Upsampling to anything other than your DAC's internal conversion rate

[sdolezalek]

If your DAC chip internally converts musical input at DSD512 or at PCM384, should you be upsampling to anything other than that and if so, aren't you just adding more conversion artificats?

 

I'm currently upconverting everything to DSD128 in HQPlayer because that is the highest DSD rate my Oppo BDP105D accepts. To me, it sounds great. But am I fooling myself?

 

I actually don't know what I would have to feed this DAC in order to forgo any further internal resampling, nor do I know if it will natively accept that format, but it seems that for those who know what that rate is and are feeding their DAC that rate and format, they should be getting the maximum benefit from HQPlayer.

 

For those of us, feeding it something that still requires further internal resamplings/conversions, I wonder just how much we are actually improving the sound vis-a-vis just feeding the DAC the original format and letting it apply its own upsampling and filtering (even if that is inferior to HQPlayers upsampling and filtering).

 

Of course, it would also really help if we could collectively come up with a chart that shows the actual native conversion rate for popular DACs.

[sdolezalek]

Unless using a DAC that's purpose designed for such use or with upsampling that can be turned off, can we be certain that any software upsampling/filtering is not making things worse than better? How do you know that by software upsampling to the DAC's max rate you are somehow bypassing the DACs own processing - maybe its just doing the job twice and making things worse as a result?

 

From my listening "feeding the DAC the original format and letting it apply its own upsampling and filtering" gets the best results.

 

Norton: I think there are a lot of folks here who would disagree with that last statement. Just read the thread on the T+A DAC8 DSD to see how many people hear the benefit of feeding it 512DSD rather than native format. The same is true of the iFi Nano and Micro DACs. I have heard that many of the Chord DACs prefer high rate PCM. So my question was not about original format versus the DAC native format, but about halfway measures.

 

I also don't see how feeding a DAC DSD512 when it uses DSD512 internally could cause it to "do the job twice." Are you suggesting it would take DSD512 convert it back to PCM 16/44 and then back up to DSD512?

[sdolezalek]

Even if a DAC supports only up to 96k PCM, it is still worth upsampling RedBook to 88.2/96k before sending to the DAC. The first step up is sonically most critical. And doing 2x conversion would already put 22.05 kHz wide space between DAC's filter and the original content, so the DACs built-in filter is already having much lesser impact. In addition, with typical DAC chip, every doubling of input rate drops out one cascade digital filter section from the built-in filter chain.

 

In addition, many devices doing digital room correction or other DSP (AVRs and such) typically resample everything with no so great filters to either 96k or 192k to apply the DSP at fixed rate. Sending these already good quality resampled data will typically also improve performance.

 

Sometimes it may be useful to choose between PCM and DSD output though. In particular with Chord DACs it is better to use highest PCM rate, since they convert DSD inputs to that same one anyway it is better to send PCM out.

 

So no worries...

 

Jussi: Thank you, I was hoping you would reply. I had forgotten that most DAC chips do this by repeated doublings from the input starting point ( i.e. 44>88>192>384) till their internal conversion rate. I assume that means they are also applying their own filters at each step? So every time I substitute a preferred HQP filter for the DAC Mfg's filter, I benefit at each doubling?

 

My other request, not of you, but of the community here, was to begin putting together a list we might all share as to what the "optimal" or "native" input rates are for at least the most popular DACs. Do you think that would be a worthwhile exercise for CA users in order to help newer HQP customers maximize their benefit from upsampling?

 

Thanks, Stephan

[sdolezalek]

Ok, after a slow start, this has gotten quite a bit more lively. Let me start by accepting the comment made by "One and a half" that trying to figure out exactly which setting, filters, etc. to use on which types of music might just be too much work for most. I have no issue with that and it makes life after the original DAC buying decision much easier.

 

For the rest of us, who like the freedom that front end software with its own upsampling algorithms and filtering choices provides, this thread was intended to provide some guidance as to where to start exploring what you like and what you don't. What I have heard so far is that there are some DACs that do seem to prefer DSD and others that seem to prefer PCM. I think Miska has made a very coherent case for why doing the filtering at much higher sampling rates is capable of reducing noise artifacts in the audible range (see post #19 above). So, based on the responses submitted so far let me take a cut at a DSD pro and PCM pro list (and please feel free to correct my mistakes and I'll republish the chart:

 

Best with DSD input: iFi, Oppo, T+A, Teac, exaSound, Fostex, Sony and most other ESS Sabre-based DACS

 

Best with PCM input: Chord DACS, NAD, Schitt (Yggy), Prism Callila

 

After this list gets a bit further along, I'll try to take a shot at also including some specific input rates for each DAC (knowing that risks much greater disagreement).

