@audioengr
It is a simple matter to use upsampling so that a gentle linear phase anti-alias filter can sit far away from the audio band and where it won’t be audible. There are ways to trick most DAC chips to employ this solution and avoid or bypass the limitations of the DAC chip built-in filter options. |
Well mathematically there is nothing wrong with upsampling and it measures much better in most cases due to the less aggressive filter and because higher sample rates tend to improve DAC linearity. Perhaps it is the way inter-sample overs are handled on upsampling DACs - since most pop rock digital music has peaks greater than 0 dbfs that cause errors in any DAC that upsamples without having extra bit depth to handle these peaks. This link below describes one error mechanism (perhaps there are others) that begin to explain why upsampling on most DACs sounds worse to you. https://benchmarkmedia.com/blogs/application_notes/tagged/inter-sample-overshttps://benchmarkmedia.com/blogs/application_notes/intersample-overs-in-cd-recordingsBenchmark make the following claim about DAC chips "Every D/A chip and SRC chip that we have tested here at Benchmark has an intersample clipping problem!" So unless manufacturers make an effort to find a solution to this oversampling DAC chip limitation then you have at least one pervasive problem with these DACs - this might be why Georgehifi dislikes all of these designs versus ladder DAC with no upsampling. Unlike many small effects at the LSB which arguably might not be audible, this occurs inter-sample over errirs occur at the MSB - so for sure it is easily audible on most pop/rock CD’s produced in the last 20 years. This red noisy signal sure looks easily audible versus the corrected green curve to me, https://cdn.shopify.com/s/files/1/0321/7609/files/Inter-Sample_Overs.JPG?v=1469682095 |
@audioengr
All the problems will be low level as the digital hash from inter-sample overs may not be audible until just after the transient. A bit like amplifier clipping where you don't hear problems on the transient itself as much as the distortion over everything else musical going on (harshness in the mid range in particular).
The last link is to a plot showing what looks like typical low level clipping distortion - distortion is low (about 60 to 100 db down) but it is broadband right across the entire spectrum which makes it very likely to be audible as harshness or glare. Notice that there is actually a distortion peak at 80Hz - and we all know that an 80Hz tone lasts an eternity compared to mid range stuff - so even if the transient is a cymbal crash at 11 KHz - you have distortion smeared broadband over at least 1/80 secs (1 cycle of 80 Hz) or more than 12 msec (which is a lot).
|
+1 Steve
In addition, it needs to be better understood that the ESS 9028 DAC is a massively parallel Delta-sigma DAC - in some implementations 512 delta-sigmas are summed in various ways to create a signal. Basically this ESS chip is a hybrid between R2R and a single 1 bit delta sigma. You can describe these chips as a massive R ladder DAC with all steps the same - instead of adding 1 + 2 + 4 + 8 (four discrete resistors on an R2R) to get a value of 15, the ESS DAC will sum up 15 discrete 1 bit delta sigmas to achieve a value of 15.
The benefit of massive single bit delta sigma DACs is that through a randomized selection process the DAC chooses randomly which delta-sigmas are summed to make the required valued. This leads to huge advantages in linearity and signal to noise over an R2R!!!
Hybrid DACs like ESS are simply better technology than an R2R by a significant margin as the specifications like THD+N show. |
All this “facsimile” and “guessing” is just marketing BS at best and misleading nonsense at worst. There is no guessing. Redbook defines the signal up to 22KHz perfectly and there is no guess when you oversample. In fact the oversample file contains only as much information and all the information as in the original file and no more as it adds nothing to the signal. It also contains the entire data in the Redbook file - nothing has been thrown out at all (as would happen in a facsimile).
