ThorstenL,
you mention that you would not use the PCM1794. Can you explain a bit why? Setting the TDA1541 aside, what would you recommend and why?
Thanks!
you mention that you would not use the PCM1794. Can you explain a bit why? Setting the TDA1541 aside, what would you recommend and why?
Thanks!
Ok, I just need a bit more clarification. with ASRC PCM1794, without ASRC than PCM1704. I am missing something. why use PCM1794 only with ASRC?
is there a device limitation, or interaction?
Patrick, not sure I understand. I checked the other thread and didn't find much in it. Please explain. I am trying to compile a list of guidelines for designing a very nice sounding high-end DAC. following these guidelines should yield a successful design 99% of the time. Are you saying that using the 9022 will accomplish this? I think I need more data to back that up. I am not going for ultra-cheap. I want ultra high-end for a reasonable cost. $300-$500 in parts and PCBs.
Yes there is , multibit like pcm1704 tolerates a lot more jitter than a pcm1794, this directly translates to numbers, the latter wont hit the datasheet spec(20+ bits) by far - without extreme caution . And the pcm1704 isnt anywhere near 20 bit anyway . : ))))
Oh, and TDA1541 is bad idea even for 16/44 , properly mastered CD-s can have dynamic range on order of 120dB . Also look at the -60 dB distortion plots. This means the more headroom left on the CD (hence 'properly mastered' ), the worse for you, silent passages sink into nonlinearity.
Ok, I think I understand. you are talking strictly from an electrical performance perspective. The PCM1704 does not benefit much from the ASRC; The PCM1794A may, in fact, require it in order to "hit" the datasheet specs.
Yes, multibit tolerate jitter very well, needs ~500 ps for 16bit , and pcm1704 isnt much more to start with, the "K" is rather sorted for "greater than 16bit" rather than "closer to 24bit" : )).
On the other hand pcm1794 wants around 50ps for 20bit, or even less. ASRC output is around 20-25ps. With spdif receiver you re lucky to hit 100 .
I've just scanned this thread and I have a question: Do you just want a DAC that meets your needs, or do you want to design and build your own from scratch?
If you really want to start from scratch, then all of the thread so far is very valid. You also have the output possibilities - eg valves or SS. However, if you just want a DAC that satisfies your needs then.... it seems to me that the Pass D1 clone that Spencer offers over in the for sale area ticks every single one of your boxes. I think you should also look at the new twisted pear buffalo implementation and also the acko DAC setup as well. AFAIK all offer balanced operation.
If it wasn't for the requirement for balanced out, then the field would be much more open. Many of the NOS DACs based on TDA1543 and 1541 are excellent sounding, and are easy enough to build.
Fran
If you really want to start from scratch, then all of the thread so far is very valid. You also have the output possibilities - eg valves or SS. However, if you just want a DAC that satisfies your needs then.... it seems to me that the Pass D1 clone that Spencer offers over in the for sale area ticks every single one of your boxes. I think you should also look at the new twisted pear buffalo implementation and also the acko DAC setup as well. AFAIK all offer balanced operation.
If it wasn't for the requirement for balanced out, then the field would be much more open. Many of the NOS DACs based on TDA1543 and 1541 are excellent sounding, and are easy enough to build.
Fran
...yes, and tda1541 were sorted for 15+ bit (S1 , S2), 1543 for 12-13 , channel matching went unsorted 😀 .
Fran, the answer is YES. I want to build from scratch. This gives me much more flexibility to try more things. Also, I plan to build 2 or 3 of these (for family and friends). Balanced is not an exact requirement, I just wanted it because I have an Aleph 2 clones which may benefit from the balanced source. I could always make balanced from a single-ended DAC. Also, I want to incorparate preamp functionality into the finished product. Input select and a volume control with remote. It is hard to get all this without doing from scratch.
The idea behind the thread is to help myself decide on what is really important to achieving excellent sound. Later, this list of "simple" guidelines could be useful to other DIY members.
My concern has always been that I would spend all this time and money; just to end-up with something no better than my Phillips SA963.
This is my way of "guaranteeing" I end up with something well worth the effort. I think most DIY members would have the same goal.
Good point! I should add something to the list about output stage choices. Right now, I think most will agree a good discrete output design is better than op-amps 90% of the time. So what about SS versus tube? There is also FET vs. BJT.
Hi,
Really basic information theory falsifies this concept. It simply not possible to add "meaning" back after it was removed. Any interpolation whatsoever will be right some of the time and very wrong at other times.
