Question about Dirac Live

[Buford TJustice]

Question regarding Dirac Live. Both with the DDRC-88 and the DDRC-24 (2x4HD upgrade), the bit depth of the processing is a full 24 bits, but the internal sampling rate of both is "only" 48khz. While, I'm not concerned with any ultrasonic information being affected, what effect does this have on higher sample rate music such as DSD, FLAC or MQA at 96khz or above (or the equivalent, re DSD)?

What would this do, from a technical (though not necessarily purely mathematical) perspective? Do the benefits in phase correction outweigh any potential penalties?

Educate me.

 

[Matthew J Poes]

Question regarding Dirac Live. Both with the DDRC-88 and the DDRC-24 (2x4HD upgrade), the bit depth of the processing is a full 24 bits, but the internal sampling rate of both is "only" 48khz. While, I'm not concerned with any ultrasonic information being affected, what effect does this have on higher sample rate music such as DSD, FLAC or MQA at 96khz or above (or the equivalent, re DSD)?

What would this do, from a technical (though not necessarily purely mathematical) perspective? Do the benefits in phase correction outweigh any potential penalties?

Educate me.

It would be great if Flak was around to address this. Basically the incoming signal is resampled to a lower sampling rate. You would lose any ultrasonic information. In practice the effect would be minimal. While the current evidence is in favor of hearing benefit from high sampling rate music, we still have no idea why and whatever differences exist are sudtle at best.

On the other hand, linear distortion is highly audible by even the most untrained listener. Dirac corrects linear distortion significantly and this makes major improvements in the sound of your system. There is nothing sudtle about it. Especially when the responses without correction look like the ones here.

 

[Buford TJustice]

It would be great if Flak was around to address this. Basically the incoming signal is resampled to a lower sampling rate. You would lose any ultrasonic information. In practice the effect would be minimal. While the current evidence is in favor of hearing benefit from high sampling rate music, we still have no idea why and whatever differences exist are sudtle at best.

On the other hand, linear distortion is highly audible by even the most untrained listener. Dirac corrects linear distortion significantly and this makes major improvements in the sound of your system. There is nothing sudtle about it. Especially when the responses without correction look like the ones here.

Gotcha. I guess I should have been more specific in my original posit; does the down sampling of a 96k/24bit source file's loss of time resolution have a perceptible negative effect on the sound quality?

I'm already guessing that the correction of linear distortions would outweigh any lost time-domain resolution, but we've got some Dirac heavyweights here. What better place to ask?

 

[Matthew J Poes]

Gotcha. I guess I should have been more specific in my original posit; does the down sampling of a 96k/24bit source file's loss of time resolution have a perceptible negative effect on the sound quality?

I'm already guessing that the correction of linear distortions would outweigh any lost time-domain resolution, but we've got some Dirac heavyweights here. What better place to ask?

What do you mean by time resolution? I'm afraid to jump into a technical answer if it isn't what you are asking.

 

[Matthew J Poes]

Gotcha. I guess I should have been more specific in my original posit; does the down sampling of a 96k/24bit source file's loss of time resolution have a perceptible negative effect on the sound quality?

I'm already guessing that the correction of linear distortions would outweigh any lost time-domain resolution, but we've got some Dirac heavyweights here. What better place to ask?

I'm actually currently testing the Dirac software which can handle up to 192khz. Since I have HAD files and since the Tidal software does MQA unfolding to 96khz I've been testing if I hear any audible effects of using lower sampling rate filters with high sampling rate content. I need to confirm with Dirac what their system does in this scenario. I created two identical filters, one peaked at 48khz and the other 96khz. I heard a noticable difference switching back and forth which really surprised me. I actually am in disbelief of what I heard, I fear the difference was a error in the process. Given the number of variables this could be a software glitch, soundcard issue, a mistake I made when creating the two filters, etc. Switching filters made an audible noise so it's also possible that the filters sound identical but hearing the blip between switching caused my brain to think it heard a difference. My guess is I was hearing the software make the sampling rate conversion. Not the sound of the sampling rate conversion just an unmuted bit of noise. I hope to find out more and see if I can report back more thoroughly In a seperate thread.

I've also heard clear differences in the sound of upsampling to 384khz PCM vs 5.8mhz DSD. I don't believe there are actually big Sonic differences here. I think what I've heard (which was clear as day to hear) was actually a problem in the upsampling process. That if we measured the output we would discover some artifact causing the difference.

