Reasonably priced good quality 6 or 8 channel USB DAC?

[Vincent Kars]

No that isn't how ASRC works. There is no relation of the ASRC clocks to the incoming spdif signal at all.

You can probably solve this by chaining multiple DACs using the word clock out of the first to slave the 2nd and Nth DACs

https://www.audiocircle.com/index.php?topic=113946.0

 

[Kal Rubinson]

You can probably solve this by chaining multiple DACs using the word clock out of the first to slave the 2nd and Nth DACs.
https://www.audiocircle.com/index.php?topic=113946.0

1. This requires a specific driver from Mytek which, afaik, is no longer available for the newer DACs/firmware.
2. OTOH, you can also do this without that firmware (but still with the clock links) but only on a recent MacOS:. https://www.stereophile.com/content/music-round-84-multichannel-mqa
3. Also, you can avoid the clock-links and use any three DACs of your choosing by getting a miniDSP U-DIO8 to split the output: https://www.stereophile.com/content/music-round-93-minidsp-ripping-sacds

 

[andymok]

I thought a better way to distribute clock signal was using T-joint?

 

[pos]

@JustIntonation, did you get a quote for the dac8+psu+usbstreamer+case custom build ?
Having the toslink in/out port exposed on the rear plate could prove useful. That means you would need a USB cable with a socket as this port is on the other side of the usbstreamer.

 

[JustIntonation]

@JustIntonation, did you get a quote for the dac8+psu+usbstreamer+case custom build ?
Having the toslink in/out port exposed on the rear plate could prove useful. That means you would need a USB cable with a socket as this port is on the other side of the usbstreamer.

Did not receive a reply yet from Okto Research.
For me personally I don't care at all about any other inputs or outputs than USB in balanced analogue outs and a nice volume knob for the chip

Btw looking at PC DSP crossovers now I'll share a list here:
https://dephonica.com/ (€70)
http://www.lupisoft.com/ekio/ ($150)
http://www.thuneau.com/alloclite.htm ($25)
https://www.audiovero.de/en/acourate.php (€340)
http://www.bodziosoftware.com.au/ ($150)
http://juicehifi.com/ ($390)
https://rruitr.home.xs4all.nl/sxq.html (free)
http://www.dsprobotics.com/support/viewtopic.php?f=3&t=2468 (develope your own crossover app it seems, $99)
https://www.vb-audio.com/Cable/index.htm (handy utility, free)
And perhaps also good is seperate EQ plugins running in a lightweight VST host, many of those around for free.

Any good ones I missed?

 

[JustIntonation]

Btw, if you don't use FIR (like I won't) and latency matters then there is further difference between the above plugins. Have yet to order them according to minimal latency.

 

[JustIntonation]

Well there's still some things to learn here regarding DAC chip sample rate and DSP processing..

As far as I can tell DAC chips perform worse at higher sample rates. Where 44.1 or 48kHz are best for low distortion and performance becomes worse at higher sample rates.
Does anybody know how much worse? And does this hold for the ES9028 chip which perform some form of upsampling to the native chip format anyhow? (what I've read so far seems to indicate that it does still matter for the ES9028 chip)

And this is in contrast to where DSP EQ / filters perform best.
First of all, the more I read about it the more it becomes clear to me that FIR / FFT filters are not the way to go for soundquality (and latency and processing requirements). IIR filters are best for absolute soundquality.
But.. A bare IIR filter has its problems at higher frequencies as by default its amplitude and phase response is 0 at the nyquist frequency. For many filters / EQ curvers it should not be 0 at the nyquist and if it is forced to be so you get asymetrical curves and a form of distortion / "digital' sound (which affects frequencies far below nyquist). The easy fix for this is oversampling and running the EQ/filters at a high sampling rate. But this clashes with the low sampling rate requirement of the best DAC performance and additionally requires good synchronous sample rate converters with additional latency and processing load.
The alternative is IIR filters which are compensated for this in both phase and amplitude / are not 0 around the nyquist. These EQ/filters exist but are in the minority. Furthermore the IIR filter should be 64bit as any errors will be magnified by feedback in higher Q.
So lots of things to look out for. Luckily this is all software and can be fidled with endlessly and most often for free.

