[SolarCetacean]

Jussi considers poly-sinc-short-mp the ideal filter for music containing strong transients such as pop and rock as far as I know.

In my experience minimum phase filters indeed improve transients but they muddy spatial cues (sensitive to time domain distortion). However most pop music doesn't contain physical space to begin with (half the instruments may even be electronic samples) so not a lot is muddied.

You seem to know your stuff, do you agree that for headphones linear phase is more important since they are linear phase devices with a single driver while for speakers minimum phase is less damaging since they are multi-driver with minimum phase crossover filters and room reflections?
I've heard someone mention that and i've been wondering about it.

I thought headphones were minimum-phase devices?

 

[plumpudding2]

I thought headphones were minimum-phase devices?

Whoops should have done a quick google search before posting, seems I misremembered

 

[camrector]

multi-driver with minimum phase crossover filters and room reflections

To correct for as much minimum phase behavior as possible, I use an analog active crossover with LR slopes for my 3ways and convolution for filters for digital room correction.
I do not even own a pair of headphones

 

[jlaako]

The step response is the response to an infinite bandwidth signal so that doesn't exactly demonstrate the effects in real music i'm afraid.

On of the good ways to see effect of particular band limiting on a single transient. Every band limiting method will affect it in a different way.

I was thinking more along the lines of taking a hires 196khz recording, decimating it to 44.1 khz using a filter similar to an adc decimation filter, and then plotting the results of upsampling it again with poly-sinc-short-mp vs sinc-L. What effects would be observable in a time vs. amplitude plot for example?

I was demonstrating this some 10+ years (or maybe almost 20?) ago with a recording I did myself. It is good to switch Audacity into dB vertical scale for this, then it is closer to how we are hearing things

But his results didn't seem to advocate for using a short minimum phase filter unfortunately.

For a good reason. When you look at step response on logarithmic vertical scale it becomes quite obvious why.

 

[jlaako]

Jussi considers poly-sinc-short-mp the ideal filter for music containing strong transients such as pop and rock as far as I know.

In my experience minimum phase filters indeed improve transients but they muddy spatial cues (sensitive to time domain distortion). However most pop music doesn't contain physical space to begin with (half the instruments may even be electronic samples) so not a lot is muddied.

This is also reflected in the manual / help page table genre suggestions...

 

[bogi]

You seem to know your stuff, do you agree that for headphones linear phase is more important since they are linear phase devices with a single driver while for speakers minimum phase is less damaging since they are multi-driver with minimum phase crossover filters and room reflections?
I've heard someone mention that and i've been wondering about it.

As usual, there are many opinions on the same thing since our listening experience is influenced by many variables.
It simply does not correlate with my experience that minimum phase filters would be insuitable for listening on headphones.
camrector has his reasons to prefer linear phase, he argued with room eq correction and I don't oppose, since I am oriented to headphone listening.
With my current SMSL D300 DAC I like minimum phase filters more than with my previous ESS based DAC, which I also fed by DSD signal from HQPlayer. I don't know why with D300 (ROHM chip) in direct DSD mode the old poly-sinc-mp filter sounds so well.

In my experience minimum phase filters indeed improve transients but they muddy spatial cues (sensitive to time domain distortion). However most pop music doesn't contain physical space to begin with (half the instruments may even be electronic samples) so not a lot is muddied.

I agree with this. Filter should be chosen according to content. For acoustic recordings containing real space information I wouldn't use short filters and I would prefer linear phase filters. But there are "modern" pop recordings missing any natural space which I cannot enjoy with longer filters when they contain fast transients. If short filters wouldn't exist I would say "bad recording". But short filters can change the situation and allow me to enjoy them. Other content like older recordings from analog era can fall somewhere inbetween. Generally I like middle length filters, both minimum phase (poly-sinc-mp) and linear phase (poly-sinc-gauss, halfband filters). They are IMO very universal.

Minimum vs linear phase is for me rather about genre. "Fast" genres - minimum phase, classical and similar - linear phase. This may be related to Jussi's example which showed how our brain create much longer post-masking than pre-masking. That's the explanation for me why short minimum phase filters work so well to eliminate ringing. No pre-ringing and masked post-ringing.

 

[MrWalkman]

Since HQPlayer is available for MacOS devices (ARM?), could there be an Android version containing at least certain filters in the future?

Maybe a collaboration with UAPP developers, to get good support for as many USB DAC as possible?

UAPP also has Tidal and Qobuz support.

 

[1laraz]

.

Dear Jussi, could you please start adding some offload to the Intel integrated graphics like Iris Xe or UHD? I believe there is a substantial fraction of HQPlayer users with integrated graphics like NUCs, laptops etc. Allowing this dormant computational power to uplift the performance will be certainly welcomed and increase the popularity of HQPlayer!

