No!
Everyone is welcome to obtain one of the earlier designs, but only in the way it was build originally.
So you are welcome to choose between the projects, this one you can influence yourself.
The former is closed by now, but still obtainable.
@ ThorstenL:
When do you exspect to return til Europe?? We really would like you to listen to our DAC, since we are pretty sure, it will make you change your fight against all CS chips.
When do you exspect to return til Europe?? We really would like you to listen to our DAC, since we are pretty sure, it will make you change your fight against all CS chips.
Hi,
No offence, but this takes our project out of your scope.
In light of your preferences, we will hardly agree upon anything I think.
We will never use any tube what so ever, just as we will not utilize any FET´s or caps if avoidable, and we will always tend to go the bipolar NFB way if possible.
If this is out of your scope, I´d recommend you begin another different project, which might be very interesting indeed.
You will always be welcome at my place, we really hope to see you here sometime.
No offence, but this takes our project out of your scope.
In light of your preferences, we will hardly agree upon anything I think.
We will never use any tube what so ever, just as we will not utilize any FET´s or caps if avoidable, and we will always tend to go the bipolar NFB way if possible.
If this is out of your scope, I´d recommend you begin another different project, which might be very interesting indeed.
You will always be welcome at my place, we really hope to see you here sometime.
Hi,
I'm likely to be in Germany for the High End and may travel around Europe a bit at the time and may make Arhus.
As for the CS Chips, do not hold your breath, remember, I did test them and others. I found in fact the sonic descriptions given (in the other thread) by some who heard your DAC to echo what I thought of the sound of the CS4398 (and of many other comparable DAC's arguably) .
I did not suggest to adopt these principles. In fact, there are enough projects like that (including ones I published) that no additional ones are needed.
I would like to see a truly state of the art DIY Project that provides really good performance without having to use "funky" engineering, something the objectivists can like because it is solid traditional engineering and something the subjectivists can like because it sounds very good.
Meanwhile I took the liberty to look up the "other" thread.
I see that your discrete analog stage is mostly a derivate of the LC Audio Zap Filter.
This is not a bad choice to start from (I have used the Zap Filter myself several times), but the limit in this circuit is the very fact that BJT's are used and the multiple series gainstages. A folded cascode would likely have been a better choice, IF this kind of direct coupled circuit with servo is considered desirable (I prefer to have my capacitor neat, au naturelle, not distorted and amplified by op-amp - which is what you get with a servo).
BTW, one of the main parasitics of BJT's (base current and base current modulation with signal) cause measurable and audible problems across the whole audio bandwidth that FET's lack. Of course, FET's have signal modulated capacitance as a problem, but that can be addressed much easier with suitable design as design for a wide enough bandwidth pushes these effects well into the Medium Wave radio band...
There are many other areas where the circuits taken from LCA are below the best possible performance, if you like and when I have more time we could discuss this.
Past that, once you have removed the noise from your shunt regulators (Hint - they do not have to have more noise than a low noise NPN Transistor [so say below 1nV|/Hz] - just get shod of the Zenners) please try them on the DAC's reference etc. pins (whichever DAC you use).
Yes, the capacitors on these pins are very audible, but getting this whole mess under active control (John Westlake is also much of a proponent of doing this) is such a step up, you may not believe your ears. And you may find the need to re-evaluate the other DAC chips after doing so.
And do avoid using the 3X7 regulators as CCS, unless you add significant size and value chokes in series to overcome the high frequency rolloff of the OLG Gain of the 3X7 which allows all the high frequency rubbish to ride right through the CCS and of course our shunt regulator regulates much less well as frequency rises, so a lot of HF garbage just rides through and can the wreak havoc in the zero NFB analogue stage.
Ciao T
I'm likely to be in Germany for the High End and may travel around Europe a bit at the time and may make Arhus.
As for the CS Chips, do not hold your breath, remember, I did test them and others. I found in fact the sonic descriptions given (in the other thread) by some who heard your DAC to echo what I thought of the sound of the CS4398 (and of many other comparable DAC's arguably) .
I did not suggest to adopt these principles. In fact, there are enough projects like that (including ones I published) that no additional ones are needed.