 

[pos]

First of all, the more I read about it the more it becomes clear to me that FIR / FFT filters are not the way to go for soundquality (and latency and processing requirements). IIR filters are best for absolute soundquality.

I would not agree on that.

In terms of sound quality:

FFT (frequency domain) convolution can be very close to direct (time domain) convolution, and even more precise for very long FIRs. This all goes down to the implementation, but good one do exist.
Now if you compare direct convolution with IIR biquads the latter is much more prone to quantization errors due to the recursion nature, and of course the longer the ringing (lower frequency and/or higher Q) the more the errors. Direct convolution it is the most straightforward and simple (brutal!) way of applying coefficients, with a simple multiplication and addition, and quantization errors are minimal.

In terms of latency:

Direct convolution does not imply any latency: one sample goes in, one sample goes out. Now the FIR itself can imply a latency, and this is typically the case with linear-phase correction, but this is not necessarily the case. If the FIR represents a minimum-phase correction (like what IIRs have to stick to) then it can be made zero latency as well.
This is an important and alas often overlooked thing: IIR can only be causal (minimum-phase or excess phase), whereas FIR can be acausal (linear-phase, etc.), *but* can also be causal (minimum-phase or excess phase). In fact FIR can do anything when it comes to magnitude/phase relation, whereas IIR is limited by causality.

Time domain (FFT) convolution processing adds its own latency (FFT buffer), but this can be much reduced using partitioning (like what BruteFIR does), and even zeroed by using direct convolution for the first part of the FIR.

In terms of processing power:

Direct convolution is easy for a DSP or an FPGA, but pretty difficult for a computer CPU.
FFT convolution is very easy for a CPU, but care has to be taken on the implementation, both in term of sound quality and processing latency.

In terms of portability / predictability:

IIR/biquad implementations are all over the place when it comes to the actual response you get for a given EQ and filter setting.
Not even talking about EQ convention here (proportional Q vs constant Q, shelving cuttoff frequency position, etc. This is a mess), but simple application of a given set of biquad coefficients. Depending on the implementation the response will vary from one device/software to another, and you cannot pass settings around and expect results to be identical, even when you can be sure both devices use the same conventions.

On the other hand a FIR literately represents the transfer function, and if the convolution process is properly implemented it will always give the same result regardless of the device/software that apply it. Direct or FFT convolution all the same, WYSIWYG!

Me personally I like direct convolution the best, and only use FIR even for purely minimum-phase corrections.
I like the brutal nature of this most simple algorithm


my .2€

 

[JustIntonation]

I would not agree on that.

In terms of sound quality:

FFT (frequency domain) convolution can be very close to direct (time domain) convolution, and even more precise for very long FIRs. This all goes down to the implementation, but good one do exist.
Now if you compare direct convolution with IIR biquads the latter is much more prone to quantization errors due to the recursion nature, and of course the longer the ringing (lower frequency and/or higher Q) the more the errors. Direct convolution it is the most straightforward and simple (brutal!) way of applying coefficients, with a simple multiplication and addition, and quantization errors are minimal.

In terms of latency:

Direct convolution does not imply any latency: one sample goes in, one sample goes out. Now the FIR itself can imply a latency, and this is typically the case with linear-phase correction, but this is not necessarily the case. If the FIR represents a minimum-phase correction (like what IIRs have to stick to) then it can be made zero latency as well.
This is an important and alas often overlooked thing: IIR can only be causal (minimum-phase or excess phase), whereas FIR can be acausal (linear-phase, etc.), *but* can also be causal (minimum-phase or excess phase). In fact FIR can do anything when it comes to magnitude/phase relation, whereas IIR is limited by causality.

Time domain (FFT) convolution processing adds its own latency (FFT buffer), but this can be much reduced using partitioning (like what BruteFIR does), and even zeroed by using direct convolution for the first part of the FIR.