 

[plumpudding2]

Thanks for responding!
The step response is the response to an infinite bandwidth signal so that doesn't exactly demonstrate the effects in real music i'm afraid.

I was thinking more along the lines of taking a hires 196khz recording, decimating it to 44.1 khz using a filter similar to an adc decimation filter, and then plotting the results of upsampling it again with poly-sinc-short-mp vs sinc-L. What effects would be observable in a time vs. amplitude plot for example?

I know Archimago did a similar experiment here;
https://archimago.blogspot.com/2018/01/audiophile-myth-260-detestable-digital.html

But his results didn't seem to advocate for using a short minimum phase filter unfortunately.

Replying to myself here, anyone that was also interested in this Jussi already did almost the same exercise over on the Audiophilestyle forums, see here:
https://audiophilestyle.com/forums/...om-mqa-thread/?do=findComment&comment=1258061

(You should create a FAQ on your website/the HQP manual so that you don't have to keep explaining these same things over and over again on different forums Jussi )

 

[ageofadz]

One question. I have 9900K and RTX 2080Ti (I have tried with CUDA offload on and off), when I do PCM 32-bit 768k I get artifacts but not with 32-bit 384k. Also I don't get artifacts with 24-bit 768k. These artifacts sound like just bits crushing. I like the sound of 32-bit 384k more but I don't know if its placebo, but I thought my 9900K would be enough to handle it.

Also is turning CUDA offload good or better to let the CPU handle all, I feel no difference in performance with both options.

Thanks!

 

[taipan254]

Number of equally weighted conversion elements corresponding to so called 6bit sigma delta DAC is 2^6 = 64. That "6bit" is only a convention to tell that there are 64 elements. It does not tell about modulator output.


Modulator output can be 1bit and serial to parallel conversion can be used to feed for example 64 conversion elements out of modulator output. That is happening at modulator rate (FIFO buffer principle). Search on web for Signalyst DSC1 open source DAC schematics. It does it this way too.

According to available sources of information camrector is true about ESS DACs architecture.

Thanks, @bogi , for your input.

Considering the moderator's comments in the A26 thread, it seems more fitting to discuss ESS DACs' conversion methodology here in the HQP thread, especially since many utilize HQPlayer to upsample with ESS DACs. @jlaako , please let me know if you think this isn't the right place for this conversation or if you have any corrections regarding our discussion. I'm eager to delve deeper into the workings of ESS DACs with upsampled PCM signals and would appreciate your insights.

Regarding the conversion elements, your points are well-taken. Given that discussion, here's an updated summary of my understanding of how ESS DACs process PCM inputs. @bogi could you highlight any discrepancies?

Big Picture: ESS DACs convert PCM signals into a multi-level bitstream, not directly into DSD. This type of bitstream modulation outputs more than 1-bit per sample and is distinctly different from traditional DSD, which strictly uses a 1-bit format.

Intermediate Format: ESS DACs use advanced Delta-Sigma modulation to convert PCM to a multi-level signal, rather than a simple multi-bit or 1-bit output:

• Higher Order Modulation: The modulator in ESS DACs is capable of generating a signal with more levels than the typical 1-bit used in traditional DSD, avoiding simple multi-bit digital signal conversion.
• Multi-Bit Quantization: Unlike simple R-2R DACs, ESS DACs produce a high-frequency PWM signal with multiple levels, offering a more sophisticated output than standard multi-bit or 1-bit DACs.
• Noise Shaping and Filtering: The multi-level output enables more efficient noise shaping and filtering, pushing noise out of the audible spectrum and significantly lowering distortion while enhancing dynamic range.

DAC Architecture: ESS uses proprietary technology combining their advanced modulator with a current steering DAC design, which differs significantly from a basic "six-bit ladder" approach. This system manages the digital-to-analog conversion with high precision and linearity.

Additionally, could you clarify your reference to the DSC1 in relation to ESS's PCM to DSD conversion process? I'm unsure how it applies here.

 

[camrector]

Thanks, @bogi , for your input.

Considering the moderator's comments in the A26 thread, it seems more fitting to discuss ESS DACs' conversion methodology here in the HQP thread, especially since many utilize HQPlayer to upsample with ESS DACs. @jlaako , please let me know if you think this isn't the right place for this conversation or if you have any corrections regarding our discussion. I'm eager to delve deeper into the workings of ESS DACs with upsampled PCM signals and would appreciate your insights.

