More on Top-notch DAC (and my hero JC)
I do not want to turn this thread into a XD0 discussion and wish to tread lightly given www.diyAudio.com's somewhat murky rules/practice regarding commercial product. Chris B started this thread with a request for input getting up the learning curve on his goal “to build a DAC that will match the performance of the very best out there and I'd like recommendations on what designs to look at”. I posted the links to XD0 because I thought the nine pages of schematics and 7200 words of description on the XD0 might further his goal. The responses below will I believe continue to further the discussion (and give me a chance to lionize my mentor/hero John Camille).
Bas wrote:
What a well documented DAC! Wonderful, I noticed you knew John Camille. I've heard quite a lot about him over at bottlehead.com. And have an "invention" of his in one of my amps. C4S...Camille's Cascaded Constant Current Source. What a pity he died so young!
John Camille was one of the most important mentors I have had the privilege to work with during my 25+ year career in electronics. How sadly ironic that such a brilliant man would succumb to a brain tumor. When I first laid eyes on John Camille he was probing the 1300VDC 200mA B+ rail of his 211 tube shunt regulated SET amp. The amp was inverted in a test fixture the bright emitter 211 tubes glowing like 100 watt light bulbs while John used a high voltage scope probe to check his B+. The 2 foot long probe with its HV flashover disks looked like a prop out of a 1930’s science fiction movie. Over that weekend as I got to know John I learned why checking a live 1300VDC circuit was no big deal. He had been a ham as a boy in the day when radio=tubes. As a teenager he worked at a local AM radio station where I expect 1300v represented just the middle range of voltages involved! His US Air Force career began as jet fighter pilot. He once told me his obsession with reliability in electronics stemmed from the fact unreliable avionics cost him his ace. During his combat tour in Vietnam he had his fifth MEG lined up in the sights of his F4’s guns and when he pulled the trigger nothing happened. Instead of the guns firing the circuit breakers tripped. After his combat tour he was in charge of maintaining radar installations in Asia. Keeping tube based radars operating and quiet in the tropics taught him more about real world circuits than I can ever hope to know. After the Air Force John worked as an electrical engineer in TI’s ‘skunk works’ on secret electronic devices for the US government. By the time I met John he was retired from TI and bring his amazing expertise to bear on high end audio electronics.
tiroth commented:
I guess I have trouble getting too excited about XD0. Correct me if I am wrong, but this is basically the "Cheap 24/96" DAC with more regulators and a different (IC instead of discrete) analog stage. Don't get me wrong, I know local decoupling and regulation can make huge improvements, but when better DIRs and ASRCs have been around for years I can't get excited about CS8420.
It is a little frustrating if one is not into the latest NOS fad...there are few designs that really break new ground. I would consider Guido's VCXO scheme and JWB's in-development DAC (better ASRC, mono mode DACs) to be in the "exciting" category for me.
For the record I designed XD0 well before I had ever heard of “Cheap 24/96”. Similarities which may exist result from us starting with a similar chip set. The fact that “local decoupling and regulation can make huge improvements” is in large part the driving force in the XD0 design. My time with John Camille drove home that point just at a time when my direct experience was making the point in my listening room. One of John’s hobbies was buying nice test equipment and tweaking it. Like taking 6-digit HP multimeters and reworking the grounds until that last digit is solid. He was always preaching the gospel of thinking RF as well as audio band, fast low noise power and ground systems more involved than typical audio practices. Concurrently I was taking the X-DAC 3.0 up to the X-DAC 3.0 Signature level and directly experiencing how attention to “local decoupling and regulation can make huge improvements”. The AD811 based regulator using SMD parts to keep current loops tiny one sees in XD0 is all about John C’s way.