In terms of processing power:

Direct convolution is easy for a DSP or an FPGA, but pretty difficult for a computer CPU.
FFT convolution is very easy for a CPU, but care has to be taken on the implementation, both in term of sound quality and processing latency.

In terms of portability / predictability:

IIR/biquad implementations are all over the place when it comes to the actual response you get for a given EQ and filter setting.
Not even talking about EQ convention here (proportional Q vs constant Q, shelving cuttoff frequency position, etc. This is a mess), but simple application of a given set of biquad coefficients. Depending on the implementation the response will vary from one device/software to another, and you cannot pass settings around and expect results to be identical, even when you can be sure both devices use the same conventions.

On the other hand a FIR literately represents the transfer function, and if the convolution process is properly implemented it will always give the same result regardless of the device/software that apply it. Direct or FFT convolution all the same, WYSIWYG!

Me personally I like direct convolution the best, and only use FIR even for purely minimum-phase corrections.
I like the brutal nature of this most simple algorithm


my .2€

Usually FIR is used for linear phase response but the more I'm reading about it, linear phase with pre-ringing is not a good thing at all. It is more audible and unnatural than natural minimal phase.
I'm also reading that the highest quality linear phase EQ implementations are actually made by IIR filters in a forward - backward manner. Not with a FIR.
Most FIR implementations have far too few taps and can't do bass EQ properly. You apparently need a LOT of taps to approach a good response and good resolution in the bass.
Here one of the pages I still have open: https://www.meldaproduction.com/text-tutorials/equalizers
And the crazy thing is, linear phase does not make any audible difference for a normal LR crossover on-axis. I've tested this with headphones. So why would anybody bother. Furthermore, the phase shifts of good speaker drivers operating in their piston range is basically completely tied to their amplitude response. So if you EQ them flat with minimal phase IIR you automatically also EQ them flat in phase. If you were to use linear phase EQ here you'd actually end up with less flat in phase drivers.
I think it's an error for anybody to use linear phase filters for speaker crossover and EQ. Other than perhaps room correction, this may be a different story (or it may not) I can't comment on this.

 

[JustIntonation]

As for FIR and convolution. These are related but two different things it seems to me?
FIR is a method for calculating the filter and implementing it. One can then turn this into an impulse for convolution. But one could also make an impulse of an IIR calculated filter.
As far as I understand it an IIR filter will give a higher quality filter for the same processing power than generating a filter by FIR (at least in the bass)?

 

[JustIntonation]

As for differences in IIR filter implementations. Yes there is great difference in VST plugins for instance. They're designed in many different ways not for scientific use but for a specific "sound". Though there are equalizers as well that specify certain filter types. For instance a Butterworth is a Butterworth, there shouldn't be any variation in that other than technical implementation of bit depth and oversampling etc. But no difference in curve (I haven't seen any so far in any case). Seems to me that this is also the case with DSP IIR, though I'm only looking at computer implemetations right now.
What I've further found is that for IIR peaking filters that their behavior is something to watch out for if their curve is not 0 at the Nyquist. This may actually not be a problem for crossover design and flattening the drivers as long as the curve is 0 at the Nyquist. But I'm not sure yet if there are other circumstances where the Nyquist response is something to take into consideration. For instance I'm assuming a lowpass which is never 0 at the Nyquist does not suffer from the same problems? And what about shelving filters? Does anybody here happen to know?

 

[pos]

Usually FIR is used for linear phase response but the more I'm reading about it, linear phase with pre-ringing is not a good thing at all. It is more audible and unnatural than natural minimal phase.

Preringing audibility is still a question in debate, as is phase shift audibility...
That said, preringing does not occur in a properly summed linear-phase crossover: the preringing of each filter part in the crossover cancels each other. Linear-phase is the way to go for crossovers.

I'm also reading that the highest quality linear phase EQ implementations are actually made by IIR filters in a forward - backward manner. Not with a FIR.

time reversal IIR is a clever trick to do linear-phase filters, but in no way it is superior to a real FIR/convolution implementation.
When you time-reverse an IIR you actually turn it into a FIR (finite) and need to overlap buffers, and this can be audible. Plus, you are limited to what IIR can do (all-pass, etc.). There is no good reason to use time-reversal nowadays.