Regarding the conversion elements, your points are well-taken. Given that discussion, here's an updated summary of my understanding of how ESS DACs process PCM inputs. @bogi could you highlight any discrepancies?

Big Picture: ESS DACs convert PCM signals into a multi-level bitstream, not directly into DSD. This type of bitstream modulation outputs more than 1-bit per sample and is distinctly different from traditional DSD, which strictly uses a 1-bit format.

Intermediate Format: ESS DACs use advanced Delta-Sigma modulation to convert PCM to a multi-level signal, rather than a simple multi-bit or 1-bit output:

• Higher Order Modulation: The modulator in ESS DACs is capable of generating a signal with more levels than the typical 1-bit used in traditional DSD, avoiding simple multi-bit digital signal conversion.
• Multi-Bit Quantization: Unlike simple R-2R DACs, ESS DACs produce a high-frequency PWM signal with multiple levels, offering a more sophisticated output than standard multi-bit or 1-bit DACs.
• Noise Shaping and Filtering: The multi-level output enables more efficient noise shaping and filtering, pushing noise out of the audible spectrum and significantly lowering distortion while enhancing dynamic range.

DAC Architecture: ESS uses proprietary technology combining their advanced modulator with a current steering DAC design, which differs significantly from a basic "six-bit ladder" approach. This system manages the digital-to-analog conversion with high precision and linearity.

Additionally, could you clarify your reference to the DSC1 in relation to ESS's PCM to DSD conversion process? I'm unsure how it applies here.

PCM enters your dac. It gets processed by 8x FIR, then the sample is copied 8 times and is fed through 3rd order IIR before being copied again 8 times and entering the modulator which produces a 1-bit DSD stream. This is converted to 32 (2^5) or 64 (2^6) bits at a time and sent to analog through those 32/64 conversion elements.
You are listening to n multi bit SDM (dsd)
DSD with your ess dac chip the signal takes a different shorter path. It goes through the 3rd order IIR without any rate conversion and then goes through the modulator to the conversion elements.


Not sure what else you want to know.
Start a new thread about ESS architecture.

 

[Ghoostknight]

It simply does not correlate with my experience that minimum phase filters would be insuitable for listening on headphones.
camrector has his reasons to prefer linear phase, he argued with room eq correction and I don't oppose, since I am oriented to headphone listening.
With my current SMSL D300 DAC I like minimum phase filters more than with my previous ESS based DAC, which I also fed by DSD signal from HQPlayer. I don't know why with D300 (ROHM chip) in direct DSD mode the old poly-sinc-mp filter sounds so well.

unsuitable is definitely the wrong word here, im also a FIR advocate but i also have to say, the difference is way less obvious with headphones, not sure what about speakers make it that different but also absolute phase is way more audible on speakers, atleast on mine, compared to DT880 headphones

tho on speakers the thing is clear for me, go with FIR EQ, it sounds way more natural and precise, depending on your eq you also notice that bass is "boosted" with FIR vs IIR because the FIR filter makes the bass notes more phase-perfect (even tho most speakers have their own phaseshift flaws, its still audible if you make things even worse with a IIR eq)

but also recently going with a overall FIR EQ with a IIR highpass, i have to say it depends! If you use FIR EQ, dont do some extreme filtering or the preringing actually makes things worse than the phaseshift itself...
since i use the Black Lion Audio 2x2 Revolution interface (which is also PCM only) i was able to actually remove the highpass again, somehow the aune x8 made subsonics really audible (or it was the burson V6 opamp) (overall balance also seem to be bass heavy in comparison)

 

[kennyb123]

seems more fitting to discuss ESS DACs' conversion methodology here in the HQP thread, especially since many utilize HQPlayer to upsample with ESS DACs

I’m glad to see you raising this here. It was said that the ESS chip will “undo” the benefits of upscaling. I would be interested to hear from a knowledgeable person as to whether this was a valid assertion.

 

[kennyb123]

I hear benefits from HQPlayer with my ESS DAC - so much so that I recently upgraded to HQPlayer 5. If it was a waste for me to spend that money then I want to hear why so I can take that into account next time I need to pay an upgrade free. I can’t see why this wouldn’t be the thread to ask these questions. I don’t get why we’re being bullied into not exploring this line of questioning. (Rhetorical remark as I do understand why.)

I honestly thought that it was absolutely worth the dollars to first buy a v4 embedded license as the PCM filters in HQPlayer are superior to those Gustard provides. When time came to upgrade to v5, I did so without hesitation once I heard the an even greater level of improvement. I find it ironic fans of HQPlayer are trying to convince people into thinking one would need a non-ESS DAC to appreciate the differences. I can’t imagine that Jussi would appreciate fewer licenses being sold.