Upsampling can make a difference in some rare cases when your dac works better with higher sampling rates. It also can make it sound worse. Most people that are hearing a difference if any almost allways assume the upsampling must be the better one. Higher numbers sound better right? Not! At least not allways. Sometimes added timing errors are mistaked with more details, happens quite a lot. Upsampling for the sake of better numbers is the dumbest thing to do if you don’t hear a difference.
Wow, I take it you two have not up sampled properly. I ain’t never going back I live in DSD512 world and always will. So much more organic and real, palpable and lifelike, dynamics, oh boy. Hangs with the best analog out there.
@Outlaw what dac are you running and what PC/Mac are you using? Have you tried lowering your up sampling to say DSD128 to see if the stuttering goes away? It could be PC related if you have an old i3 or low i5 processor. I do 512 but run new generation i7’s, or AMD 1700x or 1800x CPU’s.
What a load of BS.
Look at the several discussions/links here -
It just doesn’t make sense. So, if you take some vinyl and rip it to DSD512 it will sound better than the original vinyl? If you take a photograph at 10 megapixels and “upsample” it to 20 megapixels it will look better? (I realize not really the same thing, but it kinda is).
Not trying to start a fight here, but upsampling just does not make sense as a means to improve quality. If it is not there in the original, it can’t be magically restored.
Perhaps following posts will give you some new ideas why people like using the HQPlayer:
Well, I would have included @andybob since he has been using HQPlayer to upsample to DSD 512 for a long time (although I think he is going for 768 PCM lately). Still, I think the point Andy will make and specifically with HQPlayer, it is not just the upsampling, but more so the different filters that can be employed.
In most cases, the DAC performs some DSP to upsample the incoming signal to its maximum sample rate (more or less) regardless of what you feed to it. There are a small class of DACs out there that do not work this way, but for the most part, most of the gear that you will encounter is doing something like this.
So upsampling is going to be happening either way. Either you do it first, or the DAC does it later. Or you do some of it first, and the DAC does the rest. Either way, the digital signal is ending up at a very high sampling rate before it is converted to analog.
So if we use the photograph analogy. The question is not “will upsampling it to 20 megapixels make it look better”. The question is “If I must display a 10 megapixel photo on a 20 megapixel display, what is the best way to scale it up?”.
And that is not a question with just one answer–in the domain of audio or in the domain of imagery. There are lots of algorithms/approaches at different (performance) cost levels with different quality implications.
The reason why software upsampling can be an enhancement is that there are far fewer performance limitations in a computer–big general purpose CPUs can run more complex and expensive algorithms than DAC chips do. Software-based upsampling opens up the opportunity to select filters/algorithms that make different design tradeoffs than the one that the manufacturer of the DAC chip made.
There is nothing inherently good or bad about the design choices made in the DAC. They were designed by practical engineers working within performance constraints just like everything else. There are always tradeoffs to be made. Software-based upsampling gives access to a more resource-rich environment where fewer tradeoffs are necessary, that’s all.
There is an awful lot in the audiophile world that truly doesn’t make sense–where no-one can do better than theorize loosely about the mechanism upon which a change might affect perceived sound quality. This isn’t one of those cases. With upsampling, we understand what we are changing, how, and why.
Thank you for your explanation. I use upsampling to DSD 256. It does sound better to my ears and that is what counts for me.
Ah, OK, I should have plead ignorance, but now I get it a little more. Thanks for that crystal clear explanation. I should have learned by now that it is always more complicated then it seems. It reminds me of HDTV, especially in the early days of 1080p panels, where you had some sort of box that received the HDTV in 480i, 720p or 1080i, and the question was: should I have the cable box upscale the signal to the native rate of the TV, or present the native rate of the signal to the TV and let the TV do the upscale to its’ native rate? Which did it “better”? I realize this is a simplification of what we are discussing with digital audio, but similar.
Pleading ignorance again, I would ask this question: Wouldn’t upsampling only really be of true value if we: 1. Knew the exact native internal sample rate and type (PCM?, DSD?) of upsampling that the DAC does, 2. External to the DAC upsampled to that exact rate and type before sending the signal to the DAC, and finally 3. KNEW FOR SURE, that if the DAC got presented its’ native internal sample rate and type that it would NOT convert/re-sample?
If we prefer the sound of the upsampling filters/processing we are doing outside of the DAC, but the DAC STILL upsamples or resamples the signal to its’ native internal rate, aren’t still hearing the internal DAC’s filters and processing anyway, now adding up the effect of the preferred processing we did externally with what the DAC does internally anyway? Maybe that is what makes it sound better to some: the cumulative effect of multiple conversions/resamplings? Maybe the effect of all this digital processing is more noise, effectively smoothing out the “digital-ness” of the signal?
It is all very interesting and I appreciate everyone indulging me in this discussion! Ironically my primary DAC is a Schiit Ygdrasil, which does not accept DSD and the literature about it makes a point about it “keeping all the original samples” (but still mentions upsampling).
Brian’s response is excellent.