[sdolezalek]

My ears/brain prefer no software up-sampling with both my Schiit Bifrost Multibit and Chord Qute EX DACs. No big surprise considering both manufacturers are already listed. That being said, both DACs, when used with my Wyred 4 Sound re-clocker, which re-samples everything to 24/96, sound fantastic.

 

Melvin: Without trying to single you out specifically, if I can use what you said above to make a broader point:

 

1. Your DACs are both doing internal upsampling and applying the filters the manufacturer chose to do so.

2. Your Wyred 4 Sound reclocker, to the extent it resamples everything to 24/96 is also applying its own and different set of filters

3. HQPlayer or Audirvana/iZotope would each also be applying their upsampling and filter methods

 

If we assume (which may not be a fair assumption) that all of the upsampling is bit perfect in that it applies no DSP or other changes to the signal except duplicating data points exactly, then the differences you hear will all be due to the choices of filters applied (chosen for you in the case of Schitt/Chord/Wyred and chosen by you in the case of HQPlayer/Audirvana).

 

Given the range of opinions expressed about filters/upsampling, I think it is fair to assume that the filters DO have an impact on the resulting sound. Now, if you chose the DAC(s) you have by comparatively listening to them against other DACs, part of what you were doing was showing a preference for that manufacturer's choice of filters. Thus, if you "love" that particular sound, it would not be surprising that you would prefer it over a set of filters that produce a different sound. For most folks who use their DACs with HQPlayer or Audirvana but chose no upsampling, I think that all they are really saying is "I like the filtering choices made by my DAC manufacturer."

 

The next level up though would be to ask, which of the above is closest to the original recording? I think Miska has clearly shown why upsampling data to a higher rate allows a set of filters to be applied (whether by him or by your DAC manufacturer) that produces fewer artifacts in the audible range (so has nothing to do with higher frequency hearing/response beyond 20kHz/steeper transients, etc. that often get referenced to say upsampling can't make a difference). What is obviously much, much harder is to show before and after plots for your specific DAC using its internal upsampling/filters versus your DAC using upsampled input with HQPlayer filters applied. The Meridian folks backing MQA are suggesting they can do something along those lines (ignoring anything else they are doing re compression or copy protection) that ultimately leads to your DAC producing an output more faithful to the original recording. Miska is doing the same in continuously testing/improving his filters.

 

That is the main reason I focus on these software players -- because as time progresses, the software filters can keep getting better (for those of you who are photographers, it is like running a picture you processed in the 1995 version of Photoshop and today rerunning the RAW version of that file through the latest version of Photoshop; it is amazing how much more can be done to the RAW original with the capabilities of the new software). In theory, the same will continue to be true of the music software as we keep learning beter and better filtering methods.

[sdolezalek]

It's never bit perfect, since part of the idea is to change the sample rate (i.e., add more bits). And this is true of DAC internal sample rate conversion and filtering as well. But now think of what Miska said regarding the fact that the data is not the signal. As you mention below, what is being done is that the (digital) data is being changed in service of getting the (analog) signal closer to the performance, whether that was live or as it came from the studio.

 

Most of these conversions are "lossy" in the mathematical sense that the original data points cannot be recovered by a mathematical operation from the converted data. A few (the Schiit filters and a couple of Miska's) are "closed form," which means the original data points can be recovered. So they are not "lossy" mathematically, but they are still not bit perfect. As to whether such filters sound better, I'll leave this to the listeners and developers.

 

Jud: as you can see my feeble lawyer/VC mind struggles with the precision of these engineering concepts What I was mentally trying to separate were those changes that should be non-destructive but perhaps also non-constructive and those that should clearly have an influence on sound (whether good or bad). So thank you for the clarifications.

 

The part I found interesting in playing around with the A+ parameters in iZotope were the necessary tradeoffs and learning much more about my own personal sensitivities to a) ringing, b) decay, c) transient attack, d) frequency flatness and others. Quite often things that sounded great on day one became tiresome after a few days or weeks. The longer I listened, the more less was more (I think you experienced some of the same in your progressive leaning toward less-invasive flatter filter slopes). It was that learning with iZotope that then allowed me much more quickly to choose which filters I liked in HQPlayer (even if I didn't know what fine-tuning parameter tradeoffs Jussi had made).

