To upsample or not...

[tonapo]

Hello all!

I have always been interested in hi-fi but I always feel I can benefit from a more technical understanding to underpin any usage or purchase choices I make. I think, over the years, I have based purchases on subjective reviews mainly, particularly from main stream press in the UK, for instance What Hi-Fi (magazine), and also on review websites. However I really enjoy the objective focus of this site, and it really resonates with me (even if I don't understand most of it!!). I am very much a consumer rather than a professional in this area.

Anyway, I am aware, and have dabled with upsampling, (or oversampling if you prefer) using software like Roon, HQ Player (and I think J River a few years ago). However I am not really sure I am clear on the benefits (is it actually worth doing??), or that I really understand it! I think my lay person understanding of upsampling is where you might increase frequency rates (from 44kHz to higher values for example), and doing this might move some of the artifacts of possible suboptimal hardware to higher frequencies, away from the audible band - I don't know if these is true?? I have read the Roon website (https://help.roonlabs.com/portal/en/kb/articles/dsp-engine-sample-rate-conversion#Overview) and the HQ Player manual and a recent article from Archimago (https://archimago.blogspot.com/2022/02/using-mele-mini-pc-for.html) among lots of other related content over the years. I think I just want my kit to be set up correctly and used to the best of its abilities.

So, I am interested in understanding the value of upsampling (or not). Should we leave this to our DACs (hardware), or should we use software solutions (Roon or HQ Player)? If we were to upsample, how far should we go? Should we upsample to PCM or DSD? I am also wary of power usage now (its gone up alot in the UK recently and will again in most likelihood), so that is also in my mind. This is relevent as upsampling needs CPU power, although this varies depending on what you want to do.

I have two DACs currently, an RME ADI-2 DAC FS on my desk and a Topping D90 in my main rack. I am currently running the RME with a 'Sharp' filter and the Topping with (as they describe it), the default Short_Delay_Sharp_Roll_Off_Filter. Any thoughts of these choices? These are obviously filter choices rather than upsampling of frequency rates but are also in my mind.

I am really happy for all comments, or pointers for reading up. As I say, my knowledge is pretty rudimentary!

Thanks for reading.

 

[Vincent Kars]

My take is a very simple one.
All DAC’s are up- or over-sampling (except of course the NOS type).
Can we defeat the up- or oversampling of the DAC?
Most of the time not.
Can we match the internal sample rate of the DAC (supposing we know that value).
Mine run at 126 kHz so the answer is no.
Do we profit by having our audio re-sampled twice?
Don’t see any benefit.
Hence I feed the DAC with the recordings at their native sample rate.

 

[AnalogSteph]

I am not entirely sure why the D90 would be using the SD Sharp filter by default rather than the regular sharp one (#1), it's not like shaving off <1 ms of latency is in any way required in a pure playback application. Mind you, you probably can't hear the phase deviation in the SD filter either.

Either way, with DACs like that it's likely to be a non-issue. (My Dell Latitude E6330's IDT onboard audio is a bit of a different story, I am pretty sure I can hear the difference between 44 and 48 and possibly even that and 96 kHz.) At the same time, both support sample rates so high that you could basically replace the DAC's digital filter with high-quality upsampling of your own in order to satisfy the OCD, especially if CPU usage remains negligible. I am using the SoX resampler plugin for Foobar2000 which is heavily optimized (using SSE3 instructions if memory serves) and uses barely any CPU even on rather ancient hardware - others may not be to the same degree.

 

[RayDunzl]

My DAC resamples to 211kHz as it goes about its business of the conversion of digital data to analog voltage waveforms.

 

[DVDdoug]

I say "not". Except yeah... If you've got a 24-bit DAC 16-bits is going to be upsampled. Otherwise, you'd lose about 40dB of volume. So instead, we have 8-bits of unused resolution on the "quiet side". (Although you rarely, if ever, get 24-bits of true-accurate resolution.)

You can't add (useful) information.* And audio is a little different from upsampling video because the output from the DAC is ALWAYS a smooth-continuous analog waveform, and that that point the analog fanatics might say it has "infinite resolution".

Plus, the guys at HydrogenAudio who do scientific, blind, ABX tests will tell you that under normal-realistic listening conditions most people can't hear a difference between a high-resolution original and a copy downsampled to "CD quality". I've never done the ABX tests but I'm pretty sure I can't hear a difference. I'm a "critical listener" and I'm bothered by defects that I can hear and I seem to notice sound defects more than most people but I don't have "golden ears".


* P.S.
Sometimes you can make a useful enhancement... There are "harmonic enhancer" effects that can add high-frequency harmonics. That might help if you've got a recording with a 12kHz sample rate, or something like that.

 

[Vincent Kars]

If you've got a 24-bit DAC 16-bits is going to be upsampled.

