Impulse/phase response of Thinkpad? [solved, thanks NTK!]

[ad8e]

I'm doing research on proximity in reverbs, where phase linearity happens to be important. I have an AudioQuest DragonFly Black 1.5, which is minimum phase and cannot be used for this purpose. But I also have onboard audio through my Thinkpad T530. Is there a way I can know whether this onboard audio is linear phase without an oscilloscope, or does someone have measurements handy? I believe most Lenovo Thinkpads use the same audio chain, from Realtek.

It also specifies a "HD Audio, Realtek® ALC3202 codec"; could such a feature introduce measurement errors, such as being non-Linear Time-Invariant?

The reason I ask is that I will be switching from USB-A to USB-C ports in the next year, and I could not find a decent low-cost DAC/headphone amplifier that would be compatible with both USB types. So I am trying to make do with the equipment I have.

 

[Blumlein 88]

You could determine this if you have a way to measure output with an ADC. Use the ADC to sample at a higher sample rate than the output of the Thinkpad. Feed it -4 db white noise. You'll get a mostly flat line that drops off at or above 20 khz. This will show you the response of the filter and give you some indication of what type it is.

On going to USB C, you might well get by just fine with one of those adapters that let you connect USB 2 or 3 to type C. I've used them on a few different items and never had them fail to work. They are something like 5 bucks. If you've purchased a phone in the last 4 years it may well have had one in the box you can use to try it out. For instance I'm typing this on a Macbook with Thunderbolt ports which are also compatible with type C. I'm using a gaming mouse into an adapter that came with my Pixel phone. Works just dandy as does a recording interface I have from Antelope audio.

Here from a review by Amir is a graph of using white noise to look at filter response. This is showing multiple filters which are switchable on the Sabaj A 10d.

 

[ad8e]

I don't have an ADC that would function for this purpose.

Frequency response graphs will not reveal whether the filter is minimum or linear phase, though I do know my Thinkpad's frequency response, thanks to a paper by Michael Goetz.

I tried googling whether USB-C DACs could work on USB-A ports with a USB converter, and couldn't find anyone claiming so, so I doubt it. There is a good chance USB-A DACs would work with USB-C ports + converter, as you suggest, but I'd rather not buy a USB-A DAC.

I really like the F1 filter you show. None of the other DACs (<$120) I was considering have such a good impulse response. The Sabaj Da3 even looks dangerous, being flat all the way to 48000 Hz. A 20 Hz signal would alias to a 44120 Hz signal, causing a 67 dB boost in energy and probably hearing damage (is that right?).

 

[NTK]

Here is the IR from my Thinkpad P1 Gen 3, headphone output. Thus linear phase in my case.

 

[Blumlein 88]

You can also do an impulse response again if you had an ADC. Those also are not definitive, but that combined with the FR lets you get an idea. Minimum phase filters will roll off early and a bit slower usually combined with a different impulse response shape to a Dirac pulse.

Here is a linear phase filter. Courtesy of Stereophile.

https://www.stereophile.com/content/lumin-p1-streaming-da-preamplifier-measurements

They more commonly look like this for linear phase.

Minimum phase will look like this.

 

[dc655321]

The Sabaj Da3 even looks dangerous, being flat all the way to 48000 Hz. A 20 Hz signal would alias to a 44120 Hz signal, causing a 67 dB boost in energy and probably hearing damage (is that right?).

I'm not sure any of that is right .
c.f. here.

 

[ad8e]

I'm not sure any of that is right .
c.f. here.

I made a post in that thread, but I don't see yet what is wrong about it. The original review post you linked does not display the frequency response above Nyquist, only commenting that it is "down just 0.2 dB at 20 kHz". (Perhaps we can continue any discussion of the Sabaj Da3 in its thread, at https://www.audiosciencereview.com/...d-to-dragonfly-black.3503/page-8#post-1230910)

 

[Blumlein 88]

I made a post in that thread, but I don't see yet what is wrong about it. The original review post you linked does not display the frequency response above Nyquist, only commenting that it is "down just 0.2 dB at 20 kHz". (Perhaps we can continue any discussion of the Sabaj Da3 in its thread, at https://www.audiosciencereview.com/...d-to-dragonfly-black.3503/page-8#post-1230910)

First off DACs have imaging not aliasing. And your idea about a power boost aren't correct. The level if it imaged at 44,080 would be the same level except the anti-imaging filter will have reduced it a lot. According to the chart above from the measurements by Amir of different filters it would image at 44,080 hz and all the filters would have reduced the level by 80 db or more.

