Yes, it gives... nice numbers...
THD (when not in excess) is not so important, IM and phase modulation distortion are more. And they are related with slew-rate. I definitivaly use current feedback OPAs for analog in my designs everywhere i can. If i cannot (DACs), i try to find fast voltage feedbacks units.
And i do not care as well for very low noise numbers as long as i do not hear-it. It can even arrange the things a little, making the sound-stage more natural..
Just have to pick some nits... Silicone usually looks better than it sounds - I assuming you meant the chemical element silicon (Si, atomic number 14) that most semiconductors are made of, and not silicone, the synthetic polymer that is used in sealants and implants.
I do remember an old cerwin vega horn tweeter from the 80's that probably used both - it was usually referred to as "Dolly Parton" due to it's distinctive shape...
That means moving the dac to the amp...... Class D is not digital, but a pwm switching analog amp.
That means moving the dac to the amp...... Class D is not digital, but a pwm switching analog amp.
Well I guess that would be true if it was entirely impossible to generate the PWM signal in the digital domain (as opposed to just difficult to do well?).
Look for Nad M2 or http://www.diodes.com/_files/audio/ze0448.pdf or MAX98356.
That's a class Z
and IIRC uses an adc in the feedback loop, and so therefore also uses the analogue domain for the feedback loop
Have a look a Bruno's original papers on the ncore and follow his logic about the circuit topology of class D. This has been discussed many times in this thread. Class D uses an analogue feedback loop and is an analogue amplifier. Other topologies are other topologies
Because with 113dB DNR the DLCP is a not quite 19 bit DAC. It depends on how one chooses to do the budgeting but, in rough numbers that's 113dB DNR - 80dB headroom above noise floor - 10 to 20dB crest factor = 13-23dB of volume control range in an optimal gain structure. I generally use around 20dB of volume adjustment but tend mostly to listen to lower crest factor material so this works for me. But digital volume control in DAC to the power amp topologies easily runs into trouble if the dynamic range requirements are wider.
Every dB of excess power amp gain is a dB less of volume control range. Since there's basically no margin to start with I'm curious to see if Hypex's analog output board addresses the scaling issues for nCore customers. The nCore's gain is set, reasonably enough, to hit its 400W output but that means it's around 20dB too high for the majority of listening done with it.
Also, while the TAS3108 has a 48 bit data path and 76 bit MAC, it's limited to 28 bit coefficients. I'm not sure if that means it implements IIRs with 28 or 48 bit feedback but, either way, there will be some numerical noise in 24 bit output for filters operating at low normalized frequency---exactly what low means depends on DSP details and the filters being used but center frequencies below 1.5 to 2kHz is probably a good rule of thumb for the DLCP. How much this matters depends in part on how Hypex has chosen to partition volume control between the DSP and DAC and what the AK4396 does with the least significant bits, neither of which is specified (and if either's been mentioned on the thread I missed it), but there is some possibility of audible artifacts. I would tend to guess inaudible or pretty minor if audible, but this is really the sort of thing that requires ABX tests.
Post SRC4382 would be better as it'd avoid the impulse response smearing from the brickwall, linear phase interpolation filters in the SRC4382, though that's a significantly more involved mod. There'll also be some asymmetry in the impulse response due to the non-integer resampling ratio, which I've found tends to reduce sound quality as well (though in the mainline case of resampling 44.1 it won't be too bad). Given one can follow PLL based clock recovery with a jitter cleaner at a lower BOM than the 4382 Hypex's choice to do sample rate conversion is an interesting one. It'd also be interesting to know if the AK4396 is operating with sharp or slow rolloff antialiasing filters. Curiously, the group delay doesn't reduce in slow roll. Maybe that's a datasheet typo or maybe AKM shifts the filter's position along the linear phase-minimum phase continuum.