I would like to see a truly state of the art DIY Project that provides really good performance without having to use "funky" engineering, something the objectivists can like because it is solid traditional engineering and something the subjectivists can like because it sounds very good.
Meanwhile I took the liberty to look up the "other" thread.
I see that your discrete analog stage is mostly a derivate of the LC Audio Zap Filter.
This is not a bad choice to start from (I have used the Zap Filter myself several times), but the limit in this circuit is the very fact that BJT's are used and the multiple series gainstages. A folded cascode would likely have been a better choice, IF this kind of direct coupled circuit with servo is considered desirable (I prefer to have my capacitor neat, au naturelle, not distorted and amplified by op-amp - which is what you get with a servo).
BTW, one of the main parasitics of BJT's (base current and base current modulation with signal) cause measurable and audible problems across the whole audio bandwidth that FET's lack. Of course, FET's have signal modulated capacitance as a problem, but that can be addressed much easier with suitable design as design for a wide enough bandwidth pushes these effects well into the Medium Wave radio band...
There are many other areas where the circuits taken from LCA are below the best possible performance, if you like and when I have more time we could discuss this.
Past that, once you have removed the noise from your shunt regulators (Hint - they do not have to have more noise than a low noise NPN Transistor [so say below 1nV|/Hz] - just get shod of the Zenners) please try them on the DAC's reference etc. pins (whichever DAC you use).
Yes, the capacitors on these pins are very audible, but getting this whole mess under active control (John Westlake is also much of a proponent of doing this) is such a step up, you may not believe your ears. And you may find the need to re-evaluate the other DAC chips after doing so.
And do avoid using the 3X7 regulators as CCS, unless you add significant size and value chokes in series to overcome the high frequency rolloff of the OLG Gain of the 3X7 which allows all the high frequency rubbish to ride right through the CCS and of course our shunt regulator regulates much less well as frequency rises, so a lot of HF garbage just rides through and can the wreak havoc in the zero NFB analogue stage.
Ciao T
there are a lot of types of high and low dielectric constant ceramics, such as X7R, X5R, NP0, X6S, Y5V, X8R, Z5U etc
To answer how good/bad is using ceramic caps you need to answer to several questions
1 what type of ceramics you use
2 what capacitance, what overall dimension
3 in what place
Judging by your posts you fail to cook opamp with such low THD 🙂
it is not true, opams is much critical to it decoupling and PSU, it is one of the reason why feedback free designs sounds better, properly cook NFB devices is not easy, this requires professionalism (only article of Bode "Network analysys and feedback amplifier design" 1946y have 700pages, and many others...)
look at Si530/531
Last edited:
the main disadvantage of 4398 is his Direct charge transfer dac (DCT DAC) and you can not do better than internal dac with his internal MOSFET opamp
Hi,
@T: No I haven't done any measurements below 12 KHz.
@Nazar_lv: Thanks for the tip to SI530. One question: Do you prefer this to SI570, since it requires no MCU for programming? SI570 can be ordered with a prefered initial frequency also. Or does it have other/better data than my initial glance of the datasheet suggests? I ask because I'm curious. I don't have any preferences.
Cheers,
mkc
@T: No I haven't done any measurements below 12 KHz.
@Nazar_lv: Thanks for the tip to SI530. One question: Do you prefer this to SI570, since it requires no MCU for programming? SI570 can be ordered with a prefered initial frequency also. Or does it have other/better data than my initial glance of the datasheet suggests? I ask because I'm curious. I don't have any preferences.
Cheers,
mkc
@ ThorstenL:
Our analog stage is in no way a ripoff from the ZAPFilter. We did try the ZAPFilter, and quickly put it away.
If you check the ZAPFilter schematic, you will see that the most common thing is, that we both use BJT's and a differential pair in the input. Else, I don't think there is that much similarity.
I argree that the idea of a ZAPFilter is fine.... But the single ended output is not in my taste. We tried to replace it with a push-pull version, which enhanced the performance dramatically.
Guess the standard ZAPFilter is well suited for thoose, who like the sound of single-ended amps, tubes and transformers and so on...
Our analog stage is in no way a ripoff from the ZAPFilter. We did try the ZAPFilter, and quickly put it away.
