Looking at the Katz review and measurements, it seems like the “single-path” one measures very poorly? Several hundred micro volt output noise?
A more fair review would be to compare with a SOTA conventional DAC in the same price range at least. Those can produce a fairly clean sine wave at -90 dBFS.
This review seems rigged to me.
A more fair review would be to compare with a SOTA conventional DAC in the same price range at least. Those can produce a fairly clean sine wave at -90 dBFS.
This review seems rigged to me.
Unlikely. Bob Katz is a very well-known and well-respected mastering engineer. He wrote what kind of turned out to be the bible on the subject: https://www.amazon.com/Mastering-Audio-Third-Art-Science/dp/0240818962 That said, he is not a dac design expert so don't expect him to know exactly why a dac sounds the way it does, nor all the different ways to measure one.
The idea of sticking two dacs together to get more bits is hardly new. Segmented dacs have been around a long time. People have also stacked dac chips like TDA1541 so that each dac plays only the top half or the bottom half of the AC waveform.
The question is more about how far the techniques can practically be taken. Apparently for the new Millennia dac they claim that when the high level dac is doing most of the loud reproduction, the lower bits of the lower level dac are being masked anyway by human perceptual limits. Therefore the claim is, the fact that lowest bits are not being heard means there isn't a problem with the exact LSB size.
Regarding "veiled or grainy" sound, didn't that come from Bill Whitlock?
Please see attached, then search for veiled or grainy. Also, seems to me there was a video of him in a discussion panel where he was asked how to know if you have veiled sound. Again IIRC, he answered to effect that you know it when its gone.
Please see attached, then search for veiled or grainy. Also, seems to me there was a video of him in a discussion panel where he was asked how to know if you have veiled sound. Again IIRC, he answered to effect that you know it when its gone.
I wasn't at all asking for a description on how it works. As one can see, the post was aimed for Jan D.
//
It probably works ok. TI has a similar feature on their ADC's called "Dynamic Range Enhancer".
It uses a PGA (programmable gain amplifier) that is part of the ADC anyway and that can adjust gain in < 1 sample.
You can select up to 30dB extra gain at a threshold of -66dBFS and anything up to (say) 2dB at -12dBFS.
As long as the analogue side has less noise then the DAC you can improve resolution at anything below the threshold.
It can of course also be applied to a DAC.
Something similar was applied guitar amplifiers as the "powersoak" in the 80's if for totally different reasons. The idea was you crank up the up the Amp (say Marshall 100W Tube Head) to give you the "overdriven" tone, the power supply sag etc. plus the tone of a 4 X 12" Marshall Cabinet in a small studio room, at ear, engineer and even guitarist friendly levels.
For iFi I took the powersoak concept and turned it into an attenuator for those pesky 130dB/1V sensitive in ear monitors that also block the ear canal and produce significant isolation from outside noise, so every bit of noise is highly audible and to get 85dB SPL @ -20dBFS you need to turn the gain down 24dB, often in the digital domain. Use a 15R + 1R "powersoak" with nice highish power thin film resistors and now 1V output get 86dB @ -20dBFS with a 1V signal with the volume at a level that gives 1V out.
If we started with an amplifier with -120dBV noise and 25dB attenuation for a 95dB SNR we now turned the situation into the equivalent of amplifier with -144dBV output noise (and 62.5mV maximum output). It sure works this way.
Next stop, why not automate the whole gig many ways of doing it. Do it right and you start with a -120dBV system and improve it significantly in the real world. It's all about scaling the "0dBFS" level.
Will it create mind blowing sonic improvements? If you believe it will, it probably will.
TU Dresden - Faculty of Psychology - We hear what we expect to hear
Thor
To remain monotonic - a key linearity requirement for a DAC - the LSB of the 20 bit upper DAC must have the same 'accuracy' (not sure that this is the correct term in this context) as the lower 7 bit DAC LSB. There is no advantage in cutting up the DAC in two pieces and then go through heroic measures to try to stich it together again to what it was before.
