I’ve been upsampling to DSD256, and then recently DSD512 on the May. Compared to PCM 1.536M DSD sounds a bit too “polite”. It seems soft comparatively and lacks some of the dynamics, sense of air, decay and note tactility that shines through with PCM. It’s more realistic, analog, less fatiguing (and somehow comes across as less processed to my ears).
This is very similar to my experience with my prior DAC the Denafrips Pontus II. At first (prior to burn in) DSD sounded better. Afterwards PCM sounded better.
Sinc-Mx LNS15 (PCM 1.536) just sounds right to my ears (and on my equipment, my Pendant SE especially shines)
On Spring 3 I get much more air with DSD512 and more natural sound. 1.5M PCM sounds a bit harsh and more “on your face” with less space, attacks are a bit flat with more “flap flap” sound rather than more authority with DSD “thump thump”. Also with PCM sound gets easily more congested on dense recordings like some Opeth, Porcupine Tree or Steven Wilson. With DSD512 it keeps flowing with clear separation even on the most dense passages.
I use 1x = poly-sinc-gauss-long, Nx = poly-sinc-gauss-hires-lp, with ASDM7ECv2. Same filters also when comparing PCM. I don’t like sinc-Mx, to me it makes the sound fatiguing and lacks some transient resolution, but it works better with classical since there’s no such frequent top-end transients there.
Note!When comparing PCM vs DSD on Holo Audio, remember to set “PCM Gain Compensation” in HQPlayer settings to -6 dB! This gives you same output level for both formats. Otherwise one could easily think PCM sounds “better”, just because it is twice louder.
These were the filters that I have been using for months, either DSD256 / ASDM7ECv2 or DSD512 / AMSDM7 512+fs on my M1 Mini. They sounded so good I had no desire to change them. I try PCM every once in awhile, along with native NOS and always end up coming back to those DSD settings. Yesterday was different however. PCM sounds vastly superior, its not harsh at all. Less congested than DSD and zero listening fatigue (which I have been noticing lately with DSD). The music sounds more analog, organic and just rhythmically correct with PCM. Stage is vast and wide but separation between the instruments is conveyed with pinpoint accuracy.
Comparisons were done using various filters (same filters were used for comparisons in each) and PCM & DSD are always volume matched.
FWIW there are quite I few people who think PCM sounds better on the Holo May. @GoldenSound for one. Not sure if that is specific to the May or if he prefers PCM on the Spring 3 as well. I believe the PCM settings I posted earlier were his preferred settings as well (at least they were at one point).
Only way to properly volume match is to set the gain compensation in HQPlayer. Then it is handled automatically by HQPlayer for you. By ear it is impossible to do accurately enough. If your amplifier has volume control with dB display, then you need to drop volume by 6 dB when using PCM compared to DSD. But it is not apples to apples still, because the amplifier’s volume control may have impact on the result. When you use HQPlayer’s compensation, the output voltage from the DAC is the same for both cases. Then you know you what you are comparing.
There are of course always personal preferences, and it may be different with different music genres.
I assume you have DAC Bits set to 20 or less in HQPlayer settings? This is important in order for LNS15 (or NS9 or NS5) to do their work and remove the non-linearity distortions from the output.
Gain compensation was set in HQP, I did not attempt to volume match by ear
DAC Bits are set to 20
In addition to personal preferences and genres of music I think the rest of the chain comes into play (SS vs Tube amp for etc) I bought a ZMF Pendant SE last month (SET Tube amp from Justin at ampsandsound) and I am giving more time to my Pendant SE than my Violectric V550 Pro. I have found I prefer different combinations of filters depending on the type of headphone I’m listening to so its not out of the question that my preferences would also change with the new amp.
At least the systems seem to be somewhat different.
My systems using the settings I was referring to:
Spring 3 is connected to pre-amp + solid-state power amp (350W to the rated 4 ohm) going to Elac speakers (with their JET ribbon tweeters) and SVS subwoofer. It is also connected to Ferrum Oor headphone amp powered by Ferrum Hypsos PSU. This one is used with variety of headphones, Shure, AKG, etc. Digital room and headphone correction in use (Oratory’s EQ settings) in HQPlayer.
Spring 2 is connected to Schiit Jotunheim and Sennheiser HD800 headphones. Digital headphone correction in use.
T+A HA 200 and T+A Solitaire P headphones. Digital headphone correction in use. This is my current primary development system.
T+A DAC 200 connected to solid-state amplifier (180W to the rated 4 ohm) and Dynaudio speakers.