The bits are being converted to analog. It is the accuracy of this entire conversion that is important and NOT the specific steps in the methodology. Manufacturers want you to believe that their specific methodology is better by appealing to the old school notion that doing things in a simple archaic fashion without technology is best. If this logic were correct we would still be using an abacus or slide rule.
|
The best explanation is the “measuring cup analogy” in Stereophile 1990. Nothing is “thrown away” - Delta-Sigma is just a different (more accurate measuring cup) method for conversion to analog. R2R methodology lost the technical battle about 20 years ago - R2R is too expensive and too inaccurate to be a leader in DAC technology. R2R relies on impossible accuracy in analog devices whilst Delta-Sigma elegantly uses mathematics to achieve superior performance in a simple robust way. The latest hybrid ESS approach - massively parallel one-bit DACs all on the same DAC chip with random selection to reduce noise is some 30dB+ better in THD+N - massively outperforming R2R for a fraction of the cost. Technical progress is wonderful! As far as I can see, the marketing of R2R is highly misleading - full of all kinds of false insinuations - “throwing away data” - simply not true, not bit-perfect (when delta-sigma is far more accurate), “facsimile” - ironically R2R is actually the poor quality analog facsimile, “guessing” - again ironic because R2R actually has built-in massive non-linearities that makes the accuracy no better than a guess where delta-sigma is many times more accurate. Here is a basic description of the different approaches: ”The same reasoning underlies the development of bit-stream decoders. The problem with a conventional digital/analog converter (DAC) is that its operation involves mainly analog processes and is therefore approximate. A 16-bit DAC contains a precision current source and an array of 16 switches. Each switch is connected to a resistor, and the resistors are supposed to be scaled in exact 2:1 ratios so that each switch, when opened, will contribute exactly twice as much current to the output as the switch below it. The switches are controlled by the 16-bit codes from the CD; thus by opening and closing in various combinations, a total of 65,536 different output values can be generated. But the topmost switch (the most-significant bit, or MSB) contributes 32,768 times as much current as the least-significant bit (LSB). If the MSB current is in error by as little as one part in 32,768, the effect of the LSB is swamped. In most CD players it is; few 16-bit DACs operate to better than 15-bit accuracy. The practical result is that most CD players are non-linear at very low signal levels, reproducing small signals at the wrong levels and with added distortion. Keep in mind that this problem arises not from the digital code itself but from small errors in an analog quantity—the current produced by the DAC for the several most-significant bits. For comparison, imagine that you were assigned to fill a bucket with a known amount of water, using measuring cups varying in size from one ounce to 64 ounces. Even if you use care in filling the largest cup, it might contain 63.7 or 64.5 ounces instead of 64; you can’t be sure that it contains exactly 64 times as much water as the smallest cup. But there is a way to obtain an exact result: use only the one-ounce cup, and transfer its contents to the bucket 64 times. The capacity of the cup may not be exactly one ounce, but as long as you fill it the same way each time, the total amount transferred will be proportional to the number of refills—an exactly linear relationship. This is the idea behind 1-bit decoding. In place of a method whose result depended on slightly uncertain analog quantities (the currents in the DAC), we have adopted a simple counting scheme—a purely digital process. Of course with a small cup you’ll have to work fast, but in modern digital electronics that’s not an obstacle.” Read more at https://www.stereophile.com/content/pdm-pwm-delta-sigma-1-bit-dacs-peter-w-mitchell#44ZdP17jRzTtUGor... |
Per Steve’s last post above, this is exactly why hybrid ESS 9028 designs are way ahead of everything.
The latest ESS designs are halfway between the R2R style and delta - sigma! The 9028 a 6 bit delta sigma by virtue of hundreds of 1 bit delta sigma on the chip!!!
|
+1 on the filters - Benchmark overrides the ESS chip filters by ensuring the chip operates at 211KHz - even for Redbook - this mean the filters actually sit far above the audio (instead of close to the Nyquist ) and be extremely gentle such that they have minimal affect on the audio band (flattest response without the usual ripple) |
@willemj You are forgetting that no DAC except the Benchmark DAC2 and 3 properly handle Redbook intersample overs that are common even on high quality CD production like Steely Dan. As it stands, I am aware of only two solutions - 1) Use a Benchmark DAC 2) Use Roon - they have implemented a 3 dB drop in digital signal level which appears to be default setting in order to prevent DACs from clipping all the time with inter sample overs see this https://benchmarkmedia.com/blogs/application_notes/tagged/dac3The fact that that not even one other DAC manufacturer corrects this issue is shocking. Clearly most DACs are just thrown together using a chip and the chip user manual and without any real testing or engineering! How else can you explain that nobody else except Roon and Benchmark care about 1129 clipping distortion events on a single high quality audiophile track “Gaslighting Abbie”! |
@gdhal
I did not read Archmago entire blog post but I agree with this “ideal filter is linear phase”! Always has been and always will be!
