Apart the larger quantizer in the Sabre (not much), and the marketing numbers, these two have more in common than not. Do you think those thd and dnr numbers matter? We have those numbers for a while now , you only have to parallel a few pcm1794 or ad1955... wow... I dont think those numbers matter at all, only comes handy if you want to resort to digital volume, you can throw away 3.5 bits worth of attenuation without a second tought, but this needs a fine tuned system, not really consumer friendly.
Last edited:
The quantizer scheme in the Sabre is actually configurable and the defaults are actually not the best for DNR or THD.
Also, the Sabre can be used mono (with phase of any single output being reversed) or even mono in parallel with other sabres if you want crazy high numbers, better than the datasheet specifies.
You can even run a saber completely single ended with 16 outputs all of the same phase mono. But that would be throwing away performance, and silly.
One reason for the excellent sound of the DAC is that the ASRC output works at incredibly high sample rate. And the higher the master clock, the higher the sample rate into the modulators is.
Also keep in mind, that you can run the DAC in 6,7, and 9 bit quantizer modes to name a few.
Also there are some tricks you can do in firmware to make the modulators perform better than they will by default.
Cheers!
Russ
Last edited:
actually its like that because the linear interpolation wouldnt work at all without such "insane" amount of oversampling. Maybe the 128x oversampling filters are cheaper in silicon terms than the polynomial interpolator ( as SRC4192) ? I just ran some tests the last day, 192khz / 12bit quantizer is already very nice IMHO, not much visible "crud" at all . Then you get away with 4x OS, but resource intensive polynomial ASRC is inevitable.
Last edited:
How "insane" is something that in practice works so well?
The sabre paper spells it all out pretty clearly and the patents even more so.
Certainly not the only approach, but effective and practical.
Cheers!
Russ
Last edited:
The scientific work is already done by mr. Otala, I do not think myself capable of adding anything to that.
The matter is not really just about op-amps, it is about feedback vs. non feedback, or at least lower feedback amounts.
Looking at practically any op-amp design exept AD844, they all need to have huge amounts of feedback to be linear within the audio range, which actually is the basic idea of opamps. They were originally made for measuring purposes and that kind of stuff, where you often need one single parameter to be awesome, i.e. noise, slewrate or so.
A few of todays opamps were intended for audio i.e. NE5532/34, OPA 627 and a handfull of others. They all have data which cannot be achieved with discrete design, and absolutely no way without using NFB.
As a result of that OP-amps has become the normal building blocks for audio gear, they always measure well, and designing af PSU for them is peace of cake, as they almost runs perfectly on AC. Designing an analog stage with op-amps takes almost no time, and it will work exactly as wanted.
So why on earth spending time trying out NFB discrete designs?
The answer is sound quality.
During development of our DAC project, we´ve made experiments with dozens of op-amps, incl. audio types, racer-op-amps, out of this world op-amps, and tried hard to make them work well. Even with completely overkill low noise supplies
Facing all the work and time wasted, we ended up discarding the whole bunch of op-amps, and settled for the DC coupled discrete design with no feedback.
What we did not know, was the time and perseverance it takes to get such an amp running perfectly. But I think it was worth it.
The sonic problem we faced using op-amps, was that no matter which one one we chose, the sound was thin, undynamic an dull compared to the discrete design although they were different. They also mostly are very much depending on cable length and type. It seems like they cannot really cope with the reactance in the cables and the next stage. Trying to drive a power amp directly showed this the most.
There was though one opamp, that most of the time, showed to do exactly this better than others, and I hardly dare to say which one, because it isn´t really very higly regarded, and it is also a cheap one, but it is used widely in professional gear.
The reason for all this, is hard to find, because first of all, you cannot have the schematics for the op-amps. The ones published in the datasheets are vere simplified.
So one will have to think in headlines to get any further. My idea is that, creating an opamp, that will always do its job, calls for a lot of compensation for less than ideal implementations.
I.e the PSRR in opamps are magnificent, but to me a better PSU would be the right solution.
The amount of feedback needed to linearise the output will always give very low THD, to me the design should be linear from the beginning, then you could eventually add some small amounts of feedback for DC stabilization or so.
And then ypu have the problem of throwing away some 100-120 dB signal level , and then reamplify every time you pass through an op-amp which is linearized with NFB.
Actually op-amps are impressing devices, if I was told about them, but without ever having seen or heard them, I´d not believe they would work at all. They are examples of very clever engineering, they include shrewed solutions to a lot of problems, which does and do not really exist in audio, as they were made for other purposes.
Almost all newer voltage out dacs sound electronic to me, faffing about with various output stages just made it sound more fake.
