...and/or the associated ringing in the Impulse response?
This also got me thinking.
Given that the FR differences that I measured (see previous post) are most probably due to the use of a gentler anti-alias LP filter in the RATOC dac vs. a more conventional "brickwall" filter in the network player's delta-sigma dac... might the "cleaner" impulse response that is likely associated to the former also have something to do with the subjective listening impressions?
Marco
This also got me thinking.
Given that the FR differences that I measured (see previous post) are most probably due to the use of a gentler anti-alias LP filter in the RATOC dac vs. a more conventional "brickwall" filter in the network player's delta-sigma dac... might the "cleaner" impulse response that is likely associated to the former also have something to do with the subjective listening impressions?
Marco
Last edited:
Same measurements but with 44.1kHz file
This time, I went and measured the frequency response using pink noise at 16bit/44.1kHz.
- (A) using the RATOC (R2R multi-bit) dac
- (B) using my network's built-in (delta-sigma) dac as a point of reference.
I then again computed the DIFFERENCE between measurements (A) and (B), and here is what came out (see attachment).
Basically, the RATOC dac exhibits the same kind of slow roll-off, but now starting lower (-1.2dB at 10kHz already), and is down by a whopping -6dB at 20 kHz!
This correlates very well with the fact that the perceived differences between the two dacs are more manifest with redbook (44.1kHz) source material vs. high-res files.
Again, does this point to the fact that it is probably the frequency response resulting from the choice of anti-alias filtering (slow roll-off vs. "brickwall") - and possibly the associated time-domain behaviour in terms of ringing - that accounts for the perceived differences, rather than the inherent conversion technology (i.e., multi-bit vs. delta-sigma)?
It seems to make a lot of sense...
Marco[/QUOTE]
This time, I went and measured the frequency response using pink noise at 16bit/44.1kHz.
- (A) using the RATOC (R2R multi-bit) dac
- (B) using my network's built-in (delta-sigma) dac as a point of reference.
I then again computed the DIFFERENCE between measurements (A) and (B), and here is what came out (see attachment).
Basically, the RATOC dac exhibits the same kind of slow roll-off, but now starting lower (-1.2dB at 10kHz already), and is down by a whopping -6dB at 20 kHz!
This correlates very well with the fact that the perceived differences between the two dacs are more manifest with redbook (44.1kHz) source material vs. high-res files.
Again, does this point to the fact that it is probably the frequency response resulting from the choice of anti-alias filtering (slow roll-off vs. "brickwall") - and possibly the associated time-domain behaviour in terms of ringing - that accounts for the perceived differences, rather than the inherent conversion technology (i.e., multi-bit vs. delta-sigma)?
It seems to make a lot of sense...
Marco[/QUOTE]
Attachments
Last edited:
Further musings on this DAC thing...
So, according to the manufacturer, the RATOC dac operates "without Oversampling and Digital filter".
So it probably just employs a slow analogue anti-imaging filter. This goes a long way towards explaining the measured FR droop, especially at the lower sampling rates.
In essence, this is not only a ladder R2R multi-bit dac, but it is also of the rare "NOS" breed, and as such it is the polar opposite of a modern, conventional sigma-delta dac which oversamples everything x4 or x8 and employs a steep "brickwall" digital low-pass filter.
NOS dacs have their followers, despite having some undeniable issues (poor rejection of out-of-band frequencies being the most prominent one, of course). Basically, they trade off FR linearity for improved transient response.
Here's an informative thread for those who might be interested.
In the end, given all the rather extreme differences above, I find it remarkable how the two dacs manage to sound VERY SIMILAR nonetheless!
Our auditory system surely is very tolerant up there ;-)
So, according to the manufacturer, the RATOC dac operates "without Oversampling and Digital filter".
So it probably just employs a slow analogue anti-imaging filter. This goes a long way towards explaining the measured FR droop, especially at the lower sampling rates.
In essence, this is not only a ladder R2R multi-bit dac, but it is also of the rare "NOS" breed, and as such it is the polar opposite of a modern, conventional sigma-delta dac which oversamples everything x4 or x8 and employs a steep "brickwall" digital low-pass filter.
NOS dacs have their followers, despite having some undeniable issues (poor rejection of out-of-band frequencies being the most prominent one, of course). Basically, they trade off FR linearity for improved transient response.
Here's an informative thread for those who might be interested.
In the end, given all the rather extreme differences above, I find it remarkable how the two dacs manage to sound VERY SIMILAR nonetheless!