Archmago states:
“The key here is to remember that within a properly bandwidth limited signal where all the frequencies are below Nyquist, a linear phase FIR filter actually does not create ringing regardless of the impulse response appearance. As I have said in the previous weeks, any decent recording will follow this rule. And if it does, then the ideal filter to use is clearly a linear phase, sharp filter that can reconstruct all the frequencies in the audio data with essentially ideal temporal resolution.”
Different filters on DACs seems to be a popular thing right now - perhaps a mania. It sounds like “more” from a marketing perspective when in fact it is often less. Minimum phase filters for example are just plain WRONG! The MQA style anti-ringing minimum phase filters make no sense. For audio reproduction, it was established more than 30 years ago that linear phase is what you need. Linear phase preserves all the relationships between the multitude of frequencies that make up the sound - use any other type of filter and you ruin the timbre!!
Benchmark of course use Linear Phase only and they disable all of the other fancy distorting filters on the ESS 9028 chip. Furthermore they upsample to 211KHz so that the filter is tricked in to attenuating only around 105.5 KHz and above - way beyond Redbook audio band and therefore a gentle filter at a high corner frequency ensure no audible effect!! Meanwhile shameless marketing departments are offering the filter options on the newer chips to their user as a “feature” even though several of the filters are plain wrong and their implementation leaves a lot to be desired compared to the Benchmark approach (and also an approach used by Steve Audioengr in his own DAC design).
|
@willemj
You are correct - it will be less of an issue but still an issue. A lot of reissues of classical and jazz are boosted to clipping level too but you are correct they are often less compressed overall.
The CD Homage To Duke by David Grusin is one of the best jazz recordings I have heard and it is recorded at a very low average level with tremendous dynamics (needs to be cranked 3x higher on the volume dial compared too all my other classical and jazz). |
|
@gdhal
For highest fidelity and accuracy of Redbook, I would recommend to do this
1) Use Roon with its default 3 dB reduction to eliminate the all too common intersample over clipping errors.
2) Use Roon or another software to upsample the Redbook data to 192KHz and 24 bit
3) Select the Linear Phase Slow filter option of your DAC
This approach will minimize the filter affect on the CD audio band to a negligible level because you have pushed the anti-aliasing filter corner way up to 96KHz which is much farther from the audio band than 22 kHz which is what your DAC will normally use if fed redbook. |
@gdhal
Thanks for this awesome gift.
I would like to hear Tower of Power live in 1975.
This was their best line up ever with Lenny Williams on vocals and Chester Thompson on keyboards. They are probably the best band in the world to see or hear live! Sooo excited!!
|
@gdhal
The logic for upsampling Redbook is well documented and justified. Redbook requires a Nyquist filter at 22KHz to prevent aliasing or image noise from 22 to 44 KHz from being included in the analog output signal. Unfortunately a sharp filter at 22 KHz is guaranteed to affect the audible sound due to the proximity of this sharp filter to the audio band. By upsampling you preserve the entire original digital information contained in redbook but you push the Nyquist far up to 96KHz for a 192KHz sample rate. This makes filtering so much easier and entirely benign. Going up to 24 bit also preserves the entire Redbook signal.
The key understanding is that upsampling preserves ALL the original information and so does an increase in bit depth from 16 to 24.
This is very different from down sampling which can alter the original audio (stuff is thrown away) and normally down sampling requires careful processing and filtering to preserve the audio (dithering being a very important processing step in down sampling, as simple truncation is highly distorting)
The key to understand is that upsampling is benign but it makes the job of filtering aliased or ghost frequencies much much more accurate. Every DAC must filter out these “aliasing” artifacts as it is inherent in sampling theory (to accurately capture a signal you need a minimum of two samples per wavelength) - anything that is sampled less than twice will appear as “ghosts” or aliasing artifacts (and these artifacts sit at frequencies above the Nyquist which is half the sample rate frequency). The sharp filters at Nyquist are essential to ensure these “ghosts” do not appear in the reconstructed analog output, however, a sharp filter close to the audible range is less than ideal as it affects the audible range also!