 

[Buford TJustice]

I'm actually currently testing the Dirac software which can handle up to 192khz. Since I have HAD files and since the Tidal software does MQA unfolding to 96khz I've been testing if I hear any audible effects of using lower sampling rate filters with high sampling rate content. I need to confirm with Dirac what their system does in this scenario. I created two identical filters, one peaked at 48khz and the other 96khz. I heard a noticable difference switching back and forth which really surprised me. I actually am in disbelief of what I heard, I fear the difference was a error in the process. Given the number of variables this could be a software glitch, soundcard issue, a mistake I made when creating the two filters, etc. Switching filters made an audible noise so it's also possible that the filters sound identical but hearing the blip between switching caused my brain to think it heard a difference. My guess is I was hearing the software make the sampling rate conversion. Not the sound of the sampling rate conversion just an unmuted bit of noise. I hope to find out more and see if I can report back more thoroughly In a seperate thread.

I've also heard clear differences in the sound of upsampling to 384khz PCM vs 5.8mhz DSD. I don't believe there are actually big Sonic differences here. I think what I've heard (which was clear as day to hear) was actually a problem in the upsampling process. That if we measured the output we would discover some artifact causing the difference.

Fascinating. I read, also, that the Dirac Software has a higher ultimate sample rate capability than what is being implemented in the MiniDSP products. I assume due to hardware/processing limitations with the MiniDSP stuff (and virtually anything else that isn't a computer; not a knock on MiniDSP at all).

I'd like to hear more of your thoughts once you suss-out what the issue between the filters was.

 

[Matthew J Poes]

Fascinating. I read, also, that the Dirac Software has a higher ultimate sample rate capability than what is being implemented in the MiniDSP products. I assume due to hardware/processing limitations with the MiniDSP stuff (and virtually anything else that isn't a computer; not a knock on MiniDSP at all).

I'd like to hear more of your thoughts once you suss-out what the issue between the filters was.

I actually commented on that and was corrected that at least one or two of the 2 channel processors do 96khz. Otherwise yes they are limited, I think by dsp chip.

I hate to make a bold claim like 96khz dirac filters sound better with 96khz content than 48khz if it's a one off odd ball occurance. A few folks have measured some of this and found that he audible differences were due to actual processing errors. Oscillations, ringing, higher distortion, noise, etc. I hope to maybe recors the output of the two scenarios and load them as musical tracks that can be played back and forth with no switching noise. Maybe have a friend listen with me.

I think Dirac makes a large noticable difference. When I had a friend over to hear for himself, he described the improvement as a lot more sudtle than implied. Clearly different, mostly better, bit only fractionally so. I am intimately familiar with the sound of my system. He was hearing it for the first time, so I think he was a little in awe of the overall sound and so has a harder time distinguishing that last few percent improvement.

How you measure makes a big difference in how good the correction filters are too, so I really want to nail that down. If I take 20 identical measurements back to back with REW, it isn't uncommon to get at least one weird measurement with a lot of ringing or an odd dip or peak. What if Dirac gets one of those measurements in it's mix and you get a bad correction curve.

See this measurement of a dirac corrected response. The fundamental is the speaker response. This had 8 measurements that were pretty close together. Changes in height and symetry, but all in one seating position. If I average 8 measurements in the same places I get a at response like the guys hear are showing (if I use one measurement point I'd also get the same flatness). That peak at 38hz is a measurement artifact. Dirac put it there. It's totally unrelated to any room modes and isn't present in other measurements without dirac. No idea why dirac did that, but my guess is a phase anomlie in the IR for at least one measurement point.

 

[Sonnie Parker]

Moved this to its own thread, as it was not directly relevant to the speakers evaluation and needs its own thread. Excellent question.

 

[Matthew J Poes]

Thanks Sonnie! I was thinking the same thing. I was actually responding to this at the airport on my phone and so couldn't move anything and went to go look right now and saw it was moved.

 

[Flak]

Let me say just for information that the UMIK-1 mic (as far as I know) only supports 48 kHz in which case the DLCT will also play the stimuli at 48 kHz in order to minimize resamplings.

With reference to miniDSP's internal processing at 48 kHz I'll cite Jakob Agren's short digest on the subject:

"The problem is that the DA converters are doing different things at different sample rates. According to the Nyquist criterion we cannot have a representation of above fs/2 in the analog signal, where fs is the sample rate. On the AD side of things it means that if we sample at 48kHz we can only represent content up to 24kHz, and that the theoretical upper limit. In practice that limit is lower. Any content that is in the analog signal that has content above 24k will still have an effect on the digital representation; this is called aliasing. To minimize the amount of aliasing in the AD stage a low pass filter is always present to make sure that there is as little content as possible above 24kHz.
Now, in the DA stage the same is still true. If we know the input data has a sample rate of 48kHz, the analog output signal can never ever contain any information above 24k (in theory, lower in practice). The DA converter may still produce things above 24k as byproducts of the DA conversion, and therefore low pass filters are applied to remove this content. If the sample rate is 24kHz and we want to have “good” content at least to 22kHz this low pass need to be really steep, and will affect the phase of audible high frequency content. This is a strong case to use sample rates higher than 48k, for instance 96kHz. If the sample rate is 96kHz the Nyquist frequency is 48kHz and applying the low pass required to remove crap from the reconstructed analog signal is now much easier; you have the whole range from 24k to 48k to wreck with your low pass, and you can hear that anyways. Go to 192kHz and you have even more “headroom” to do this operation.
But... this used to be a real problem back in the past. Now however any decent quality DAC will use internal upsampling to a really high rate (384k or 768k is not uncommon), apply digital low passes there, then do the DA conversion. That is, the DAC chips themselves deal with this internally and is not dependent on the source material having a high sample rate to ease the DA conversion"

miniDSP itself would answer appropriately about the DDRC-24 and DDRC-88 but I'm confident
Flavio

