Reasonably priced good quality 6 or 8 channel USB DAC?

[pos]

DAC equipped with ASRC do resample to the highest sampling freq they can use, and this is a logical choice.

 

[andreasmaaan]

Yes I know but the thing is when the resampler is the same as the incoming sample rate or a close multiple or division of it then you will get all kinds of audible artifacts from the clocks not being in sync.

Could you explain what artefacts you mean exactly?

 

[JustIntonation]

Could you explain what artefacts you mean exactly?

Did a quick google search but can't find it.
I read a good whitepaper about it I think 10-15 years ago with actual measurements etc explaining this in depth. Only thing that stuck from this was to never use an ASRC at the incoming sample rate or a low integer multiple/division of it.
But rest assured it is a real thing. And Bruno (and many other DAC manufacturers) apparently also thinks so and he surely knows his things.

 

[JustIntonation]

I have done this, you can make the filter minimum phase. I have done this as a test to try and reduce latency for theatre duties. Short version is It really doesn't sound as good. Poorer imaging and less clarity.

Please see this file: https://we.tl/t-GiNDRUuPWy
It is part of a test I did. One file (original) is the original music, the other file is a simulated LR16 (96dB/oct) crossover at 2kHz simulated on-axis (so an allpass filter at this angle). The group delay around the crossover freq is 4 times the wavelength of 2kHz.
Of course this is a more extreme example than a LR4 filter. But I cannot tell any difference at all between the two files.
Can you? Note that two songs, one starting around 1:36 and the other around 2:36 have a deep soundstage. I cannot tell any difference in soundstage size or depth between the two files.
If you use ABX for Foobar2000 or something similar you can test this properly and switch back/forth etc.

 

[JustIntonation]

Please see this file: https://we.tl/t-GiNDRUuPWy
It is part of a test I did. One file (original) is the original music, the other file is a simulated LR16 (96dB/oct) crossover at 2kHz simulated on-axis (so an allpass filter at this angle). The group delay around the crossover freq is 4 times the wavelength of 2kHz.
Of course this is a more extreme example than a LR4 filter. But I cannot tell any difference at all between the two files.
Can you? Note that two songs, one starting around 1:36 and the other around 2:36 have a deep soundstage. I cannot tell any difference in soundstage size or depth between the two files.
If you use ABX for Foobar2000 or something similar you can test this properly and switch back/forth etc.

Oh well actually and now I just read this:
https://en.wikipedia.org/wiki/Group_delay_and_phase_delay#Group_delay_in_audio
Which claims a 2 period group delay is the threshold of audibility at 2kHz.. Perhaps tested with a pure impulse? So that would correspond to a LR8 (48dB/oct).
At 500Hz it is 1.6 period corresponding to about LR6 (36dB/oct).
So perhaps the safest setting for minimum phase would indeed be LR4 (24dB/oct). This does infact make a case for linear phase or a mix of linear phase and minimal phase with a total group delay of 1 period (to be on the safe side) if one uses high order crossover filters. Even though in my example I can't hear a difference with the steep 96dB/oct apparently there are conditions in which the LR16 minimal phase group delay is audible at 2kHz.

edit: yes that makes sense actually. My little test was for simulating the phasing and ringing off-axis and the noise at the end was actually the clearest for this. For testing audibility of group delay an impulse or a low frequency saw or square waveform would be best to test the limit of audibility.. I'll redo a test for this later.

 

[March Audio]

I will listen when I get a chance.

Your major concern with FIR appears to be pre ringing. Not wanting to get into the debate about audibility, but with acourate you can look at the impulse signal and see the amount of pre ringing. you can also adjust the amount of phase compensation.

https://www.audiovero.de/pdf/AcouratePRCen.pdf

here is a snip from the above document. How concerned are you about the pre ringing you see here

Hers is one of my own measurements (a random one I grabbed out of a pile of ones I did). How concerned are you about this level of pre ringing?

or this one

 

[Blumlein 88]

Did a quick google search but can't find it.
I read a good whitepaper about it I think 10-15 years ago with actual measurements etc explaining this in depth. Only thing that stuck from this was to never use an ASRC at the incoming sample rate or a low integer multiple/division of it.
But rest assured it is a real thing. And Bruno (and many other DAC manufacturers) apparently also thinks so and he surely knows his things.