If you check the ZAPFilter schematic, you will see that the most common thing is, that we both use BJT's and a differential pair in the input. Else, I don't think there is that much similarity.
I argree that the idea of a ZAPFilter is fine.... But the single ended output is not in my taste. We tried to replace it with a push-pull version, which enhanced the performance dramatically.
Guess the standard ZAPFilter is well suited for thoose, who like the sound of single-ended amps, tubes and transformers and so on...
I just picked up the ESS 9018 board at KvK.... Next week I will start implementing our analog stage. We have very high expectations on this, since we both have a feeling, that Dustin (Who made the ES9018), has made a very nice DAC chip.
If the ES9018 performs well, there is NO DOUBT that we will be using this in the project. Then I guess no one will point fingers at us, because of some religious problem with CS chips. ES9018 has got absolute best performance on the marked. THD: -120dB!!
If the ES9018 performs well, there is NO DOUBT that we will be using this in the project. Then I guess no one will point fingers at us, because of some religious problem with CS chips. ES9018 has got absolute best performance on the marked. THD: -120dB!!
It has the same internal circuit but 530 is XO and 570 is programable XO
2nd, 3rd harmonics below -120db also have PCM1794A/1792, AD1955, SM5865
Hi Nazar_lv,
Thanks, for the reply.
Yes, my initial response was that they where the same die.?
I also saw "Hurtigs" entry about the -120dB. I did manage to secure a few AD1955 and also AD1896. As an ex Analog Devices employee, I think I should carry the flag.
Truth told. I care less for CD players. Given the uncertainty of optical drives, I much rather want a network streamer. Hopefully "T" wont kill me for that remark. But that is how I feel.
I have had this on the drawing board for some time, but time has not permitted me to pursue this. Being a SW/HW guy, this feels natural.
Best regards,
mkc
Thanks, for the reply.
Yes, my initial response was that they where the same die.?
I also saw "Hurtigs" entry about the -120dB. I did manage to secure a few AD1955 and also AD1896. As an ex Analog Devices employee, I think I should carry the flag.
Truth told. I care less for CD players. Given the uncertainty of optical drives, I much rather want a network streamer. Hopefully "T" wont kill me for that remark. But that is how I feel.
I have had this on the drawing board for some time, but time has not permitted me to pursue this. Being a SW/HW guy, this feels natural.
Best regards,
mkc
I disagrree.
Any LC design I´ve ever heard is a bad state to start from.
And our design is pretty different from his.
IMHO there is no alternative to BJT´s, FET´s just sound to much like "black background".
We like our shunts pretty well, and they are fortunately pretty noiseless.
How set in stone is the use of the src4192? Have folks built a circuit to audition other chips, for instance CS8420/21/22 ?
Hi Kurt,
You mean they have too little background noise? I cannot see that as a "bad thing"...
Hmm, how do you definde "pretty noiseless"? The schematic that was posted showed three 4.7V Zenner diodes in series to set the voltage and no measures taken to remove their intrinisic (and quite high) noise. The only noise that needs to be present is that of the NPN Transistor, use one with the right geometry (I like MMBT4401) and noise can be very low.
Anyway, it's your design, I'm merely suggesting some small ways to improve things.
Ciao T
You mean they have too little background noise? I cannot see that as a "bad thing"...
Hmm, how do you definde "pretty noiseless"? The schematic that was posted showed three 4.7V Zenner diodes in series to set the voltage and no measures taken to remove their intrinisic (and quite high) noise. The only noise that needs to be present is that of the NPN Transistor, use one with the right geometry (I like MMBT4401) and noise can be very low.
Anyway, it's your design, I'm merely suggesting some small ways to improve things.
Ciao T
Hi,
I ain't got nothing against network streamers. But I have been highly optimised PC based sources for years (but ones including coverart and decent graphics plus touchscreen), especially with big collections they absolutely rule.
I have yet to find a Network Streamer who's interface is not SO BAD, it has me cursing all the time. I have one for the bedroom system, actually, a forerunner of the Squeezebox Boom by Linksys. It gets used very rarely.
When I was trying to use a friends Squeezebox I got so frustrated I walked up the stairs to the PC to cue up the tracks on the PC, as it was pretty much impossible on the SB itself.