Yup. The issue is that Mr Nyquist and his theorem notwithstanding the sample rate of CD is too low. It is fine for sinewave testing at 20KHz, but realistically for anything above 7KHz the waveform is effectively discarded by the sampling process (and 7KHz sine, triangle and square waves do sound different). Dither, oversampling, upsampling and other method's cannot restore what is lost, they at best can turn everything into sinewaves..
On the other hand, if we sample at 64KHz (the original Decca digital system was 18/64 IIRC) we get at least up to 10KHz and sampling at double speed (88.2/96KHz) gets us up to 15..16KHz. Probably even Decca's system was the critical bit better.
Well, we may debate the waveform etc. all we will. For a fun bit, the late Peter Baxandall opined on many occasions that 16/44 NON OVERSAMPLED PCM was essentially sonically transparent. And guess what, he was essentially right as well. The same cannot be said for later oversampled noise shaped systems of theoretically superior spec. operating at 16/44.
Ciao T
Really basic information theory falsifies this concept. It simply not possible to add "meaning" back after it was removed. Any interpolation whatsoever will be right some of the time and very wrong at other times.
Yup. The issue is that Mr Nyquist and his theorem notwithstanding the sample rate of CD is too low. It is fine for sinewave testing at 20KHz, but realistically for anything above 7KHz the waveform is effectively discarded by the sampling process (and 7KHz sine, triangle and square waves do sound different). Dither, oversampling, upsampling and other method's cannot restore what is lost, they at best can turn everything into sinewaves..
On the other hand, if we sample at 64KHz (the original Decca digital system was 18/64 IIRC) we get at least up to 10KHz and sampling at double speed (88.2/96KHz) gets us up to 15..16KHz. Probably even Decca's system was the critical bit better.
Well, we may debate the waveform etc. all we will. For a fun bit, the late Peter Baxandall opined on many occasions that 16/44 NON OVERSAMPLED PCM was essentially sonically transparent. And guess what, he was essentially right as well. The same cannot be said for later oversampled noise shaped systems of theoretically superior spec. operating at 16/44.
Ciao T
Hi,
The PCM1794 is a 13 Bit DAC (68 Level multilevel modulator & 128 * Delta Sigma), right now I am listening to a DAC that offers true 16 Bit (+/- 0.5 LSB) resultion, why would I want to downgrade? 😛
I want an upgrade to at least 20 Bit in "dbA", not a downgrade to 13 Bit..
I wrote an article illustrating the differences between DAC's in the context of discussing tube stages for Digital Audio over a decade ago.
I have recently updated it and submitted it for publication @ ETM, but it has not made the spring issue. In there I discuss the differences and also the use of Tube stages with different types of DAC's with some additions to my original designs shown a decade ago.
Maybe I get it published here instead?
Ciao T
The PCM1794 is a 13 Bit DAC (68 Level multilevel modulator & 128 * Delta Sigma), right now I am listening to a DAC that offers true 16 Bit (+/- 0.5 LSB) resultion, why would I want to downgrade? 😛
I want an upgrade to at least 20 Bit in "dbA", not a downgrade to 13 Bit..
I wrote an article illustrating the differences between DAC's in the context of discussing tube stages for Digital Audio over a decade ago.
I have recently updated it and submitted it for publication @ ETM, but it has not made the spring issue. In there I discuss the differences and also the use of Tube stages with different types of DAC's with some additions to my original designs shown a decade ago.
Maybe I get it published here instead?
Ciao T
Ok, so we seem to have a debate about NOS vs. OS.
I see no clear cut answer, it is a matter of which evil one "sees" as the greater. ASRC conversion and oversampling have great benefits from a strict technical standpoint. ASRC results in a very low jitter data stream, although the jitter may be "transposed" into something else contained in the actual data rather than time domain. oversampling reduces the DAC images and reconstruction filter requirements to remove them. With the combination of ASRC and OS the DAC images are even more attenuated and the external filter needed is very minimal.
However, I think it is true to say the following:
1) If the PCM stream is low jitter and the DAC images are not detrimental to the sound or following electronics; than there is no reason to use ASRC or OS. NOS is completely sufficient in achieving excellent sonic results, because the underlying recording only has 16/44 worth of information to begin with.
would you all agree with this statement?