I think you slightly misunderstood this. Though you aren't altogether wrong. Using even multiples in resampling didn't filter out any jitter in the clocks. By using ASRC (asynchronous sample rate conversion) with non-even multiples, you could very effectively filter jitter as well as change sample rates. It is why Benchmark products did the unusual thing in their DACs of resampling all input rates to 110 khz rates internally (I think it was 110 khz). Even 192 khz inputs were resampling to 110 khz. As I recall the two benefits were Benchmark could better handle intersample overs and filter jitter to the point it was exceptionally low even with jittery inputs. In current Benchmark DACs they resample to 211 khz. They had a paper about this and it may be what you read a decade or so ago.

 

[JustIntonation]

I think you slightly misunderstood this. Though you aren't altogether wrong. Using even multiples in resampling didn't filter out any jitter in the clocks. By using ASRC (asynchronous sample rate conversion) with non-even multiples, you could very effectively filter jitter as well as change sample rates. It is why Benchmark products did the unusual thing in their DACs of resampling all input rates to 110 khz rates internally (I think it was 110 khz). Even 192 khz inputs were resampling to 110 khz. As I recall the two benefits were Benchmark could better handle intersample overs and filter jitter to the point it was exceptionally low even with jittery inputs. In current Benchmark DACs they resample to 211 khz. They had a paper about this and it may be what you read a decade or so ago.

Thanks. I believe the paper I read was by Lavry but I could be mistaken..
I thought it was something that the jitter in the case of transfer to ASRC could give audible imaging or ahh I don't remember what. But real audible frequencies in the result and that this was much reduced by using an ASRC that wasn't a low integer multiple (or perhaps indeed an even multiple as you say). I remember the 110 kHz of the Benchmark indeed.
Untill now I didn't realize Benchmark and Lavry were doing something unusual, I thought everybody did it when using an ASRC before the DAC (most DACs weren't using one back then). So I wasn't surprised to see it in Hypex and Bruno's designs but rather was surprised to see that MiniDSP are not doing it..

 

[Blumlein 88]

Thanks. I believe the paper I read was by Lavry but I could be mistaken..
I thought it was something that the jitter in the case of transfer to ASRC could give audible imaging or ahh I don't remember what. But real audible frequencies in the result and that this was much reduced by using an ASRC that wasn't a low integer multiple (or perhaps indeed an even multiple as you say). I remember the 110 kHz of the Benchmark indeed.
Untill now I didn't realize Benchmark and Lavry were doing something unusual, I thought everybody did it when using an ASRC before the DAC (most DACs weren't using one back then). So I wasn't surprised to see it in Hypex and Bruno's designs but rather was surprised to see that MiniDSP are not doing it..

It is quite possible it was Lavry. They had a similar systems of ASRC called CrystalLock.

 

[JustIntonation]

I will listen when I get a chance.

Your major concern with FIR appears to be pre ringing. Not wanting to get into the debate about audibility, but with acourate you can look at the impulse signal and see the amount of pre ringing. you can also adjust the amount of phase compensation.

https://www.audiovero.de/pdf/AcouratePRCen.pdf

here is a snip from the above document. How concerned are you about the pre ringing you see here

View attachment 17186

Hers is one of my own measurements (a random one I grabbed out of a pile of ones I did). How concerned are you about this level of pre ringing?

View attachment 17187

or this one

View attachment 17188

My biggest objections against linear phase are theoretical regarding the pre-ringing, and latency. I will be using my speaker system for making music as well as listening to it. I could make 2 presets, one for making music one for listening and editing etc where latency isn't an issue.
But I still think minimal phase is the "right" way to do things. Be it 24dB/oct LR4.
I'll do testing regarding group delay audibility later (perhaps when the speaker is finished as right now I'm without a good headphone allready sold it to help fund my speaker build)

Btw, are those measurements with both speakers running at the same time?
If not and this is of one speaker, then there is a fair amount of direct reflection interference visible in your measurements which can be reduced with more absorption at early reflection points it seems to me?

 

[March Audio]

IThe pre ringing IMO is not a significant issue and can be dealt with with judicious use of the processing controls in Accurate. The measurement above show this.

Latency however you will not overcome.

I cannot remember the conditions of the measurements, these are just a couple of randoms I grabbed to demonstrate the lack of pre-ringing. However the room is not an issue. Carefully treated to bring reverberation time down evenly across the frequency range so direct and reflected sound isn't overly different. If you over do absorption it affects this balance. The other effect of too much absorption at first reflection point is that it reduces the sense of spaciousness. Those B&Ws werent there for the measurements.

Mixture of absorption and diffusion including ceiling.

