Is bad upsampling the only reason why HiRes can sound better?

I heard this claim in a podcast recently: A guy who has designed DACs said that HiRes sounding better is not because there’s any additional audible content in the files but because DACs are far from perfect at upsampling due to restricted computing resources, and that the artefacts they produce in the process are reduced when the upsampling factor is reduced due to higher source sampling rates.

Personally, I think there might be some truth to that. Not sure if it’s the only reason, but it might at least play a role.

The difference is tiny, anyway, and the recording quality has far more influence than the sampling rate, I know, we don’t have to discuss that here.

If you think HiRes makes a difference: Do you think DAC upsampling might be a major reason for that?

 

That’s always been the promise of upsampling, though it requires a dac not to do any internal conversion afterwards.

Something like hqplayer which claims to do more (e.g. dac correction) seems more like eq but I haven’t tried it so don’t know. Ultimately all you should care about though is what sounds better to you.

 

Well, every DAC that’s not a NOS DAC upsamples internally, up to MHz frequencies in case of a D/S DAC. The theory would be, if you have something that does that better (like HQPlayer, not necessarily like Roon or sox or similar), the more of the task you can move to a computer, the better, i.e. feed it with the highest possible frequency.

 

Very interesting question! sorry for the shitpost I just wanted to add excitement

 

So if a DAC has switchable OS/NOS modes, I guess it would be better to have the source upscale and run things through NOS?

I’m just thinking that some DACs may be built with better OS capability than non.

I would think that an RPi would be better; however all that processing done upstream might introduce more noise. I don’t know the science but I’m sure there’s got to be some trade offs with upsampling whether at the source or at the DAC.

edit: appended

 

I think it depends. On the DAC as well as on the software used for external upsampling.

I guess that what makes expensive DACs worthwhile (probably, I didn’t compare any myself) is good components on the one hand, like a good power source and clock, good amplification etc. On the other hand, it’s good upsampling.

In my personal experience, using HQPlayer to feed a shitty DAC like the SMSL D-6 I have (not in active use), produces great results that easily surpass DACs 3-5x the price if those don’t get the HQPlayer treatment. But it only becomes really breathtaking with a better equipped DAC.

Also, the Meier Daccord DAC I have (no active use anymore, either) can only be used with up to 192 kHz. It has solid upsampling, I’d say: Roon upsampling sounded exactly the same to my ears as what the DAC did by itself (minimum phase, linear phase was slightly different), but HQPlayer provided an improvement. With its good components, I would have loved to hear what HQPlayer could do to its sound if it allowed higher sampling rates and, it being a delta sigma DAC, especially DSD. As it was, the D-6 fed with HQPlayer DSD sounded better to my ears.

Anyway, I digress.

There’s lots of audiophile speculation about what it might be that makes HiRes content sound better. As the additional content should not be audible, the idea that the upsampling process could be the culprit makes sense to me.

Which would also mean: The better the DAC, the less difference does HiRes material make—which conforms to my personal experiences with HQPlayer.

What are your experiences?

 

To add to yours, I have in the past preferred the Yggdrasil's (A2 is my version) internal "super combo burrito" filter with some music, and with other music source files I have proffered to up-sample to 192 using HQPlayer, and yes even Roon's (sox based I beleve) linear phase filter. Now it is true in this situation Schiit's combo burrito filter is still being applied, but with 44.1/48 the processing to 192 seems to have the greater effect. Granted these actions are subtle, the last 1.2% and all that, but you can hear it. It does not change the overall character of the Yggdrasil and defiantly falls into the category of "minor tweak". I have also played around doing this with low/mid-fi dacs (e.g. various iFi products) and there as well it can make a "minor" difference, usually finding it better to just upgrade the dac itself then try to get substantially better sound using up/resampling.

I mostly listen to a Holo Cyan now, re-sampling everything fed to it to 11.2mhz SDM (DSD). This dac is designed for this circumstance, so really it is a hardware "backend" for an HQPlayer "front end" - it is 'half a dac' as has been said. HQPlayer is sort of the ('ultimate'?...probably) replacement for the processing limitations of all dacs since their front ends are always design/cost limited in some shape or fashion. Thing is, you can't really bypass this limitation in most dacs due to built in design and hardware limitations. Even many NOS dacs have inputs that severely limit rates as in your Meier experience (or my Yggdrasil).

In the end (as usual) it is about synergy, the algorithmic "front end" working together with the hardware "backend" producing something that has to be judged together.

 

My conviction is that is due to the greater passband of the filter. With 44.1 kHz, when you want to have a flat response to a theoretical 20 kHz and then full attenuation at 22.05 kHz, well, that’s a really steep brick wall filter within 2 kHz introducing quite some ripple and group delay. At 48 kHz, you’ve already got double the bandwidth for the filter to do that, so an easier filter with less artifacts as a result.

This also explains to me why external upsampling can often sound better: it pushes on-chip filters out of the audible band.

Something Bruno Putzeys wrote that I’ve always thought interesting in a good way:

1. The aliasing problem can be solved at once, anywhere in the audio chain, using a single lowpass filter that enters stop-band before the alias band ie before 0.4535fs. A good place to do this is at reproduction or before final dithering. This means that halfband filters as in 1 may be used throughout without deleterious effects. I find that running a CD through an ultrasteep filter (pb to 18.5kHz, sb from 20kHz, eliminating all aliases that were created anywhere in production) results in an improvement in contrast, depth and precision of the stereo image.
[...]
3. As we're already increasing sampling rate in view of the group delay problem, we can push it further to obtain a more natural roll-off after 20kHz.