Most FIR implementations have far too few taps and can't do bass EQ properly. You apparently need a LOT of taps to approach a good response and good resolution in the bass.

I agree that hadware direct convolution engine can be limited, but with an openDRC you can already do almost anything you need in a real world application. Software FFT convolution has more than enough taps for anything you would like to do.

Here one of the pages I still have open: https://www.meldaproduction.com/text-tutorials/equalizers
And the crazy thing is, linear phase does not make any audible difference for a normal LR crossover on-axis. I've tested this with headphones. So why would anybody bother.

Phase shift audibility is one thing, and it is indeed subtle (as many things hifi ), but FIR is certainly not limited to building linear-phase crossovers! FIR can do anything IIR can, with less quantization errors, and can do many other things like very complex EQs without penalties (the more IIR biquads you add the more errors you get, whereas in FIR only the final response matters), specific crossover slopes, steep (brickwal if you like), directivity-aware slopes (Horbach Keele), etc.

Furthermore, the phase shifts of good speaker drivers operating in their piston range is basically completely tied to their amplitude response. So if you EQ them flat with minimal phase IIR you automatically also EQ them flat in phase. If you were to use linear phase EQ here you'd actually end up with less flat in phase drivers.
I think it's an error for anybody to use linear phase filters for speaker crossover and EQ. Other than perhaps room correction, this may be a different story (or it may not) I can't comment on this

Of course EQ should always be minimum-phase (but you can do this in FIR no problem ), linear-phase EQ are a nonsense as far as sound reproduction is concerned.
That said a multi-way loudspeaker is not a minimum-phase device, because of the crossovers, and these can be addressed with FIR.

 

[pos]

As for differences in IIR filter implementations. Yes there is great difference in VST plugins for instance. They're designed in many different ways not for scientific use but for a specific "sound". Though there are equalizers as well that specify certain filter types. For instance a Butterworth is a Butterworth, there shouldn't be any variation in that other than technical implementation of bit depth and oversampling etc. But no difference in curve (I haven't seen any so far in any case). Seems to me that this is also the case with DSP IIR, though I'm only looking at computer implemetations right now.
What I've further found is that for IIR peaking filters that their behavior is something to watch out for if their curve is not 0 at the Nyquist. This may actually not be a problem for crossover design and flattening the drivers as long as the curve is 0 at the Nyquist. But I'm not sure yet if there are other circumstances where the Nyquist response is something to take into consideration. For instance I'm assuming a lowpass which is never 0 at the Nyquist does not suffer from the same problems? And what about shelving filters? Does anybody here happen to know?

If you measure the output of all these IIR implementation you will certainly see differences.
First as already said care must be taken regarding EQ and filters convention (Q convention for EQ, and Fc convention for shelving are especially cumbersome), but even when this is taken care of difference can (and will) remain, especially in the high frequency range.

 

[JustIntonation]

Thanks for your information!
I'll look into it further. I'm not convinced yet So far it looks to me that with equal processing power (minimal phase) IIR beats (minimal phase) FIR.
As for high frequency response differences, yes this is as I described for how non-zero Nyquist frequency peaking filters are handled. But I've compared and measured Butterworth filters in both Equilibrium set to IIR and ProQ2 set to natural phase, and while indeed they have a different method for Q the results when this is set properly is identical. (though have not looked at the bit level).
As for linear phase for crossovers, I don't think this is needed at all. This is not audible. I'm not saying phase is never audible, it is in many circumstances but not for LR crossovers. The differences become only audible off-axis where linear phase will give pre-ringing off-axis and minimal phase will only give the less audible post-ringing. I did test this myself and heard differences but didn't test it well enough to see which one I personally would prefer. But looking at it now, I think a minimal phase crossover is preferable as it sounds exactly equal to linear phase on-axis but should sound more natural off-axis.

