Towards a better DAC - introducing the HEC
Posted 14th June 2010 at 04:16 AM by abraxalito
Updated 8th September 2011 at 03:05 AM by abraxalito
Updated 8th September 2011 at 03:05 AM by abraxalito
Doug Self's comprehensive work on power amplifiers has been an inspiration to me over a number of years. In it, he coined the notion of the 'blameless amp' as one which was beyond reproach, measurements-wise. So I pondered over running with the notion of 'blameless DAC' along the same lines. Initially this sounded attractive, even though not entirely original. Google found a reference to it already here on diyaudio but could not show me the actual posting.
I decided on reflection that the 'blameless DAC' would be not such a good idea. That's partly because for me, measurements are not the goal of audio design. Enjoyable sound is what does it for me, not vanishingly low THD figures. So a 'blameless DAC' could only be blameless if measurements are God, so to speak. For someone like me for whom auditory experience is all, I decided I'd need another term for my DAC project.
I've settled instead on the term 'highly evolved converter' or HEC for short. This recognises that all designs are really a process - and 'highly evolved' means its been undergoing the tweak process for a considerable length of time. Its also 'highly evolved' in the sense that I've been following DAC technology for several decades (its almost 30 years since the birth of CD) and have some awareness of the strengths and shortcomings of various approaches. 'Highly evolved' is a more modest claim than 'blameless' as it recognises that there may be future discoveries about DACs which suggest enhancements to be made - the HEC is not the end of the road for DACs, merely a milestone along the way.
Frederick Brooks in his altogether excellent 'Mythical Man Month' has a article entitled 'No silver bullet' regarding the process of software design. I'm also of a like persuasion that there's no 'silver bullet' in DACs, no single solution or technological advance which will bring about an order of magnitude improvement. Brooks also has interesting observations on how software gets written which I feel are applicable here too. Originally he notes that software was written. Later on, it got built. But now he feels the time is ripe for a third metaphor - software should in future be grown. In other words, what Brooks is saying is what us tweakers have known for years - the fun is in improving something which already is going. Making something from scratch is considerably less rewarding because the payback is a long time coming. With a tweak, gratification is fairly instant.
The starting point for developing my HEC is this design from Taobao AD1955 DAC board
Choosing a DAC chip
You'll notice that this board I selected has an Analog Devices AD1955. The story how I came to begin with that chip is rather a long one, involving lots of experience with earlier generation converters. One of the influences is that I believe the designer of this chip to be Bob Adams, a name I came across in the late 1980s as the designer of the legendary dbx oversampled A/D then used by Chesky Records. If anyone knew how to design a decent sounding oversampled DAC I figured he'd be in with a shout. The technical arguments are that, in short, pure multibit converters suffer from fairly poor low-level linearity and single bit converters suffer from terrible noise floors (containing correlated tones aka noise modulation). So the best chance of a good sounding DAC has to be a multibit noise-shaped device which puts pay to both of these difficulties. Of course, BB/TI make their PCM179X range of DACs along very similar lines but the ADI/Bob Adams cachet was too strong to resist.
Listening to the unevolved board
My reference DAC until the purchase of this AD1955 was the PCM179X in my computer's Asus soundcard. On initially firing up the Taobao board (feeding it SPDIF from the PC) it was clearly nowhere in the same league as the Asus. Soundstage depth was indistinct and imaging rather messy - no engagement, no drawing in to the music with the AD1955. So I began to ponder why this may be so, looking more closely at the implementation. I realised that I hadn't looked at all closely before, because it struck me for the first time that there were no decoupling caps around the AD1955 at all! Pulling up the datasheet I found that, sure enough, decoupling caps were specified for both the digital and analog supplies. I was surprised that the AD1955 system worked at all in the total absence of such caps. It was then that it dawned on me that the designer of this board was probably no professional engineer, rather an audio hobbyist so I figured this probably wouldn't be the only howler. Looking carefully at the grounding confirmed that, whilst he'd certainly been aware of the star grounding concept, he wasn't experienced enough to apply it consistently (neither was I when I worked in the industry incidentally, so this isn't to trash the guy).
Improving the power supplies
Fortunately for the potential of this board, there's a GND power plane under the digital circuitry so I just had to figure how to make the best use of this in introducing some decoupling caps. I decided that drilling holes in it would be the best solution, then soldering wires through the holes to act as vias. For this I went shopping for a 0.7mm drill bit which was fun with my level of Chinese (almost non-existent). I also needed some SMT decouplers - I settled on 10uF Z5Us as these have ultra-low ESRs. Their drawback though is the strong voltage coefficient of capacitance, which means that a 10uF 10V rated part might be less than 5uF when running at 5V. Still they're so small and cheap that paralleling isn't a problem.
...
I decided on reflection that the 'blameless DAC' would be not such a good idea. That's partly because for me, measurements are not the goal of audio design. Enjoyable sound is what does it for me, not vanishingly low THD figures. So a 'blameless DAC' could only be blameless if measurements are God, so to speak. For someone like me for whom auditory experience is all, I decided I'd need another term for my DAC project.