 

[ad8e]

First off DACs have imaging not aliasing. And your idea about a power boost aren't correct. The level if it imaged at 44,080 would be the same level except the anti-imaging filter will have reduced it a lot. According to the chart above from the measurements by Amir of different filters it would image at 44,080 hz and all the filters would have reduced the level by 80 db or more.

I understand what you mean by imaging and I will use your terminology here. Maybe my word "aliasing" is unusual in consumer audio.

Note that the chart you are referring to is of a different product, the Sabaj A 10d, all of whose filters I agree are good enough to avoid the effect I am talking about. The product I am discussing, the Sabaj Da3, has a different anti-imaging filter whose amplitude response is nearly flat all the way to 48000 Hz.

I disagree with you that the energy of an imaged signal is the same as the original signal, and I believe this to be a basic fact of DSP. To make this accessible, consider this analogy: a 2 Hz sample rate. We feed it a 1 Hz sine wave: +1 and -1, alternating. We output this with a typical anti-imaging filter with an approximate 20 kHz sinc. One of the images is a 10001 Hz sine tone at maximum volume (+1 and -1). This is extremely loud, even though the original 1 Hz sine wave is inaudible. This demonstrates that the imaged signal gets a significant power boost.

If the math is too hard, then just listen to the attached wav file of two sines, one at 50 Hz and one at 10000 Hz. They have the same height, but the 10 kHz is much louder. You can call it loudness curves if you like, but it comes from transferred power.

Your 22070 Hz is a mistake, but I cannot correct this error because I do not know where it comes from. I can only suggest that using the z-domain unit circle, aliases flip along the x-axis, or travel a full revolution around the circle, but 22070 Hz is a 180 degree rotation which is not imaged/aliased. So 20 can alias with -20 (= 44080) or 44100+20, but not Nyquist+20. You figured it out without me, excellent. (But the other person hasn't yet, so I'll leave this here.)

 

[Blumlein 88]

I understand what you mean by imaging and I will use your terminology here. Maybe my word "aliasing" is unusual in consumer audio.

Note that the chart you are referring to is of a different product, the Sabaj A 10d, all of whose filters I agree are good enough to avoid the effect I am talking about. The product I am discussing, the Sabaj Da3, has a different anti-imaging filter whose amplitude response is nearly flat all the way to 48000 Hz.

I disagree with you that the energy of an imaged signal is the same as the original signal, and I believe this to be a basic fact of DSP. To make this accessible, consider this analogy: a 2 Hz sample rate. We feed it a 1 Hz sine wave: +1 and -1, alternating. We output this with a typical anti-imaging filter with an approximate 20 kHz sinc. One of the images is a 10001 Hz sine tone at maximum volume (+1 and -1). This is extremely loud, even though the original 1 Hz sine wave is inaudible. This demonstrates that the imaged signal gets a significant power boost.

If the math is too hard, then just listen to the attached wav file of two sines, one at 50 Hz and one at 10000 Hz. They have the same height, but the 10 kHz is much louder. You can call it loudness curves if you like, but it comes from transferred power.

Your 22070 Hz is a mistake, but I cannot correct this error because I do not know where it comes from. I can only suggest that using the z-domain unit circle, aliases flip along the x-axis, or travel a full revolution around the circle, but 22070 Hz is a 180 degree rotation which is not imaged/aliased. So 20 can alias with -20 (= 44080) or 44100+20, but not Nyquist+20. You figured it out without me, excellent. (But the other person hasn't yet, so I'll leave this here.)

You appear not to realize that higher frequencies also spend less time on each wave of the signal. So at the same voltages power is independent of frequency. How loud it sounds is a whole different topic not having a bearing on power.

You are basing the idea the DA3 is flat to 48 khz without the sample rate of the graph you showed. Most likely it was 96 khz and the DA3 has some sort of filter. It would be flat to around 48 khz at a 96 khz sample rate. Without knowing for certain that is the most reasonable thing to expect because Sabaj makes pretty good gear and I don't see them making something without an anti-imaging filter. Also aliasing is in ADCs and imaging is in DAC's. It isn't a matter of consumer terms differing. It is far too common in consumer publications to talk about aliasing in DAC's, but it isn't a thing actually. I'm not really being too pedantic I don't think just pointing out a common incorrectly used term.