Apart from this sharp near audio band filtering issue due to the low sample rate, CD Redbook is a near perfect audio distribution format - especially when carefully produced. Only recently has analog equipment achieved the equivalent of 21 bit resolution making 16 bits a further limitation of CD redbook. |
@gdhal Never seen the Grateful dead but I watched a movie documentary and I love their live jams. Listen to them on Tidal a lot. Tom Petty does a great live jam called Two Men Talkin’ that seems to be a tribute to Grateful Dead - do you know what is the original Grateful Dead jam this is based on? https://youtu.be/HU9VBQBFQ_w |
@gdhal
That would be it “The other one”. Petty played Friend of the Devil live often also. Nearly every musician is a fan of Jerry Garcia - no denying tremendous talent.
@maplegrovemusic
I have college degrees in engineering and physics. Don’t use any of it for work but I remember stuff and read up on audio regularly. I was also in AES Audio Engineering Society for many years. Time Series Analysis used in Geophysics is exactly the same mathematics as digital audio and I took courses just on that! I studied atmospheric physics in depth too but I better stop there... as nobody likes to hear rational discussion on that topic which has become a new cult religion.
|
@gdhal Another advantage of upsampling redbook is on a ladder resistor network DAC or multi-bit DAC. Upsampling with Roon or another software should shift ghosts much higher in frequency - where the ghost frequencies can act a bit like random noise which can actually help linearize the DAC response. Of course the ghosts are ultimately filtered out by the final filter but not until after they have helped improve linearity. The way to think of this is that the high frequencies force the DAC to use a wildly different bank of resistors on nearly every sample or at a minimum with much more variety of network choice than would be used if there was only low frequency CD redbook data - this means individual resistor network non-linearities are converted to uncorrelated noise which is orders of magnitude less audible! Here is the full theory - in fact the author claims this is the primary benefit of upsampling! http://www.mlssa.com/pdf/Upsampling-theory-rev-2.pdf |
Lol - Dark matter? Try Brian May lead guitarist of Queen / he is actually an astrophysicist! I know almost nothing about that. I did read Stephen Hawkings original Black Hole Radiation paper when in high school but never pursued that side of Physics as fascinating as it is - I must have read that in the days when Scientific American was actually Scientific - it has now devoted into populist Cult Cargo science as Feynman would say if he was still alive. |
“I don't know how many of you have played with power cords. But they make a bigger difference than one would think.”
Absolutely! And you hit it on the nail “they make a bigger difference than one would think”!
So sad that even most high end equipment has such poorly designed power supplies that special power cords become necessary! Myself I simply get rid of electronics that doesn’t work properly without these kind of band aids. |
So far I have not found any sound quality problems with Roon - you may want to try it. I have known for a long time that most softwares (especially Amara Jriver and others) tend to colour the sound. I don’t know if Steve is aware but the SRC converters in Apple core audio are terrible - it is very important to make sure no conversion takes place in core audio (iTunes has its own much better SRC converter built in to the software). I used a software called Bitperfect to ensure Core Audio automatically matched the file sample rate back when I used iTunes. Even Tidal has implemented the ability to control Apple core audio to ensure good sound quality.
My overall approach is to totally avoid all SRC in software and to disable all software processing (even volume control) and filtering - perhaps that is why I don’t have all the typical problems everyone reports. Software engineers or programmers are like DAC designers - most haven’t a clue when it comes to digital signal time series processing (these are full semester courses in 3rd or 4th year engineering). Software can work one day and then an update comes along and trouble happens and one might not notice....best to turn everything off. |
+1 @beewax
I agree. I would suggest to use the cleanest high performance DAC I could find for the most resolving line level source signal and then use a tube preamp and roll tubes to taste for the musical taste. This approach is much more flexible. Buying a DAC with tubes built in is kind of limiting or restricting yourself... |
@beewax
+1 I agree. My strategy if I were to try to put some methodology to the task would be to
1) go for the most jitter immune, cleanest, lowest distortion and highest SNR DAC available. This would obviously be a new DAC within the last year or two. It might be too lean and analytical for the desired goal of “musicality” and engagement tie tapping but you start with the bare truth and a reference signal.