I see no clear cut answer, it is a matter of which evil one "sees" as the greater. ASRC conversion and oversampling have great benefits from a strict technical standpoint. ASRC results in a very low jitter data stream, although the jitter may be "transposed" into something else contained in the actual data rather than time domain. oversampling reduces the DAC images and reconstruction filter requirements to remove them. With the combination of ASRC and OS the DAC images are even more attenuated and the external filter needed is very minimal.
However, I think it is true to say the following:
1) If the PCM stream is low jitter and the DAC images are not detrimental to the sound or following electronics; than there is no reason to use ASRC or OS. NOS is completely sufficient in achieving excellent sonic results, because the underlying recording only has 16/44 worth of information to begin with.
would you all agree with this statement?
Is this true 16bit DAC available to me? If so, what are you referring to? If not, what would you recommend that I can get my hands on easily.
Also, I agree the internal architecture of the PCM1794 is not fully 24 bit. However, it does achieve 24bit performance using "other" techniques (delta sigma). Are you saying that the result of this is a sonic degradation? Please elaborate if possible.
Hi,
Why? It performs to within fractions of dB to the limits of 16 Bit Audio.
Only HDCD after decoding and even then more like 20 Bit. Further, most microphones do not even offer a usable 16 Bit dynamic range AND unless I use noise modulation (aka Dither) I cannot encode any signal below 90.3dB on 16 Bit Digital Audio. If I use Dither I actually reduce the dynamic range by about a further 6dB (or 3.7dB for ideal TPDF Dither).
Of course, I can remove all this noise (Dither) by averaging many samples, which allows me to measure sinewaves below -90.3dB, for pseudo stochastic signals (music) the result is reduced real dynamic range.
BTW, as the lowest signal possible with CD (undithered) is -90.3dB (and ideally has equal RMS and Peak Levels) and as the maximum signal on CD has an RMS level of -3dB the dynamic range of CD is 87dB, not 120dB, or 96dB or whatever other number. 😛
They are very close (most of the audio bandwidth < 0.5dB) to the absolute limits of 16/44 digital audio. What's your beef. 😀
Ciao T
Why? It performs to within fractions of dB to the limits of 16 Bit Audio.
Only HDCD after decoding and even then more like 20 Bit. Further, most microphones do not even offer a usable 16 Bit dynamic range AND unless I use noise modulation (aka Dither) I cannot encode any signal below 90.3dB on 16 Bit Digital Audio. If I use Dither I actually reduce the dynamic range by about a further 6dB (or 3.7dB for ideal TPDF Dither).
Of course, I can remove all this noise (Dither) by averaging many samples, which allows me to measure sinewaves below -90.3dB, for pseudo stochastic signals (music) the result is reduced real dynamic range.
BTW, as the lowest signal possible with CD (undithered) is -90.3dB (and ideally has equal RMS and Peak Levels) and as the maximum signal on CD has an RMS level of -3dB the dynamic range of CD is 87dB, not 120dB, or 96dB or whatever other number. 😛
They are very close (most of the audio bandwidth < 0.5dB) to the absolute limits of 16/44 digital audio. What's your beef. 😀
Ciao T
Last edited:
ThorstenL,
I am interested in this tube stage vs. SS item. I myself have never heard any tube gear (a shame, I know). What would you say the big difference is? Is it just that tube stages give a "more predictable" musical sound. where you need to take much more care to get a SS design to sound musical.
Can you post your tube I/V stage design?
I am interested in this tube stage vs. SS item. I myself have never heard any tube gear (a shame, I know). What would you say the big difference is? Is it just that tube stages give a "more predictable" musical sound. where you need to take much more care to get a SS design to sound musical.
Can you post your tube I/V stage design?
Hi,
Actually, I found the key issues with TDA1541 S1/S2 etc. relate to the normally free-running DEM oscillator frequency. Link that DEM logic suitably to BCK and the differences go away and all TDA1541 measure "S2".
I implemented that about 5 years ago in the CD-Player I helped design using TDA1541A.
Ciao T
Actually, I found the key issues with TDA1541 S1/S2 etc. relate to the normally free-running DEM oscillator frequency. Link that DEM logic suitably to BCK and the differences go away and all TDA1541 measure "S2".
I implemented that about 5 years ago in the CD-Player I helped design using TDA1541A.
Ciao T
Hi,
For that i am afraid you need to use the DAC others work on, to use the same DAC chip as your Philips.
BTW, why not first tweak the Philips. You already have a player where you can eliminate or significantly reduce jitter issues and likely a I2S signal you can tap off.