As I experienced the blooming of the CS8412/CS4328 chip set used in X-DAC 3.0 as each refinement of the supporting circuits I noticed an interesting correlation with reviews in the press. Of course my little effort, like the DIY community at large, was completely below the radar of Stereophile & TAS. There were a lot of commercial offering using the CS8412/CS4328 chip set because in its day it was the first two chip solution to a hi-fi audio DAC. The opinion soon grew around this combo that it always sounded good but not quite class A SOTA. And that is what I experienced in initial implementations. Then a couple of companies like Timbre Technologies TT-1 (http://www.soundstage.com/rise02.htm) and Entec introduced DACs using CS8412/CS4328 with all-out high end design and tweaking applied. Suddenly the sound was SOTA. Back home I experienced similar transformations taking my CS8412/CS4328 equipped X-DAC 3.0 up to the X-DAC 3.0 Signature level. My point is there are rewards to lingering over a given chip set long enough to get its full measure before chasing the latest parts.
Engineering and design are all about the trade-offs. Let me close this already too long post comparing a couple of trade-offs in recent digital audio chips.
*”Dynamic Range”
Of course this is not an exhaustive listing of parts. I believe it does illustrate two points. These are:
1. Of the DACs introduced after the CS43122 only the PCM1702 betters it signal to noise ratio (SNR). Newer is not necessarily better as the goal is often to make the chips smaller, cheaper, or add features of no interest to DIYers and the high-end audiophile.
2. After the CS8420/CS43122 digital audio chips went to smaller less DIY friendly packages. Now if we are going to play with digital audio in 2004 dealing with SMD is a fact of life. I like SMD (surface mount devices) because I can get my circuit boards really tight with tiny current loops. But one does have to consider actually building the thing. With passive parts I like to stop at 0805 (with a few 1206 when I must) because I can reliably hand solder these using tweezers. The SOIC28L package used on CS8420 and CS43122 is 17.9mm long x 7.5 mm wide with 1.27 mm pin spacing (for those who like me still think in inches that is ~ 0.7 x 0.3 with 0.05” pin spacing). The SSOP and QSOP packages used in later devices are similar over all size 10 mm x 4.4 to 7.5 mm. The real challenge is these use a 0.65 mm pitch on the pins. That’s 0.025 inches between pin centerlines. This is no big deal for professionals who solder SMDs all day. My personal experience is I can solder SOIC packages 100% good and going to SSOP/QSOP my success rate drops to the 75-80% range.
I do not want to turn this thread into a XD0 discussion and wish to tread lightly given www.diyAudio.com's somewhat murky rules/practice regarding commercial product. Chris B started this thread with a request for input getting up the learning curve on his goal “to build a DAC that will match the performance of the very best out there and I'd like recommendations on what designs to look at”. I posted the links to XD0 because I thought the nine pages of schematics and 7200 words of description on the XD0 might further his goal. The responses below will I believe continue to further the discussion (and give me a chance to lionize my mentor/hero John Camille).
Bas wrote:
What a well documented DAC! Wonderful, I noticed you knew John Camille. I've heard quite a lot about him over at bottlehead.com. And have an "invention" of his in one of my amps. C4S...Camille's Cascaded Constant Current Source. What a pity he died so young!
John Camille was one of the most important mentors I have had the privilege to work with during my 25+ year career in electronics. How sadly ironic that such a brilliant man would succumb to a brain tumor. When I first laid eyes on John Camille he was probing the 1300VDC 200mA B+ rail of his 211 tube shunt regulated SET amp. The amp was inverted in a test fixture the bright emitter 211 tubes glowing like 100 watt light bulbs while John used a high voltage scope probe to check his B+. The 2 foot long probe with its HV flashover disks looked like a prop out of a 1930’s science fiction movie. Over that weekend as I got to know John I learned why checking a live 1300VDC circuit was no big deal. He had been a ham as a boy in the day when radio=tubes. As a teenager he worked at a local AM radio station where I expect 1300v represented just the middle range of voltages involved! His US Air Force career began as jet fighter pilot. He once told me his obsession with reliability in electronics stemmed from the fact unreliable avionics cost him his ace. During his combat tour in Vietnam he had his fifth MEG lined up in the sights of his F4’s guns and when he pulled the trigger nothing happened. Instead of the guns firing the circuit breakers tripped. After his combat tour he was in charge of maintaining radar installations in Asia. Keeping tube based radars operating and quiet in the tropics taught him more about real world circuits than I can ever hope to know. After the Air Force John worked as an electrical engineer in TI’s ‘skunk works’ on secret electronic devices for the US government. By the time I met John he was retired from TI and bring his amazing expertise to bear on high end audio electronics.