So, please make a network streamer with a really good user interface and at least 7" Touchscreen. Oh yes, and with a really good SPDIF Outputput and I2S with MCK option. I'll take two at least.
Ciao T
I ain't got nothing against network streamers. But I have been highly optimised PC based sources for years (but ones including coverart and decent graphics plus touchscreen), especially with big collections they absolutely rule.
I have yet to find a Network Streamer who's interface is not SO BAD, it has me cursing all the time. I have one for the bedroom system, actually, a forerunner of the Squeezebox Boom by Linksys. It gets used very rarely.
When I was trying to use a friends Squeezebox I got so frustrated I walked up the stairs to the PC to cue up the tracks on the PC, as it was pretty much impossible on the SB itself.
So, please make a network streamer with a really good user interface and at least 7" Touchscreen. Oh yes, and with a really good SPDIF Outputput and I2S with MCK option. I'll take two at least.
Ciao T
What set in stone is that SRC4192 and its sibilings do 16* oversampling before the 4point curve fitting procedure (the actual SRC function ). Whereas ESS does , what, 2000* OS and 2 point (linear curve fitting). Anagram did both of these, the polynomial version looks worse on paper than the TI, and the 2000* version was reserved for XY hiend customers.
I should mention that setting slow roll of using an ASRC is a bad bad idea , once the input sample rate is low, like 44.1 , the EvilSRC ( with its crude 🙄 28, 32bit DF ) and its curve fitting function going to follow the aliased waveform shape, and that must look ugly, even though its goin to be noise shaped at the end, but still, I dont quite get the obsession with the bit perfect thing. These people who preach the bit perfect thing usually use either NOS or sigma delta (inherent noise shaping) , also SRC don't qualify for the asynchronous buzzword, as it is rather evil (not bit perfect), but what is ?
If you would use 16x OS pcm1704 I would understand you like your waveforms as they are, but hey, NOS ? S-D ? Can you please upload those AP2 plots with SRC and the resulting IMD ?
So messing in time domain (SRC) is inherently wrong, textbook anti alias filtering alike, but how about messing with amplitude? Thats okay? Can we do that ?
...
Hi,
Probably. IMNSHO and IME setting "sharp rolloff" using an ASRC is as bad an idea though... 😛
It is not so much "bit-perfect". First, based strictly on listening and on using DAC's that offer a direct conversion of numbers into analogue signals (not ones that use what is effectively noise density modulation) it appears that any of the available digital filters has the effect of more or less subtly (mostly less subtly) degrading the perceived sound quality.
As the CD-Players that I helped design include the option to make that comparison in a direct and "fair" (as in "all else being equal") quite a few people had the chance to try this. So far I am unaware of anyone who, having the option directly on the remote and hence being able to very easily select classic DF style oversampling, ASRC Style upsampling or NOS selected any of the digitally filtered versions as superior or preferable for most music.
This can be done D2D. When i get back to the Lab and around to digging out the TI Eval PCB for their ASRC I'll happily oblige.
Actually, both are wrong. The issue is that CD lacks a sufficient sample rate to resolve higher frequency signals. We can now do several things.
One is to oversample and create "pretend" extra information (which of course based on simple information theory can NEVER be actual information) and we can oversample by a varity of ways including integer and non-integer relations between sample rates, we can use very few taps or very many taps on the filter and so on ad nausaeum.
Or we can accept that we simply have a distorted signal coming in and live with the distortion in the amplitude domain. Or we can use special analogue filters.
My experience is that I quite like (sonically) some digital filters (eg. HDCD PDM100/200) and that I like Non-Os as well. I also quite dislike most digital filters but find that "slow rolloff" filters (such as originally used by Wadia and Pioneer and now found on many DAC's) seem subjectively less bothersome.
Again, to a degree it would be desirable, I think, to give the builder of a DIY DAC the choice to have or not have the ASRC and to have or not have a Soft Rolloff Digital filter (especially given that it is pretty trivial to implement) and all of that in a really well designed and inexpensive to build DAC. That way everyone can choose what suits them sonically. And if the majority ends up without ASRC and with a slow rolloff filter by choice, after evaluating the options, where is the harm in that?