 

[Matthew J Poes]

Let me say just for information that the UMIK-1 mic (as far as I know) only supports 48 kHz in which case the DLCT will also play the stimuli at 48 kHz in order to minimize resamplings.

With reference to miniDSP's internal processing at 48 kHz I'll cite Jakob Agren's short digest on the subject:

"The problem is that the DA converters are doing different things at different sample rates. According to the Nyquist criterion we cannot have a representation of above fs/2 in the analog signal, where fs is the sample rate. On the AD side of things it means that if we sample at 48kHz we can only represent content up to 24kHz, and that the theoretical upper limit. In practice that limit is lower. Any content that is in the analog signal that has content above 24k will still have an effect on the digital representation; this is called aliasing. To minimize the amount of aliasing in the AD stage a low pass filter is always present to make sure that there is as little content as possible above 24kHz.
Now, in the DA stage the same is still true. If we know the input data has a sample rate of 48kHz, the analog output signal can never ever contain any information above 24k (in theory, lower in practice). The DA converter may still produce things above 24k as byproducts of the DA conversion, and therefore low pass filters are applied to remove this content. If the sample rate is 24kHz and we want to have “good” content at least to 22kHz this low pass need to be really steep, and will affect the phase of audible high frequency content. This is a strong case to use sample rates higher than 48k, for instance 96kHz. If the sample rate is 96kHz the Nyquist frequency is 48kHz and applying the low pass required to remove crap from the reconstructed analog signal is now much easier; you have the whole range from 24k to 48k to wreck with your low pass, and you can hear that anyways. Go to 192kHz and you have even more “headroom” to do this operation.
But... this used to be a real problem back in the past. Now however any decent quality DAC will use internal upsampling to a really high rate (384k or 768k is not uncommon), apply digital low passes there, then do the DA conversion. That is, the DAC chips themselves deal with this internally and is not dependent on the source material having a high sample rate to ease the DA conversion"

miniDSP itself would answer appropriately about the DDRC-24 and DDRC-88 but I'm confident
Flavio

Hi Flak, good to have you stop in.

Is what you are saying true or the generation of correction filters? That is, if you measure the response using a 48khz sampling rate that the correction filters can only pass 48khz? I assumed that the correction filters were generated at high sampling rates based on the source info. That the correction filters would only correct the range of the measured response, but the filter would still pass on the information in the musical signal that is beyond the speakers correction range?

Unless I misunderstood it sounds like you are saying that the use of a umik would limit the musical signal that could be corrected to the Mic measurement of 48khz sampling (thus 24k response).

I agree with the rest of what you are saying, that pushing the low pass filter beyond the audible range by a large margin is likely the most audible effect with high sampling rate. Having said that, the meta-analysis I alluded to contained studies that accounted for that and found that content with information beyond the 24khz limit was in the high sampling tracks. That these tracks were discernable from standard sampling rates. They obviously don't prove we hear beyond 20khz and they don't prove why they are discernable. They do prove that upsampling doesn't give the same results.

This is a tangential topic but the meta-analysis is interesting to me. It begs alot of new questions. It establishes a strong Evidence base for encoding music at high sampling rate, but it also provides a lot of new questions as to why that would matter. It wasn't clear to me that these studies were capable of reproducing the high frequency content which makes me think audibility isn't the issue, but the reconstruction filter can't be the only answer either.

Here is the study in question:

http://www.aes.org/e-lib/browse.cfm?elib=18296

 

[Flak]

Hi MjPoes,
thanks for your link to that detailed study and your interesting comments.

When mentioning the UMIK-1 specs I was referring only to the generation of correction filters, in other words the UMIK-1, even if bandwidth limited, is a valid solution for the purpose...
there is nothing to correct beyond the audibility range.

The subject of perception of high resolution audio is a different one and at first sight it seems to me that the referenced meta-analysis has (correctly) tried to look only into the possibility of perceiving a difference... it does'nt (and probably cannot easily) get into judging whether those eventual subtle differences are for the better or worse (the latter being a possibility in my personal opinion)

Anyhow others at Dirac certainly have much more knowledge, I'm focused on large improvements in the audible range that don't require such an in depth research in order to find out if they can be perceived.

