It is best set HQPlayer output fixed 48k in that case. DAC Bits to 16 and TDPF or Gauss1 dither.
You’ll anyway need a filter to convert 44.1k sources to 48k and the apodizing function still works.
4 Likes
Coming from PCM sound and having “converted” myself to DSD upsampling, I rediscovered the Sinc-Mx and Sinc-L filters.
They sound completely different in DSD and I am trying to understand why. (Any clue, @jussi_laako?)
In PCM Sinc-L sounds clear, transparent, sharp, but also potentially edgy. In DSD it is dense and completely “analog”, all the “digital” signature is gone for good, but it is also much softer and more relaxed. That was a big surprise to hear.
If I do the math right, Sinc-L has 32 million taps from redbook to DSD256, while it only has 2 million taps from Redbook to PCM756. Sinc-Mx would have about 16 million taps from 44k to 256x. Does the Sinc-L softness have any relation to the number of taps? I no longer buy into the “linear PSU running HQP” and “audiophile PC” nonsense (well, I don’t think I do!), though some would argue that unoptimized PC = more computational errors = more softness.
However, I wonder what could be the reason for this huge change in sound signature. Gauss-long, for example, remains Gauss-long whether in PCM or DSD mode (except for the DSD sound signature and modulator part).
In any case, I don’t mind a 20-second wait to use sinc filters. They are that good!
In particular, I found that the Sinc-Mx sounds absolutely hypnotic and sublime in 7ECv2. Unlike the Sinc-L filter, it retains the sharpness of its PCM version. It is very captivating! Have you
tried it, @MistaLovaLova? I’d be curious to know how it compares to your beloved Gauss-long.
2 Likes
sinc-L is a type of filter that in a way improves as function of length. To keep steepness similar and delay constant, it scales with ratio just like most others. While most other filters retain same quality regardless of length scaling, like sinc-Mx for example. So it could be that this length increase lessens it’s characteristic “flavour”. To test this theory, you can try the shorter variants of sinc-L, such as sinc-Ls at DSD rates. If it comes again closer to how sinc-L sounds at PCM it confirms this theory. Not necessarily the case, but just a theory of possible reason.
2 Likes
And simply, poly-sinc-gauss-hires-lp for all : 1x and Nx… 44.1 to 96 in imput ?
Jussi, have you experimented with lower order modulators? Especially for the higher rates DSD512X and DSD1024X. With my limited knowledge about dsd modulators I think that at these rates less noise shaping is needed. A third order modulator might be enough? I’m asking because I seem to prefer the 5th order modulators, they sound a bit warmer with higher tonal density when I compare them with 7th order.
A simple modulator would also mean less processing power needed, and DSD1024x can be reached with less powerful hardware.
Very interesting explanation, thanks!
I too hear the difference between Ls/L. Sm sounds like an in between, but closer to Ls.
Sinc-L @ 32 M taps is very unique. I would describe it as the emblematic “velvet sound”. Everything seems to stretch. It is tonally rich, fluid and linear, there are no rough edges or accumulations on the edges of the notes. It makes music flow like like sand in a Zen garden! Although it lacks punch for some albums, it offers a very relaxed and hypnotic presentation which is especially interesting for longer listening sessions.
Could you please explain the word “quality”? Does this mean that the 16M taps of the Sinc-Mx @ x256 don’t improve the filtering quality as much as the 16M taps of the Sinc-L?
1 Like
Yes, of course I have. I think the 5th order and 7th order is good balance. I’ve considered 6th order in the middle as well, but not implemented such yet.
If the modulator order gets too low, it causes some additional issues. And if it gets too high, it will demand a very elaborate conversion/analog filter setup and there’s no particular benefit in such.
So I rather use the higher rates to improve bandwidth and/or dynamic range. Those adaptive modulators can adjust to the rate changes for example.
It shouldn’t be too bad nowadays with non-EC modulators?
2 Likes
It is worth to remember though, that it is non-apodizing. So the source material matters a lot, since it will not correct the source errors.
Usually number of taps don’t really relate to quality, only steepness (roughly, when you double the conversion ratio, you need to double the number of taps too to retain same steepness). sinc-L group is exception to this rule as it is more complicated for that case. sinc-L group quality depends on the length too, not only, but it is notable part of it.
2 Likes
It’s been a while since I’ve done PCM to DSD comparisons. I never spend too much time doing that as the initial “wow factor” of PCM’s attack and force quickly turns into a realisation that it’s most likely an illusion due to sounds being presented more in the form of a “wall of sound” (relatively speaking, of course, when compared to DSD and bear in mind I have a delta-sigma DAC). As a result I hardly ever use PCM nowadays.
