I don’t know what your EE/signal processing training is, but you have said two things on this thread that are plain wrong.
About upsampling/digital filtering: unique reconstruction of the original signal is only possible for band-limited signals when the receiving D2A has an exact copy of the band-pass filter used to band limit the source signal before A2D. That’s of course not the case in the real world where the source filter conditions are only approximately known and vary among source material. As a result, unique reconstruction fails, that is, there will be multiple analog waveforms compatible with the samples. This is nothing to do with nonlinearity, it’s the math of the sampling theorem. Thus all DACs have to make assumptions about how to reconstruct the signal, whether with a digital filter/upsampler, or in the analog side as in NOS DACs.
20 vs 24 bits: This has nothing to do with sampling, but only with noise level. 20 bits is more than enough for human ears, there’s nothing in the bottom 4 bits of an alleged 24 bit recording that is anything but source dithering. The only point to use 24 bits is to provide headroom for DSP either in the studio or in room correction at the destination, as DSP does arithmetic with samples. (Well, another reason is to score “mine is bigger than yours” points with people who don’t know any better). And BTW, for quite a while Yggy DSP has been dithering, and originally it did not truncate, it rounded.
I was referring to the original signal as the digital file - not the original analog signal before A2D. I agree that nothing can reconstruct this original analog signal accurately - even if you had the original A2D band limiting filter I doubt the reverse process would work well in practice even if it is theoretically possible (you would need incredible precision like 64 or 128 bits or more considering the band limiting filter is usually a brick wall)
Good to hear that Yggy is fixed now and does dithering - I was shocked to read it did truncation or rounding (which amounts to the same) as it shows ignorance on the part of the designer with respect to digital.
As for nothing in the 4 LSB on a digital music file except noise. I agree that most of the time this should be the case. However, there are products out there now with 21 bit resolution which is starting to get close to 24 bit. In any case, quantization noise from a 20 bit truncation or rounding could be quite nasty and audible compared to random noise (of course only at high volumes)
Good discussion. Sounds like you have some good training in time series analysis.
BTW I have heard KEF Ref 1 on an all ARC setup with a JL sub to fill in the ultra LF and I was highly impressed! Congrats on a great setup!
I remember this being said about 16 bits not that long ago…
Similarly, 16 Million colours were all the human eye could see (24 bit true colour) and all that a decent graphics card needed, but now we are up to 48 bits…
You say that truncation noise at 20 bits could be audible, but given the energy for the LSB at maximum acceptable volume, that seems a real stretch, for which I’ve never seen any evidence.
I have no idea and you are probably correct; I was just making the point that once someone sets absolute’ that is all you’ll ever need’ limits on something, progress overtakes them at some point and reasons for higher limits are found.
True.
But on this matter I remember a review somewhere (Stereophile?) of a 2x300W amplifier, they raved about its low noise level, completely black backgrounds — and Unusually, they specified it not just in V or dB, but as 21.4 bits.
21 bits below 300 watts, mind you.
And this was a record, the lowest noise they ever measured.
Excellent point. It is indeed a very small signal and provided the music is mastered properly with peaks close to MSB then we are talking a signal level which is -120 dB from the peaks. So the music would have to be mastered really low in level which would be quite unusual.
I’m no expert, but I don’t expect upsampling to change the sound of itself. I use upsampling to create additional bandwidth in order to use a wider variety of reconstruction filters. Digital filters are always a trade off between steep and short, the various ways of addressing that can sound different.
Yes, that brings up another point about dithering. At recording/mastering, the low-order several bits (at least 4 but more likely more) are noise, and in fact they are likely to be “shaped” with an appropriate filter, that is, dithered. In other words, the low-order bits of the digital source you get are randomized anyway. What truncation or rounding can do to that is just a slight change in noise shaping at sub-audible power. Mike Moffat, the Yggdrasil designer, has been in the DAC business for decades, in fact he designed some of the earliest high-quality DACs at Theta Digital. He’s also an iconoclast who likes to mess with the hi-end industry and audiophiles. Interpreting what he has written about this issue, my guess is that he went for the simpler coding (not RNG, …) of a deterministic word length shrinking originally because it made no audible difference and it poked at the bear, but ultimately it was too much hassle to keep explaining so they rewrote the firmware to dither, even though it makes no audible difference.
Not sure if you @Sloop_John_B are still looking, but I just got the Ayre QX-5 Twenty.
Oh man… when I thought it was hard to improve anything in my system, the QX-5 brought everything to an amazing level. Absolutely delightful! Not exaggerating folks, I have been listening all my favorite tracks today for hours. It’s like listening to them for the first time. I am stunned! And trust me, it’s hard to stun me with this stuff.
This is not my first DAC. Far from it. Have gone through many, the latest being the PS Audio Directstream.
I call whoever says all DACs sound the same (over $100 computer chip level) utterly BS! What the f&*”# have you been drinking!?
In terms of the original question, definitely consider the Meridian 818v3. Does most of what you require and does it very well indeed. DSD is over DoP, though, but MQA is fully supported.
Also, someone else suggested the DCS Rossini. This is excellent also, although I don’t think it’s a full MQA renderer, just does the first unfold, I think. Don’t know of the relative importance of this aspect to you, though.
Doesn’t play well with Roon just at the moment MQA wise (no EQ etc possible). Meridian are looking at a fix but they can be a bit glacial. DSD512? I think not. Not sure how the SQ would stack up against the current arrangement, I wouldn’t back it. Not sure if it’s pretty, it’s quite a chunk for a few boards and a PS.
Not quite right. Roon solves the problem of MQA and DSP EQ.
But the 818v3, like much Meridian gear, cannot do the second-stage rendering of pre-unfolded MQA, in spite of being fully capable of unfolding+rendering — it doesn’t recognize the rendering signals. There are rumors that Meridian is working on a fix.
Nonetheless, I use it with DSP room correction all the time, and when MQA content shows up, Roon unfolds it and I am content.
It has great strengths, and some limitations, and a high price.
I always finds reviews interesting, personally I use my own ears. It sounds very good. It doesn’t sound as good as my T+A Dac 8 in my system and I understand the Holospring to be a step further.