So why not just make a small PCB with suitable clock (send back to the player), reclocker (to kill the jitter from LIM in the CDP), DAC, final powersupplies and analog stage? Using mostly SMD chips and both sides of the PCB such a "sneaky" DAC can be very small and it helps to get the sound you want from the gear you already payed for.
First, there are few "good" discrete designs. Using the RIGHT kind of IC (not neccessarily the latest audiophile favourites) usually outperforms discrete designs, subjectively and objectively. Very few people ever publish modifications with the "right" chips (in fact I only remember Walt Jung).
Of course, if implemented as well and better than these IC's discrete stages perform even better.
I like tubes very well, they make "good" sound easy to get. They do have some (usually very mild) colorations even when applied ideally, but the colorations are generally for want of a better word "musical", in the opposite of the way in which ASRC and Delta/Sigma are "amusical". 🙄
Possibly an aquired taste. 😀
I grew up with tube radios and all the early solid state stuff I got to hear as Child sounded positively NASTY next to the beautiful old german Valve Radio, with it's polished dark wood case and the "sound register" switches marked "Opera", "Jazz", "Speech" and so on and lets not forget the green magic eye and the excotic station names on the Shortwave section of the tuning Window. 😎 It does leave impressions.
Ciao T
For that i am afraid you need to use the DAC others work on, to use the same DAC chip as your Philips.
BTW, why not first tweak the Philips. You already have a player where you can eliminate or significantly reduce jitter issues and likely a I2S signal you can tap off.
So why not just make a small PCB with suitable clock (send back to the player), reclocker (to kill the jitter from LIM in the CDP), DAC, final powersupplies and analog stage? Using mostly SMD chips and both sides of the PCB such a "sneaky" DAC can be very small and it helps to get the sound you want from the gear you already payed for.
First, there are few "good" discrete designs. Using the RIGHT kind of IC (not neccessarily the latest audiophile favourites) usually outperforms discrete designs, subjectively and objectively. Very few people ever publish modifications with the "right" chips (in fact I only remember Walt Jung).
Of course, if implemented as well and better than these IC's discrete stages perform even better.
I like tubes very well, they make "good" sound easy to get. They do have some (usually very mild) colorations even when applied ideally, but the colorations are generally for want of a better word "musical", in the opposite of the way in which ASRC and Delta/Sigma are "amusical". 🙄
Possibly an aquired taste. 😀
I grew up with tube radios and all the early solid state stuff I got to hear as Child sounded positively NASTY next to the beautiful old german Valve Radio, with it's polished dark wood case and the "sound register" switches marked "Opera", "Jazz", "Speech" and so on and lets not forget the green magic eye and the excotic station names on the Shortwave section of the tuning Window. 😎 It does leave impressions.
Ciao T
HDCD is old news, even semi professional audio editing tools can do effectively 20bit DNR 16bit mixdowns.
Check out this "REAL WORLD SCENARIO" , how they supposed to make 16bit media in this age ( and Im not talking about 20 - 25 years ago, when they had 8th order analog filter in front of nyquist rate ADC - and you know that) . Now we have those placeholder filters in front of ADC chips -doing nothing. OK audio is recorded at 6bits 5mhz, then downsampled into something like, say 24 / 96khz . If they do digital signal processing further on, even that 96khz is oversampled once or twice. Yay. Now the worse part: they have to press CD-s. Since 44.1 and 96 is not common, they make something like 11mhz, put a filter to 20khz so the stuff wont fold back. If they are really pushing the limits , they use something like an ideal filter, but for people like you an average linear phase filter is too much , so lets not go there 🙂) . Anyway, this is not all that different what they do with 88.2 -> 44.1, the filter is still there for the same reason, and thne, playing back that CD with a NOS dac, you get ... now what... square alike pre and post ringing ! : ))) ... and since there was no effective anti aliasing in front of the ADC chip, theres awful lot of stuff there : )) . This is the killer impulse response of the NOS thing. And it sells for megabucks. : )
Check out this "REAL WORLD SCENARIO" , how they supposed to make 16bit media in this age ( and Im not talking about 20 - 25 years ago, when they had 8th order analog filter in front of nyquist rate ADC - and you know that) . Now we have those placeholder filters in front of ADC chips -doing nothing. OK audio is recorded at 6bits 5mhz, then downsampled into something like, say 24 / 96khz . If they do digital signal processing further on, even that 96khz is oversampled once or twice. Yay. Now the worse part: they have to press CD-s. Since 44.1 and 96 is not common, they make something like 11mhz, put a filter to 20khz so the stuff wont fold back. If they are really pushing the limits , they use something like an ideal filter, but for people like you an average linear phase filter is too much , so lets not go there 🙂) . Anyway, this is not all that different what they do with 88.2 -> 44.1, the filter is still there for the same reason, and thne, playing back that CD with a NOS dac, you get ... now what... square alike pre and post ringing ! : ))) ... and since there was no effective anti aliasing in front of the ADC chip, theres awful lot of stuff there : )) . This is the killer impulse response of the NOS thing. And it sells for megabucks. : )
Ok guys, lets dial this down a bit. It is a great debate, but I have many other items which I think need to be discussed besides the technical performance of different DACs.