Sinc-L is special, in my opinion. I keep coming back to it, depending on the mood and source material. It is majestic, grand, full-bodied, authoritative and deep/holographic-sounding. There are days (and even weeks) when I prefer the gauss-long filter because it sounds more natural, less sharp, and puts me in the middle of the performance (albeit at the cost of the soundstage becoming more intimate and kind of changing it shape a little, wrapping itself around my ears), but I have had so many great listening sessions with the sinc-L. I must admit than on days when my ears are ready for the full force of a recording, the sinc-L can make the gauss-long filter sound a little muddy/dirty in comparison.
I actually like your description of “velvety” to explain this. I’d say that gauss-long achieves the sensation of realism by making things smooth, a little darker, with sounds having a little less clang to them, whilst sinc-L is like a super-high-res, super-clean and vivid 8K TV (in this analogy the gauss-long would be a slightly less-resolving TV with a low-setting energy power mode switched on which actually keeps your eyes from getting tired and overwhelmed, potentially making the experience better in the long run). Which one is ultimately better? Hard to say, I like the TV analogy because for me it is hard to deny that sinc-L would appear to be the more advanced, “correct”-sounding filter. But when we enter the realm of sheer music enjoyment and making trade-offs in order to find that elusive equilibrium - then gauss-long is pretty convincing.
It surprised me that you found sinc-L smooth compared to sinc-M(x). I am not a big fan of sinc-M filters because they smooth things over just a little bit too much for me, making them sound a bit “off” to my ears. Gauss-long does that too but it also does other things at the same time and the final effect is somehow convincing, whilst sinc-M(x) always sounds to my ears like a tweaked version of another filter (which it kinda is, I guess). I’d say that it’s the M filter that provides smoothness over everything else whereas the L filter, whilst never sharp as such, is more of a vivid, no-compromises type of sound. It’s interesting how we perceive them differently.
1 Like
The difference between 5th and 7th is just right, a nice choice based on personal preferences. I don’t think something in the middle would really add something.
The better modulators all seem to provide a cleaner, faster and more spacious sound. It might be a fun next challenge to aim for a organic and warm sounding modulator for high rate SDM. Although I must say that DSD5v2 256+ at 512X already has a very nice tonal density.
Maybe we are not comparing the same things? Try 44.1k to DSD256/ASDM7ECv2.
Both Gauss-Long and Sinc-Mx are sharper and more dynamic than Sinc-L. But there’s something magical about the Sinc-L filter with 32M taps (in PCM mode it’s a different story : Sinc-L is the sharpest filter).
All my comparisons in DSD are done using DSD256 and ASDM7ECv2. I think I’m starting to understand what you mean about Gauss-Long sounding a little bit sharper, but Sinc-M(x) still sounds very smooth to me and that is ironically the main reason why I’m not a big fan. I love when individual sounds “pop out” of the background and Sinc-L simply does it so much better than M(x). This could be a preference thing, though. Some people seem to love gear that presents everything smoothly perhaps at the cost of not being so technically proficient (i.e. making smaller or bigger sacrifices - apparently Meze headphones are like that) whereas for me it would put me to sleep (although it might be nice for background listening whilst doing something else).
However, having said that, if I’m having a “sinc-only day” (which has been the case recently, actually…) then for recordings with lots of apodising errors and also recordings that for some reason sound sharp in the treble area (not always linked to apodising errors) I will switch to M(x) just to make listening less fatiguing. I guess that if the source material is less-than-ideal then we might need to tweak it a little bit to make it more pleasant to the ear.
For well-recorded tracks, Sinc-L with the above DSD settings is still capable of making my jaw drop. I actually spent a few hours today using these settings and despite knowing my gear really well, it still blew my socks off on quite a few occasions… 
2 Likes
You might give closed-from and poly-hb a try…
You might not like them on your system but the filters you mentioned sound super bloated on mine…
I think I know what you mean by “bloated” as Gauss-Long doesn’t seem to suffer from this, but then it loses some of the punch, vividness and holography (although somehow it usually sounds more natural which is why I spent a long time using it).
Thanks, I’ll give them a try, I’ve not used either of those yet.
2 Likes
2 Likes
@Eminent_One I appreciate that it might be seen as a little off-topic; however, to me it is directly connected to making the most out of what the HQPlayer offers.
