Upsampling vs. Room Correction

[wgscott]

Although I have mostly redbook files, I also have some "higher resolution" files with authentic content beyond 22kHz.

 

My playback system is a Mac mini connected to a DAC (Teac NT503) via USB. I typically use Dirac Room Correction software, as a plug-in with Audirvana (playback software that allows the use of AU plugins as well as optional up-sampling and filtering).  The Dirac AU plug-in can be run in such a way that it is the last to process the signal before being out-put to the DAC.

 

The room correction target curves (orange) I have generated always have a very steep drop-off around 17 kHz.

 

 

 

Does this sharp cut-off behave in essentially the same way that a (higher-frequency) redbook brick wall filter would behave, and if so, is it therefore more likely to produce audible aliasing artifacts?  Would up-sampling (within the Audirvana playback software), prior to application of this filter, make this worse, or ameliorate aliasing artifacts? Should the window for the filter be adjusted so it only goes up to (say) 15kHz? Note that the actual signal above 10kHz is greatly attenuated to begin with (presumably my speakers and/or Class D amp are at fault). I can hear (barely) to about 15-16kHz, fwiw; certainly no higher.  I am far more worried about the audible range than anything else.

[vl]

2 hours ago, wgscott said:

Although I have mostly redbook files, I also have some "higher resolution" files with authentic content beyond 22kHz.

 

My playback system is a Mac mini connected to a DAC (Teac NT503) via USB. I typically use Dirac Room Correction software, as a plug-in with Audirvana (playback software that allows the use of AU plugins as well as optional up-sampling and filtering).  The Dirac AU plug-in can be run in such a way that it is the last to process the signal before being out-put to the DAC.

 

The room correction target curves (orange) I have generated always have a very steep drop-off around 17 kHz.

 

 

 

Does this sharp cut-off behave in essentially the same way that a (higher-frequency) redbook brick wall filter would behave, and if so, is it therefore more likely to produce audible aliasing artifacts?  Would up-sampling (within the Audirvana playback software), prior to application of this filter, make this worse, or ameliorate aliasing artifacts? Should the window for the filter be adjusted so it only goes up to (say) 15kHz? Note that the actual signal above 10kHz is greatly attenuated to begin with (presumably my speakers and/or Class D amp are at fault). I can hear (barely) to about 15-16kHz, fwiw; certainly no higher.  I am far more worried about the audible range than anything else.

 

I am a Dirac Live user implemented on miniDSP and the results are excellent.  The most important gain is the optimization of the impulse response of the system.

 

The left most point and the right most point of the target curve are the amplitude response points of your system before correction.  They are not user movable.  

 

I find for my system the far field amplitude response (multi point measurements) as measured by Dirac Live has the same high frequency roll off as yours.  If I choose a flat response to 17 KHz, the 17 KHz point and the right most point defines the resulting slope.  The near field measurement (1 meter, single point measurement) was much flatter.  

 

I ended up with two target curves.  One is like yours, down about 2 dB at 17 KHz.  The other one follows the general response of the speakers, so the roll off is gentler.  I find the latter to have a more realistic spectral balance compared to my favorite local concert hall.  With one measurement set you can customize four target curves to suit your taste.  

 

My music source is 95% ripped Red Book CDs.

[Jud]

I don't know the answers to the filter convolution questions.  (That's a topic I'm curious about too.)

 

I'm just commenting in answer to your first question that I believe a sharp cutoff would produce ringing at the cutoff frequency rather than aliasing or imaging above and below it. 

[wgscott]

Thanks.  Those are interesting suggestions. 

 

My listening/measuring position is about 16 ft from the speakers, hence the HF roll-off (augmented, no doubt, by my equipment).  You are right -- near-field is much flatter.

[adamdea]

18 hours ago, wgscott said:

Although I have mostly redbook files, I also have some "higher resolution" files with authentic content beyond 22kHz.

 

My playback system is a Mac mini connected to a DAC (Teac NT503) via USB. I typically use Dirac Room Correction software, as a plug-in with Audirvana (playback software that allows the use of AU plugins as well as optional up-sampling and filtering).  The Dirac AU plug-in can be run in such a way that it is the last to process the signal before being out-put to the DAC.

 

The room correction target curves (orange) I have generated always have a very steep drop-off around 17 kHz.

 

 

 

Does this sharp cut-off behave in essentially the same way that a (higher-frequency) redbook brick wall filter would behave, and if so, is it therefore more likely to produce audible aliasing artifacts?  Would up-sampling (within the Audirvana playback software), prior to application of this filter, make this worse, or ameliorate aliasing artifacts? Should the window for the filter be adjusted so it only goes up to (say) 15kHz? Note that the actual signal above 10kHz is greatly attenuated to begin with (presumably my speakers and/or Class D amp are at fault). I can hear (barely) to about 15-16kHz, fwiw; certainly no higher.  I am far more worried about the audible range than anything else.

I use DRC but not dirac. But here's my best effort.

I'm not sure I understand the question, even allowing for aliasing artefacts to mean something like filter ringing .

Surely the target curve is gernated by you- why are you putting in that dip at the end?