The Sabre used in the Buffalo's was the only new type of chip which lacks this fake electronic signature to my ears
For diy, Buffalo32s is my favourite, for commercially ready built dac AYA II is my favourite, I've tried Benchmark, Chord and various others, I didn't like them, they all sounded processed
If I had to pick between AYA II or Buffalo32s I'd go with AYA, it does not quite have the detail of Buffalo especially in the highs but some instruments especially piano sounded more real imo
Lets face it....
Kurt you spent so long on your output stage technology has passed you by. You need to address your DAC stage and start over it sounds like rather than brow beating TPA incredibly nice designs. If your DAC is so much better let the world hear it and we will be the judge. All of this fodder that you are spreading is making a smell around here
Kurt you spent so long on your output stage technology has passed you by. You need to address your DAC stage and start over it sounds like rather than brow beating TPA incredibly nice designs. If your DAC is so much better let the world hear it and we will be the judge. All of this fodder that you are spreading is making a smell around here
Kurt you spent so long on your output stage technology has passed you by. You need to address your DAC stage and start over it sounds like rather than brow beating TPA incredibly nice designs. If your DAC is so much better let the world hear it and we will be the judge. All of this fodder that you are spreading is making a smell around here
You make it sound like he's gold plating a turd
Almost all newer voltage out dacs sound electronic to me, faffing about with various output stages just made it sound more fake.
The Sabre used in the Buffalo's was the only new type of chip which lacks this fake electronic signature to my ears
For diy, Buffalo32s is my favourite, for commercially ready built dac AYA II is my favourite, I've tried Benchmark, Chord and various others, I didn't like them, they all sounded processed
If I had to pick between AYA II or Buffalo32s I'd go with AYA, it does not quite have the detail of Buffalo especially in the highs but some instruments especially piano sounded more real imo
Hi Leo,
Thanks a lot for your reply!
Best regards
Peter
Kurt you spent so long on your output stage technology has passed you by. You need to address your DAC stage and start over it sounds like rather than brow beating TPA incredibly nice designs. If your DAC is so much better let the world hear it and we will be the judge. All of this fodder that you are spreading is making a smell around here
Regardles of the DAC chip, you still have to use an analog stage and postfiltering, through wich the whole thing must perform.
You can compare this to buying a state of the art cartridge for your recordplayer, and then settle with a standard headamp. That won´t work.
Maybe the ESS chip might be very good, but so are ie. CS8416, CS4398 and AD1896. The 32 bit part seems to become the standard in the future, as Cirrus did do a 32 bit ASRC, and both BB and AKM 32 bit DAC chips. Surely AD and CL will make ASRC´s and DACs featuring 32 bit, until then we settle for 24, which I think is quite sufficient.
Guys,
You are being too rough on Kurt and Hurtig. Their arguements have some merit and maybe if I do not agree totally with everything they say most of it has be proved to me in practice. Most designs are good as their weakest link.
Personally I do not care much for op-amps, but they have their uses..........consider a black box with a few control pins to set gain and not much more or a discreet design where there is access to every component or parameter and you will see what I mean. Yes, IC's are getting better but ........
Jam
You are being too rough on Kurt and Hurtig. Their arguements have some merit and maybe if I do not agree totally with everything they say most of it has be proved to me in practice. Most designs are good as their weakest link.
Personally I do not care much for op-amps, but they have their uses..........consider a black box with a few control pins to set gain and not much more or a discreet design where there is access to every component or parameter and you will see what I mean. Yes, IC's are getting better but ........
Jam
I have not a Buffalo, but I made a board built around the ESS 32bits, and tried it with an IVY from TP.
I would say my implementation of the DAC is superior to the first buffalo run (lot more on decoupling, trace layng, using 4 layers etc etc) but that has yet to be proven...
And it won't be, as using outboard I/V WON'T give you the max specs given by ESS. I reached 120dBA dynamic range though, so not bad anyway. Given that bottleneck, I suppose the first TP run and my DAC are of very close performance (but they do not use the same generation ESS chips, that are claimed to have a different sound).
So I can tell you about an ESS 32 bits dac chip and the IVY used together.
I didn't liked the IVY. I checked it up with an audioprecision, it gave me 2dBA more dynamic than voltage mode, THD seemed the same, but it was not the same distortion as voltage mode.
The IVY was less punchy, less treble extension. Probably less fatiguing too on the other hand. A more "standard" sound. Be warned this is purely subjective, you might prefer it!
But that is, on seperate boards! The new Buffalo probably addresses this problem.
I would say: really go to the new Buffalo. I hadn't the chance to hear it, but it looks like a killer, and, as a designer working around the same chip, I see reasons why it may be very good. Of course, has to be tested and listened to.