Flavio

 

[Matthew J Poes]

Hi Flavio, I completely agree with your assessment. Good to know it works as advertised.

I don't get hung up on these small differences. You will see my first comment was that the linear distortion correction of Dirac far outweighs any minor losses from resampling.

My own experiences were not with sound quality benefit. It was with odd behaviors from mismatches in the sampling frequency causing resampling and possibly causing odd sounds.

My view on the higher sampling and bit rate is not so much I think it sounds far superior so we should do it. It's that it doesn't hurt anything hen done right, and in todays world, it doesn't need to cost anything more. I see no reason not to do it. If it sounds better, great. If it doesn't, well who cares, it didn't hurt anything to do it. The streaming capabilities and storage capabilities of today can more than handle this. The DA and AD converts are already high bit rate and high sampling rate. Just makes sense to me. The fact that the overall preponderance of evidence is that its a noticeable difference as well is all the more reason to do it.

 

[AudiocRaver]

...The subject of perception of high resolution audio is a different one and at first sight it seems to me that the referenced meta-analysis has (correctly) tried to look only into the possibility of perceiving a difference... it does'nt (and probably cannot easily) get into judging whether those eventual subtle differences are for the better or worse (the latter being a possibility in my personal opinion)

Anyhow others at Dirac certainly have much more knowledge, I'm focused on large improvements in the audible range that don't require such an in depth research in order to find out if they can be perceived...

Flavio,

Thanks for your input. Based on my own experience, I agree that the differences (1) with higher res tracks, or (2) through resampling to a higher rate, when audible at all, are minute, that even being able to detect them at all usually requires a quick A-B comparison, in itself not a straightforward matter. And further agree that the difference is not easily judged as "better" or "worse", but simply as "very slightly different."

...I don't get hung up on these small differences. You will see my first comment was that the linear distortion correction of Dirac far outweighs any minor losses from resampling.

I COMPLETELY agree, the benefits of frequency and phase response correction far outweigh any minute downside. I have yet to conclusively hear a downside myself. Earlier versions of the Dirac Calibration tool were project-file compatible between the nanoAVR version (48k sample rate) and the full PC version (native sample rate), and I once did a complete calibration on a nanoAVR, then imported the project with measurements into the PC version and created filters for that version there. Switching between them (nanoAVR vs. PC), I thought there was a subtle difference, but it was VERY subtle, I could not claim to do it again in a real A-B test comparison.

Newer versions of the Dirac program do not have that project level compatibility, as far as I know.

...My view on the higher sampling and bit rate is not so much I think it sounds far superior so we should do it. It's that it doesn't hurt anything hen done right, and in todays world, it doesn't need to cost anything more. I see no reason not to do it. If it sounds better, great. If it doesn't, well who cares, it didn't hurt anything to do it. The streaming capabilities and storage capabilities of today can more than handle this. The DA and AD converts are already high bit rate and high sampling rate. Just makes sense to me...

My view exactly. It it is easy and cheap and should give some benefit with no downside, then why not? No need to prove anything. Just DO IT!

The fact that the overall preponderance of evidence is that its a noticeable difference as well is all the more reason to do it.

Hmmm, still thinking about that one.

 

[Matthew J Poes]

Flavio,

Thanks for your input. Based on my own experience, I agree that the differences (1) with higher res tracks, or (2) through resampling to a higher rate, when audible at all, are minute, that even being able to detect them at all usually requires a quick A-B comparison, in itself not a straightforward matter. And further agree that the difference is not easily judged as "better" or "worse", but simply as "very slightly different."



I COMPLETELY agree, the benefits of frequency and phase response correction far outweigh any minute downside. I have yet to conclusively hear a downside myself. Earlier versions of the Dirac Calibration tool were project-file compatible between the nanoAVR version (48k sample rate) and the full PC version (native sample rate), and I once did a complete calibration on a nanoAVR, then imported the project with measurements into the PC version and created filters for that version there. Switching between them (nanoAVR vs. PC), I thought there was a subtle difference, but it was VERY subtle, I could not claim to do it again in a real A-B test comparison.

Newer versions of the Dirac program do not have that project level compatibility, as far as I know.



My view exactly. It it is easy and cheap and should give some benefit with no downside, then why not? No need to prove anything. Just DO IT!



Hmmm, still thinking about that one.

See the JAES Meta-analysis I linked. They calculated the overall effect size across all studies which were of sufficient quality and methodology for inclusion and found an overall effect in favor of hearing a difference. The studies results are fairly conclusive for what they are. They don't tell us that people prefer this difference or why we hear a difference. Just that across all studies (of high quality) a difference was reliably heard greater than chance. I've done a few Meta-analysis for my day job and feel this was well done. The article is a good read.

 

[AudiocRaver]

Thank you, that clarifies your comment for me and makes perfect sense.