That doesn't really make sense to me. Why would they bother to include a resampler when 24/192 DAC chips are so cheap?
https://www.audiosciencereview.com/...hdmi-audio-extractor-some-measurements.38590/
The DAC is the CS4354. It will do 24/192 without resampling. THD+N is just -86dB, though.
https://www.audiosciencereview.com/...hdmi-audio-extractor-some-measurements.38590/
The DAC is the CS4354. It will do 24/192 without resampling. THD+N is just -86dB, though.
IIUC licensing for full resolution HDMI audio is limited to big companies selling things like expensive AVRs. Low cost HDMI converters are not licensed for and aren't supposed to be able to decrypt full resolution audio. There may be some hacks going on in China, but if so its not clear how they work and or whether they just resample again to fool you.
True. But most of the content people usually express interest in involves HDCP. The converters are usually designed for those customers. The internal dacs in the converters are junk. The SPDIF/TOSLINK might be okay, but is no doubt jittery due to how it is clocked. An ASRC can only attenuate some of that. In general its hard to see how it could be a very Hi-Fi solution. Wouldn't use it myself nor recommend it. Asynchronous USB or its equal is the only serious game in town if FIFO buffering is not acceptable. But, yes, I agree that HDMI itself could be used despite remaining undesirable in some ways.
Last edited:
I guess most people would rather spend more money. Not me, I'm broke 😀
Just meant to point out that the option is there and it would technically fulfill the requirements mentioned.
Just meant to point out that the option is there and it would technically fulfill the requirements mentioned.
I am afraid that two separate devices are very hard to synchronize sample-wise. I do not see any way to issue the start-DMA command to two devices at the very same moment. They will run synchronously, but the first 8 samples will be shifted against the second 8 samples.
Uh Gents, many comments, thanks, interesting to read this thread (don't stop please) 😉
What's more interesting for me is the opinion of some guys, who (at least one of them) develops a complex room correction software here.
According to him, the clocks being out-of-sync with eachother is a valid event, however, the effect of it shall be comparable to how the "volume" of my head is increasing and decreasing as my heart beats and some of my veins on my temple pulse to this beat under my skin. So to put it simple, according to him I simply shouldn't care that much and using 3 commercial (end-user, non-pro) DACs of my choice is simply not an issue regarding their clocks being out of sync, the real effect is sooo negligible.
Any opinions ? (Or better: experience ?)
I accept if (with time) slightly differing clocks are an issue, but if it's a negligible issue in a home theater setup, it's not an issue then.
What's more interesting for me is the opinion of some guys, who (at least one of them) develops a complex room correction software here.
According to him, the clocks being out-of-sync with eachother is a valid event, however, the effect of it shall be comparable to how the "volume" of my head is increasing and decreasing as my heart beats and some of my veins on my temple pulse to this beat under my skin. So to put it simple, according to him I simply shouldn't care that much and using 3 commercial (end-user, non-pro) DACs of my choice is simply not an issue regarding their clocks being out of sync, the real effect is sooo negligible.
Any opinions ? (Or better: experience ?)
I accept if (with time) slightly differing clocks are an issue, but if it's a negligible issue in a home theater setup, it's not an issue then.
Hadn't thought of the DMA aspect. How did you arrive at the 8 sample offset? Is that the sampling rate of 192k divided by 24k HDA clock?
I can imagine that in some cases, an offset of 1/24000 second might be acceptable. Then again, if there's a need to sum signals coming from different output devices, that's probably no good.
@Vortex Thanks for bringing Cavern to my attention! I had never heard of it before and now I have some reading to do.
Gerne !
Cavern is an amazing tool, I've used it only for some ABX blind testing of different opus bitrates but some guys here use this tool with such great results they all seem to be 'blown away' by the results it can bring to a normally average HT chain. I also intend to dig deeper into the tool and all it offers, learn how to use it, measure, calibrate, fine-tune and then see what it brings but the time is missing yet. 😏
Maybe worth beginning here with Room Correction (through Equalizer APO) but I think it has many other use cases too.
Cavern is an amazing tool, I've used it only for some ABX blind testing of different opus bitrates but some guys here use this tool with such great results they all seem to be 'blown away' by the results it can bring to a normally average HT chain. I also intend to dig deeper into the tool and all it offers, learn how to use it, measure, calibrate, fine-tune and then see what it brings but the time is missing yet. 😏
Maybe worth beginning here with Room Correction (through Equalizer APO) but I think it has many other use cases too.
It depends. Do you care a lot about imaging and or ITD localization?
https://en.wikipedia.org/wiki/Interaural_time_difference#:~:text=The interaural time difference (or,sound source from the head.
