Maybe you can dispel a notion that gets talked about around DAC chips: implementation. For example, I have heard the Dangerous Source Convert-2 DAC uses the same AD1955 chips as the DC-1, however since the implementation of those chips is different, you have two DACs that don't sound identical even when using the same chip. Is this audiophile BS? Cause I read impressions from other audiophiles who access to a wide assortment of DACs and they use all that fancy audiophile lingo like: "this is more resolving, holographic, better separation of instruments, better timbre, etc".
When you say "a more accurate output", how do you go about assessing that? What makes an accurate output poor or good, or very good, or excellent? It seems to me the analog boards themselves would be crucial in this, yes?
And while we are at it, just how much of the internal components of the DAC play a role in how that sounds from the analog output? Is it 60% DAC chip, then 30% of that in the digital filters, then 10% in whatever else? With so many DACs on the market, why should we as consumers, who are after high fidelity, not consider all the internal components and how they are implemented?
I certainly don't have enough exposure with DACs as you do Keith, this is why I'm asking you these questions. I would trust your answers to these questions than 99% other random people over the internet cause you have the experience.
>>> That is entirely true - although the details are very complex and it's not possible to assign percentage values. For example, the AD1955 includes built-in oversampling filters - which are actually quite good (the guys who designed the chip do know what they're doing), but can also be used with external filters. If you use the internal filters, you have a few options, But, if you use external filters, and design your own, you could end up with a very good result - or a very bad result. So, at least in terms of that aspect of the design, DACs that use an AD1955, and use the internal filters, tend to sound somewhat similar, while DACs that go the external-filter route are all over the map. Other areas of the circuitry, like the power supplies, and the analog output circuits, are also quite important, and you have lots of options to get them right - or not.
>>> There's also an aspect of this that is purely market driven. If you look in the AD1955 Application Notes you'll find a schematic for how the manufacturer suggests you use their chip. They will even give you layouts you can use to manufacture the circuit board - free of charge and free to use. And, if you follow that schematic carefully, you will end up with a DAC that performs extermely well. However, it will also perform, and sound, very much like every other DAC, made by everyone else, who follows that schematic. So, while this DAC will work really well, it's difficult to sell it. ("We use the AD1955, carefully following the AD plans, and it works really well, just like identical units made by a dozen of our competitors. We hope you like our paint job better." This is not a formula for a successful product... unless you want to join the proverbial race to the bottom and sell yours cheaper.)
>>> It would be most accurate to say that the DAC chip can be a limiting factor, but most modern high quality DAC chips are capable of delivering very good, and very similar, performance - if you design everything else correctly. One slight but notable exception is the ESS Sabre DAC chips. For a variety of reasons, while most other high quality DAC chips tend to sound quite similar, if you use Sabre chips according to manufacturer's recommendations, they tend to have a slight "house sound", which makes them sound somewhat different than most others (they tend to emphasize high frequency detail - and so to sound either "very detailed" or "somewhat grainy and etched" - depending on whether you prefer that sound or not.) Not all units that use Sabre DAC chips have this sound, or to the same degree, and it can be avoided, but it is audible to some degree in many products that use the Sabre line of DAC chips.
>>> The way you define "a more accurate output" is very simple.... You want what comes out of your DAC to sound as much like the original analog source as possible. Unfortunately, since we rarely have access to the original analog source, this can be very difficult to achieve in practice. And, no, an analog tape, or a vinyl album, is simply a poor quality copy of the analog source.... and not the original. (You can simply consider how accurately a digital copy of an analog album sounds like the album... as long as you have no illusions beyond that.)
Thank you for the PDF document, I will review it this weekend.
I believe based on the AD1955 spec sheet, the oversampling rate is 8x. Unless that rate is variable based on the incoming data and how that chip is programmed, as you said.
>>> Yes, on the AD1955, you have many options for exactly how this is handled, and they are programmed by the firmware you use with it. This means that, unless you actually dissect the source code of the firmware, or measure it in operation, you're not going to be able to tell which options are being used. To be quite honest, I'm not sure exactly what options we chose on the DC-1 (and I've never considered it to be terribly important). In general, one tries to select the highest operating sample rate that the chip handles well, whcih results in different multiples being applied to different incoming sample rates.