When I'm doing my PC based crossover I'll measure the whole thing to see the level of error and share this.

 

[pos]

The problem is processing power indeed, or more specifically the lack of dedicated convolution hardware out there.
An FPGA implementation could do almost any amount of taps you would ever need, but there is simply no market for it because... because the industry feels it is not needed and IIR is sufficient. There are too many misconceptions about FIR, like being limited to linear-phase (hence high latency) corrections, or even brickwall (was still the case not that long ago...). FIR is the future for sure, but it should/could already be the present

As for high frequency response differences, yes this is as I described for how non-zero Nyquist frequency peaking filters are handled.

magnitude and phase will never by 0 at Nyquist. It can be close for low order/low frequency corrections, but it is asymptotic non the less. In rephase I use a trick to make phase a 180° multiple at Nyquist and avoid a lot of ringing ("closest perfect impulse" centering option).

The differences become only audible off-axis where linear phase will give pre-ringing off-axis and minimal phase will only give the less audible post-ringing.

Looks like there is no shortcut for a properly summed crossover with matched directivity drivers

More importantly, there is another major advantage of linear-phase crossovers over minimum-phase ones: phase matching is as easy a matching a flat line, whereas with minimum-phase filters you need to match phase shifts, which is not easy and one of the main pitfalls designers face when buidling an acoustical crossover.
And things only get (much!) more complicated when you go from a 2-way to a 3-way designs, and up, because phase shifts of a crossover point influence the other ones, and you cannot build each crossover in isolation.
Linear-phase acoustical crossover construction is a piece of cake in comparison, regardless of the number of drivers.

 

[March Audio]

Just,

I'm with Pos on this. Linear phase Fir. I don't share your concerns about pre ringing or its audibility. When you tested this yourself can I ask how controlled the test was? If it wasn't blind and if it wasn't with identical frequency response (different filters roll of differently into the audio band) then I would disregard the conclusions.

Processing power. If you are using a PC, as opposed to dedicated Dsp device, and latency is not a concern, then its a non issue. Can be done on a low power PC with plenty of low frequency resolution. Its only when you try to tackle latency that it gets problematic.

With your plug in approach and iir, I think you are going to run into a lot of problems. Pos has gone through this already, I would just reiterate his points.

From someone who has been through all of this just grab something like Acourate and save yourself a lot of time and pain

If I can talk subjectively about my results, dangerous territory I know, but I will give the response of others that have heard my speakers which are 3 way Fir. The consistent stand out comment is regarding the soundstage. Multiple people have said that they have never heard stereo like it, regardless of system price. I never achieved this with passive or iir XO. Also comments about clarity, dsp is a vast improvement over passive in this respect.

Oh one other thing you can very easily do with software like Acourate is multiple convolution. I. E. perform driver response correction first. Measure at close distance and correct so it is "correct" as possible coming out of the speaker in the first instance. You can then perform additional corrections on top of this if you wish to adjust for room issues. My experience is that if its right coming out of the speaker in the first place then it makes a big difference with very little need to do anything beyond tackling significant room modes.

 

[JustIntonation]

Regarding linear phase FIR vs minimal phase IIR for the Linkwitz-Riley crossover filters:

Linear phase FIR vs minimal phase IIR on-axis:
Sound: exactly the same.
Processing: FIR requires much more processing power for the same sound quality / resolution.
Latency: IIR minimal latency, FIR very large latency when processing with the same sound quality.

Linear phase FIR vs minimal phase IIR off-axis:
Sound: pre-ringing and post ringing for FIR, only post-ringing for IIR. Without saying which sounds better they for sure sound different off-axis and in my opinion there's something wrong with pre-ringing on a logical level. You will hear the ringing leading up to a sound, how unnatural is that?
Processing: same as described above for on-axis.
Latency: same as described above for on-axis.

I mean, how clear a picture do you want?
IIR clearly wins in this particular application!
It ties in on-axis sound, it wins in off-axis sound, it wins in processing power requirements and it wins in latency. It doesn't loose in any of the categories.