Also, if you play music around the clock, eventually small differences in dac clock frequencies can get the audio channels noticeably out of sync with each other.
Usedto be a problem in the early days of home recording DAWs that supported multiple USB sound cards for recording. Using cheap sound cards, some people had trouble with sync over the course of recording one song.
Last edited:
Sorry, it should have read "8 channels offset", due to joining two 8ch interfaces.
Who doesn't care ?
But I just look around and all digital-inside active speakers (studio, PA, etc). works this way: they have totally independent own amplifiers, DAC-s inside, everything - and they still do wonders in pairs while their DACs are not synced.
Do sound engineers seem to suffer from unsynced clocks of these in-speaker integrated DACs ?
What I also get as a feedback to this topic is: it doesn't really matter for different sides. It matters when crossing digitally within e.g. a 2-way active speaker. There the DAC's clock is crucial but these 2-ways have 2ch integrated DACs anyway so clock is the same within a speaker. But between the two separate speakers, e.g. L and R, they work totally independently from eachother and are still usable and in fact work very well for hours without issues. So, I'm still curious what's going on ..
Let's take another example: a pair of Eve SC208 active speakers.
Apart from the power cord, there's only an analog XLR input on both - and that's it.
Inside there's and ADC, a DSP, a 2ch DAC behind the DSP and 2 amp channels driving the speakers respectively, all these on a common PCB.
So there're practically separate and parallel AD and DA conversions in these speaker pairs with no clue of eachother, nor common clocks whatsoever and this still works as a concept.
Same applies for dozens of active studio monitors your music is mixed on and plenty of PA speakers too. Are they all blindfold and don't recognize this design flaw - if that's really an issue ? (So the question still applies).
A member on another forum has exactly these kind of speakers and according to him they're switched on since more than 2 months now without having been switched off. (Of course a random glitch in the mains power, a shorter 1-3 second power outage can reset the 'timer'). He also thought okay, let's test if this DAC clock drift can be detected by the human ear or not, hoping that for some days or even weeks, the clocks are already way off of eachother, according to the theory. He switched the speakers off and back on again (with plenty of milliseconds, actually full seconds difference between them) and experienced no improvement at all, although DAC clocks shall be now closer to eachother than ever before (assuming the power-on time between them was the same like now).
So plenty of weirdo stuff around DAC clocks NOT being synced at all and still, it seems to work for many. And in this case (and also other cases), an ADC just comes in as an extra spice into the story.
That's my confusion around this whole topic.
Apart from the power cord, there's only an analog XLR input on both - and that's it.
Inside there's and ADC, a DSP, a 2ch DAC behind the DSP and 2 amp channels driving the speakers respectively, all these on a common PCB.
So there're practically separate and parallel AD and DA conversions in these speaker pairs with no clue of eachother, nor common clocks whatsoever and this still works as a concept.
Same applies for dozens of active studio monitors your music is mixed on and plenty of PA speakers too. Are they all blindfold and don't recognize this design flaw - if that's really an issue ? (So the question still applies).
A member on another forum has exactly these kind of speakers and according to him they're switched on since more than 2 months now without having been switched off. (Of course a random glitch in the mains power, a shorter 1-3 second power outage can reset the 'timer'). He also thought okay, let's test if this DAC clock drift can be detected by the human ear or not, hoping that for some days or even weeks, the clocks are already way off of eachother, according to the theory. He switched the speakers off and back on again (with plenty of milliseconds, actually full seconds difference between them) and experienced no improvement at all, although DAC clocks shall be now closer to eachother than ever before (assuming the power-on time between them was the same like now).
So plenty of weirdo stuff around DAC clocks NOT being synced at all and still, it seems to work for many. And in this case (and also other cases), an ADC just comes in as an extra spice into the story.
That's my confusion around this whole topic.
Nope, the reason why it all 'just works' is because the AD and DA converter do in fact share a clock. They are always in perfect sync, by design.
A lot of speakers use the Analog Devices ADAU series, in these, the AD and DA converter are even on the same chip, together with the DSP. Any clock drift going in will also be there going out. That way, the whole topic is a lot less critical.
So, you are saying that the ADC, DSP, and DAC in these speakers all have difference clocks? Not!
It's more that likely that they all run from the SAME clock. Also, they are not made up of buffered samples and software being scheduled by a computer operating system along with lots of other program but rather are hardware based DSP that does nothing else besides crank samples. These are completely different scenarios. Not even comparable.
It's more that likely that they all run from the SAME clock. Also, they are not made up of buffered samples and software being scheduled by a computer operating system along with lots of other program but rather are hardware based DSP that does nothing else besides crank samples. These are completely different scenarios. Not even comparable.