Adding 8 zero bits to get 24 bits without chancing the sample rate is not called up-sampling imho.
An example is SPDIF. It is a 32 bit word with a 8 bit preamble and a 24 bit payload.
When sending 16 bit over SPDIF we do have to append 8 zero bits but we don't call it an up-sampling protocol

 

[BeerBear]

I think my lay person understanding of upsampling is where you might increase frequency rates (from 44kHz to higher values for example), and doing this might move some of the artifacts of possible suboptimal hardware to higher frequencies, away from the audible band - I don't know if these is true?

Yes. If you send 44.1kHz to hardware and its DAC filter is suboptimal, it could result in some high frequency roll off, imaging, ringing, or a combination of these. And this could be audible, in theory, because it happens at around 20kHz.
If you instead upsample the audio first in software, say, to 88.2kHz, these DAC filter artifacts are less likely to be audible, because they would happen at around 40kHz (and it's just silence up there anyway). And obviously the software upsampling has to be good enough for this to make sense.

Whether this ever results in an audible improvement is a different question...

 

[tonapo]

Thanks for the replies.

Do we profit by having our audio re-sampled twice?
Don’t see any benefit.
Hence I feed the DAC with the recordings at their native sample rate

I wanted to avoid the possibility of re-sampling twice. I did some research before about this, and from what I understand, with an input rate of 352.8k and higher, you are bypassing the internal filters in respect of the ADI-2. And I think, although I am not sure, I think this is true for the Topping too.

At the same time, both support sample rates so high that you could basically replace the DAC's digital filter with high-quality upsampling of your own in order to satisfy the OCD, especially if CPU usage remains negligible.

I can use Roon to upsample pretty easily as it has DSP funtionality (and my Roon Core runs on a dedicated Intel NUC). I can set Roon to upsample with a 'Max PCM rate (power of 2)' (or anything in between) - so I am currently listening to 44khz content upscaled to 705.6khz which the NUC seems quite happy with. The CPU use looks to be around 10% - 15% utilised on one core - I think! So if the CPU use seems fairly negligble, then is it worth doing? Or, am I just wasting power??

Whether this ever results in an audible improvement is a different question...

Thats what I am not sure of. Maybe I need to test it. Maybe I could co-opt/coerce/pay-off my wife or one of the kids to help?? I think if I avoided headroom management in Roon, I could negate volume issues (I could check music before to see if it clips when upsampled). I could keep it simple if I stuck with the same filter in Roon (Precise, Linear phase) and just test based on the upsampling (or not)?? Someone else (so not me) could then sit with the tablet and turn filering on/off when they choose and I woundnt know.

I suppose I would be testing to see if I can tell the difference between no upsampling compared to when upsampling (if I cant tell, then not worth doing, well, with Roon at least). Although I think I would have an issue as the filters would be different, i.e. when not upsampling I would be using the filter in the DAC and when upsampling, I would be using the filter in Roon. I need to check to see if the RME has a NOS mode I think, then I could ask Roon to use its filter whether upsampling or not.

 

[dc655321]

I suppose I would be testing to see if I can tell the difference between no upsampling compared to when upsampling (if I cant tell, then not worth doing, well, with Roon at least).

You will need to be careful that the levels match between the original and upsampled versions: even v. small differences in volume are enough to bias the test (ignoring other confounding factors).

 

[Zensō]

However I am not really sure I am clear on the benefits (is it actually worth doing??)

There‘s your answer. If you can’t easily hear a difference (I don’t expect you would), then why bother?

 

[jasonhanjk]

If I remember correctly, up sampling is an old tech to improve audio quality.
It does not bring benefit to current dac.

New DACs running on 44.1kHz and 48kHz are excellent.

 

[tonapo]

I was wondering about blind testing. I enjoyed watching Amirs video on the subject. I think I could test Roon upsampling blind fairly easily (testing to see if I can tell the difference) with some help from my other half. I think I could control most variables, but I am not sure of volume balancing. I can set Roon to reduce the signal by -5dB and apply it whether upsampling or not this avoids clipping on the upsampled music. I actually tried quick a/b testing, switching the sample rate converter on and off on Roon. Even though this was sighted I don't think I could tell the difference between upsampling (power of 2, to 705khz or 768khz generally) with no upsampling (generally with 44.1khz content). I think the two variables, volume challenges aside, would be the higher pcm rate and also the two filters, one in the DAC (RME ADI-2) and one used by Roon. I used 'Sharp' in the RME and 'Precise, Linear Phase' in Roon. I still couldnt tell the difference. I might fiddle around a bit more but I wonder if its worth the effort now.

I do think though I would like to try HQ Player, blind, against a no upsampled set-up, but there are alot more variables to consider, not least the volume. More thought needed...

 

[ADU]

I like to try to keep an open mind on stuff like this. But I'm not sure I understand what benefits upsampling to such high sample rates could have, especially if the only benefit is an increase in the content's frequency range (beyond the range of human hearing).

