Hi Jussi, thanks for the long response. I find it awesome that you’re willing to educate the users of your product online!
It absolutely did not occur to me that the ringing is introduced at the sampling stage, not at the reconstruction stage. I figured that if you have music that does not contain any frequencies above 20khz, you would have an appropriately band-limited signal that satisfied the Nyquist sampling theorem.
A closer inspection of Aliasing - Wikipedia clarifies that this is not the case and a proper bandlimited signal is only obtained after taking an infinite amount of samples. This makes it much more plausible that a filter that is so extremely close to the Nyquist filter still produces time-domain errors.
Yes, because RedBook itself is already a compromise. It needs band-limiting and a lot of content is hitting this band limiting. You need to choose what kind of compromise is best for a particular case. And then quality of implementation defines how good the end result is within the boundaries set by the compromise.
Having perfect time and frequency domain performance of band-limited content at the same time is mathematically impossible due to inverse relationship. My biggest efforts have gone into designing filter algorithms that can get as close to the impossible as possible, while at the same time having many other aspects of the filter performance as good as possible. Length wise, this is not super short nor super long. But length is just one of the many aspects of filter, there are many many other properties that matter.
Since different people are sensitive to different aspects of the sound and listen different kind of content, there’s no “one size fits all” filter. Instead it is always combination of personal preference and type of content.
From technical point of view, for example lot of DAC chips have stop-band attenuation of 100 dB or less, giving around 16 - 17 bits reconstruction accuracy, and about 0.01 dB pass-band ripple. While some of the HQPlayer filters have stop-band attenuation of over 300 dB, giving around 50 bits reconstruction accuracy, and less than 0.000000001 dB pass-band ripple.
But you have alternatives. Using DSD provides you a sampling system that doesn’t require band-limiting the source signal. Also using high enough sampling rate for PCM allows you to avoid band-limiting the source signal. You still have other challenges creating perfect reconstruction at the D/A stage, but you don’t need to worry about the filter ringing aspect anymore, so you get one big win with hires already. (for PCM a lot of content requires at least 192k sampling rate to avoid band limiting, and some even more, but 352.8k aka DXD is safe in this respect)
I use “classical” architecture for digital chain:
ROCK (10i7 NUC ) → HQPe Server → NAA (8i7 NUC) → DAC
Here is my detailed write-up on HQPlayer Embedded 12900K build:
As for the sound of it, ofcourse it is hugely subjective, but what I perceive is enlarged sound stage at DSD1024 rates and noticeable change (to the better I suppose) in imaging inside this sound stage. 7ECv2 modulator adds high degree of timbre realism and “body”. In this regard 5ECv2 sound similarly attractive but “thinner” on my system.
Currently with 7ECv2 I have only been successful with DSD1024, no x48, But I’m very happy as I can run most of my favorite “go-to” filters. 1x: short-mp / xla / ext3 and Nx: short-mp / gauss-hires-lp/mp
One downside for poly-sinc filters is that they have noticable initialisation times at DSD1024. Like short-mp for 1x: takes about 3-4 minutes at initialisation. Anecdotically - I tried to play long-lp filter and the source was 96/24. I had my shower and had my lunch and 1h later restarted the PC it was still initialising…
You can use also ASDM7ECv2 with ESS if it sounds better. ASDM5ECv2 is just a recommendation based on the ESS DAC chip properties and typical DAC design around ESS chips.
In some systems, ASDM7ECv2 may perform better still.
I have a fairly basic question:
How do all the HQP filters work in general?
My intention here is that since the music data is somewhat fixed, adding something seems hard to me.
Is it possible to do something because the app does the oversampling and perhaps duplicates the data so that you have the extra data to add the filter’s data?
Let me know if you can come up with a simple explanation to a lay guy.
I really enjoy it the HQPlayer app, as I don’t need anymore to go to my computer and change the filters there. Can do on my phone very easy and convenient.
I would suggest if possible to be able to have option to select different output devices. Like if I have 2 NAA, I can switch HQPlayer DAC instead of do it on computer. Many thanks
I’m using it with jazzy music, acoustic guitars and some pop.
I’m very pleased for the DSD256, although my ifi NEO can do DSD512, my Mac Mini M1 can’t do the same filters 7ECv2, only DSD5v2, but my ears still like more DSD256 with 7ECv2.
I did glimpse at a NUC i9 to use as HQPlayer only to squeeze the filters limits of DSD512 . But not yet!!
I have not been with xtr filters for a long while, but it is surely interesting to come back. I cant say they are better than my “ordinary” set, clearly different.
I can see the processing power though, some cores are in constant 90s. And probably GPU would’ve come helpful here, set up hicks up time to time…
PC: NUC is a nice platform but equally is fairly limited in performance. You already can see that stepping up from 256 to 512 is not necessarily always good, because the modulator setting matters a lot. I dont really know what can be achieved with 9th gen NUC, thought.
EDIT: Just tried “ordinary” short-mp filter on the same source. Yep, now I remember why! xtr at first glance gives a feeling of full body but than slightly artificial. short-mp imaging seems to me noticeably cleaner. but, hey, these are where our personal preferences kick in…
will be the best outcome.
and change the filters there. Can do on my phone 
very easy and convenient.
to use as HQPlayer only to squeeze the filters limits of DSD512 
. But not yet!!
.