IR - reflections over 20dB down

reverberation - pretty much in line with EBU 3276 studio

 

[andreasmaaan]

Oh well actually and now I just read this:
https://en.wikipedia.org/wiki/Group_delay_and_phase_delay#Group_delay_in_audio
Which claims a 2 period group delay is the threshold of audibility at 2kHz.. Perhaps tested with a pure impulse? So that would correspond to a LR8 (48dB/oct).
At 500Hz it is 1.6 period corresponding to about LR6 (36dB/oct).
So perhaps the safest setting for minimum phase would indeed be LR4 (24dB/oct). This does infact make a case for linear phase or a mix of linear phase and minimal phase with a total group delay of 1 period (to be on the safe side) if one uses high order crossover filters. Even though in my example I can't hear a difference with the steep 96dB/oct apparently there are conditions in which the LR16 minimal phase group delay is audible at 2kHz.

I agree with the gist of this, although the figures on the wiki page come (it seems) from Blauert, whereas a few later studies have found slightly lower thresholds. I read all the studies I could find at one point when I was doing the designs for my own range of speakers. My conclusion at that time was that LR4 above about 700Hz had never been shown to be audible under any test conditions (including headphones, impulse signals, etc.). One study by Genelec did show LR4 to be audible at low-mid frequencies, although that study used headphones and impulse signals. It also seemed from the studies I could find that, below 200-300Hz, our ears again become far less sensitive. Not much research has been done in this frequency range, however.

A member on this site did some of his own studies which are unpublished, in which he found even lower thresholds in some circumstances. His studies were not as scientifically rigorous in that he found statistically significant results only after being selective about the data. But his experimental procedure was otherwise sound. His subjective view was that linear phase filters could sound more neat and compact with some music.

The only study I recall reading in which FIR filters had been investigated concluded that up to 600 taps did not reduce sound quality (although this study didn't investigate the question of absolute audibility). I didn't look into this in as much depth as I was already satisfied that for my purposes LR4 filters would perform adequately (inaudibly) and the delay caused by FIR filters was unacceptable - not to mention the additional cost.

One possible advantage of FIR filters is that the off-axis interference region is going to be much narrower if the slopes are relatively steep. That's nothing to sneeze at

Anyway, your conclusion that LR4 is "safe" (apart perhaps from in the 300-700Hz range) is the same conclusion I came to.

EDIT: the Genelec study showed that LR4 might be audible at low-mid frequencies. Recalling it in more detail now, that study didn't test for any single frequency range in isolation, so the results in that region may have been caused by group delay higher up.

 

[JustIntonation]

I agree with the gist of this, although the figures on the wiki page come (it seems) from Blauert, whereas a few later studies have found slightly lower thresholds. I read all the studies I could find at one point when I was doing the designs for my own range of speakers. My conclusion at that time was that LR4 above about 700Hz had never been shown to be audible under any test conditions (including headphones, impulse signals, etc.). One study by Genelec did show LR4 to be audible at low-mid frequencies, although that study used headphones and impulse signals. It also seemed from the studies I could find that, below 200-300Hz, our ears again become far less sensitive. Not much research has been done in this frequency range, however.

A member on this site did some of his own studies which are unpublished, in which he found even lower thresholds in some circumstances. His studies were not as scientifically rigorous in that he found statistically significant results only after being selective about the data. But his experimental procedure was otherwise sound. His subjective view was that linear phase filters could sound more neat and compact with some music.

The only study I recall reading in which FIR filters had been investigated concluded that up to 600 taps did not reduce sound quality (although this study didn't investigate the question of absolute audibility). I didn't look into this in as much depth as I was already satisfied that for my purposes LR4 filters would perform adequately (inaudibly) and the delay caused by FIR filters was unacceptable - not to mention the additional cost.

One possible advantage of FIR filters is that the off-axis interference region is going to be much narrower if the slopes are relatively steep. That's nothing to sneeze at

Anyway, your conclusion that LR4 is "safe" (apart from in the 300-700Hz range) is the same conclusion I came to.

Wow, good information! Thanks!
Planning LR4 at 150-160Hz and 1.4-1.5kHz myself.

Yes the little test I did showed that reduced off-axis interference because of narrower region (tested up to LR16 96dB/oct) outweighs increased ringing of minimal phase filters. Haven't tested this with linear phase filters though. And also didn't test any effects of off-axis dropoff of the mid driver on this.

 

[andreasmaaan]

Wow, good information! Thanks!
Planning LR4 at 150-160Hz and 1.4-1.5kHz myself.

Yes the little test I did showed that reduced off-axis interference because of narrower region (tested up to LR16 96dB/oct) outweighs increased ringing of minimal phase filters. Haven't tested this with linear phase filters though. And also didn't test any effects of off-axis dropoff of the mid driver on this.