As Google mentioned: A DAC is termed monotonic if the analog output always increases or remains constant as the digital input increases. If the DNL is less than -1 LSB, the DACs transfer function is non-monotonic.
Jan
Last edited:
Yes, supposedly caused by RF. Here it is supposedly caused by something totally different. Also "wife in the kitchen" in audiophilia can be used to refer to various supposed audible improvements.
CS43131 and CS43198 use some secret algorithm to enhance the DR. This has been reported to cause audible and measurable issues (e.g. https://www.audiosciencereview.com/...gh-unbalanced-output-power.62024/post-2275689).
Help me sort out the minus sign...
"..... A DNL of greater than -1 LSB will guarantee monotonicity and no missing codes."
Page 3 and onwards gave some good insight of the concept... https://www.ti.com/lit/an/snaa077c/...05956&ref_url=https%3A%2F%2Fwww.google.com%2F
Anyway, where the 2 DACs meet, around that digital input level to the system - stepping from higher to lower digital representation must never yield in an analog increasing value out of the system in order to be said to be monotonic. There is a Sirius chip that don't seem to be monotonic when coming from one of the directions.... due to other reasons than the architecture discuss here I believe...
Remaining "constant" would also meet the requirement of being monotonic - which I think is a too loose for audio (?) but seem more useful in a digitised control ADC/DAC system where things is still apparently OK when constant but can go really bad if reversing...
//
Monotonic is a very clear English word, and is used correctly in the definition of monotinic DACs. If you step the digital input from 0 to the max code, at each step, the analog output must either increase or not change, but never go down at any step.
If that is the situation, the DAC is monotonic.
If not, your analog audio signal would decrease when the digital input increases, and such a DAC is unfit for high quality audio.
So when you go from the max step of the 7 bit DAC to the 1st step of the 20 bit DAC, your analog output change has the same limits as when the 7-bit DAC steps from 0 to 1, or when the digital input steps from, say 2000 to 2001. Cutting up the DAC and splicing the parts later doesn't change this ugly fact.
Jan
If that is the situation, the DAC is monotonic.
If not, your analog audio signal would decrease when the digital input increases, and such a DAC is unfit for high quality audio.
So when you go from the max step of the 7 bit DAC to the 1st step of the 20 bit DAC, your analog output change has the same limits as when the 7-bit DAC steps from 0 to 1, or when the digital input steps from, say 2000 to 2001. Cutting up the DAC and splicing the parts later doesn't change this ugly fact.
Jan
Funny, I thought Denon resolved this problem back in the 90's when they added 2 Bit DAC's made from logic gates and resistors to existing commodity 16/18 Bit DAC Chip's to create their 18/20 Bit CD Players. For PCM I really don't see what the kerfuffle is about, it's ancient tech. Nothing is unusual or difficult.
Bitshift the right bit's into both DAC's, attenuate one DAC by 2^x and you have a very large number of bit's.
I once did this as concept for TDA1541. The TDA1541 has 110dB SNR @ digital silence. Parallel 8 pcs and we get 119dB SNR. It's pointless of course, because TDA1541 is a 16 Bit DAC.
But if we add TDA1541 # 9, we can load the lower 16 Bit of a 32 Bit word into this DAC. We could entierly ignore the MSB and only use the lower 15 Bit.
Use a 0.1% tolerance resistor of 1R as I/U conversion, use a "T" attenuator with a very high value build out resistor to inject the correct current into the main TDA1541 Bank Output. Full scale on the main bank is 32mA PP, we just need to scale the MSB current so it is 1/2 LSB current of the TDA1541 * 8, bob''s your uncle. A Trim may be necessary (it was was with the Denon DAC's.
Now to combine modern DAC's where we have more Bit's than ENOB will need a different variation on the same theme. It's not exactly trivial, but it's a pretty basic engineering problem.
More interesting question is where do you find 27 Bit recordings BTW? Even 24 Bit recordings are commonly random noise in the bottom 6 LSB.