tiroth commented:
I guess I have trouble getting too excited about XD0. Correct me if I am wrong, but this is basically the "Cheap 24/96" DAC with more regulators and a different (IC instead of discrete) analog stage. Don't get me wrong, I know local decoupling and regulation can make huge improvements, but when better DIRs and ASRCs have been around for years I can't get excited about CS8420.
It is a little frustrating if one is not into the latest NOS fad...there are few designs that really break new ground. I would consider Guido's VCXO scheme and JWB's in-development DAC (better ASRC, mono mode DACs) to be in the "exciting" category for me.
For the record I designed XD0 well before I had ever heard of “Cheap 24/96”. Similarities which may exist result from us starting with a similar chip set. The fact that “local decoupling and regulation can make huge improvements” is in large part the driving force in the XD0 design. My time with John Camille drove home that point just at a time when my direct experience was making the point in my listening room. One of John’s hobbies was buying nice test equipment and tweaking it. Like taking 6-digit HP multimeters and reworking the grounds until that last digit is solid. He was always preaching the gospel of thinking RF as well as audio band, fast low noise power and ground systems more involved than typical audio practices. Concurrently I was taking the X-DAC 3.0 up to the X-DAC 3.0 Signature level and directly experiencing how attention to “local decoupling and regulation can make huge improvements”. The AD811 based regulator using SMD parts to keep current loops tiny one sees in XD0 is all about John C’s way.
As I experienced the blooming of the CS8412/CS4328 chip set used in X-DAC 3.0 as each refinement of the supporting circuits I noticed an interesting correlation with reviews in the press. Of course my little effort, like the DIY community at large, was completely below the radar of Stereophile & TAS. There were a lot of commercial offering using the CS8412/CS4328 chip set because in its day it was the first two chip solution to a hi-fi audio DAC. The opinion soon grew around this combo that it always sounded good but not quite class A SOTA. And that is what I experienced in initial implementations. Then a couple of companies like Timbre Technologies TT-1 (http://www.soundstage.com/rise02.htm) and Entec introduced DACs using CS8412/CS4328 with all-out high end design and tweaking applied. Suddenly the sound was SOTA. Back home I experienced similar transformations taking my CS8412/CS4328 equipped X-DAC 3.0 up to the X-DAC 3.0 Signature level. My point is there are rewards to lingering over a given chip set long enough to get its full measure before chasing the latest parts.
Engineering and design are all about the trade-offs. Let me close this already too long post comparing a couple of trade-offs in recent digital audio chips.
Code:
Mfg. Part # SNR Package
---------------------------------------------------------------
TI PCM1738 106 SSOP(DB)28
TI PCM1702 127 QSOP-20
TI PCM1792 117 SSOP(DB)28
TI SRC4190 125* SSOP(DB)28
AD AD1955 120 SSOP(DB)28
CS CS43122 122 SOIC28L
CS CS8420 128* SOIC28L
CS CS4398 120 28-TSSOP*”Dynamic Range”
Of course this is not an exhaustive listing of parts. I believe it does illustrate two points. These are:
1. Of the DACs introduced after the CS43122 only the PCM1702 betters it signal to noise ratio (SNR). Newer is not necessarily better as the goal is often to make the chips smaller, cheaper, or add features of no interest to DIYers and the high-end audiophile.
