Our daughter now lives in a log cabin on a 7 acre property with nothing but trees that looks a bit like your picture.
She left when I was still a child.
She left when I was still a child.
If a very quiet listening room is about 30 dBA, then 162 dB dynamic range brings us to ~192 dBA. As a point of comparison, the solid rocket boosters on the SLS make about 175-180 dBA 15 metres from the pad. Subwoofers by Northrop Grumman could have low Spouse Acceptance Factor, however. The room catching fire also might be a disadvantage.
Mind you, at that point the sound energy is about 10 kilojoules per cubic metre. Imagine about ten space heaters heating up only your body.
Nice numbers on the DAC but perhaps not overly practical.
Nice numbers on the DAC but perhaps not overly practical.
I talked to some pro beta users of this DAC at AXPONA and they were really impressed.
We now have one in house to use as a reference.
We now have one in house to use as a reference.
There is still too much unknown about the dac. Does it use dac chips? If so, which ones? Does it use I/V stages? How is it clocked? Is random noise intermodulation with the audio signal being measured? Soundstage depth cues? Lateral localization cue precision? Typical measurements don't tell the whole story. Impressed reviewers without any details of what the dac was compared with, what test recordings were used for comparisons, was DSP used for system and or room correction and if so to what bit-depth were the calculations, how were any differences in and or qualities of sound evaluated (ITU standards?), etc., isn't all that informative either. IOW, right now all we have is a number for dynamic range, and a vague description that some pro beta users were impressed.
Last edited:
https://positive-feedback.com/reviews/hardware-reviews/imersiv-d1-dac/
The first review and measurements. I won't say much more but Katz isn't the only person that won't send his back.
In 35 years of doing this I have been impressed by two or three things. This is one.
The first review and measurements. I won't say much more but Katz isn't the only person that won't send his back.
In 35 years of doing this I have been impressed by two or three things. This is one.
Katz seems to be comparing the Millennia dac with "standard" dacs, such as SMSL? If so, getting better sound than that isn't such a big trick. For one thing, standard measurements simply don't measure everything, and what they do measure isn't exactly the way humans hear. That's especially true for dacs, which are a very different animal from an amplifier.
If this is the first time Katz heard a really good dac, not just one that measures well at ASR, then of course he wouldn't want to send it back. But what makes it really good has to be a lot more than simply dynamic range and THD, especially as they are typically measured.
If this is the first time Katz heard a really good dac, not just one that measures well at ASR, then of course he wouldn't want to send it back. But what makes it really good has to be a lot more than simply dynamic range and THD, especially as they are typically measured.
Last edited:
They forget to measure one thing: a signal that triggers the muting of 7 bit high path every cycle so we can check proper merging of both signal paths.
So lets measure a tone just above this switching level, lets say -40dBFS.
What about the stuff that makes stereo properly holographic based on a two condenser mic recording in good concert hall, what separates instruments from each other, what makes accurate bass, soundstage width and depth (and their precision reproduction, not just an added effect), etc.? None of that is captured by standard measurements. At ASR they will claim all those are solved problems so only AP measurements matter. The difference is those guys can't hear and Bob Katz can hear. But apparently he still doesn't understand the limitations of his own measurements so he is likely attributing what he hears to factors he happens to know how to measure. Common mistake.
True standard measurements don't tell the whole story. The measurements are the best I have ever seen though.
If you can get orders of magnitude better as John has that is really something.
Katz is one of about a couple dozen of the top industry engineers listening on recordings they know or have made themselves.
They have access to the best available and this is better. John has indeed delivered a revolutionary product.
If you can get orders of magnitude better as John has that is really something.
Katz is one of about a couple dozen of the top industry engineers listening on recordings they know or have made themselves.
They have access to the best available and this is better. John has indeed delivered a revolutionary product.
If they can show that the path switching is working properly. Its that part where others had the most doubt of and its not measured....
Not for standard measurement dynamic range. However, I can get most of what Katz described hearing (as opposed to a "standard" dac). Actually, a former colleague of yours (starts with letter J, and who NP once described as still having one of the best pairs of ears in the business) has heard my diy dac. He described it as "very impressive," pretty much for the having the same audible characteristics as Katz described. Some parts of it are true SOA; however what makes it sound the way it does is mostly not about dynamic range. IMHO, its probably more about low "close-in signal-correlated noise," which is one of those things that's hard measure quantitatively, and something pretty much only a concern in data converters (and maybe in some class-d amplifiers). Also, spatialization is more about about very precise reproduction of things like ITD lateral localization cues, and precise reproduction of depth localization cues. Again, things hard to measure quantitatively.