Okay, there are different scenarios that have be evaluated separately.
In the case of multiple USB dacs, that's one case. In the case of, say, a self powered speaker with analog and SPDIF/AES inputs that's another case.
Where clock differences may tend to show up the most is in case of multiple asynchronous USB dacs. That's because asynchronous USB is clocked according the local clock inside dac, at whatever exact frequency it happens to be. In that case the timing of audio output is entirely depending on the local dac clock.
In the case where a speaker has an analog input, then the internal ADC, DSP, and DAC all have to keep up with analog input, but with some processing latency. If all the speakers are the same then the latency should same for each speaker and thus more or less cancel out.
In the case where a speaker has a SPDIF/AES input, most likely that will run though an ASRC to reclock it to local crystal clock. Again the incoming audio stream is primarily what determines the output timing. ASRC merely resamples the digital data to keep up with incoming digital audio stream.
More generally when there is a self powered speaker with analog and or SPDIF/AES inputs and a Class-D amplifier inside, what comes out is already pretty damaged sound. You may not know it, and or may not believe it, but if you listen the same music played back without all that processing on a good system its trivially easy to hear the difference.
Consider the economics: a Holo May dac costs a little more than $5k, or more than $2.5k per channel. What does a top of the line Prism Sound ADC cost? Are the best data converters perfectly transparent? Not really. What about the DSP? is it 32-bit or 64-bit or more processing? How many taps do the filters use? What are the specs for the internal ASRC if SPDIF/AES is used? As good as SRC4392 or AK4137? No. Does the class-D amp suffer from problems such as the hysteresis distortion shown by Bruno Putzeys? Probably. How much does one speaker cost? Is it enough to cover the cost of as close to transparent as we know how to do? Very Unlikely.
In the case of multiple USB dacs, that's one case. In the case of, say, a self powered speaker with analog and SPDIF/AES inputs that's another case.
Where clock differences may tend to show up the most is in case of multiple asynchronous USB dacs. That's because asynchronous USB is clocked according the local clock inside dac, at whatever exact frequency it happens to be. In that case the timing of audio output is entirely depending on the local dac clock.
In the case where a speaker has an analog input, then the internal ADC, DSP, and DAC all have to keep up with analog input, but with some processing latency. If all the speakers are the same then the latency should same for each speaker and thus more or less cancel out.
In the case where a speaker has a SPDIF/AES input, most likely that will run though an ASRC to reclock it to local crystal clock. Again the incoming audio stream is primarily what determines the output timing. ASRC merely resamples the digital data to keep up with incoming digital audio stream.
More generally when there is a self powered speaker with analog and or SPDIF/AES inputs and a Class-D amplifier inside, what comes out is already pretty damaged sound. You may not know it, and or may not believe it, but if you listen the same music played back without all that processing on a good system its trivially easy to hear the difference.
Consider the economics: a Holo May dac costs a little more than $5k, or more than $2.5k per channel. What does a top of the line Prism Sound ADC cost? Are the best data converters perfectly transparent? Not really. What about the DSP? is it 32-bit or 64-bit or more processing? How many taps do the filters use? What are the specs for the internal ASRC if SPDIF/AES is used? As good as SRC4392 or AK4137? No. Does the class-D amp suffer from problems such as the hysteresis distortion shown by Bruno Putzeys? Probably. How much does one speaker cost? Is it enough to cover the cost of as close to transparent as we know how to do? Very Unlikely.
Clock synchronization isn't a problem depending on what the different devices are doing.
If you've got two separate active speakers going ADC/SPDIF > DSP > DAC > amplifier > drivers and each speaker contains its own separate set of electronics then this isn't a problem.
Yes the two clocks in each speaker will be different and you will end up with a slight delay difference between the left and right but this is akin to moving one of them back or forward by a tiny bit. It's not really consequential.
In this case the important part is that the DSP/DACs of an individual speaker are all synchronised. This means that the phase alignment between the drivers and crossover is maintained. They are all linked to one common DSP.
If you had a situation where one DSP was on the tweeters and another DSP was on the woofers the difference between the two would translate into moving the drivers themselves a bit every time you potentially press play and that will not do as your driver integration would go out the window.
Of course with both of those scenarios the audio is going to be fed to each speaker via a common source, either via the ADC (from say a vinyl player) or SPDIF (from a CD player) so the two channels would never drift over time.
With an asynchronous USB device the PC is slaved to the DAC and the clock inside it tells the PC when to send over new data. Obviously if you've got two different USB DACs, with their different clocks, then they will drift apart over time and, I'm assuming, the faster one would be ever so slightly higher in pitch too. Probably undetectably so mind you.