Those are similar slopes/frequencies to what I settled on (80-100Hz and 950-1250Hz). My choices were dictated by driver properties (distortion, power handling, directivity, etc), group delay audibility, off-axis interference, and C2C spacing (as I know yours are too).

 

[JustIntonation]

Those are similar slopes/frequencies to what I settled on (80-100Hz and 950-1250Hz). My choices were dictated by driver properties (distortion, power handling, directivity, etc), group delay audibility, off-axis interference, and C2C spacing (as I know yours are too).

Do you have a description and pictures somewhere of your speakers? Curious

 

[andreasmaaan]

Do you have a description and pictures somewhere of your speakers? Curious

I'm still prototyping at the moment, so no pictures... But I can happily describe them here.

They are basically higher-end PA speakers designed with reference to psychoacoustic principles. They use drivers from B&C, amps from Pascal, and DSP/DACs from ALLDSP.

The waveguides are designed in-house and are conical elliptical / constant directivity. They avoid discontinuities with the compression driver exit and minimise diffraction in the transition region along the lines suggested by Geddes. Directivity is 90° x 60° in the upper-mids and highs, transitioning to omni in the bass. The baffle and enclosure shape ensure the transition is not too abrupt.

The boxes are 15mm properly reinforced birch ply, and the ports are cylindrical, flat-walled and flared at both ends (this is the case for both mains and subs).

The models comprise:

• 12" woofer, 1.7" voice coil / 1" exit compression driver, F3 of 57Hz, 4pi max SPL 126dB (100-20,000Hz)
• 15" woofer, 2.5" voice coil / 1.4" exit compression driver, F3 of 49Hz, 4pi max SPL 132dB (100-20,000Hz)
• 18" subwoofer, ported, F3 of 36Hz, 2pi max SPL 132dB (about 122dB before port air velocity reaches 14m/s)
• 21" subwoofer, ported, F3 of 29Hz, 2pi max SPL 132dB (about 126dB before port air velocity reaches 14m/s)

The figures I've given may not seem to compare that well to some manufacturer's claims, particularly the max. SPL figures, but I take into account sensitivity loss due to baffle step, etc (it's typical to simply take the 2pi sensitivity of the woofer and add it to 10 * log(power handling)).

The main selling point relative to other similar PA speakers is the use of very high quality components and the attention to detail, especially in terms of the waveguides, ports, and cabinets. I'm in Europe, and retail prices will be in the range of 1700-3000€ depending on the model. Mark-up is pretty low by large manufacturer standards but it's a labour of love for me, and in any case I don't expect these will have a huge market outside a few small-medium sized venues and clubs that take sound quality very seriously.

Anyway, I could go on for hours about these... Maybe better to wait til I actually put the prototypes into production and then I can post a link to more detailed information and measurements, which I will publish on the website.

 

[JustIntonation]

I'm still prototyping at the moment, so no pictures... But I can happily describe them here.

They are basically higher-end PA speakers designed with reference to psychoacoustic principles. They use drivers from B&C, amps from Pascal, and DSP/DACs from ALLDSP.

The waveguides are designed in-house and are conical elliptical / constant directivity. They avoid discontinuities with the compression driver exit and minimise diffraction in the transition region along the lines suggested by Geddes. Directivity is 90° x 60° in the upper-mids and highs, transitioning to omni in the bass. The baffle and enclosure shape ensure the transition is not too abrupt.

The boxes are 15mm properly reinforced birch ply, and the ports are cylindrical, flat-walled and flared at both ends (this is the case for both mains and subs).

The models comprise:

• 12" woofer, 1.7" voice coil / 1" exit compression driver, F3 of 57Hz, 4pi max SPL 126dB (100-20,000Hz)
• 15" woofer, 2.5" voice coil / 1.4" exit compression driver, F3 of 49Hz, 4pi max SPL 132dB (100-20,000Hz)
• 18" subwoofer, ported, F3 of 36Hz, 2pi max SPL 132dB (about 122dB before port air velocity reaches 14m/s)
• 21" subwoofer, ported, F3 of 29Hz, 2pi max SPL 132dB (about 126dB before port air velocity reaches 14m/s)

The figures I've given may not seem to compare that well to some manufacturer's claims, particularly the max. SPL figures, but I take into account sensitivity loss due to baffle step, etc (it's typical to simply take the 2pi sensitivity of the woofer and add it to 10 * log(power handling)).