Last edited:
What is not measured 'that humans hear'? I mean - we have a waveform on the recording, and we want it reproduced. There is no hidden orthogonal component that is not measurable. Its not even at a difficult frequency or amplitude range to measure accurately.
I have been looking at this some time ago.
The problem I have with it is this.
You design a 7 bit DAC that only handles the lowest level signals, with good linearity.
Then you design another DAC to handle the higher signal levels, with good linearity.
You splice the two DAC outputs at the right ratio and voilà, a 27 bit DAC.
But however you dice this, the high signal level DAC must have at least the same linearity as the low signal level DAC.
That means that the high signal level DAC must have the same LSB and LSB accuracy as the low signal level DAC LSB.
At that point, you realise there is no difference with a single DAC with the full 162dB dynamic range.
And that can't be build.
I have no doubt that it sounds great, but that is a different issue.
Jan
The problem I have with it is this.
You design a 7 bit DAC that only handles the lowest level signals, with good linearity.
Then you design another DAC to handle the higher signal levels, with good linearity.
You splice the two DAC outputs at the right ratio and voilà, a 27 bit DAC.
But however you dice this, the high signal level DAC must have at least the same linearity as the low signal level DAC.
That means that the high signal level DAC must have the same LSB and LSB accuracy as the low signal level DAC LSB.
At that point, you realise there is no difference with a single DAC with the full 162dB dynamic range.
And that can't be build.
I have no doubt that it sounds great, but that is a different issue.
Jan
Look at the noise skirts at the base of spectral lines. The consist of AN and PN (amplitude noise and phase noise) that has become intermodulated with the audio signal. Now the noise is part of the audio signal, not a harmonic, and not a stationary noise floor. When that correlated noise is of a random distribution then it becomes all the more insidious and difficult to measure quantitatively.
Why does it happen in dacs? Its because every dac and every ADC needs two analog reference signals, an analog time reference and an analog voltage reference. Both of those signals are multiplicative with the audio signal, not additive. Multiplication in the time domain is what produces intermodulation products, as well as modulation sidebands. When the noise and its intermodulation products are of a repetitive type, then you have a chance to see it as spurious spectral line on an FFT. When the noise is random, the the noise power is distributed across multiple bins (which on an FFT means it looks like simple additive noise, which it is not).
The difference between what happens in dacs and what happens in amplifiers is that for amplifiers it is assumed that HD/IMD is the result of a slightly curved stationary transfer function. That means amplifiers are considered to be only weakly nonlinear systems. In comparison, modern dacs are strongly nonlinear systems. Yet, standard measurements are always interpreted as though the DUT is only weakly nonlinear.
Getting back to dacs and FFTs, although noise skirts can be visualized in a high res FFT, trying to figure what they correspond to in the time domain can be vexing. Part of the reason for that is because spectral plots only show the magnitude of frequencies that have some average correlation with a bin frequency. Phase information is discarded which means that the crest factor of noise is unknowable. Also, the fact that FFTs are a measure of average signal correlation with bin frequencies over the length of an acquired dataset, means that non-steady state sine wave signals (what ESS calls non-PSS signals) do not necessarily show up at their full intensity even if they do appear as spurs. A loud, but one time chirp of a bin frequency signal during an FFT acquisition will only show up as having an amplitude which is averaged over the entire acquisition dataset.
Moreover, not all random noise is gaussian. There is more than one type of random noise bell curve, and in most of the curves tail events are more common than if the random distribution were a Normal one. It means you can have what looks like a constant low level noise floor on and FFT, say, white noise, but white noise can sound like popping and or frying sounds depending on the underlying process that is creating the white noise. In the most extreme case constant low level white noise on an FFT can be a Dirac impulse in the time domain. IOW, it can be a very loud one time POP event.
The main point I'm trying to make here is that dacs and typical audio measurement are actually more complex than we usually assume. IMHO, a deeper understanding is needed if we are to understand why even "standard" dacs can sometimes sound different from one another. Its also necessary if we are to understand how to design better dacs more like the kind Bob Katz would prefer not to have to return. They won't all necessarily need to have huge dynamic range measurements for that either.
Last edited:
What measurement system do he use - do we know? Its not so easy to measure at such low levels..
Apparently a Lyra 2 mic amp to raise the levels.... https://beta.prismsound.com/products/lyra2/tech-specs/
I wonder if mr Katz use the exact equal HW path for both cases... when such extraordinary findings are obtained it would have been prudent to make a really detailed description of the setup. If not for educational purposes.
//
Last edited:
I have a little problem following your analysis.... I would assume the two different DACs could be optimised around handling signal levels equal to half its output - if so, there would be quite different circumstances for the two DACs. I can't quite follow your reasoning around accuracy... - please help me?
//