If I remember rightly, you used to have a Gustard X26 Pro (which is what I am currently using, in NOS mode) and now have a Holo Audio May. Comparisons between the two when talking about PCM seem pointless - I would suspect that the May would devour the Gustard (although perhaps I’m wrong and the actual difference is not that big…).
I am, however, wondering if there’s any substantial difference when upsampling in DSD, i.e. any difference that is audible and clearly enhances the experience. I am aware that on paper the ESS chip will be held back by its internal processing (whatever internal processing actually takes place for DSD), but, for instance, volume control that the Gus/chip offers is beneficial to me because I like my amplifier in by-pass mode so as to by-pass its volume control (I wouldn’t risk using software volume control only - imagine what a single glitch could do to your ears!).
I don’t know if you did direct DSD comparisons between the two DACs, using the exact the same DSD settings. If you did, it would be great if you could share the results as the price difference is absolutely huge and it makes me wonder if it’s actually reflected in performance when using the devices with the HQPlayer as opposed to on their own (again - not comparing PCM as this would appear to clearly favour the May).
I have Gustard X26 Pro , Dsd better than Pcm in my system .(I hear difference because i use ribbon tweetter) .I use Hq player+Qobuz+library
1 Like
I’ll try my best. I have to preface though that I never had X26P and May at the same time. In between I did have R26 and it did overlap with X26P and May. I only use NOS mode in all, but never listened/tested/preferred NOS itself without HQP.
With that out the way… a bit of context:
I only use LCD5
-X26P/U18 via i2s/Oor/Hypsos all with Purple fuses/Convolution/PEQ (3.3 low shelf)/PCM, 705/768, sinc-MX/LNS15
-R26P/U18 via i2s/Oor/Hypsos all with Purple fuses/Convolution/PEQ (3.0 low shelf)/DSD 256, AMSDM7 512+fs, ext 2
-R26P with Purple fuse/NC400 monoblocks/Freya S/Convolution/PEQ (~2.0 low shelf)/DSD 512, AMSDM7 512+fs, poly-sinc-gauss-hires-lp
-May/NC400/Freya S/Convolution/PEQ (.75 low shelf)/DSD 512, AMSDM7 512+fs, poly-sinc-gauss-hires-lp
-May/Bliss/Convolution/No PEQ/DSD 512, AMSDM7 512+fs, poly-sinc-gauss-hires-lp
-May/Bliss/Convolution/No PEQ/New PC 13900k/DSD 1024, AMSDM7EC 512+fs, poly-sinc-gauss-hires-lp
The biggest difference I can describe off memory for lack of better words are X26P was more 2D, less dynamic, like a tonally pleasing plane of all sounds, excellently balanced tone with eq’d bass shelf to give density and weight to sounds in soundscape.
This isn’t meaning it sounded bad, I did declare at that time it was endgame. It made me feel music for the first time where I experienced frisson (goosebumps, skin orgasms) and a sound so engaging and addictive.
Compared to now, X26P was more etchy from memory, probably from going through stock OS filter along with HQP. It wasn’t bad, just different is the best I can describe.
As for warranting the price, if you’ve heard the effects like expansion in spatial separation going up in rates as in soundscape fullness, then this is even greater when going to DSD1024 and it’s even more addictive and intoxicating with my preferred combo. 1024 gives off a density and pressure similar to adding a bass shelf. What I like about this combo is the vocals and accompanying ensemble become divine with high notes resonating inside me to give those frisson effects as it did in the past but now with all the elements flowing into a musical cascade that gives songs the emotion it tries to express through just instruments and lyrics.
Of course this could be music/gear dependent too. So YMMV.
This may read as non audio related but it is this emotional engagement that I find the most important now. Not just sound hitting my ears or tonalityas that can be adjusted to preference.
Other reasons why I chose to keep May are the remote, less boxes where I lost remote volume on R26 in NOS as it is fixed.
5 Likes
Thank you for such detailed description, @Eminent_One
There is one immediate “issue” that caught my attention before I even read about your listening experiences - it appears that you only ever used your X26Pro in PCM. Was that actually the case? It would be a shame since in my case switching to DSD changes things (for the better) by a really significant margin, so I wouldn’t be surprised if the Gus came up short against your other gear when it was being fed PCM whilst others DSD.
Did you ever try it with DSD? I am aware that even if you did you still wouldn’t have been able to compare it directly to your other gear, but I am somehow certain that a memory of that experience would have stayed with you (I think it’s that good) 