2. After the CS8420/CS43122 digital audio chips went to smaller less DIY friendly packages. Now if we are going to play with digital audio in 2004 dealing with SMD is a fact of life. I like SMD (surface mount devices) because I can get my circuit boards really tight with tiny current loops. But one does have to consider actually building the thing. With passive parts I like to stop at 0805 (with a few 1206 when I must) because I can reliably hand solder these using tweezers. The SOIC28L package used on CS8420 and CS43122 is 17.9mm long x 7.5 mm wide with 1.27 mm pin spacing (for those who like me still think in inches that is ~ 0.7 x 0.3 with 0.05” pin spacing). The SSOP and QSOP packages used in later devices are similar over all size 10 mm x 4.4 to 7.5 mm. The real challenge is these use a 0.65 mm pitch on the pins. That’s 0.025 inches between pin centerlines. This is no big deal for professionals who solder SMDs all day. My personal experience is I can solder SOIC packages 100% good and going to SSOP/QSOP my success rate drops to the 75-80% range.
A 8's and tiroth's additions and corrections to my post above set me off on a survey of some of the latest SOTA digital audio chips. Here are the results. As always we need to take specifications off data sheets in proper context and with a grain or two of salt. See the notes I have added as a first level attempt to illustrate some of the conditions and trade-offs. If anyone has a favored part in the 24 bit/192k/DSD/PCM class they want to add call it out.
COMMENTS:
A 8 did catch me in a bifocal moment posting incorrect specifications for the PCM1702 above. I have dropped this part from this posts table because it is a 20 bit DAC and rather old news. I did include the PCM1704 because it will do 24 bits at 96k and it represents that dying breed (alas) of DACs featuring bipolar plus AND minus analog power supplies. Specs be damned if we really want to plumb the depths of +20 bit reproduction it seems parsing up 10 volts on the analog side has GOT to be better than 5. Get 'em before they are gone.
As to the "PCM1792, 132dB dynamic range and THD -108dB". This reminds me of an old saying "on a good day, downhill, and with a tail wind".
Seriously I have included two lines for the PCM1792 in this table, one using the part in stereo mode and the second in the differential mono mode yielding the 132dB figure. Note that to get there the TI engineers use dual differential DACs and analog stages setup to 9 volts output at full scale. A 9 volt out DAC sets me off on flights of fancy involving bespoke custom stereo systems using zero gain or step-down transformer ¡¥preamps¡¦ and low gain power amps. However in the real world all a 9 volt out DAC gains you is double the parts cost and a screwed up gain structure if used with typical amps. I find it sadly ironic that all the post PCM1704 DACs listed above rely on single +5 volt analog rails and then to get back the lost potential the suggestion is to use two of the chips in differential mode, essentially the same thing without the benefits of integrating the part onto a single die. Perhaps I should look at the glass as half full and award points for added flexibility.
Finally on the subject of voltages I have added a column listing the part¡¦s voltage requirements and notes on that subject. Obviously the digital side of digital audio is under going an evolution to 3.3 volt (and even lower) rails. No doubt the MPEG decoder and DD/DTS VLSICs used in DVD players and set top A/V decoders require the lower rails not to melt down. It does raise the issue of selecting parts not only for functionality and performance but also compatibility with the rest of the parts in one¡¦s DAC. For example tiroth¡¦s suggestion of the SCR4192 as the specification king of SRCs brings with it the issue of getting this 3.3 volt part talking to other chips. Analog Devices seems to of considered this in the AD1896 as have Wolfson in their DACs. In XD0 I ¡¥cheat¡¦ and run the CS43122 Vd at 5 volts rather than 3.3. This is closer to the edge of the max than I like to design but I consider the lesser of the evils. Back in the day when we had to resort to glue logic to get Yamaha DIR talking to NPC filter to Burr-Brown DACs we found it added jitter, ground bounce, and other nastiness. I am loath to repeat that history adding 3.3 to 5 or 5 to 3.3 volt translators. So I will repeat A 8¡¦s call ¡§anyone yet tried a SRC4192 and two DSD/PCM1792¡¨ or other designs to share using all or partial 3.3 volt logic? It looks to me the TI parts will steadfastly refuse to talk directly to 5 volt parts. Anybody with some bright ideas on integrating these with legacy parts? Or do we start with a clean sheet design enforcing 3.3 volts for digital and 5 volts analog rule?