Well after all that (unintersting?) talk, let me tell you the ESS based DACs are of ultra high quality. They are so neutral than you may think the sound is colored -not at all, it's the dac you are comparing it to that's not keeping up.
Expect very large treble extension, ultra high quality low sound (man you have to listen to some music with real infrabass with that dac!), the best stereo imaging I heard so far, and no veil (probably thanks to jitter reduction).
Dynamic Range is a good thing, but what makes these DACs different are their THD, IMD and other distortion related qualities. That's why you can't beat them.
Please note that, being absolutely neutral, you can play anything from techno to classical through it.
My 2 cents
Nicolas
(sound engineer, and working on boards based on ESS products)
I would say my implementation of the DAC is superior to the first buffalo run (lot more on decoupling, trace layng, using 4 layers etc etc) but that has yet to be proven...
And it won't be, as using outboard I/V WON'T give you the max specs given by ESS. I reached 120dBA dynamic range though, so not bad anyway. Given that bottleneck, I suppose the first TP run and my DAC are of very close performance (but they do not use the same generation ESS chips, that are claimed to have a different sound).
So I can tell you about an ESS 32 bits dac chip and the IVY used together.
I didn't liked the IVY. I checked it up with an audioprecision, it gave me 2dBA more dynamic than voltage mode, THD seemed the same, but it was not the same distortion as voltage mode.
The IVY was less punchy, less treble extension. Probably less fatiguing too on the other hand. A more "standard" sound. Be warned this is purely subjective, you might prefer it!
But that is, on seperate boards! The new Buffalo probably addresses this problem.
I would say: really go to the new Buffalo. I hadn't the chance to hear it, but it looks like a killer, and, as a designer working around the same chip, I see reasons why it may be very good. Of course, has to be tested and listened to.
Well after all that (unintersting?) talk, let me tell you the ESS based DACs are of ultra high quality. They are so neutral than you may think the sound is colored -not at all, it's the dac you are comparing it to that's not keeping up.
Expect very large treble extension, ultra high quality low sound (man you have to listen to some music with real infrabass with that dac!), the best stereo imaging I heard so far, and no veil (probably thanks to jitter reduction).
Dynamic Range is a good thing, but what makes these DACs different are their THD, IMD and other distortion related qualities. That's why you can't beat them.
Please note that, being absolutely neutral, you can play anything from techno to classical through it.
My 2 cents
Nicolas
(sound engineer, and working on boards based on ESS products)
"Dynamic Range is a good thing, but what makes these DACs different are their THD, IMD and other distortion related qualities. That's why you can't beat them."
that sounds like a whole lot of (...) to me, d-s dac are inherently linear and grant those numbers by design . Also Dustin admitted 128 DNR using his FPGA prototype while the first dac samples werent as good , hm. So I consider this chip a mixture of 95% DSP, and 5% delta sigma ,DEM.
that sounds like a whole lot of (...) to me, d-s dac are inherently linear and grant those numbers by design . Also Dustin admitted 128 DNR using his FPGA prototype while the first dac samples werent as good , hm. So I consider this chip a mixture of 95% DSP, and 5% delta sigma ,DEM.
Last edited:
Delta Sigma may be linear by design, just read at the datasheet and you'll find that THD of the ESS part is lower than any other DAC chip, would it be delta/sigma or anything else.
Especially when it was just out. For comparison, I'll use the last ESS 32 bits:
Now, consider the new 1792/1794 TI part. It has 132dB DNR in mono mode vs 135dB DNR for the ESS. But it is -108dB THD+N minimum, increasing with sampling frequency, when the ESS is -120dB constant. At 192kHz, the TI's THD+N went up to -96dB
And that is the last TI beast to compete with the ESS. WM8741 is -100dB THD+N.
The 32 bits PCM1795 is 123dB DNR and -106dB THD+N at best, down to -96dB THD+N.
The well known pcm1704 is "120dB SNR, 112dB DNR" (if someone can explain me how a peak can be below a mean...) and -102dB THD at best (K grade), down to -90dB (U grade, worst spec).
And, appart from the amount of THD, is the spectrum and IMD. ESS claims their dac are excellent at this, and show graphs. I didn't measured it, but clearly heard it.
All this may seem very low number, but I hear distortion down to very low level, as do my sound engineer and musicians friends. As to the average joe... I don't know, but difference sounds really obvious.
And before someone jumps in: yes, no amp is capable of handling these values and even are a few decades away. Difference still is obvious: I suppose this is because the distortion created by a DAC and all his DSP is not at all the same as the one produced by an amplifier.