I've settled instead on the term 'highly evolved converter' or HEC for short. This recognises that all designs are really a process - and 'highly evolved' means its been undergoing the tweak process for a considerable length of time. Its also 'highly evolved' in the sense that I've been following DAC technology for several decades (its almost 30 years since the birth of CD) and have some awareness of the strengths and shortcomings of various approaches. 'Highly evolved' is a more modest claim than 'blameless' as it recognises that there may be future discoveries about DACs which suggest enhancements to be made - the HEC is not the end of the road for DACs, merely a milestone along the way.
Frederick Brooks in his altogether excellent 'Mythical Man Month' has a article entitled 'No silver bullet' regarding the process of software design. I'm also of a like persuasion that there's no 'silver bullet' in DACs, no single solution or technological advance which will bring about an order of magnitude improvement. Brooks also has interesting observations on how software gets written which I feel are applicable here too. Originally he notes that software was written. Later on, it got built. But now he feels the time is ripe for a third metaphor - software should in future be grown. In other words, what Brooks is saying is what us tweakers have known for years - the fun is in improving something which already is going. Making something from scratch is considerably less rewarding because the payback is a long time coming. With a tweak, gratification is fairly instant.
The starting point for developing my HEC is this design from Taobao AD1955 DAC board
Choosing a DAC chip
You'll notice that this board I selected has an Analog Devices AD1955. The story how I came to begin with that chip is rather a long one, involving lots of experience with earlier generation converters. One of the influences is that I believe the designer of this chip to be Bob Adams, a name I came across in the late 1980s as the designer of the legendary dbx oversampled A/D then used by Chesky Records. If anyone knew how to design a decent sounding oversampled DAC I figured he'd be in with a shout. The technical arguments are that, in short, pure multibit converters suffer from fairly poor low-level linearity and single bit converters suffer from terrible noise floors (containing correlated tones aka noise modulation). So the best chance of a good sounding DAC has to be a multibit noise-shaped device which puts pay to both of these difficulties. Of course, BB/TI make their PCM179X range of DACs along very similar lines but the ADI/Bob Adams cachet was too strong to resist.
Listening to the unevolved board
My reference DAC until the purchase of this AD1955 was the PCM179X in my computer's Asus soundcard. On initially firing up the Taobao board (feeding it SPDIF from the PC) it was clearly nowhere in the same league as the Asus. Soundstage depth was indistinct and imaging rather messy - no engagement, no drawing in to the music with the AD1955. So I began to ponder why this may be so, looking more closely at the implementation. I realised that I hadn't looked at all closely before, because it struck me for the first time that there were no decoupling caps around the AD1955 at all! Pulling up the datasheet I found that, sure enough, decoupling caps were specified for both the digital and analog supplies. I was surprised that the AD1955 system worked at all in the total absence of such caps. It was then that it dawned on me that the designer of this board was probably no professional engineer, rather an audio hobbyist so I figured this probably wouldn't be the only howler. Looking carefully at the grounding confirmed that, whilst he'd certainly been aware of the star grounding concept, he wasn't experienced enough to apply it consistently (neither was I when I worked in the industry incidentally, so this isn't to trash the guy).
Improving the power supplies
Fortunately for the potential of this board, there's a GND power plane under the digital circuitry so I just had to figure how to make the best use of this in introducing some decoupling caps. I decided that drilling holes in it would be the best solution, then soldering wires through the holes to act as vias. For this I went shopping for a 0.7mm drill bit which was fun with my level of Chinese (almost non-existent). I also needed some SMT decouplers - I settled on 10uF Z5Us as these have ultra-low ESRs. Their drawback though is the strong voltage coefficient of capacitance, which means that a 10uF 10V rated part might be less than 5uF when running at 5V. Still they're so small and cheap that paralleling isn't a problem.
...
Total Comments 8
Comments
-
More please...
Posted 15th June 2010 at 12:13 PM by wakibaki -
Hi , I have that AES paper about the 18bit AD, want a direct link?
Posted 15th June 2010 at 12:41 PM by tritosine -
Posted 15th June 2010 at 02:27 PM by abraxalito -
Posted 15th June 2010 at 02:28 PM by abraxalito -
Posted 15th June 2010 at 05:20 PM by Nazar_lv -
Quote:
Also have an Apogee G "low phase dispersion" anti alias filter module , the one they swapped the switched capacitor 8th orders with. In the circuits you have , how the resistor network looks? Maybe some photos would suffice if it isnt enclosed.Posted 15th June 2010 at 07:52 PM by tritosine
Updated 15th June 2010 at 07:56 PM by tritosine -
Great, thanks for that link. Shame it doesn't have the full schematic though
Quote:
Quote:Also have an Apogee G "low phase dispersion" anti alias filter module , the one they swapped the switched capacitor 8th orders with.
Quote:In the circuits you have , how the resistor network looks? Maybe some photos would suffice if it isnt enclosed.Posted 16th June 2010 at 12:51 AM by abraxalito -
Posted 20th July 2010 at 02:02 PM by regiregi22