 

I also have to assume that in choosing some of these filters we are "making up" for limitations elsewhere in our system -- for example a bookshelf sized set of speakers might seem much larger with a bit more reverb and slower decay. With my giant Magnepans, exactly the opposite was true -- I tended toward leaner, more nuanced presentations.

 

All that being said, I do think there is value in us collectively trying to come up with a list of starting points for different DACs. Put differently, I'd like to make sure I'm going through filter choices that make the greatest possible impact because I have chosen a DSD512 or 24/384 starting point wherever that is possible and thus knowing where to start then pushes the learning over to the other filter choices rather than choice of upsampling.

[sdolezalek]

For example rock recorded in studio puts demands on the transient response, so a minimum phase filter that doesn't have pre-ringing is usually good fit, since there's usually not much real acoustics involved anyway. While classical music recorded in real acoustics puts demands to the sound field/space, so a linear phase filter is usually good fit, since there are no strong transients or at least very few.

 

Given that most music library software (like Roon) includes "Genre" tagging, would it be possible to create a genre-specific filter assignment rather than having to stop HQPlayer, change the filter and restart? I understand it would have to create a gap in playing as the filters are changed, but the ability to change filters "on the fly" is particularly relevant for those of us where the processing computer is in a completely different part of the house than our music system.

[sdolezalek]

Guys: The last two pages of this post are a complete waste of time. Wilhelm: if you want to have an argument with Miska, start your own post, but please do not ruin what was a very helpful set of posts here until you intervened.

 

NOW, can we get the conversation back to the original discussion???

[sdolezalek]

Where does the PS Audio Directstream DAC fit into all of this? It internally upconverts to 10XDSD, even higher than HQPlayer DSD512. And for whatever it's worth, Stereophile A+ rating. Not bad for $3K-4K on the used market.

 

Is the consensus that HQ Player upsampling via a powerful PC and played through a DSD512 capable DAC is giving even better results than the Directstream?

 

I'm curious to hear whether anyone has used HQPlayer with the PS Audio Directstream. My guess is that over time, particularly as newer chips become available, more and more manufacturers will transition to high multiple upsampling and the use of gentler filtering methods. The question will be whether software makers, like Miska, will constinue to stay ahead, in part because their product cycles can be measured in weeks, not years.

 

Separately, as Miska said in one of the prior posts, choice of the best filter might ultimately come down to the nature of the music input and its genre. If that is the case (and I'll avoid the issue of whether filter changes can be made automatic) then you might still get more out of software upsampling because of your ability to fine tune it to your listening tastes and music collection.

[sdolezalek]

I have recently purchased the Bel Canto Dac2.7. This was on my short list of DAC's to choose from that are just a PCM only DAC. I have no DSD material so I have no vested interest into a DAC that does anything but PCM.

 

Now as far as Upsampling goes I'm currently using Hqplayer on a trial basis to see if I can hear a difference that I like. I've tried several settings but I've settled on poly-sinc-shrt-mp/NS9. Everything is upsampled to 192 this is the DACs highest setting. During this trial period I've compared this to Roon by itself, Jriver with no upsampling and upsampling done with Jriver.

 

What I have found is that I prefer no upsampling from any external software. When the DAC is feed the raw signal there seems to be more weight (bass) and a bit more spaciousness around the high ends. Prior to this I used and still have the Peachtree DACIT. I upsampled everything with that, and I liked it a lot. The Bel Canto DAC is just in another league. I will continue to use HQplayer until my trial is over just to experiment, why not, it's free. I'd like to thank Miska for all his explanations on the how's and why's his software effects the signal and sound that you may hear. I find it hard enough to describe what I'm hearing, but I know what I like in my environment.

 

I think for me, the Bel Canto 2.7 may be one of those DACs that dosen't benefit from external upsampling. Remember that this is not a DSD dac, the 2.7 only decodes up to 192. Maybe that's to subtle of a change for a night and day difference. As I stated earlier I could hear a change, but for me, I like it without. Others may not.

 

Thanks Chris for cleaning up this thread, I think this is very helpful for anyone experimenting with upsampling.

 

Shawn: Interestingly in a review of your Bel Canto, I found the following Quote: "The manufacturer recommends converting files to 24/176.4 PCM using a high performance playback software and makes persuasive arguments/measurements in their white paper, stating that “24/176.4 PCM is inherently a lower noise, wider dynamic range playback than DSD”.

 

So that adds this DAC to the PCM preference lst, but I thought it interesting that Bel Canto would want DSD files converted to 24/176.4 when it otherwise upsamples everything to 24/192. I assume that it also accepts 176.4 without further upsampling and that may be a better integer multiple of the DSd frequency.