Recipe for perfect PCM:
Specify a sample rate well above twice the audio band, e.g. 192kHz (lower is arguably acceptable too but since we're slowly standardising on 192 who cares)
Specify all interpolation/decimation filters as halfband, 0.4fs to 0.6fs transition.
Put exactly one non-halfband lowpass filter in the chain (e.g. at replay or before final dithering) that enters stopband at 0.4fs. Specifying its passband at 20kHz will allow for a very smooth roll-off and hence very short and practically ringing-free impulse response.

Source: https://forums.stevehoffman.tv/threads/sacd-fundamentally-flawed.26075/page-3

Myself, having dabbled with HQPlayer, I’ve happily settled on my Schiit megacomboburritofilter, but happily pay extra for 24/48 masters if I can.

 

I tried a lot of HQplayer, but generally find it not worth the "improvement". Sure soundstaging changes, and little details here and there come out, but I find the sound always loses some edge and mass for a smoother, rounder, gentler presentation. Sure you can tweak it to get more body, but also with the use of filters, I always felt the blackground suffers. There is this nagging feeling there is a very transparent thin layer that makes it sound "faker" to me. I feel this is the case for most upsampling, even in roon. I would just choose a dac of my choosing and stick with the native sound. native format. I haven't tried dacs that have these features included like the Wandla GSE with 2 hqp filters along with all its plugin enhancements and software, but I'm in the camp of keeping the signal chain as simple as possible.

On the subject of Hi-res files....I feel they are mastered differently with the sound usually brickwalled harder and louder, with less dynamic range. There have been some instances where I see the merit, but its not consistent enough to fully attribute to just higher sampling rate.

 

Unless your using a NOS capable dac in pure NOS mode, then the 'native sound' of the dac your listening to is filtered & upsampled. Also, if your trial of HQPlayer was not with a NOS dac, then what you heard was both the HQPlayer filtering/upsampling & the internal filtering/upsampling of the dac you were using...just laying that out in case it was not clear.

 

They were used in NOS and my observations still hold. The post is a bit rude imo, with the assumption that I'm dumb enough to try to listen to filters while overlapping them.

 

In a word: yes, and it's not just DAC oversampling. Keep in mind that the same kind of brick wall is necessary to decimate a high-resolution intermediate into a 44.1KHz sampling rate.

The whole point behind DSD as I understood it is that no digital reconstruction filter is supposed to be necessary. Simply splat the 1-bit samples out and filter in the analog domain using the filter required by the Scarlet Book. Nevermind that an analog 5th-order filter is going to have a fair amount of variability to it compared to running the same filter digitally! But the other advantage this provides is that provided a 5th-order modulator was used to encode the DSD (which is not a fair assumption anymore), the noise floor will remain constant from the 50KHz corner frequency of the filter out to the Nyquist at 1411.2KHz. This isn't something you typically see with DACs these days, sadly.

I think it also would have been better if the Red Book standard had included standard coefficients for at least the first half-band oversampling filter (which doubles from 44.1KHz to 88.2KHz), but no doubt that would have been seen as onerous at the time. To the extent that MQA had any value at all, enforcing uniformity on the reconstruction filters used by various DACs is a good idea; it's too bad that the one they chose is so misguided. Nevertheless, an audio standard that attempted to solve this problem while keeping the space advantages of standard-resolution content would still be useful. Just make it 48KHz instead of 44.1KHz so the filter doesn't have to have quite so many taps to get decent attenuation by (or just after) Nyquist.

DAC chipset manufacturers could also use a kick in the pants to encourage them to support more filter taps, particularly on their higher end DACs. Likewise, DAC product manufacturers could consider actually using the substantial computational resources of the XMOS chips they're putting in there instead of treating them as glorified USB bridges. Many things could be done better with little additional cost. At some point, when you buy a product like a Schiit or a Chord DAC, you're paying for the manufacturer to care about the engineering instead of just pumping out product after product. The rest of the market will just take whatever's in the reference design and ship it.

 

Relax, no offense was intended, it just wasn't clear. I myself originally messed around with HQPlayer without NOS thinking I could still draw some worthwhile conclusions and of course you can I suppose, sort of stumbling upon something that is useful for your particular situation, but the fact is it's not an honest test of up/re-sampling in general.

 

That first bit above has always been my position, and why I've sought out > 44.1Khz, preferably 88 or 96 KHz recordings.

Recently I did a few experiments digitizing some LPs and comparing to the CDs I have or my own conversions to CD quality, 48KHz/24b, and 96Khz/24b. (Ppl said early '80s CD were crap but that's what I had to hand).

What I found was that, for me, most of the difference came from going from 44.1KHz/16b to 48Khz/24b; could not hear much more going up to 96Khz. What I think I heard was that 24b captured certain instruments' sound better, particularly big band horns, which is what I was comparing.

I'm somewhat annoyed because I was hoping I wouldn't hear anything.

I don't think it's just the oversampling artifacts that are an issue; I think the recording process matters much more (what a revelation ). Having provenance information on our recordings would be a big win.

 

I don't think that all older CDs were crap. I do have a few that sounded to me much the same as the vinyl. On the other hand, I've heard relatively later CDs that were so awful that I scuttled back to my digitisation, scratches'n'all. In those instances, I think that the mastering was the key. Given... the same material.

My suggestion for ABing such things is, if you have a turntable and vinyl, make your own digitisations, and compare those. That way, the analogue material and the entire chain remains identical.

In the instance in my mind, the modern CD was a perfect example for audiophool nerds who claim that digital music has no life. It really did sound that way! But my own digitation (at a modest sample rate) had all the life of the vinyl.

 

That's what I ended up doing and my conclusions match yours.

It was an interesting and revelatory experience.