Our auditory system surely is very tolerant up there ;-)
Also, here's a long argument for NOS dacs originally published by Ryohei Kusunoki in the Japanese magazine MJ in 1996:
NOS DAC
That is an amazing different in FR, I've never seen anything like that on a DAC. But if one of them is using a low order analog filter, That does seem possible.
Do you have a way to test the DAC with sweeps and without going thru the speakers. By that I mean a good soundcard with line level inputs? The line level input on a laptop would probably be good enough. I'm a little puzzled by the smooth, smooth graph. That's unusual with pink noise. Perhaps the averaging is very long?
Do you have a way to test the DAC with sweeps and without going thru the speakers. By that I mean a good soundcard with line level inputs? The line level input on a laptop would probably be good enough. I'm a little puzzled by the smooth, smooth graph. That's unusual with pink noise. Perhaps the averaging is very long?
If I read correctly, marco_gea used the input mic as a "line" input. Is this right?
Reading some datasheets from digital filters (eg. DF1706), the "slow" roll-off perhaps match with this response. Since I use the DF1704 filter in my PCM DAC, I measured the inpulse response (years ago). The less ringing in the square wave is evident with "slow' setting. A difference in time domain, for sure.
Reading some datasheets from digital filters (eg. DF1706), the "slow" roll-off perhaps match with this response. Since I use the DF1704 filter in my PCM DAC, I measured the inpulse response (years ago). The less ringing in the square wave is evident with "slow' setting. A difference in time domain, for sure.
Yes, I measured the whole chain using my calibrated mic into the sound card.
Any effect due to the amp, speakers, room etc. is zeroed-out when computing the difference between the two measurements.
According to the manufacturer, "The RAL-24192DM1 doesn't use the DF1704 8 x Oversampling and Digital Filter chip". Instead, I think the "raw" output of the PCM1704 r2r dac chip is just passed through a mild analogue filter. I think this explains the -6dB @20kHz with 44.1kHz material: approx. -3dB are due to the inherent sin(x)/x function, plus -3dB at the corner frequency of the analogue low-pass filter @ Fs/2.
This link explains the basic process well, I think: DAC frequency response
Last edited:
The RATOC manufacturer don't make it clear if it's S-PDIF decoder block also have a buit-in digital filter, or if operate in NOS mode: "It's Internal firmware control PCM1704 directly and bring out pure and natural sound from the DAC"
So, directly means the absence of external digital filter only, or is about NOS (the complete digital filter absence)?
A simple measurement with digital generated squarewave and a 'scope will reveal the truth.
So, directly means the absence of external digital filter only, or is about NOS (the complete digital filter absence)?
A simple measurement with digital generated squarewave and a 'scope will reveal the truth.
Well, on the manufacturer's web page it does describe it as "without Oversampling", so I would tend to think it is NOS.
Additionally, on this Japanese language page they say:
"The PCM 1704 is controlled by software without using a dedicated controller (DF 1704). We do not use 8 times oversampling and digital filter, but by doing the original operation of multi bit, natural sound with little distortion is reproduced."
But I have just sent them an e-mail to ask for a definitive clarification.
I'll let you all know if they reply.
Last edited:
Thanks for the test, Marco, and for confirming that it is indeed the change in FR that you are responding to by subjectively classifying it as smoother sounding. That's exactly what a slightly rolled off top end sounds like. There might be ringing, but the way we perceive sound is via the FR, and so, the FR manifestation of the ringing is what we would hear.
Can you also measure just the DAC response, like Pano said? The 6 db roll off seems a bit much, and I wonder if the directivity of your speakers is playing some role here.
The next step would be to make the two DAC (only) responses the same (do not correct the measured speaker response) and see if it gets rid of the relative subjective differences between the two DACs.
I am reading "A More Perfect Heaven" by Dava Sobel, a book on how Copernicus changed our understanding of the cosmos. I am at an interesting point in the book and just reading something last night, which was funny, and sort of related to what we're seeing here. Hopefully, I won't give too much away. In the book, a learned professor of mathematics travels far to see Copernicus because he is very impressed by his work and his "theory" that the sun is at the center and the planets spin around, and not only that, but the earth spins on its own axis as well. When he finally meets him, and they start talking, he realizes that Copernicus really thinks that the earth is spinning. He is dumbfounded and says something like, "Everyone knows your theory, but you don't think it is reality that the earth spins, do you?" He stomps on the ground and says, "Look, nothing happens, it does not move."
This was funny to me because here we have one of the most learned mathematician of his time, who purportedly "understands" the new theory, but yet refuses to believe it is reality. It's kind of similar to all of us knowing the theory that FR explains it all and yet not believing that it is reality. All the evidence is there and yet we do not want to believe it.
