Dirac with DAC having a variety of up-sampling and filtering options

[wgscott]

I just got a new DAC that has a bewildering array of up-sampling and filtering options.

 

cf: http://www.computeraudiophile.com/f6-dac-digital-analog-conversion/i-need-learn-about-conversion-and-digital-filters-28166/

 

Briefly, should I turn up-sampling off, or choose the least invasive filter, or what is the best recommendation for settings when making initial calibration measurements, or should I make a separate set of measurements for each combination of filters and up-sampling regimes, or does it even matter?

[wgscott]

If I understand correctly, I think the least problematic option would be to apply the PCM FIR SHARP filter during measurements:

 

"An FIR filter with a steep roll-off is used to sharply cut signals out- side the audio band."

 

(Naively, I would have thought turning off all filters would be the way to go, but it appears to be potentially the most problematic in terms of introducing noise, delay, aliasing artifacts, roll-off, etc.)

[Fitzcaraldo215]

If I understand correctly, I think the least problematic option would be to apply the PCM FIR SHARP filter during measurements:

 

"An FIR filter with a steep roll-off is used to sharply cut signals out- side the audio band."

 

(Naively, I would have thought turning off all filters would be the way to go, but it appears to be potentially the most problematic in terms of introducing noise, delay, aliasing artifacts, roll-off, etc.)

 

Sometimes devices with a lot of options and settings are a problem rather than problem solver.

 

If it were me, I would turn Dirac off and listen to various filter settings for an evening, trying to narrow it down to the two or three, maybe even four, I liked best with a sampling of music. Then, assuming you hear a noticeable difference, even though small, I would run the Dirac calibration mike sweep for each of those. I would load those, up to 4, into Dirac and listen to each some more. Once I had a favorite, I would just leave everything that way and totally forget about the other settings and calibrations.

 

Yes, mike positioning on the multiple sweeps could alter things even more than the DAC filter setting. So, try to be as precisely repeatable as you can in mike positioning for each sweep. Likely, you will not be perfect, though, in positional repeatability. But, so what? You will still be electing to use the best sounding filter setting + Dirac calibration you have heard. Fortunately, Dirac calibrations do not take very long.

 

I think you need to proceed this way because the DAC filter settings will likely alter top octave frequency and impulse response. So, the sound might be somewhat different with each, and should be recalibrated. But, it is also possible that the sonic differences with different DAC filters will converge and be minimal with Dirac EQ, because of the imposition of its target curve and impulse response overriding the filter settings in sonic significance. Even so, you will have learned something important.

 

Then, I repeat, totally forget about all this microscopic listening you have done, just sit back and enjoy the music, never revisiting the DAC filter settings issue again.

[SteveS1]

Sound advice. I only have two filter options and would have been happier with none. It's a ball ache to do the different filters but once you do, run with the preference and forget the other options. Insanity lies nearby.

 

If I ever change my DAC it would be for one with no options.

[wgscott]

I decided to do the measurements with the steep cutoff FIR filter, as it is the one that creates the least number of potential artifacts within the measurement window. Some of the others, for example, result in roll-off and/or aliasing above 10kHz, so I chose the least invasive of the options (and the one most likely used by the manufacturer if I had a DAC with no options (i.e., one choice).

[Jud]

I decided to do the measurements with the steep cutoff FIR filter, as it is the one that creates the least number of potential artifacts within the measurement window. Some of the others, for example, result in roll-off and/or aliasing above 10kHz, so I chose the least invasive of the options (and the one most likely used by the manufacturer if I had a DAC with no options (i.e., one choice).

 

"Least number of artifacts" depends on what you test for. If it's ringing, a filter without a sharp cut will produce less of those artifacts.

 

You should also look into what Dirac's filters are. Are they minimum phase? Would another minimum phase filter work better with them?

[wgscott]

Thanks, Jud. BTW, I got some mail last night!

 

I'm going by this: Archimago's Musings: MEASUREMENTS: Digital Filters and Impulse Response... (TEAC UD-501) It looks like the slower cutoff of the minimum phase filter produces more aliasing artifacts. As you were pointing out to Bob in the other thread, there is a cost-benefit trade-off for all of these. My idea is to start with the least invasive filter simply for the purpose of measurement, and then proceed from there. If I wind up settling on, say, DSD upsampling and filtering, I can always go back and remeasure under those conditions.