Pausing for a moment with the pub racing over data sheets I can announce I have posted the 4-layer PCB layout for XD0 here:
http://www.audiocraftersguild.com/XD0/XD0_downloads.htm
Sorry for the large size of the PDF files. I set the graphics resolution rather high so we can zoom into details while the beta testers and I check this before it gets released for fab.
At this juncture some may ask why I continue to pursue the CS8420/CS43122 chipset with the newer parts listed above available. The answer is I love the way they sound! As the image below shows I had this chipset running early in its life cycle as the X-DAC 3.24 upgrade. It is a very sweet combo.
Code:
[B]
Mfg. Part # SNR DNR THD Va/Vd Package note
DACs[/B]
AD AD1955 120 120 -110 5/5 SSOP(DB)28 1
AKM AK4383 110 110 -94 5 20-TSSOP 1,11
AKM AK4395 120 120 -100 5/5 28-VSOP 7
CS CS43122 122 -102 5/3.3 SOIC28L 2
CS CS4398 120 -107 5/3.3 28-TSSOP 1
TI PCM1738 117 117 0.0004% 5/3.3 SSOP(DB)28 3
TI PCM1704 120 112 0.0008 +-5/5 SOIC 20 4,7
TI PCM1792 127 127 0.0004 5/3.3 SSOP(DB)28 5
TI PCM1792 132 132 0.0004 5/3.3 SSOP(DB)28 6
TI DSD1702 106 106 0.0015 5/3.3 QSOP-20 1
Wolfson WM8716 112 112 -92 3.3 - 5 SSOP 28 7
Wolfson WM8718 111 111 -100 3.3 - 5 SSOP 20 7
Wolfson WM8740 117 117 -104 3.3 - 5 SSOP 28 7
Wolfson WM8740 120 117 -104 3.3 - 5 SSOP 28 7,8
[B]SRCs[/B]
AD AD1896 137to142 137to142 -123to-133 5 or 3.3 SSOP 28 9
AKM AK4124 132to140 -124to130 3.3/3.3 VSOP 30 9,10
CS CS8420 128 -117 5/5 SOIC28L
TI SCR4192 138to144 -137to140 3.3 SSOP(DB)28 9,10
TI SRC4190 125 -125 3.3 SSOP(DB)28 9,10
NOTES:
1. Dual mode PCM+DSD part
2. Vd can be 5V, CS43122 PCM only, pin compatible CS4397 adds PCM+DSD
3. Digital inputs are 5 volt tolerant.
4. Spec for K-grade parts
5. Stereo 2v RMS out mode, add 2 bits to SNR & DNR spec in 4.5v mode,
6. Mono 9v RMS out mode.
7. PCM only
8. Dual parts in mono differential mode.
9. Specification ranges depending on sample rate I/O ratios.
10. No 5V I/O, appears to be compatible with 3.3 or lower logic levels only.
11. Single power input pin for Va and Vd.COMMENTS:
A 8 did catch me in a bifocal moment posting incorrect specifications for the PCM1702 above. I have dropped this part from this posts table because it is a 20 bit DAC and rather old news. I did include the PCM1704 because it will do 24 bits at 96k and it represents that dying breed (alas) of DACs featuring bipolar plus AND minus analog power supplies. Specs be damned if we really want to plumb the depths of +20 bit reproduction it seems parsing up 10 volts on the analog side has GOT to be better than 5. Get 'em before they are gone.
As to the "PCM1792, 132dB dynamic range and THD -108dB". This reminds me of an old saying "on a good day, downhill, and with a tail wind".