The only point not covered is the load on the computer possibly causing noise issues with a direct connected DAC. That can be greatly reduced or solved by using a dedicated light streamer after the computer doing the heavy processing.
Roon has RoonBridge for this purpose and HQPlayer has NAA for this purpose.
My HQPlayer box isn’t worked hard at all, 10% without CUDA and less with. Personally I just use a SOTM USB card in my HQPlayer box which has special isolation and allows me to power it via battery. It also allows me to kill the 5+ to my DAC as it doesn’t need it.
So, whoever you spoke with gave you a bunch of BS. As has already been mentioned and I complain about in the posts I referenced, it’s probably a problem with a CPU short on horsepower.
What type of machine are you using, out of curiosity?
It has been explained here pretty good, and your “unknowing” logic does make sense, but you have to hear it done correctly and then you will begin to understand. I use the T+A dac it does no internal up sampling just plays what it gets. I send the dac a HQP up sampled stream generated usually from a Aiff or Flac 16/44 file at 22.6 MHz or 24.6MHz depending on source frequency (44 or 48K based), this frequency is 512 times the base frequencies and then the complex algorithms come in and do their thing, which requires a very powerful CPU’s. The output on a good source can be breathtaking.
A lot of dac’s as mentioned do their own internal upsampling. The Chord Dave I think up samples everything in its big FGPA chips to something like DSD2048. So sending a 512 signal to that may not be of benefit.
Bottom line once you hear 512 done right you may not want to go back either.
Outlaw has been posting about problems in his system for quite some time on the CA site.
The problem needs to be logically found.
- Does it happen with Roon and HQPlayer doing DSD256 up-converting?
- If it happens with both of the above does it happen with a different DAC capable of DSD256?
3.1) If it happens with other DACs, the issue could be a bad USB cable, USB port or the computer.
3.2) If it only happens with Outlaws DAC and not others but his DAC plays back actual source DSD256 content correctly it could be how the DSD is created and the DAC just being incompatible with it. Maybe try different modulators, DSD5 instead of DSD7.
Very often, these facts can be learned by reading the data sheet for the DAC chip and for popular DAC families, the parameters are essentially public information.
Even if you don’t, but the general architecture of the DAC is known (there are only a few), you can make an educated guess about what might help.
DACs usually work in multiple stages.
So for example, the audio may be upsampled 44.1kHz -> 88.2kHz with one filter, then 88.2kHz -> 176.4kHz with another, 176.4 -> DSD64 with a third, DSD64 -> DSD256 with a fourth stage. Or something like that.
This is done because it is much cheaper than doing it in one go. This is mainly because after the first stage, very lenient (read: inexpensive) filters can be used.
From a mathematical precision + filter choice point of view, the earlier stages are more important than the later ones. So even if you can’t get all the way up to the final output rate of the DAC in software, upsampling can still bypass the earlier (most important) parts.
I can see how you might think that, but it isn’t consistent with the math. There is ample precision both in DACs and in software based upsamplers to perform their work without pushing arithmetic error anywhere near to an audible place, even with several stages of processing and accumulated error. Especially in software, the stopband rejection of these filters is often well beyond the point where it could–from an arithmetic standpoint–introduce the filter’s noise floor into the signal.
Jeff I asked Outlaw in the 5th post what his dac was and about his PC and he has gone dark since the first post. It was more on topic back then. Shame, lots of folks willing to help.
My dac is a moon neo 280d dsd.Computer windows 10 .CPU Xeon. E5 2620 v2.Graphics card is a GTX 980 ti for CUDA off load.Problem is on and off.Last 2 times had no issues.This is the lates email I received.Up sampling regular 16/44.1 FLAC files doesn’t benefit the sonic characteristic.
It simply gives add them more weight by adding ‘0’ all over, hence the burden loading up your CPU.The strange thing when I get the pounding sound only out of the right speaker.Even when I pause music the sound continues.
The Dac is plenty capable of up to DSD256, not sure why MFG responded as they did. The CPU tho is probably the issue, especially the stuttering. It is a very slow CPU at 2.6Ghz turbo, 2.1 Ghz at base speed so the issue probably resides there. One the plus side you have lots of L3 cache, a good thing, and 6 cores. I would suggest dropping the up sample rate to DSD128 or even DSD64 and see if the noise continues. Possibly even try PcM 352 as the upsample target the CPU barely breaks a sweat at that rate. If the lower sampling rates solve the noise then the CPU is the culprit.
Really, did you even bother to read the link I first gave you?
No? Let’s try again -
Bottom line - your Xeon is probably too slow for upsampling to DSD256. I can get DSD256 out my dual Xeon, but DSD512 results in the same symptoms you complain about.
An excerpt from the above link -
Thanks for everyone’s help.I just don’t understand how last night and all day today upsampling to dsd 256 working perfectly without a hiccup.
Maybe not the case here but in my world that usually means that there has been a software update somewhere that requires a firmware update on a piece of gear. For a thorough look at hi-res audio concepts and the pros and cons of upsampling and such things, The Recording Academy (Grammy folks) have created a document that was created by a committee of folks who very likely have created much of your favorite music along with consultation from many active creators of hi-res music.