2) Get the most resolving SS power amp - Benchmark ABH2 looks impressive. Again the power amp should be powerful and transparent - a Bryston, Pass Labs, Krell - there are plenty of choices.
3) Roll preamps and roll preamp tubes until you find the desired sound.
I think the strategy of using using every component (DAC, Preamp and power amp) to ALL create or add musical warmth is a never ending uncontrolled dog’s breakfast.
Another alternative would be to stick with one manufacturer - go all ARC for example - at least there is a good chance that the equipment has synergy to begin with, |
@audioengr
+1 Yes it is the filtering that creates the dead effect. Still one of the most problematic issues with any DAC. Marantz house sound is dark and dead. I have heard plenty of great sounding SACD but Marantz is not to my taste.
@georgehifi
DS or R-2R can both sound fantastic if done well. DS currently has the edge in performance especially the hybrid designs that have 6 bits in depth. Old DS with only one bit depth suffered from high amounts of noise but was besutifully linear. R-2R has much less noise but suffers from linearity issues. The hybrid designs that are in the all the modern ESS chips are a great blend of the positives from both - they are highly linear and low noise too.
Modern DS chips are basically akin to a 6 bit R-2R - they function just like R-2R with their output level a function of the number of selected individual DS that are summed up to make up the output - exactly like a ladder DAC.
Noise is what drives the need for heavy filtering....
|
Perhaps it is the harmonic distortion from an R-2R DAC that listeners like - kind of like tubes - measures poorly but sounds subjectively better...
With two types of resistor perhaps the distortion is more in the even harmonics.
All I can say is that for sure the harmonic distortion is way higher on R-2R than DS DACs due to the non-linearities introduced by the finite accuracy of the ladder resistors.
If true, then even low cost R-2R might sound great because higher cost R-2R will likely have less harmonic distortion... |
Here is a thesis paper explaining why R-2R DACs have lots of harmonic distortion. Looks like the distortion is consistent across all harmonics. https//pdfs.semanticscholar.org/9a52/fcba6dd87371974b3c146dbb5a3a00ffee74.pdf These students were sponsored by NAIM to investigate ways to reduce this non-linearity harmonic distortion that is inherent to R-2R designs. The 2nd harmonic will give this kind of DAC a richer sound. The higher odd harmonics will tend give an R-2R a slightly harsh edgy sound in the treble. My guess is that R-2R DAC manufacturers will use filtering to roll off the highs in order to reduce the odd harmonic harshness in the upper treble. Subjectively this might be preferable to a higher performance DS DAC. |
@gdhal
Even harmonic distortion is pleasing. It is higher order odd harmonics above 7th which are not pleasing and harsh sounding.
It looks like the R-2R DAC has equal amounts of even and odd harmonic distortion. The 2nd harmonic being the most important. Above 7th odd harmonics is where it becomes disagreeable to the ear. If the audio is filtered at higher frequencies (most older NOS designs have gentle roll off from analog output filters) then the lower harmonics will dominate and it could sound subjectively excellent.
This might be the key reason R-2R sound could be preferred for the same reason tubes are highly regarded for their wonderful sound...
A complete lack of harmonic distortion is about as dry and clean sounding as it gets - the latest ESS DS chip has extremely low 2nd and 3rd harmonic distortion (-130dB in the mid range) which is exactly where you get most of the harmonious pleasing warmth in tubes.
|
+1 Dlcockrum
I think that one has the be careful about a DAC like the T&A DSD 8 when most of the sound is from what looks like poorly implemented filters.
I think the fixation on DACs and their sound is a mistake. Get a good high quality tube preamp with the right tubes and create the tailored sound you want. Simply Connect a high performance SS DAC - any number of Stereophile class A+ will do and many others that Stereophile hasnt reviewed.
Carefully crafted optimal sound does does not need to come out of one box and one box alone.
Synergy anyone?
A high synergy 10K system will beat a poorly matched 100K system. |