Seriously I have included two lines for the PCM1792 in this table, one using the part in stereo mode and the second in the differential mono mode yielding the 132dB figure. Note that to get there the TI engineers use dual differential DACs and analog stages setup to 9 volts output at full scale. A 9 volt out DAC sets me off on flights of fancy involving bespoke custom stereo systems using zero gain or step-down transformer ¡¥preamps¡¦ and low gain power amps. However in the real world all a 9 volt out DAC gains you is double the parts cost and a screwed up gain structure if used with typical amps. I find it sadly ironic that all the post PCM1704 DACs listed above rely on single +5 volt analog rails and then to get back the lost potential the suggestion is to use two of the chips in differential mode, essentially the same thing without the benefits of integrating the part onto a single die. Perhaps I should look at the glass as half full and award points for added flexibility.Finally on the subject of voltages I have added a column listing the part¡¦s voltage requirements and notes on that subject. Obviously the digital side of digital audio is under going an evolution to 3.3 volt (and even lower) rails. No doubt the MPEG decoder and DD/DTS VLSICs used in DVD players and set top A/V decoders require the lower rails not to melt down. It does raise the issue of selecting parts not only for functionality and performance but also compatibility with the rest of the parts in one¡¦s DAC. For example tiroth¡¦s suggestion of the SCR4192 as the specification king of SRCs brings with it the issue of getting this 3.3 volt part talking to other chips. Analog Devices seems to of considered this in the AD1896 as have Wolfson in their DACs. In XD0 I ¡¥cheat¡¦ and run the CS43122 Vd at 5 volts rather than 3.3. This is closer to the edge of the max than I like to design but I consider the lesser of the evils. Back in the day when we had to resort to glue logic to get Yamaha DIR talking to NPC filter to Burr-Brown DACs we found it added jitter, ground bounce, and other nastiness. I am loath to repeat that history adding 3.3 to 5 or 5 to 3.3 volt translators. So I will repeat A 8¡¦s call ¡§anyone yet tried a SRC4192 and two DSD/PCM1792¡¨ or other designs to share using all or partial 3.3 volt logic? It looks to me the TI parts will steadfastly refuse to talk directly to 5 volt parts. Anybody with some bright ideas on integrating these with legacy parts? Or do we start with a clean sheet design enforcing 3.3 volts for digital and 5 volts analog rule?
Pausing for a moment with the pub racing over data sheets I can announce I have posted the 4-layer PCB layout for XD0 here:
http://www.audiocraftersguild.com/XD0/XD0_downloads.htm
Sorry for the large size of the PDF files. I set the graphics resolution rather high so we can zoom into details while the beta testers and I check this before it gets released for fab.
At this juncture some may ask why I continue to pursue the CS8420/CS43122 chipset with the newer parts listed above available. The answer is I love the way they sound! As the image below shows I had this chipset running early in its life cycle as the X-DAC 3.24 upgrade. It is a very sweet combo.
An externally hosted image should be here but it was not working when we last tested it.
Hello Norman,
I don't really disagree with what you say but still want to point out that using some of the newer chips could improve things.
Regarding the 3.3 vs 5 volt issue, there is an "simple" path (if one can handle ssop) to use the DIR1703-SRC4192-(DF1706)-PCM1792 (in stereo or monomode depending how far one wants to go) without glue logic and get to -108dB THD and -127dB SNR.
I realize this does not automatically give sonic qualities but still, if you applied the same care and attention as you and a few others put into their designs perhaps you could push performance a bit further then before eventhough very few would actually be able to benefit from it.
/Michael
I don't really disagree with what you say but still want to point out that using some of the newer chips could improve things.
Regarding the 3.3 vs 5 volt issue, there is an "simple" path (if one can handle ssop) to use the DIR1703-SRC4192-(DF1706)-PCM1792 (in stereo or monomode depending how far one wants to go) without glue logic and get to -108dB THD and -127dB SNR.
I realize this does not automatically give sonic qualities but still, if you applied the same care and attention as you and a few others put into their designs perhaps you could push performance a bit further then before eventhough very few would actually be able to benefit from it.
/Michael
RAKK dac pic
Price 250$ ....I'm gonna get one. Will compare with my DDDAC1543 and report back to you's.
An externally hosted image should be here but it was not working when we last tested it.
Price 250$ ....I'm gonna get one. Will compare with my DDDAC1543 and report back to you's.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.