I'd actually rather have a good 18 Bit Current Steering DAC using bipolar and ECL Logic, mono, capable of 1MSPS than a 27 Bit DAC. Sadly nothing like this in production. One could use 4 X TDA1541 and effectively use them as the "MSBS" in a 4 Bit thermometer DAC for the MSB (all get loaded the LSB's below equally, but 0/1/2/3 DAC's load zero data according to the needs of the 4 MSB).
Thor
@jan.didden ,
please have a look for posts by user @Signalpath over at ASR for more info about how this DACs work.
This is not a physical 7-bit and physical 20-bit DAC combined, rather the implementation uses standard DS DAC chips always at their full resolution.
The main idea is to switch off the upper DAC once input signal is below -45dBFS so that the lower DAC can work in full resolution and not be disturbed by the noise of the upper DAC. The mixing of the two DACs signals is passive, using resistors. See US9871530, Fig.3.
This thing does neither have 162dB Signal-To-Noise Ratio, nor 28bit actual resolution nor does it have lower distortion. It "only" offers an intelligent increase of dynamic range to allow to reproduce very small signal levels more accurately, The switching is basically glitch-free, in contrary to the simple scheme of changing analog output gain of a single DAC and inversely changing the digital input level. And the transition is monitored with an ADC so that gains and offset of the two paths can be controlled dynamically to facilitate the glitchless switching.
please have a look for posts by user @Signalpath over at ASR for more info about how this DACs work.
This is not a physical 7-bit and physical 20-bit DAC combined, rather the implementation uses standard DS DAC chips always at their full resolution.
The main idea is to switch off the upper DAC once input signal is below -45dBFS so that the lower DAC can work in full resolution and not be disturbed by the noise of the upper DAC. The mixing of the two DACs signals is passive, using resistors. See US9871530, Fig.3.
This thing does neither have 162dB Signal-To-Noise Ratio, nor 28bit actual resolution nor does it have lower distortion. It "only" offers an intelligent increase of dynamic range to allow to reproduce very small signal levels more accurately, The switching is basically glitch-free, in contrary to the simple scheme of changing analog output gain of a single DAC and inversely changing the digital input level. And the transition is monitored with an ADC so that gains and offset of the two paths can be controlled dynamically to facilitate the glitchless switching.
I don't really care what is inside the 'black box DAC'.
Information and engineering is what it is, you can't fool Nature on basic principles.
Harry Nyquist figured it all out for us, decades ago.
And I challenge you to produce a passive resistive summer that is accurate to say 1:1.000.000, which is 'only' 120dB.
Pie in the sky, but clever marketing.
Jan
Information and engineering is what it is, you can't fool Nature on basic principles.
Harry Nyquist figured it all out for us, decades ago.
And I challenge you to produce a passive resistive summer that is accurate to say 1:1.000.000, which is 'only' 120dB.
Pie in the sky, but clever marketing.
Jan
So conceptually it's the reverse of TI do on ADC's for like a decade or so with their "dynamic range enhancer".
Interesting patent. Northrop Grumman.
Thor
???
The range difference is ~ 45dB, so a ~ 200:1 resistive mixer, and it only has to be stable.
Output impedance is ~ 2R, so the resistive mixer would be 400R:2R
Maximum output is ~ +24dBu, so the switchover happens at ~ -22dBu.
Assume the signal is well below -45dBFS. Only the lower DAC is operating, the upper DAC receives a zero signal and its output -- input to the 2R of the mixer -- is shorted to GND,, not contributing noise. When the lower DAC starts to approach its headroom limit, the upper DAC is switched on at some point, but not outputting signal only adding noise. Upon further increase of level the digital input signal is crossfaded to the upper DAC in such a way that the lower DAC does never max out. Since the device seems to monitor the two DAC output signals and the mix it can handle any static or dynamic tolerances, and there is a calibration feature.
That's at least what I think I have understood.
I haven't read AES paper #21106 -- not an AES member anymore -- but the inventor says everything is detailed there for those who care.
But, as noted, posts here contain a lot of valuable tidbits (look in the "minimum noise levels" thread, too).