The main selling point relative to other similar PA speakers is the use of very high quality components and the attention to detail, especially in terms of the waveguides, ports, and cabinets. I'm in Europe, and retail prices will be in the range of 1700-3000€ depending on the model. Mark-up is pretty low by large manufacturer standards but it's a labour of love for me, and in any case I don't expect these will have a huge market outside a few small-medium sized venues and clubs that take sound quality very seriously.

Anyway, I could go on for hours about these... Maybe better to wait til I actually put the prototypes into production and then I can post a link to more detailed information and measurements, which I will publish on the website.

Wow very cool!
Btw, perhaps I have a link for you which can be helpful for cabinet construction choice:
http://www.hsi-luidsprekers.nl/index.php/projecten/12-kast-materialen-onderzoek If you translate it from Dutch to your language with Google.
I myself will use a combination of OSB (maaybe cement bonded particleboard) on the inside and MDF on the outside, probably both 19mm thick, and additional bitumen on the inside and proper reinforcing / bracing / ribs. And personally stuffing with wool. If you look at measurements these things really matter a lot, not sure if 15mm plywood is optimal even with proper reinforcements but perhaps in your situation weight matters as well for PA speakers?

 

[andreasmaaan]

Wow very cool!
Btw, perhaps I have a link for you which can be helpful for cabinet construction choice:
http://www.hsi-luidsprekers.nl/index.php/projecten/12-kast-materialen-onderzoek If you translate it from Dutch to your language with Google.
I myself will use a combination of OSB (maaybe cement bonded particleboard) on the inside and MDF on the outside, probably both 19mm thick, and additional bitumen on the inside and proper reinforcing / bracing / ribs. And personally stuffing with wool. If you look at measurements these things really matter a lot, not sure if 15mm plywood is optimal even with proper reinforcements but perhaps in your situation weight matters as well for PA speakers?

Thanks! That link is excellent, I hadn't seen such a concise and detailed set of measurements on this topic before.

Actually, for the first prototype boxes we did experiment with bitumen sheets adjacent to the birch ply walls, and also with various thicknesses of birch ply. There was definitely some benefit, but in the end we found that reinforcement was the best bang for buck/weight, and it was difficult to get below our target cabinet weight using other combinations of materials.

And I didn't mention it in the other post, but we are also using 100mm high-density rockwool to partially line the cabinets. So I guess the system we've settled on is most similar to example 18 in the link you shared (although using a slightly lower birch ply thickness and less wool), which is acceptable for our purposes.

And our baffles are 18mm birch ply, which I also forgot to mention before!

 

[JustIntonation]

Thanks! That link is excellent, I hadn't seen such a concise and detailed set of measurements on this topic before.

Actually, for the first prototype boxes we did experiment with bitumen sheets adjacent to the birch ply walls, and also with various thicknesses of birch ply. There was definitely some benefit, but in the end we found that reinforcement was the best bang for buck/weight, and it was difficult to get below our target cabinet weight using other combinations of materials.

And I didn't mention it in the other post, but we are also using 100mm high-density rockwool to partially line the cabinets. So I guess the system we've settled on is most similar to example 18 in the link you shared (although using a slightly lower birch ply thickness and less wool), which is acceptable for our purposes.

And our baffles are 18mm birch ply, which I also forgot to mention before!

Ah yes if weight is an issue then bitumen isn't the best stuff
As for Rockwool, it works best in the center where the air is moving, close to the walls the air is hardly moving only pressure changes and Rockwool doesn't do much there.
If you have a ported system then you need a clear air path from the woofer to the port so harder to put something in the center but even more importantly if you have a ported system you cannot use Rockwool or fibreglass! The port will spew out mini particles, very harmful and irritating, for ported systems there are other materials which do not produce dust particles. Rockwool and fibreglass are only for closed boxes.
For a good laugh and cry see this:

 

[andreasmaaan]

Ah yes if weight is an issue then bitumen isn't the best stuff
As for Rockwool, it works best in the center where the air is moving, close to the walls the air is hardly moving only pressure changes and Rockwool doesn't do much there.
If you have a ported system then you need a clear air path from the woofer to the port so harder to put something in the center but even more importantly if you have a ported system you cannot use Rockwool or fibreglass! The port will spew out mini particles, very harmful and irritating, for ported systems there are other materials which do not produce dust particles. Rockwool and fibreglass are only for closed boxes.
For a good laugh and cry see this:

Yeh, that's true. The rockwool is not flush with the walls but separated by an air gap (it's attached to the bracing, actually). We've used rockwool in the prototypes so far, wrapped inside cloth, but one thing I've been meaning to do is talk to an expert in this area to clarify whether the cloth is adequate for containing the rockwool particles when air moves through it. If not, we'll have to switch materials.. Is this something you know a bit about?