Last edited:
hmm well ****, I hadnt anticipated that type pf DNR/analogue performance.... but the point is, if you cant trust it to do volume, how can it be trusted to do DSP, since attenuation is a basic ability expected of a crossover.
that would be enough range for me still and i'd like to see how much easier it would be to better it with 3-4 channel balanced analogue control (12-16deck). in the real world....
that would be enough range for me still and i'd like to see how much easier it would be to better it with 3-4 channel balanced analogue control (12-16deck). in the real world....
Last edited:
Because there's a difference between attenuating in and out of band signals. The stopband of any crossover evenually hits some sort of analog or digital noise floor (more often than not it's analog---113dB DNR with 0dBFS being 2.9V is 6.5uV RMS of audio band noise or roughly a 3nV/sqrt(Hz) noise density, which means one has to pay attention to intrinsic noise and noise gain in op amps, power amps, and so on in the downstream analog chain not to screw it up) but it's by design that the noise floor be reached as it's intentional to reject the information in that part of the signal. Conversely, dropping the in band signal towards the floor is undesirable as the intent is to retain as much in band information as is feasible.
It may be more useful to compare digital volume to level adjustments in XO+EQ to match driver efficiencies. If one's being good about managing bit depth within the DSP, it's often feasible apply the XO filters first (which drops the peak signal level since some of the signal is rejected) and then boost the levels of the least efficient drivers. This is equivalent to turning the volume up so it actually increases bit depth. In speakers with enclosures levels are usually differ only by a few dB so the improvement from level matching is typically only a fraction of a bit. But, hey, it helps a little. Dipoles can end up with 15 or 20dB of boost at the low end of a passband, depending on the baffle configuration, so it's possible to pull out several "extra" bits at certain frequencies if the DSP's internal resolution allows it. The 48 bit paths in the DLCP's TAS are more than sufficient for this---32 bits is sufficient.
I don't expect to be a customer either, but from what I can tell Hypex is the most capable engineering organization in this space (and, while only having an advisory role on the DLCP, Bruno is one heck of a smart guy whom I've considerable respect for). So, for those of us who are building our own equivalents, I think it's pretting interesting to look at Hypex's implementation. And, well, this is the thread about Hypex's implementation.
how is this part of a different process? thats all part of the crossover, it would be part of the analogue XO, why is it separated here? youve just drawn a line to divide a process so you can disagree.... there will also be points along a slope that are not fully deleted but are attenuated, digital is steep, but its not THAT steep, corrections for driver response dips/peaks will also use the same process....
agreed, Bruno will go down as a legend in the field, whether I buy his products or not my system likely would not be as it is, if it werent for him, his papers are superbly written as well, quite the package. i'm just saying I have no interest in reading the tech notes with a view to defending, or dissecting the pitfalls of this specific design, I have enough work and reading to do to make sure my own build works out as planned.
if it wasnt an interesting design (the digital side is of more interest to me) I wouldnt be here reading about it, though there really isnt much info yet, just bickering...
of course you have to make sure the analogue stages are up to the job, even moreso with digital attenuation, but thats a given, I dont need to go to special efforts because of that, I would do that anyway.
using analogue attenuation in my system would be an idiot thing to do and to match the quality of it for 16decks....$$$$ forget it... makes remote control much more difficult too
Last edited:
I think it's an interesting and entirely valid one---it's more than a 2x cost riser from USD 300 to euro 550 at the moment, and that's not including Hypex's add on boards. Unfortunately I'm not aware of any hardware breakdown for the 2x4 or 2x8 similar to what you've posted for the DLCP (though it has been a few months since I last searched DIYA for this info). The ADAU1701 in the 2x4 is not as capable as the DLCP and there are some good quality measurements of the 2x4 over in the miniDSP forum that show miniDSP did a good job of getting what they could out of the ADAU1701. So I'm inclined to see this as mainly a DLCP versus 2x8 kind of thing. Unfortunately I don't own a 2x8 or have any plans to buy one so I'm afraid I can't be much help in sorting this one out. But if anyone who has a 2x8 is inclined to share measurements or take a look at the board that would be cool.