Can you also measure just the DAC response, like Pano said? The 6 db roll off seems a bit much, and I wonder if the directivity of your speakers is playing some role here.
The next step would be to make the two DAC (only) responses the same (do not correct the measured speaker response) and see if it gets rid of the relative subjective differences between the two DACs.
I am reading "A More Perfect Heaven" by Dava Sobel, a book on how Copernicus changed our understanding of the cosmos. I am at an interesting point in the book and just reading something last night, which was funny, and sort of related to what we're seeing here. Hopefully, I won't give too much away. In the book, a learned professor of mathematics travels far to see Copernicus because he is very impressed by his work and his "theory" that the sun is at the center and the planets spin around, and not only that, but the earth spins on its own axis as well. When he finally meets him, and they start talking, he realizes that Copernicus really thinks that the earth is spinning. He is dumbfounded and says something like, "Everyone knows your theory, but you don't think it is reality that the earth spins, do you?" He stomps on the ground and says, "Look, nothing happens, it does not move."
This was funny to me because here we have one of the most learned mathematician of his time, who purportedly "understands" the new theory, but yet refuses to believe it is reality. It's kind of similar to all of us knowing the theory that FR explains it all and yet not believing that it is reality. All the evidence is there and yet we do not want to believe it.
Agreed. To be honest, it's a fine line between "smoother" and "duller". On some tracks, I find it an improvement, on others not so much.
OK, I know that the two aspects (FR and ringing) are intimately related, BUT couldn't the time domain difference per se be somewhat responsible for what we hear, too?
That would be a bit of a pain in the neck to do, because I have already installed the dac in the system, and it's awkward to reach behind it now, without moving a lot of stuff around.
Yes, initially I was surprised too, but then I thought: -3.8dB at Fs/2 are due to the intrinsic sin(x)/x response of the dac process itself (which, unless compensated for in the anti-imaging filter, is universal and unavoidable). The additional ~2-3 dB of attenuation may well be due to the "shoulder" of the gentle (-6 dB/oct??) analog low-pass filter...
The directivity of my speakers (or any other speaker- or room-related effect) has NOTHING to do with what I measured, because the plot I posted is the result of mathematically calculating the DIFFERENCE between two measurements (respectively using my other "Reference" dac and this one), so any speaker- or room-induced alteration is automatically zeroed out in the process.
Yes, I was thinking of trying this. I could do it by connecting my PC to the RATOC dac (instead of feeding it by SPDIF) and doing the appropriate pre-equalization in the PC itself. But it would be tricky to compensate exactly for the FR "droop", and also I'd have to figure out a way to switch between the two chains (network dac vs. PC>RATOC dac) seamlessly in order to be able to A/B them in real time...
Interesting quote. And yes, I think you have a point there!
I'm still a little intrigued, however, by the possible relative importance of this whole "transient response" vs. "frequency response" trade-off.
In a nutshell, the network player's dac seems to be more conventional in being optimized to prioritize the latter (FR), while the RATOC dac seems to have been designed in order to prioritize the former (TR).
Both sound good to me, but I wonder which might be the "better" approach from a theoretical standpoint, considering how our auditory system actually works.
And in any case, at the higher sampling rates, the issue becomes sort of moot, because then one can "have their cake and eat it too" so to speak.
It isn't quite that simple within Stereo. True, FR largely determines perceived tonal balance.
But phase response does matter for our perception.
Just try Pano's phase shuffler as an example. All it does is change phase and not FR, but it does change our perception. This is an easy example, where we change the phase of the left and right speaker to be different. Which ultimately changes our perception by changing the perceived frequency balance at the ears. But it proves that phase does change our perception, while FR is still the same.
Our room can change phase too, so it becomes an important part in the chain.
FR is king in tonal balance, but phase response does matter in our perception of that FR. As long as we listen to 2 speakers and have 2 ears.
I went ahead and modelled the theoretical sin(x)/x attenuation with Fs=44.1kHz, and then tried concoct a LP filter in order to match the measured frequency response of the RATOC dac. Turns out, a 6th order Bessel at approx. 38 kHz (= 0.86*Fs) does the trick (see attachment).
So maybe this is what they are doing in there. No digital filter, followed by an analog 6th order Bessel? (in fact, a 4th order Bessel with a slightly lower corner frequency seems to fit just as well, so it could also be that. But a 1st or 2nd order filter it definitely is not).