Even if you had that situation as long as you were sensible with the channel allocation you'd be okay. For example use one 8 channel device for your main two loudspeakers. If you have a 3 way main speaker pair using DSP crossovers then use 6 channels from one device for that.
If you're doing phase and time alignment keep in mind that different DACs use different digital filters. These all have, potentially, different delays. Let's say you had two identical USB DACs that were completely clock synchronised across a suitable interface. Let's say they also have swappable digital filters. You swap one from a fast to slow digital filter and the sound immediately changes. Swapping the filter could alter the delay by say 30 audio samples. At 48kHz this is like moving a speaker 20cm. If you had DAC one running the tweeters and DAC two running the midrange you'd completely change the driver integration.
If the DMA start from one DAC occurs 8 samples ahead of another one then it's always going to be a couple cm ahead of the next DAC in terms of speaker position. This isn't of any real consequence if each individual speaker is run from its own DAC. It's not any real consequence either if the same DAC is always ahead by a couple of cm, you can compensate for that in the design. However if the DAC that starts first always changes whenever you start and stop an audio stream that would affect driver integration if two separate DACs were used for say the tweeters and midranges of one loudspeaker.
If you've got two separate active speakers going ADC/SPDIF > DSP > DAC > amplifier > drivers and each speaker contains its own separate set of electronics then this isn't a problem.
Yes the two clocks in each speaker will be different and you will end up with a slight delay difference between the left and right but this is akin to moving one of them back or forward by a tiny bit. It's not really consequential.
In this case the important part is that the DSP/DACs of an individual speaker are all synchronised. This means that the phase alignment between the drivers and crossover is maintained. They are all linked to one common DSP.
If you had a situation where one DSP was on the tweeters and another DSP was on the woofers the difference between the two would translate into moving the drivers themselves a bit every time you potentially press play and that will not do as your driver integration would go out the window.
Of course with both of those scenarios the audio is going to be fed to each speaker via a common source, either via the ADC (from say a vinyl player) or SPDIF (from a CD player) so the two channels would never drift over time.
With an asynchronous USB device the PC is slaved to the DAC and the clock inside it tells the PC when to send over new data. Obviously if you've got two different USB DACs, with their different clocks, then they will drift apart over time and, I'm assuming, the faster one would be ever so slightly higher in pitch too. Probably undetectably so mind you.
Even if you had that situation as long as you were sensible with the channel allocation you'd be okay. For example use one 8 channel device for your main two loudspeakers. If you have a 3 way main speaker pair using DSP crossovers then use 6 channels from one device for that.
If you're doing phase and time alignment keep in mind that different DACs use different digital filters. These all have, potentially, different delays. Let's say you had two identical USB DACs that were completely clock synchronised across a suitable interface. Let's say they also have swappable digital filters. You swap one from a fast to slow digital filter and the sound immediately changes. Swapping the filter could alter the delay by say 30 audio samples. At 48kHz this is like moving a speaker 20cm. If you had DAC one running the tweeters and DAC two running the midrange you'd completely change the driver integration.
If the DMA start from one DAC occurs 8 samples ahead of another one then it's always going to be a couple cm ahead of the next DAC in terms of speaker position. This isn't of any real consequence if each individual speaker is run from its own DAC. It's not any real consequence either if the same DAC is always ahead by a couple of cm, you can compensate for that in the design. However if the DAC that starts first always changes whenever you start and stop an audio stream that would affect driver integration if two separate DACs were used for say the tweeters and midranges of one loudspeaker.
the AD and DA converter do in fact share a clock
No, the left channel's AD-DA stack has - obviously - absolutely nothing to do with the right channel's AD-DA stack, 2 fully independent active speakers.
A lot of speakers use the Analog Devices ADAU series
Hypex FA series-powered DIY 3-way boxes. AKM ADC -> some kind of DSP (maybe ADAU, not sure) -> AKM DAC
Markw4, 5th element, uh, I processed your posts, complex but I understand (I think), thank you for the detailed explanations.
But if you use a digital self clocked signal as source then the differents stacks share a common clock signal, they are synchronised ( theorically and if implemented as it should).
That is one of the advantage of Dante: along the network you can choose the clock you prefer ( each gear as a different generator and there is differences between them) as master and sync other gear without the need of a dedicated physical clock line ( wordclock).
Works great in practice and native system latency is low ( 0,8ms recommended for the gear i own) and up to a limit not used chanel count related ( up to 16i/o at 96khz/24bit with my gear).