 

How the filters might interact also has me thinking. Dirac is "mixed-phase" rather than minimal phase. I may (for example) be better off having Audirvana upsample/filter the incoming audio to the Dirac maximum (modulo 2, i.e., 176 or 192 kHz), apply the Dirac filter, and then pass that to the DAC with no further up-sampling or perhaps DSD upsampling/filtering.

[Jud]

Thanks, Jud. BTW, I got some mail last night!

 

I'm going by this: Archimago's Musings: MEASUREMENTS: Digital Filters and Impulse Response... (TEAC UD-501) It looks like the slower cutoff of the minimum phase filter produces more aliasing artifacts. As you were pointing out to Bob in the other thread, there is a cost-benefit trade-off for all of these. My idea is to start with the least invasive filter simply for the purpose of measurement, and then proceed from there. If I wind up settling on, say, DSD upsampling and filtering, I can always go back and remeasure under those conditions.

 

How the filters might interact also has me thinking. Dirac is "mixed-phase" rather than minimal phase. I may (for example) be better off having Audirvana upsample/filter the incoming audio to the Dirac maximum (modulo 2, i.e., 176 or 192 kHz), apply the Dirac filter, and then pass that to the DAC with no further up-sampling or perhaps DSD upsampling/filtering.

 

You know enough to repeat what you've read, but you haven't really taken it on board yet. For example, the following is self-contradictory:

 

As you were pointing out to Bob in the other thread, there is a cost-benefit trade-off for all of these. My idea is to start with the least invasive filter simply for the purpose of measurement

 

You can't tell which is the "least invasive filter" without knowing (i.e., measuring) all the applicable tradeoffs. Have a look at SRC Comparisons for iZotope 64-bit SRC, the sample rate conversion software used by Audirvana, and look at the tradeoffs among the various test measurements as the parameters are adjusted from "Steep, No Alias" to "Intermediate Phase." You aren't just adjusting frequency response.

 

Just starting off as you are, playing with this many interacting variables is a recipe for making yourself crazy while not producing the optimal outcome. I would strongly suggest only worrying about Dirac later, once you have determined which internal or external filtering you prefer.

[wgscott]

@Jud:

 

I apologize if I seem to be disregarding your advice. I am not.

 

Playing back redbook bit-perfectly (no Dirac or pre-dac upsampling), I cannot reliably distinguish between filter options. If I turn off all dac upsampling and filtering, I do hear a difference (it sounds worse), but even then it is slight. I should probably do it with headphones. In other words, I could probably live a happy life with any of the filtering options.

 

What I want to do is this:

 

(1) Upsample to 176 or 192 kHz with Audirvana.

(2) Apply Dirac room correction to the 176 or 192 kHz stream

(3) Take the up-sampled, frequency/phase room-corrected audio stream into the DAC and output analogue without destroying the impulse response (phase) correection that Dirac imposed.

 

To do all that, I need to do two things:

 

(1) Make room measurements and create the filters. To do that, I have to play the test signal through the DAC. To make those results meaningful, especially the impulse response correction, I think (I may be wrong here) I need to use a linear phase rather than min phase DAC filter for the measurement, even if I never use it again.

 

(2) Apply filters without undoing the impulse response correction imposed by Dirac, so I need to understand what the various options in the DAC do.

 

Basically, I want to have Audirvana and Dirac do most or all of the processing if possible.

[Fitzcaraldo215]

I think Jud is right. You are overthinking this, dazzled and bewildered by all the adjustments you now have. And, some filtering in the DAC needs to be there one way or the other. You cannot eliminate it, just tailor it to tradeoffs according to a particular "school of thought". They do something totally different from the filters in Dirac, but they may have an unpredictable combined result with Dirac, maybe even an insignificant one. That might be particularly true if you use the standard Dirac target with that top octave rolloff particularly above 15k, which is what I prefer myself.

 

Maybe this is much ado about not much at all sonically, once you finally recalibrate with Dirac. We know Dirac is a good and essential thing in the final analysis.