Then there is question of what filter is actually being implemented in order to generate that target curve. This is not clear. It's down to dirac

Leaving that to one side, if you have a sharp filter then there is the possibility of ringing, but I suspec ttha the filters implemented will be minimum phase and in any event are not really as sharp as a decent anti aliasing fitler because -15dB of attenuation is not as dramatic as a really good anti alising filter.

In any event I can't see why you are doing this.

[wgscott]

16 hours ago, vl said:

 

The left most point and the right most point of the target curve are the amplitude response points of your system before correction.  They are not user movable.  

 

 

 

 

27 minutes ago, adamdea said:

In any event I can't see why you are doing this.

 

 

I am not.  See quote above.

 

27 minutes ago, adamdea said:

I suspec ttha the filters implemented will be minimum phase

 

They are not.

[audiventory]

19 hours ago, wgscott said:

is it therefore more likely to produce audible aliasing artifacts?

 

In my opinion:

1. artefact are not desirable,
2. ringing is not big matter for music signal,
3. goal #1 is most important.

As rule, music signal hasn't sharp transients between neighbour samples. So, ringing happen rare.

Also, I never see serious ringing to ear impact researches.

But to reduce objective distortions, I'd recommend to reduce ringing maximally, as it possible, after optimal alias suppressing.

 

If we try to reduce ringing, we may:

1. reduce band in higher frequencies,
2. occupy band above half of sample rate.

When both these things are happens in ultrasound range (above 20 kHz), it is not so scary.

 

[adamdea]

45 minutes ago, wgscott said:

 

 

 

 

 

I am not.  See quote above.

 

 

They are not.

 Actually you are somewhat responsible for the sudden downturn at the end by specifying a 17KHz response which is in fact way higher than the actual (pre-correction) response, leaving a big gap between 17 kHz target and 20kHz reality which is filled in by the software. Anyway the real issue what Dirac in fact does in response. On the face of it you seem to be asking dirac to boost the region 10-17kHz and I would have thought that the curve would be better off sloping down a bit more . I'm not sure what dirac does with the rightmost portion of the target-perhaps it ignores it since it seems to be a bit of an unintentional artefact.

In any event, since the end point at 20kHz is apparently the pre-correction response, I can see no reason why it would be brick walling at 20kHz which seems to resolve your initial query.

[PeterSt]

On 9/27/2018 at 4:43 PM, adamdea said:

In any event, since the end point at 20kHz is apparently the pre-correction response, I can see no reason why it would be brick walling at 20kHz which seems to resolve your initial query.

 

On 9/26/2018 at 9:13 PM, wgscott said:

I also have some "higher resolution" files with authentic content beyond 22kHz.

 

Bill, unless things changed, Dirac does not operate beyond 20KHz. Thus, you can feed it with Hires material all right, but it will always "cut off" as per your suggestion. Whether that is at 17KHz I don't know (you could imply this yourself somewhere as the stop band (width) or something - I have no real experience with Dirac).

[Bob Stern]

You'd have to ask Dirac whether the frequency response correction they implement actually follows your target curve all the way to 20 KHz.

 

Did you choose the 17 KHz inflection point, or is it mandated by Dirac?  Certainly a steep cutoff between 17 KHz and 20 KHz would create ringing at 17 KHz, so you should delete the 17 KHz inflection point if Dirac allows you to do so.

 

The ideal target curve according to most room correction experts is a constant slope (on a logarithmic frequency axis) from some low frequency inflection point to 20 KHz.  Opinions differ widely on the inflection frequency, ranging from 20 Hz to 2000 Hz.  The attenuation at 20 KHz should be in the range of 6 to 10 dB.

[Wilderness]

I have the TEAC NT 503 DAC, too, and I am very pleased with it.  I tried all of the DAC's various filters and options.  The best I have tried is the DSD option, which converts PCM digital files I send to it to 11.2 or 12.2 MHz DSD.  Then with the DSD option I also selected the 50kHz cut-off filter; this combo makes the digital files sound a bit like vinyl.

 

I have a mix of Apple 256, CD rips, and high resolution files.  The only files that don't sound fully fleshed out are some classic rock, but they are enjoyable, with the limitation being my small speakers more so than the DAC.

 

My room is fairly neutral.

 

My speakers are powered with a class AB amp.

 

I have two 12 inch subs.

 

My listening position is 9.5 feet from the speakers.

 

I am sending the digital files from my iMac and JRiver without any upsampling or digital processing.  A Wyred 4 Sound Remedy reclocker converts all of the digital files to 96kHz. before they get to the DAC.

 

I use a Wyred 4 Sound uLINK to convert USB to optical.

 

I have never tried room correction software.

 

I understand the limitations of my speakers, but I feel that most of what I listen to now sounds pleasing, much better than with the low budget DAC I had been using.  When I get better speakers I will still use the TEAC DAC.

 

Have you tried the DSD option with the 50kHz. filter?  When I first tried it, I thought it was so rich and full that it was too much of a good thing to the point that it even sounded distorted.  Then after five hours or so the combo settled in and I haven't felt the need to fiddle with any of the other options.