Saturday, July 15, 2017

The Dacs I use now are Nearly time aligned (within 0.44 msec)

I finally tested delay, using my ancient test file LoudPolarityTest.wav, which outputs a large scale signal (not -20dB, but I can't remember how high, perhaps -5dB) postive only narrow pulses at about 20 Hz.  I couldn't get a clear oscilloscope picture of both dacs at the same time using SpeakerPop, which has much lower frequency of "pops".

At 96kHz, the Emotive and Denon DVD-5000 DACs are nearly perfectly aligned.  I found the two traces line up more perfectly adding 0.06msec to the Emotiva Stealth DC-1--which apparently has 0.06msec less latency at 96kHz than the Denon (I'm lucky it's not a much larger difference).

But at 44.1kHz, the alignment slips apart a bit, and the Emotiva there leads by about 0.50 msec.  So there is a 0.44 msec difference between the two excess latencies, which is really the important factor in the system design (since I can always align the different ways to accomodate the constant latency differences, at least up to about 300 msec...and in fact I must adjust latency anyway because of speaker offsets, though if doing so by guesswork as I mostly do I would need to keep this in mind).  But here the constant latency is actually quite trivial at 0.06 msec (less than one inch sound travel) but the variable latency at 0.44 msec (at more than 6 inches) may be worth thinking about, if not always compensating for.

Initial offset at 44.1kHz, about 50 msec
Aligned to compensate in DEQ

This isn't perfect, but it's fairly trivial.  I can't say my time alignments are necessarily even accurate to  0.44 msec, they may not be "accurate" to 1 msec, though I try for 0.02 msec accuracy when I do such things (not frequently) by measurement, and I have believed rightly or not that I can hear balance differences down to 0.02 msec.  I can adjust down to 0.01 msec.  I haven't done a measured alignment in years and it could be as much as 2msec off, but I'd guess it is still within about 1msec.  Nowadays I set the alignment by guesswork mostly.

However, the lenght of the Acoustat being what it is, time alignment at some point is going to differ as much as 8 msec from time alignment at another part on the same speaker, fwiw.

I could just let this slide, adjusting for best response at 96kHz and leaving that to be that.  Or, I could split the difference, leaving a barely audible if at all difference of 0.2 msec on each side (equivalent to  about 2 inches of air distance) which is even less noticeable, or I could resample everything to 96 kHz, or I could adjust if I am being super critical.  Is it more important that the tweeters align with the panels (you'd think so based on the frequency relationships, but not on the large length of the acoustats, which is going to make any precise alighnment moot, and it's barely audible anyway) or the subs?  Right now with DC-1's on subs and tweeters, those are always in the same (mis-) alignment, and they are both point sources, which maybe matters.

I had a lot of concern when just looking at 44.1 whether they were actually within 0.5 msec, or it was 50 msec times some number plus 0.5 msec.  But since they line up nearly perfectly at 96kHz, it seems likely there's only a small extra at 44.1 as I observe, and the possibilities of lining up perfectly at 96kHz by chance even though there's a huge difference...seems very unlikely.  (I spent much time thinking about this though.)

Testing the RD-V1, which I used for a year before getting the Audio GD Dac 19, it has 0.32 msec less latency than the DC-1, amazing for such an ancient device (there's the Apogee clock, which must be an asynchronous receiver system).  AND the latency stays constant at both 44.1 kHz and 96 kHz with the DC-1, which suggests they must both have constant latency (and it was the DVD-5000 et al that have variable latency).

Since I live in the old fashioned world where sources rule because of SPDIF and AES, and Dac's aren't the master with slave control over the source, I put the RD-V1 as my digital transport for CD's, figuring the more accurate clock more important to me there.  But, look what it could do as a DAC, perfect time alignment at every rate.

Sadly my recently purchased RDC-7 has blown outputs, that might have been even better than RDV-1 because balanced output and dual differential 1704's.  But perhaps I was wrong, actually, in believing the RDV-1 does not have dual differential, as I just read somewhere that it does.

The RDV-1 does not beat the DVD-5000 however in RMAA testing.  In fact it's considerably behind in S/N and dynamic range, and about exactly the same in THD.  So it would not necessarily be a better mid way dac.

So we have this ultimate jewel configuration of dual differential 1704's in 19 year old DAC's, lots of them, it seems, not as esoteric as Wadia or even Levinson.  They vary with state of repair or something.  And the analog circuitry, which can generally be assumed to be equivalent to 5534's, which are actually pretty good, I think 5534 distortion is typically below 0.0005%, but not as good as LM 4562.  Compared to OPA2134...a very common choice...I'm not sure the 5534's are clearly worse, they're about the same level.  The OPA 211 are best of all.

Or maybe my poor measurements of my first DVD-9000 are not a matter of state of repair.  The second Denon DVD-9000 measures almost exactly the same 95.8dB S/N ratio, dynamic range, frequency response, the same except distortion in the first DVD-9000 at 0.0014% was actually lower, the second DVD-9000 pretty much matches most of my older dual differential 1704 stuff at 0.003%.

The second DVD-9000 also shows the same hashy looking highest octave noise (but below -110dB) that the first DVD-9000 does.  I'm beginning to wonder if that's an artifact of the PMI 100 digital filter being used.  I love the sound of the DVD-9000 on HDCD's, that was my original reason for having it, but then I decided it was great on other discs too, except my first wouldn't play DVD-Audio because of a problem Denon Service couldn't fix.  Well maybe the smoothing digital filters are actually producing a kind of noise.  Perhaps it even sounds good while measuring poorly.  But this is all speculation now, though the DVD-9000's are the only players to show that noise, and the only ones with real PMI 100's, it may be coincidence.

Here's a picture showing the 9000's and the 5000, and more:

Clearly the DVD-5000 is the winner as an R2R dac in my collection now.  The DVD-9000's both show identical weird top octave noise during the slapped together RMAA test.  However there's no reason not to use the newest DVD-9000 as an HDCD player despite higher nominal measured THD which may be measurement artifact from loading differences.  The two 9000's differ by the second one having a tad more third harmonic, which contributes to only a tiny rise in IM (because, strangely enough, not an even harmonic).  Meanwhile the 5000 is far cleaner at the top, you can see the noise threshold all the way to 20k at least with only a few peaks, but it has more 2nd harmonic, still quite low at -106dB, a flaw I would not consider too serious.  The 9000's apparently cancel that better, but have more distotion elsewhere.  The 5000 does have a bit more power supply ripple though still -120dB.

The RDV-1 does show much lower jitter as a transport, right at my limits of measurement around 170ps.  I think that's what you get from simply tracking a SPDIF stream, the limits of SPDIF itself so to speak, though SPDIF receivers can then reduce it to almost zero using PLL or asynchronous interpolators.  The clock inside the RDV-1 is what ultimately matters, and that could be considerably better than 170ps but no worse.  Meanwhile, the second DVD-900 measures 500 ps playing the same 880 Hz disc (my current test for this, such as it is), clearly worse, not as good as SPDIF itself.

No comments:

Post a Comment