[wgscott]

Again, my concern is to understand what the filter choices actually do to the impulse response. I still have no idea for the two DSD filters. Any ideas other than telling me not to worry my pretty little head with the effects of discontinuities and inflection points?

[Jud]

dazzled and bewildered by all the adjustments you now have

 

[Jud]

@Jud:

 

I apologize if I seem to be disregarding your advice. I am not.

 

Playing back redbook bit-perfectly (no Dirac or pre-dac upsampling), I cannot reliably distinguish between filter options. If I turn off all dac upsampling and filtering, I do hear a difference (it sounds worse), but even then it is slight. I should probably do it with headphones. In other words, I could probably live a happy life with any of the filtering options.

 

What I want to do is this:

 

(1) Upsample to 176 or 192 kHz with Audirvana.

(2) Apply Dirac room correction to the 176 or 192 kHz stream

(3) Take the up-sampled, frequency/phase room-corrected audio stream into the DAC and output analogue without destroying the impulse response (phase) correection that Dirac imposed.

 

To do all that, I need to do two things:

 

(1) Make room measurements and create the filters. To do that, I have to play the test signal through the DAC. To make those results meaningful, especially the impulse response correction, I think (I may be wrong here) I need to use a linear phase rather than min phase DAC filter for the measurement, even if I never use it again.

 

(2) Apply filters without undoing the impulse response correction imposed by Dirac, so I need to understand what the various options in the DAC do.

 

Basically, I want to have Audirvana and Dirac do most or all of the processing if possible.

 

I'd go about the process of reaching the goal differently, but I am not the boss of you, and more importantly I don't get to say how you have the most fun with your new box. If you eventually do find yourself getting stressed out versus enjoying yourself, then you might want to try variations one at a time rather than in multiples.

 

I'd also gently suggest that somewhere along the line you try the DAC's upsampling off with (1) A+ upsampling to 352.8 or 384KHz, and (2) HQPlayer upsampling to DSD256 with various of its filters and modulators (there's a free trial), and see what you might like. I know HQPlayer does convolution with room correction filters, but I have never tried it and don't know how it's done.

[Jud]

Again, my concern is to understand what the filter choices actually do to the impulse response. I still have no idea for the two DSD filters. Any ideas other than telling me not to worry my pretty little head with the effects of discontinuities and inflection points?

 

You're going to get the best idea what these settings do by playing with the iZotope settings with the DAC's internal filtering and Dirac off, because this will allow you to experiment with what I'd call "pathological" settings, i.e., the difference will be so overwhelmingly audible no one would ever use such a filter in a production DAC. And then the answer to what settings like pre-ringing (minimum to linear phase), steepness, cutoff frequency (sq rt of 2 over 2 point), etc., do is "*That*, but much less of it."

 

Because seriously, we don't even have common audio terminology for a lot of this stuff, like how changes to phase settings sound, or how it sounds when a filter's impulse response "rings." Miska says some filter ringing could "smear" transient response over the time taken by a 1KHz wavelength, i.e., a thousandth of a second. Do you think you are going to consciously notice a thousandth of a second difference in a transient if you listen *really* closely? That's why I'm suggesting not to bother with A/B. If it removes some excitement on a subconscious level, maybe you'll notice it as feeling a little less like tapping your feet or something.

 

But if you have all other filtering off and you make a substantial adjustment to iZotope's pre-ringing (phase) settings I guarantee you will hear it, though we may not have the words to describe the change.

 

Edit: For the two DSD filters, feed the DAC DSD256 if you can, DSD128 at least unless your computer can't manage it, and use the higher (is it 150?) filter. If you feed the DAC DSD64, try the lower (50KHz?) filter.

