here's my take on this: A CD has 16 bits, and in that context accuracy refers to making sure "zero" is reliably close to zero volts output and that the step size between each bit is as even as possible.

Well, insofar as this chip choice is final (and I'm pretty sure it is), there are a number of reasons for using the 1702 above the 1704.
First let's see what we DON'T need to worry about.
1. Both chips are capable of the same maximum (oversampling) sample rate.
2. Bipolar zero drift and gain drift are identical.
3. Low level linearity is the same.
4. Typical dynamic range for both is same, quoted at 110dB, as is the idle SNR at -120dB
5. THD+N, well this is given differently for each chip
1702 is given as -92dB.
The 1704 looks far more impressive at 0.0025%, but then 20*log(0.0025/100) = -92.04 so once again they are the same. We are going to use the higer grade chip (J at -96dB) but the figures are the same here too.
6. Settling time identical, ie same speed.


Given the above the 1702 has a number of obvious strengths
1. It's through hole. The 1704 is SMT which means it can't be easily installed in a kit.
2. It's cheaper, meaning same performance to the builder at lower cost.
3. It has a better typical low level gain error (by a factor of two), although this is mitigated by what I discuss below.
4. Clocking in a 24bit signal out for each 16 bits in means running the DAC clock 20% faster than required for a 20bit DAC. Higher clock rates mean MORE JITTER, so we can squeeze a little more performance out here as well.

Since we are only going to put a 16-bit signal through it, so the question arises, if they are the same in all other respects why do we need the "resolution?"

I can, and will go on a little conceptual digression here, and welcome any comments on this.
If we put a 16bit signal into a 20 bit DAC that means that 24dB of the headroom and output swing is wasted, lowering the SNR. Double that for 24 bits. So to maintain low noise 16 -> 20 involves a gain of 24dB and 16 - > 24 involves a gain of 48dB to use to full headroom of the chip. These 6dB multiple gain jumps can be done digitally and losslessly because they involve only powers of 2 in the gain function. Incidentally a side bonus of using at least 6dB digital gain in a DAC with more bits than we actually need means we never actually use the lowest bit of the DAC meaning we avoid the worst of any non-linearities present which are usually at the extremes of the range.

Now the current output range of the two chips is identical for both BUT THE 1702 SWINGS THE SAME RANGE IN FEWER STEPS (don't worry about this being fewer, there are already nearly a million more than we need for CD!). So all else being equal the apparently "coarser" use of 20 bits means each bit in the ORIGINAL SIGNAL is a larger proportion of the output range. So a larger movement minimises the contribution of any small deviation from perfection (as listed in the figures above). This means better linearity, gain error and dynamic response. Of course this can be overcome using some DSP techiniques, for instance more gain in the reconstruction filter, but why add the expense, especially when it limits the choice of filter chips?


The 11 points listed were pretty much what drove the choice with points 1 and 4 of the second list being the big clinchers.

There are VERY GOOD reasons for using more than 16 bits in a CD player DAC, provided you don't use too many. But that's another story I might tell later...

T.

[AKSA]

Ben,

Thank you for the heads up on the DAC choice.  Seminal analysis, nicely written.  

Raj,

Good question.  Let me offer a few points Ben and I mulled over on this very topic.

1.  A tube output stage will have much higher noise, possibly by as much as 15-20dB.  This is due to the intrinsic noise issues of tubes.
2.  A tube output stage will have a high source impedance, typically 120R for a 10mA cathode follower (with unity gain).  This is typically two orders of magnitude higher than the Zout of a well designed emitter follower, which is around 2R.  High source impedance confers far greater sonic influence from the interconnects, and this is a bad thing.
3.  Aspen already offers a tube hybrid preamplifier which is specifically voiced for CD and AKSA power amplifiers.  Adding another tube stage is overkill and will likely be too much of a good thing.  Remember, tubes are coloured, their contribution is euphonious but somewhat distortive, and with all the design effort going into high resolution, a further unity gain tube stage would partly negate our DAC and analog processing design efforts.
4.  A tube requires a filament supply and a carefully filtered high tension supply.  This considerably adds to R&D costs, production cost and complexity, and threatens budget, increasing kit price beyond what the customer is prepared to pay.  Don't laugh!  this is VERY important!
5.  A tube output stage is covering ground occupied already by the GK1.  This is not a commercially sensible approach;  the last thing we should be doing is obviating the GK1, which was specifically designed and voiced for a SS analog output stage on a DAC.

I hope this answers your question!  None of these decisions is taken lightly;  all have technical, commercial and marketing repercussions and must be thought through.  The idea is to deliver what people want, to keep it inexpensive, and to offer more value sonically, technically and in terms of component quality than people expect.  Any other approach is doomed from the outset, and a healthy dose of paranoia helps in making these decisions.

Cheers,

Hugh