I would like to capture larger themes in high end audio design and not too much detail. Just as much as is needed to draw a decent and mostly correct conclusion.
Here is the list at this point:
1) PCM1704 is a great sounding DAC. Worth the cost over the newer and cheaper DACs from BB, ADI, CS, Wolfson, AKM, etc.. However, it may be possible to achieve great sound with almost any of the new high end chips. If you only want 16/44 the TDA1541A is as good as they come, but is pretty hard to get.
2) Shunt regulators are well worth the extra effort due to the most times bad sonic signature of almost all large value electrolytic decoupling caps. The shunt regulator circuit should have low output impedance over a wide bandwidth and only incorporate small value high quality (film) capacitors at the output. These should be used for the analog sections and the DAC reference if available.
3) ASRC is usually a bad thing, but may in some cases yield some benefit. Certainly if the incoming jitter is very high the ASRC could be useful to clean it up. However, a 2nd PLL could also do the same thing without adding it's own signature to the signal. Putting it in the design would be nice to be able to try for one's self, but a BYPASS option should be included.
4) The WM8804/5 is a cut above the rest in terms of jitter rejection at the SPDIF interface. In fact, a secondary PLL may not even be necessary when using this reciever due to its very low jitter clock recovery architecture.
5) If the PCM stream is low jitter and the DAC images are not detrimental to the sound or following electronics; than there is no reason to use ASRC or OS. NOS is completely sufficient in achieving excellent sonic results, because the underlying recording only has 16/44 worth of information to begin with
6) Discrete analog stage for I/V is 90% more likely to sound good over an op-amp design as long as implemented correctly. BJT vs. FET?, SS vs. Tube?
Let's see if we can probe into #6 a bit now. I think the debate over NOS/OS/ASRC etc. has been done many times already on this forum.
I would like to capture larger themes in high end audio design and not too much detail. Just as much as is needed to draw a decent and mostly correct conclusion.
Here is the list at this point:
1) PCM1704 is a great sounding DAC. Worth the cost over the newer and cheaper DACs from BB, ADI, CS, Wolfson, AKM, etc.. However, it may be possible to achieve great sound with almost any of the new high end chips. If you only want 16/44 the TDA1541A is as good as they come, but is pretty hard to get.
2) Shunt regulators are well worth the extra effort due to the most times bad sonic signature of almost all large value electrolytic decoupling caps. The shunt regulator circuit should have low output impedance over a wide bandwidth and only incorporate small value high quality (film) capacitors at the output. These should be used for the analog sections and the DAC reference if available.
3) ASRC is usually a bad thing, but may in some cases yield some benefit. Certainly if the incoming jitter is very high the ASRC could be useful to clean it up. However, a 2nd PLL could also do the same thing without adding it's own signature to the signal. Putting it in the design would be nice to be able to try for one's self, but a BYPASS option should be included.
4) The WM8804/5 is a cut above the rest in terms of jitter rejection at the SPDIF interface. In fact, a secondary PLL may not even be necessary when using this reciever due to its very low jitter clock recovery architecture.
5) If the PCM stream is low jitter and the DAC images are not detrimental to the sound or following electronics; than there is no reason to use ASRC or OS. NOS is completely sufficient in achieving excellent sonic results, because the underlying recording only has 16/44 worth of information to begin with
6) Discrete analog stage for I/V is 90% more likely to sound good over an op-amp design as long as implemented correctly. BJT vs. FET?, SS vs. Tube?
Let's see if we can probe into #6 a bit now. I think the debate over NOS/OS/ASRC etc. has been done many times already on this forum.
- Status
- Not open for further replies.