Right, the AD1955 is a multi-bit DS chip which I assumed handled 4-5 bits. But maybe this is part of my contention with DS chips in the first place in that if you are sending 16 bits of data to the DAC chip, why break that down into fewer bits? I know this is asking more so the methodology of DS (or SD), but why wouldn't you want to keep the original 16 bit samples?
>>> The reason for doing this is simple. Because of how DACs work internally, as you design a DAC chip to handle more bits, it becomes exponentially more difficult to design one that has good linearity, and the same relationship holds true as the frequency increases. So, for example, the multi-bit DAC chip used in Yggdrasil, which is barely able to deliver 21 bits of linearity, costs around $100 retail - each. In contrast, the actual DAC part of the chip inside the AD1955 handles four or five bits, with near perfect linearity, at a much higher speed, for a tiny fraction of that cost. And, even when you count all the extra processing necessary to divide a 24-bit sample into 5-bit chunks, submit them to the 5-bit DAC, and then combine them again afterwards, you end up with far better overall performance for a small fraction of the cost. Even beyond that, by using the same small, fast, accurate, DAC to handle the data in smaller chunks, then merge the chunks together afterwards, the small errors that do occur tend to average out, increasing the accuracy and linearity even further. So, putting aside several other aspects of performance, you can get a D-S DAC chip for $2 that delivers better linearity at 24 bits than the $100 chips they use in Yggdrasil can deliver at 21 bits. (We're not talking abut errors that happen in the five bit chip but not in the 16- bit chip. We're talking about errors that simply go uncorrected in the 16 bit chip, are smaller in the 5-bit chip to begin with, then are reduced even further when they are corrected.)
>>> When you start talking about R&D costs and such you need to be very careful of context. The military spent a lot of money developing that DAC chip for certain specific purposes... but those purposes did not include using it for audio equipment. One of the reasons Schiit audio had to spend so much money on their own R&D was that the chip they chose wasn't really intended for audio applications at all. As a result, they had to expend a huge amount of effort "adapting it" and "getting it to fit". In contrast, the AD1955 was also developed by a major chip design company, who spent a lot of money and resources perfecting it, and it
WAS designed specifically for use in audio equipment - so it works well without all that extra effort..
>>> What you need to try very hard to do is to separate "aesthetic considerations" from practical ones. The math involved in a D-S DAC probably sounds beautiful and elegant to a mathematician. However, to the rest of us, it seems like the process involves an awful lot of unnecessarily complex, and somewhat dubious, mathematical manipulations. What you need to remember is that most of that happens for a reason. There isn't some evil scientist who just enjoys mincing numbers and rearranging them. The reason D-S DACs were developed in the first place was that multi-bit DACs are very expensive and don't actually work very well. As it so happens, in modern digital circuitry, "math is cheap". A computer that can perform a billion calculations a second costs less than a decent 16 bit DAC. And, by leveraging that fact, and using math to get the best performance from parts that are very economical, a D-S DAC is able to deliver superior performance for a much lower price. (And it's interesting how many audiophile companies try so hard to convince us that getting better performance for a lower price is somehow cheating - or that "there must be a catch. )
The more I read about the delta-sigma (and sigma-delta) method, the more it makes me wary to keep in my system. Again all depends on the results like you said. Even the Convert-2 is very good but that's more so the implementation. I don't know, I've heard R2R ladder DACs and I'm more drawn to the overall sound. Something like a Yggdrasil, which is a hybrid R2R, requires some serious R&D. I mean they are using AD chips that the military uses for targeting systems or medical use.
On the other hand, DS is cheap to buy, design and build.
That's a funny analogy with the chicken nuggets and I wonder if the way DS minces the data produces that hashy treble glare. I've yet to hear a good Sabre implementation that doesn't have that nasty treble glare. I'm not a fan. Sure those DAC products may measure well but they sound like ass.