Is vinyl capable of producing frequencies up to 384 kHz? And could we even hear them, if it did? Is there equipment (including speakers) that can deliver a reliably flat response up to such ranges? And are there ways to accurately measure that?

Using a simple frequency sweep, my admittedly somewhat damaged hearing seems to tap out somewhere in the mid-teens kHz-wise, which is well below this. And most frequency response measurements only go from about 10 or 20 Hz up to about 20 kHz.

 

[firedog]

You might hear a slight difference when upsampling. It's possible your DAC has a sweet spot at which it operates at it's most accurate.

HQP gives you endless ways to "flavor" the sound with its upsampling and filtering options. Again, you might hear a slight difference between some of the options.
Will you hear a significant difference in blind testing? Maybe not. Maybe a slight one.

I play with it sometimes and hear some differences. Can't say it's "better". It's different. Generally I'll listen that way for a while for a change of pace and then go back to no upsampling.
It's worth a try and some experimenting. I wouldn't think it's a big deal. When you read the audiophile sites and press they make it sound like it makes some massive difference. It does - in their minds.

 

[Sokel]

About upsampling.
Has anyone else noticed anything like this?See the spikes at 8Khz and about 16Khz when I enable Sox in foobar (mod and mod 2 for the two families of 44.1 and 48).
Mic is in the listening position 2.8m away,so it's not something insignificant and no music playing,just the ambient (noisy) environment.

No Sox

SoX enabled

 

[Nobleman]

I say "not". Except yeah... If you've got a 24-bit DAC 16-bits is going to be upsampled. Otherwise, you'd lose about 40dB of volume. So instead, we have 8-bits of unused resolution on the "quiet side". (Although you rarely, if ever, get 24-bits of true-accurate resolution.)

You can't add (useful) information.* And audio is a little different from upsampling video because the output from the DAC is ALWAYS a smooth-continuous analog waveform, and that that point the analog fanatics might say it has "infinite resolution".

Plus, the guys at HydrogenAudio who do scientific, blind, ABX tests will tell you that under normal-realistic listening conditions most people can't hear a difference between a high-resolution original and a copy downsampled to "CD quality". I've never done the ABX tests but I'm pretty sure I can't hear a difference. I'm a "critical listener" and I'm bothered by defects that I can hear and I seem to notice sound defects more than most people but I don't have "golden ears".


* P.S.
Sometimes you can make a useful enhancement... There are "harmonic enhancer" effects that can add high-frequency harmonics. That might help if you've got a recording with a 12kHz sample rate, or something like that.

I've been trying to get High res 24/192 audio out from Apple Music and dabbled with the Midi config of my Mac Mini which I am using as the music server. I wanted to set the Midi config output to 24/192 from default 24/48 so that I will get a constant 24/192 output from my Topping D10s. I did not expect to 'hear' much of a difference when a 44.1 source gets upsampled to 192k in realtime using software but unfortunately I can and to me its very apparent (Magnepan 1.6/Audiolab 6000A) I thought I had solved the mystery of getting 24/192 out from Apple Music routed through my Topping D10s but no joy...It sounds like the mid range is missing detail or compressed maybe due to the filtering. Also you lose some amplitude/volume output Id say 10% likely in filtering process . OSX Midi (Sound) output configurator allows you to toggle between different sampling rates available through the output source in real time (milli seconds delay when changing rates) which is very convenient when testing. Would someone test this setup and share their findings please?

 

[jasonhanjk]

I did not expect to 'hear' much of a difference when a 44.1 source gets upsampled to 192k in realtime

You mean 44.1k upsample to 88.2k or 176.4k?

 

[Chrispy]

Why would anyone use a higher sample rate particularly just for playback?

 

[Blumlein 88]

Why would anyone use a higher sample rate particularly just for playback?

If filters slightly effect the sound, then upsampling might make a bit of a difference.

Of course this blind test of 44.1 khz vs 88.2 khz sample rates makes one wonder.

https://www.researchgate.net/publication/257068631_Sampling_Rate_Discrimination_441_kHz_vs_882_kHz

Free download of the paper.

They found some could discern 88.2 khz vs downsampled 44.1 khz using a very good downsampler. But no one had results indicating they could hear native 44.1 khz vs 88.2 khz recorded concurrently. No processing, and good quality gear used for this. One could surmise the downsampling was audible.

 

[Chrispy]

If filters slightly effect the sound, then upsampling might make a bit of a difference.

Of course this blind test of 44.1 khz vs 88.2 khz sample rates makes one wonder.

https://www.researchgate.net/publication/257068631_Sampling_Rate_Discrimination_441_kHz_vs_882_kHz

Free download of the paper.

They found some could discern 88.2 khz vs downsampled 44.1 khz using a very good downsampler. But no one had results indicating they could hear native 44.1 khz vs 88.2 khz recorded concurrently. No processing, and good quality gear used for this. One could surmise the downsampling was audible.

Hard to believe.