[wgscott]

OK, I took your advice, and did a semi-subjective test. I used the Q-sound effect in "Perfect Sense" to try to hear differences, first with various up-sampling and filter combinations (or none) in the DAC, using just redbook bit-perfect playback in iTunes (and then in Audirvana, just to be sure). My conclusion from this is that the spatial sound effect is best preserved by either PCM 8X up-sampling with sharp FIR filtering, or DSD up-sampling and filtering (with no audible difference between the 50Hz and 150Hz cuttoff options). Surprisingly, it wasn't that much worse with all the upsampling and filtration turned off. Leaving it off, I then played around with Audirvana, and achieved consistent results. Briefly, only the pre-ringing setting seemed to make much of a difference in terms of focusing the image; 1.0 being the most focused, and 0.0 being the most diffuse. Then using Dirac subsequent to Audirvana up-sampling with the best settings, the DAC settings seemed to make very little difference. I think I convinced myself that either PCM sharp FIR filtering or DSD (further) up-sampling and filtering sounded the best in terms of minimally de-focusing the image, but the differences were so slight there is a good chance that observation is contaminated by expectation bias. One thing that does seem clear (and that I had previously observed) is that the Q-sound effect is actually degraded somewhat by the application of Dirac, but I was able to minimize this simply by doing up-sampling, first, in Audirvana, rather than subsequently in the DAC.

[Jud]

OK, I took your advice, and did a semi-subjective test. I used the Q-sound effect in "Perfect Sense" to try to hear differences, first with various up-sampling and filter combinations (or none) in the DAC, using just redbook bit-perfect playback in iTunes (and then in Audirvana, just to be sure). My conclusion from this is that the spatial sound effect is best preserved by either PCM 8X up-sampling with sharp FIR filtering, or DSD up-sampling and filtering (with no audible difference between the 50Hz and 150Hz cuttoff options). Surprisingly, it wasn't that much worse with all the upsampling and filtration turned off.

 

This is why very gentle, minimal filtering (e.g., Ayre or my homebrew A+ settings) can often work fairly well - a lot of the noise is above a frequency we can hear, and what does come through may just make the sound "warmer."

 

Leaving it off, I then played around with Audirvana, and achieved consistent results. Briefly, only the pre-ringing setting seemed to make much of a difference in terms of focusing the image; 1.0 being the most focused, and 0.0 being the most diffuse. Then using Dirac subsequent to Audirvana up-sampling with the best settings, the DAC settings seemed to make very little difference. I think I convinced myself that either PCM sharp FIR filtering or DSD (further) up-sampling and filtering sounded the best in terms of minimally de-focusing the image, but the differences were so slight there is a good chance that observation is contaminated by expectation bias. One thing that does seem clear (and that I had previously observed) is that the Q-sound effect is actually degraded somewhat by the application of Dirac, but I was able to minimize this simply by doing up-sampling, first, in Audirvana, rather than subsequently in the DAC.

 

Undoubtedly the pre-ringing setting has the greatest effect. If what you heard re Q-sound is real, then it is an interesting demonstration that initial oversampling with a linear phase filter (iZotope settings in A+) before handing off to the intermediate-phase filters used by Dirac helps to preserve the phase effects that are the basis of Q-sound.

 

Edit: By the way, adjusting the cut-off frequency by fairly small amounts (a few tenths) will likely have a pretty audible effect as well. I'm supposing this would be due to the cut actually starting in the audible range and just reaching the "knee" at the cut-off.

[wgscott]

I wouldn't want to go to the mat defending that observation ...

[Jud]

I wouldn't want to go to the mat defending that observation ...

 

Nope, but who needs to. If you think you like it better, you're the one listening.

[wgscott]

Even if it is accurate, it is a very limited-scope test of one phenomenon, that is artificial to begin with. I just find it is easier to make observations using extreme test examples. However, the more extreme the test example, the more of a stretch it is to claim it is reverent to every-day listening.

[Jud]

Even if it is accurate, it is a very limited-scope test of one phenomenon, that is artificial to begin with. I just find it is easier to make observations using extreme test examples. However, the more extreme the test example, the more of a stretch it is to claim it is reverent to every-day listening.

 

Of course. But obviously you cared enough about the Q-Sound (Waters' "Amused to Death"?) to notice Dirac messed it up, so I'm happy it subjectively seems to you to work better.

[wgscott]

It's kind of an interesting "sound-stage" illusion, and it is very easy to lose the image just by tilting your head or positioning yourself wrong, so I think it is fairly sensitive for detecting phase errors.

 

I've also noticed on the new version of Amused to Death (both DSD and converted PCM), I cannot get the illusion of Hal's voice 90° to the left. (The one that works for me best is the live album.)