Attachments
There was a blind test on Computer Audiophile with files recorded from the analog outputs of the Yggydrasil vs the XXXHighEnd NOS Phasure DAC using PCM1704, the ADC was a very high quality Pacific Microsonics which is native PCM. The recording was a modern jazz band with some higher frequency primary and harmonics from the sax and drum kit. Eliminating the people that couldn't hear a difference in the files when it was revealed which file was which the Phasure turned out to be almost universally favored and the subjective responses (before the results were revealed) was that they preferred the file which was "smoother" (paraphrased down from similar adjectives).
It turned out the Phasure DAC was rolling off the high end quite a bit, similarly starting at around 10 KHz or perhaps lower. The discussion turned pretty uncivil after some point, so I'm not sure what the DAC designer of the Phasure had to say about it.
It turned out the Phasure DAC was rolling off the high end quite a bit, similarly starting at around 10 KHz or perhaps lower. The discussion turned pretty uncivil after some point, so I'm not sure what the DAC designer of the Phasure had to say about it.
Thank you for this, I found it very interesting.
In my case, I am finding that the rolled-off highs are a bit too "polite" for my taste, and I actually often prefer my "reference" sigma-delta dac...
That's all with 44.1kHz material. With high res, the differences are so subtle as to be essentially inconsequential.
Probably something along the lines of the impossibility of designing a good sounding filter any other way.
Those who got upset should get out more: this is old, old, old news.
Yes, of course phase can be used to change perception. It is used in reverb filters to add a sense of space, and I'm sure it is used in other recording/mastering tools as well. Pano's phase shuffler is a clever tool where the phase of the L and R speakers are jumbled (or shuffled) relative to each other so that they aren't the same, producing an improved FR response and a central phantom image.
However, the phase response of a speaker by itself is not so important (at least in my opinion) for a speaker to sound good. That is to say, a flat phase response is not audibly better than one with a 24db/oct LR crossover. Or, any one type of phase response is better than another. You can try this now with tools like rePhase and DRC by flattening the phase response independently of the FR. I have tried it and couldn't find a preference. Make sure to do it blind, so that when you switch, you do not know whether you're listening to the modified version or the baseline version.
It would largely depend on the room if you do hear a difference or not.
Try it the other way around. Have your bass section play before the tweet and try again.
Don't use headphones, you've got to feel the energy and hear it.
Do make sure your room is free from any mayor reflections for at least 20 ms.
This is a fun read too for crossovers in the vocal range: Siri's Killer Note
Try it the other way around. Have your bass section play before the tweet and try again.
Don't use headphones, you've got to feel the energy and hear it.
Do make sure your room is free from any mayor reflections for at least 20 ms.
This is a fun read too for crossovers in the vocal range: Siri's Killer Note
Last edited:
Yes, if the timing is grossly out of sync, then yes, it is audible. I've had it happen to me a few times because the FIR correction produces different delays for my subs and mains, and you have to get them somewhat aligned, otherwise it sounds odd. However, these are gross errors, on the order of 10s and 100s of milliseconds. But the delays or phase warps produced by crossovers are so minimal that they are unlikely to be audible.
I read that note you linked from Troels. Seems to me like he is expressing his subjective opinions on the different on- and off-axis frequency responses produced by different orders of crossover. The conclusion I do not agree with. It is not that one crossover order is preferable over another because of its so called "timing" benefits. But that different crossover orders produce different frequency responses and those are audible. How two drivers combine depend upon their inherent responses, relative position on the baffle with respect to the listening position, crossover point, order, and type. At the end of the day, its not the crossover order but the on- and off-axis response that matters. When we express subjective opinions, we are simply responding to the FR. Nothing more, nothing less. Now, I'm sure I'll be run out of the room with this notion, but it's been well established by folks much smarter than me, and who've spent their entire lives researching this topic.
This takes us back to the DAC that Marco commented on. Would be interesting to see if he ran any more subjective tests on it.
I read that note you linked from Troels. Seems to me like he is expressing his subjective opinions on the different on- and off-axis frequency responses produced by different orders of crossover. The conclusion I do not agree with. It is not that one crossover order is preferable over another because of its so called "timing" benefits. But that different crossover orders produce different frequency responses and those are audible. How two drivers combine depend upon their inherent responses, relative position on the baffle with respect to the listening position, crossover point, order, and type. At the end of the day, its not the crossover order but the on- and off-axis response that matters. When we express subjective opinions, we are simply responding to the FR. Nothing more, nothing less. Now, I'm sure I'll be run out of the room with this notion, but it's been well established by folks much smarter than me, and who've spent their entire lives researching this topic.
This takes us back to the DAC that Marco commented on. Would be interesting to see if he ran any more subjective tests on it.