I am using a Don Sachs Model 2 6SN7-based linestage with a Don Sachs Kootenay KT88-based amplifier, and a Holo Audio May KTE set to 20 DAC bits when using PCM. I am using Tekton Double Impact speakers. Upsampling to DSD always sounds better than upsampling PCM to higher rate PCM.
I guess “terrible” for NOS Redbook is a little strong. But when I compare how rough that sounds to simply upsampling Redbook 2x it is not much of an exaggeration. Going 16x is MUCH better and going to DSD256 is another level again.
Yes, that is good approach from sonic point of view. It retains the filter’s characteristic properties throughout.
For DSD-to-PCM case choice of PCM Noise Filter and PCM Conversion are the most critical sonically, in this order. Followed then by further processing.
I’ve personally spent quite a bit of time trying to choose noise filter from “low”, “high-order”, “wec2” and “medium”. For DSD64 I would be likely mostly settling for “medium” these days. But the optimal choice for noise filter would also depend on what is the output gear.
I’m back to DSD, and I have a reasonable explanation what for what my obviously broken ears were hearing.
I switched from DSD256 / ASDM7ECv2 to DSD512 / AMSDM7 512+fs shortly after getting my Holo May KTE. I convinced myself DSD512 sounded better than DSD256, even without the Extended Compensation. (M1 Mini can’t do DSD512 with EC)
Going back and comparing these two now, what I was attempting to attribute to a difference between PCM and DSD was more likely the differences between DSD with and without Extended Compensation
I’m back to Gauss-long (1x) & Gauss-hires-lp (Nx) DSD256 / ASDM7ECv2. This has everything I am craving, without the shouty in your face characteristics of PCM using sinc-Mx
I do flip between Gauss-long and Gauss-xla for 1x, but today Gauss-long has more punch to my ears so I’m sticking with it.
Yes, 16 bits based on that. But note that the deviation starting from -90 dB is due to test signal reaching (analysis) noise floor. You can recognise this from the smooth upward curve. Real non-linearity would be flipping around 0.
For better figure, this would need to be measured in a different way. I will do it if I get one on my test bench at some point. But 16 bits is safe choice. 16 bits with high rate and noise shaper designed for this purpose already gives you plenty of dynamic range in audio band (much more than 96 dB).
I’ve just tried using this combo (gauss-long and guass-hires-lp), in my case using PCM.
My initial impressions were really good - things were smooth (transients?), soundstage felt even more spacious. Overall, not much to dislike.
But the more I listened, the more I realised that it lacked the body, meatiness and depth that I got from the Sinc filters. To be fair, I don’t think it would be that noticeable on lower-tier gear that I used to own (like the Hifiman Arya or even the HD800) but reproduction of depth is so good on the current headphones (Utopia, HE1000se) that it becomes apparent very quickly. The Gauss filter compared to Sinc L/M reminds me of what comparing he HD800 to HEKse felt like (not in terms of bass response, of course, just depth) - the former presents big, spacious sound, but it does so by pushing everything farther away from the listener and there’s not that much depth in terms of the relationship between/positioning of individual notes. The Gauss is big and impressive but more of a 2D canvass compared to Sinc’s more holographic presentation.
Judging by how much less time it takes for music to start playing when using the Gauss filter, I suppose it’s got far fewer taps than the Sinc filters.
It’s been good fun experimenting with different filters (there’s still so many I haven’t tried). So far, Sinc-L is still the most convincing/advanced to my ears, but I must admit that I appreciate what apodising does to recordings where the apodising errors counter reaches high numbers. Then I switch to Sinc-M and things get smooth again. Really nice. I recommend checking out Wiz Khalifa’s “On my level” on Sinc-L and then Sinc- M - the error count is very high and the sound is much smoother on Sinc-M.
I am, however, struggling to understand why even on recordings with 0 apodising errors there is still a noticeable difference in sound between Sinc-L and Sinc-M. Perhaps it’s due to L’s much higher tap count.
Is there any difference between Sinc-M and Mx when listening at 705.6KHz? I understand that their number of taps are supposed to be the same at that rate, would everything else be identical too?
sinc filters also use bigger processing block sizes that also affects delay.
Initialization time depends a lot on the filter type. Regular poly-sinc category can take a long time.
Because they are completely different kind of filters. So even with same number of taps they sound different. You can compare this with sinc-M vs sinc-Ll at 16x conversion ratio.
These filters cannot reach reconstruction or time domain accuracy of poly-sinc-gauss-long/xl/xla though.
They are steeper, if you like extremely steep brickwall filters. But that comest with cost in time domain length. Time and frequency domain are related through 1/x relationship.
sinc-S and sinc-M(x) are similar to poly-sinc-ext2 and poly-sinc-ext3. So you could try to compare to those ones if you like sinc-M.