>>> That's one of those "perception issues".... in other words you are being played. Sabre DACs have "that treble glare" because their designers wanted them to sound that way... it is essentially their trademark sound. (And, yes, a lot of people actually
like that sound.) And, likewise, many manufacturers of multi-bit DACs go out of the way to tailor the way
they sound to meet the expectations of
their customers. Nobody is going to make a bright sounding multi-bit DAC... because, well, nobody would buy it.
>>> There are in fact situations where certain technologies lend themselves to specific types of flaws... for example, with true non-oversampling DACs, it is very difficult to avoid an audible roll off in high frequency response. However, most of the differences you hear nowadays are more likely due to either deliberate "tailoring" of the sound, or are side effects of specific filter choices, made either for that reason, or because they serve some other purpose. (For example, many designers are quite convinced that filters with less pre-ringing sound more natural, and filters with that characteristic often sound different than other types of filters.) The "real audible differences between D-S and multi-bit DACs", if they exist at all, are incredibly subtle.... Also, lest we forget, virtually any piece of digital audio you ever hear will have been converted from analog to digital initially using some form of D-S technology (multi-bit A/D converters have been considered to be pretty much obsolete for a long time - and many of the original ones performed very poorly).
Interesting. I don't even use the USB input in the DC-1 (my PC chain is USB->Schiit Eitr->RCA/BNC coaxial digital cable->BNC connection on DC-1). Most USB audio solutions I've heard are mostly junk, either they introduce too much haze, audible pops, clicks, glitches, etc. Compared to a good AES/SPDIF transport, USB can't match it. To use your analogy, it's like sending the head chef at a Michelin star restaurant chicken nuggets and have him try to create a dish worthy enough not to earn a scathing review by the food critic sitting at the table. I try to avoid USB at all possible. Now supposedly there is some rumors that Schiit has finally figured out a USB solution that will surpass SPDIF, but like I said we'll see when it arrives.
I overlooked that the DC-1 had an ASRC option, because I usually just ignore any USB inputs in general when it comes to audio gear.
So from what you are saying, the ASRC will ignore all incoming jitter and then re-samples as if the source inputted the original samples as if it was jitter-free. This sounds pretty good, so this makes me wonder if I should ditch my USB decrapifier solution.
What are your thoughts on ASRC, in particular to sonic quality, Keith?
>>> In theory the ASRC should remove any audible effects of jitter... and so render any sort of "decrapifier" largely moot... and I wouldn;t bother with one. However, I cannot rule out the possibility that the decrapifier
could reduce USB ground noise, or do something else which
could produce some sort of audible benefit. And, even though the effect of the ASRC should be inaudible on a signal that is clean to begin with, they do sometimes seem to produce a vary slightly audible change. (In theory the ASRC delivers and outut that is "what it should be" to within about -130 dB.)
Correct about the recordings I have exists at a certain sample rate.
What I meant by setting my music library to a certain rate is I don't want my media software to upsample into a certain rate I don't want it to. So if the Bifrost Multibit retains the original 16-bit samples, I will make sure the output is set to 16/44.1k. If I have files that are 24 bit, I'll let the software downsample them...in theory, I suppose. But yes I do not want my software to be doing any conversion except what should be done inside the DAC.
Oh and if I use ASRC in the DC-1, since the original samples are being discarded, none of this would matter.
>>> That's not exactly true. Every sample rate conversion involves some filtering... so it always makes sense to avoid unnecessary conversions.
(If the software introduces a small error when it resamples, then the ASRC introduces another tiny error, you now have twice as much error.)
In fact, when downsampling is performed by high quality software, the result is usually inaudible... )
Likewise, since upsampling cannot create new information, it cannot improve the quality of your recording.
However, since the process incolves filtering, which may produce small but audible changes...
it's possible that, in a particular case, you may find the result to sound different and potentially pleasing.
Thank you kindly for this information.
As always Keith, thank you so much for your thorough replies. They have been enormously helpful. I hope Emotiva is paying you well.
)