Ciao T
Probably. IMNSHO and IME setting "sharp rolloff" using an ASRC is as bad an idea though... 😛
It is not so much "bit-perfect". First, based strictly on listening and on using DAC's that offer a direct conversion of numbers into analogue signals (not ones that use what is effectively noise density modulation) it appears that any of the available digital filters has the effect of more or less subtly (mostly less subtly) degrading the perceived sound quality.
As the CD-Players that I helped design include the option to make that comparison in a direct and "fair" (as in "all else being equal") quite a few people had the chance to try this. So far I am unaware of anyone who, having the option directly on the remote and hence being able to very easily select classic DF style oversampling, ASRC Style upsampling or NOS selected any of the digitally filtered versions as superior or preferable for most music.
This can be done D2D. When i get back to the Lab and around to digging out the TI Eval PCB for their ASRC I'll happily oblige.
So messing in time domain (SRC) is inherently wrong, textbook anti alias filtering alike, but how about messing with amplitude? Thats okay? Can we do that ?
Actually, both are wrong. The issue is that CD lacks a sufficient sample rate to resolve higher frequency signals. We can now do several things.
One is to oversample and create "pretend" extra information (which of course based on simple information theory can NEVER be actual information) and we can oversample by a varity of ways including integer and non-integer relations between sample rates, we can use very few taps or very many taps on the filter and so on ad nausaeum.
Or we can accept that we simply have a distorted signal coming in and live with the distortion in the amplitude domain. Or we can use special analogue filters.
My experience is that I quite like (sonically) some digital filters (eg. HDCD PDM100/200) and that I like Non-Os as well. I also quite dislike most digital filters but find that "slow rolloff" filters (such as originally used by Wadia and Pioneer and now found on many DAC's) seem subjectively less bothersome.
Again, to a degree it would be desirable, I think, to give the builder of a DIY DAC the choice to have or not have the ASRC and to have or not have a Soft Rolloff Digital filter (especially given that it is pretty trivial to implement) and all of that in a really well designed and inexpensive to build DAC. That way everyone can choose what suits them sonically. And if the majority ends up without ASRC and with a slow rolloff filter by choice, after evaluating the options, where is the harm in that?
Ciao T
I think if you know as much, you can do an initial oversampling on the PC beforehand after examining whether theres
- natural frequency range extension above 18khz (this counts as textbook filtering )
- just crude above 18khz (that would qualify as a job for an apodising filter )
- nothing above 18khz ( and this would work with slow roll off IMHO)
You end up with 32bit 88.2 khz , and its your choice you keep 24bit or render 16bit with "psychoacustically optimised wordlength reduction " (... and thats perfectly safe compared to how CD's are made).
- natural frequency range extension above 18khz (this counts as textbook filtering )
- just crude above 18khz (that would qualify as a job for an apodising filter )
- nothing above 18khz ( and this would work with slow roll off IMHO)
You end up with 32bit 88.2 khz , and its your choice you keep 24bit or render 16bit with "psychoacustically optimised wordlength reduction " (... and thats perfectly safe compared to how CD's are made).
Last edited:
Hi,
Now, this is strictly personal of course (so no recommendation for impementation with this project), I have tried upsampling with my PC (my digital source is PC of course and my DAC accepts up to 96KHz and my PC outputs that also bitperfect) I have to say I prefer the unprocessed file, subjectively (DAC of course has only analogue filters).
I also prefer files with genuinely higher than CD sample rate though, but "upsampled" CD standard files sound not like the real high resolution copies, in fact, straight CD sounds more like the high resolution copies than upsampled CD...
But again, that's just me.
Ciao T
Now, this is strictly personal of course (so no recommendation for impementation with this project), I have tried upsampling with my PC (my digital source is PC of course and my DAC accepts up to 96KHz and my PC outputs that also bitperfect) I have to say I prefer the unprocessed file, subjectively (DAC of course has only analogue filters).
I also prefer files with genuinely higher than CD sample rate though, but "upsampled" CD standard files sound not like the real high resolution copies, in fact, straight CD sounds more like the high resolution copies than upsampled CD...
But again, that's just me.
Ciao T
- Status
- Not open for further replies.