And you can do a test yourself like I did. Hell I can upload some files if you wish.
For an on-axis LR up to 96dB/oct (I didn't test any further) you cannot in any way tell the difference between original and crossover with the IIR phase shift as a result of the crossover. I mean they sound 100% identical. Tested this with great headphones and great DAC and great headphone amp. And I'm not the only one who says this, I later read mr Linkwitz himself did this same test and came to the exact same conclusion. (I also read that certain other phase modifications ARE audible, but a LR crossover, nope truly 0 audibility).
But feel free to test this yourself.

As for off-axis ringing differences. I did a quick listen to off-axis simulation with 96dB/oct LR with IIR vs linear phase FIR and the FIR off-axis ringing sounded quite different from the IIR off-axis ringing. Everybody can hear this it was quite obvious. I did not test which one I prefer or if one is more audible than the other. But again logic tells me that pre-ringing is something quite weird, you will hear ringing BEFORE a sound is made. For instance a percussion sound.. That's not natural or correct to me. (It's different for a DAC where the pre-ringing is out of the audible range)
So I very much stand by what I said. I think it's an error to use linear phase EQ for crossovers.

As for the problems of IIR. They're not so big as I understand. And nothing which isn't fixed in more properly programmed IIR filters, 64bit and with either compensation programmed in for non 0 Nyquist responses in both phase and amplitude (like for instance Equilibrium has and several other plugins) or by oversampling.

 

[andreasmaaan]

Without saying which sounds better they for sure sound different off-axis and in my opinion there's something wrong with pre-ringing on a logical level. You will hear the ringing leading up to a sound, how unnatural is that?

Intuitively what you say makes sense, but psychoacoustic studies show that pre-ringing - as long as it is not sufficiently early / high in level - is masked by the signal in exactly the same way post-ringing is.

In other words, as long as the pre-ringing is below the masking threshold, it is inaudible. This is of course also the case for post-ringing (notwithstanding that the audibility threshold for post-ringing is higher).

A typical temporal masking graph is shown in this link, although the values vary with SPL and frequency.

Otherwise I agree with you. The group delay caused by typical IIR filters has never been shown experimentally to be audible.

 

[JustIntonation]

I never achieved this with passive or iir XO. Also comments about clarity, dsp is a vast improvement over passive in this respect.

Yes DSP vs passive is well known. Many problems with passive crossovers.
But as for comparing linear phase with minimal phase for the crossover. Have you done this comparison with Acourate? You should be able to do it now with Acourate correct? Just change the crossover part from linear phase to minimal phase doesn't matter if it's FIR or IIR. I can tell you you will hear no difference whatsoever in soundstage except perhaps for the slightly negative contribution of the off-axis pre-ringing going into the room sound.
As I read you previously used a NanoSHARC or MiniSHARC from MiniDSP? These have a seriously flawed design, the ASRC for instance is running at 96kHz, who does this? This is the same as 96kHz sample rate and a multiple of 48kHz it will give serious problems / audible errors when playing these sample rates. And I can only guess what else wasn't done right, their cheaper models son't exactly measure well either (see for instance Amir's measurements of the 2x4HD (or was that the 4x10HD?)).

 

[JustIntonation]

A typical temporal masking graph is shown in this link, although the values vary with SPL and frequency.

Thanks!
Actually, I was simplifying things by saying minimal phase IIR has only post-ringing.
Minimal phase IIR actually has short pre-ringing and long post-ringing. Pretty much like the masking graph shows!
One can use the masking graph to construct the maximum Q of a minimum phase filter which is still masked.
If one then turns that exact same filter with the same Q into a linear phase filter the graph shows that much of the pre-ringing will no longer be masked.
So masking indeed says that linear phase is more audible than minimal phase.

I think the error is thinking that 2 drivers and a crossover should have a linear phase as if its one driver for it to be perfect and that nature did a bad trick on music coming from multi-driver loudspeakers Nature didn't do a trick, minimal phase LR crossovers are perfect. And nature itself uses minimal phase EQ, be it analogue EQ or the filtering of sound traveling through air etc.