Especially when it was just out. For comparison, I'll use the last ESS 32 bits:
Now, consider the new 1792/1794 TI part. It has 132dB DNR in mono mode vs 135dB DNR for the ESS. But it is -108dB THD+N minimum, increasing with sampling frequency, when the ESS is -120dB constant. At 192kHz, the TI's THD+N went up to -96dB
And that is the last TI beast to compete with the ESS. WM8741 is -100dB THD+N.
The 32 bits PCM1795 is 123dB DNR and -106dB THD+N at best, down to -96dB THD+N.
The well known pcm1704 is "120dB SNR, 112dB DNR" (if someone can explain me how a peak can be below a mean...) and -102dB THD at best (K grade), down to -90dB (U grade, worst spec).
And, appart from the amount of THD, is the spectrum and IMD. ESS claims their dac are excellent at this, and show graphs. I didn't measured it, but clearly heard it.
All this may seem very low number, but I hear distortion down to very low level, as do my sound engineer and musicians friends. As to the average joe... I don't know, but difference sounds really obvious.
And before someone jumps in: yes, no amp is capable of handling these values and even are a few decades away. Difference still is obvious: I suppose this is because the distortion created by a DAC and all his DSP is not at all the same as the one produced by an amplifier.
Last edited:
you should compare it to AD1955 and those "scrambler" implementations like Anagram, or http://www.accuphase.com/pdf/dp-67_e.pdf
pcm1792 = Burr Brown heritage, "advanced segment" (partly multibit)
Pcm1704 = whole lot of multibit
not really apples to apples.
Im sad the ESS marketing department is free to roam and inject this FUD . Why cant they tell , hey, it really worths the money ? I think it does, but those -120 numbers are ...well... (...)
pcm1792 = Burr Brown heritage, "advanced segment" (partly multibit)
Pcm1704 = whole lot of multibit
not really apples to apples.
Im sad the ESS marketing department is free to roam and inject this FUD . Why cant they tell , hey, it really worths the money ? I think it does, but those -120 numbers are ...well... (...)
Last edited:
Well, Twisted Pear measured -118 or -117 dB THD on their latest implementation AFAIR...
And they admit (and TP and I do too) that the limit on THD is the I/V converter.
Regarding the Accuphase it is speced at 0.001% worst case. That is -100dB! I
THD+N of the AD1955 is -110dB.
Anagram Quad is 126dB DNR, -116dN THD, that is very close to ESS (and not yet up to!). They don't exactly cost the same, too xD
Numbers published by ESS are real and correct, even if I could not go up to the max specs I knew why and measured it with an audioprecision, TP has got better results and reasons are obvious why. Even in suboptimal setups like mine or first buffalo, the results you obtain, measure, and hear, remain superior to most other DACs. Not anagram in these case, but the Anagram will have the same requirements WRT design and layout, and will even be more regarding at the quality of the input digital signal, the ESS decoding stage being quite one-of-a-kind.
Please understand, I don't sell and do not work for ESS nor TP, I only had a few boards for spare that I sold a fair price, and I don't have any more so I really don't care. Just do not point the ESS marketing department for lying, when they just didn't. 135dBA mono mode (-60dB FS) -120THD+N (0dB FS) are facts, measured by an AP, and if you have precise enough instrumentation stuff, you should be able to measure it too.
EDIT: I seen the -144dB THD+N number on the scrambler 2 datasheet, that seems strange, and does not correspond to what the AP measurements show at the end of the paper...
And they admit (and TP and I do too) that the limit on THD is the I/V converter.
Regarding the Accuphase it is speced at 0.001% worst case. That is -100dB! I
THD+N of the AD1955 is -110dB.
Anagram Quad is 126dB DNR, -116dN THD, that is very close to ESS (and not yet up to!). They don't exactly cost the same, too xD
Numbers published by ESS are real and correct, even if I could not go up to the max specs I knew why and measured it with an audioprecision, TP has got better results and reasons are obvious why. Even in suboptimal setups like mine or first buffalo, the results you obtain, measure, and hear, remain superior to most other DACs. Not anagram in these case, but the Anagram will have the same requirements WRT design and layout, and will even be more regarding at the quality of the input digital signal, the ESS decoding stage being quite one-of-a-kind.
Please understand, I don't sell and do not work for ESS nor TP, I only had a few boards for spare that I sold a fair price, and I don't have any more so I really don't care. Just do not point the ESS marketing department for lying, when they just didn't. 135dBA mono mode (-60dB FS) -120THD+N (0dB FS) are facts, measured by an AP, and if you have precise enough instrumentation stuff, you should be able to measure it too.
EDIT: I seen the -144dB THD+N number on the scrambler 2 datasheet, that seems strange, and does not correspond to what the AP measurements show at the end of the paper...
Last edited:
- Status
- Not open for further replies.