Measurements of MiniDSP 2x4HD

[March Audio]

I will fill comments later when I have time but here are the raw measurements of the DAC output

0dBFS - 1.88V RMS

-6dBFS

-60dBFS

Noise Floor

-90.31 24bit waveform

Jitter

IMD SMPTE

Image Filter

Linearity - WTF???? Need to double check measurement

 

[March Audio]

OK, looks like it picked up on a Thesycon XMOS driver that was installed for another dac. Thats been removed and I have updated the 2x4HD to the latest firmware, drivers and software. Things look different!

0dBFS - Harmonic distortion is so-so.

-6dBFS

-60dBFS - High noise floor

-90.31 dBFS 24bit waveform - Noisy

Jitter - definitely a few things going on here, even if its at a low level

More close in phase noise and bizarrely missing the J-Test spectrum components! WTF!!!!!

SMSL M8

IMD SMPTE - Not the worst I have seen but

Image filter

Linearity - significantly different from previous but still not good

I think this beast needs more investigation.

 

[RayDunzl]

I think this beast needs more investigation.

Ok, well, out here in the real world my new 2x4HD is doing work, driving the cheezewoofers with an 80Hz lowpass, still duplicating the mains, though.

Filled in my 40-50Hz hole pretty well (on music).

Haven't readjusted the overall room correction filter yet.

Initial results:

Green (unsmoothed), old sweep measurement, providing a baseline of what I had going before adding the 2x4HD.
Red, peak RTA of music at listening position (string bass has the 41Hz low).
Black, RTA at the time of the snapshot.

Compare to an old example (Feb 25):

Same Green but has 1/12th octave smoothing in this snap of some other musical content.

 

[RayDunzl]

I think this beast needs more investigation.

How about a measurement using the Analog input, since that is how I'm driving it right now.

 

[stunta]

It would be interesting to find out if the DAC (when using digital input) is the same as the DAC in the ADC->DAC path.

 

[Blumlein 88]

About your J-test results here:

It could be the speed difference in clocks of the QA vs DUT. When you view close in even 100 ppm difference (common to be this large or larger) will spread the base of your Jtest signal. I suppose noise might obscure the missing small signal spikes. All that is happening in such cases is the actual frequency as seen by the QA is not 12 khz, but some fraction that fits off-center in the FFT bin so low levels spread to adjacent bins. Once you start looking at close in jitter you need to fix the relative speeds or your results will mislead on close in jitter.

So to start looking close in you need to correct for the clock rate differences. I sometimes correct the speed with a speed change function in Audacity. Current Audacity only corrects to the nearest 10 ppm. Earlier versions would correct to the nearest 3 ppb. Another option is do the first Jtest then use it to see what speed change is needed. For instance if you found the recorded signal was 100 ppm slow you could repeat the test with 12,001.2 hz for the Jtest which is 100 ppm faster. The ADC will see it as 12,000 rather precisely.

Here is the digital J-test, vs the same signal sped up by 100 ppm. The sped up version is in hot pink. There actually is no close in jitter spreading the base of the tone. The tone is simply the wrong frequency. The difference is too small to notice in the full 20 khz graph.

 

[amirm]

Another solution is to just not dig so deep into the noise floor.

 

[March Audio]

Another solution is to just not dig so deep into the noise floor.

We are only down to -125dBFS, thats not exactly deep and, it is indicative of something being wrong as I have not seen it on any other DAC.

On a purely subjective POV I rejected this as a solution for my DSP speaker as there just was something not quite right about the sound. Very difficult to articulate what it is.

 

[March Audio]

OK senior moment over. A very simple explanation, I used the wrong J-Test, 24 bit instead of 16bit DOH! Nothing of concern here.

Overall the performance of the DAC section appears to be pretty average. It has a high noise floor and poor linearity in comparison to many cheap DACs

 

[March Audio]

About your J-test results here:

It could be the speed difference in clocks of the QA vs DUT. When you view close in even 100 ppm difference (common to be this large or larger) will spread the base of your Jtest signal. I suppose noise might obscure the missing small signal spikes. All that is happening in such cases is the actual frequency as seen by the QA is not 12 khz, but some fraction that fits off-center in the FFT bin so low levels spread to adjacent bins. Once you start looking at close in jitter you need to fix the relative speeds or your results will mislead on close in jitter.

So to start looking close in you need to correct for the clock rate differences. I sometimes correct the speed with a speed change function in Audacity. Current Audacity only corrects to the nearest 10 ppm. Earlier versions would correct to the nearest 3 ppb. Another option is do the first Jtest then use it to see what speed change is needed. For instance if you found the recorded signal was 100 ppm slow you could repeat the test with 12,001.2 hz for the Jtest which is 100 ppm faster. The ADC will see it as 12,000 rather precisely.

Here is the digital J-test, vs the same signal sped up by 100 ppm. The sped up version is in hot pink. There actually is no close in jitter spreading the base of the tone. The tone is simply the wrong frequency. The difference is too small to notice in the full 20 khz graph.
View attachment 11483

The shape of the skirt is affected by the window function, its not really the effect of not being centre bin, although obviously this is also affected by the window type and its leakage and the spectral resolution. That will create a centre amplitude difference, but the ultimate spread should be the similar. The leakage is there whether you are centre bin or not. It shouldnt go from no leakage to significant leakage. So I am more than a bit surprised by your audacity result there, I will test it out.

I had been using Kaiser windows on ARTA, but the QA doesnt have this option. I have been using Hann window in all the measurement apart from linearity. (flat top), so it was really a comparison to the SMSL which had a thinner skirt indicating higher close in phase noise on the miniDSP.

https://en.wikipedia.org/wiki/Window_function

 

[Blumlein 88]

OK senior moment over. A very simple explanation, I used the wrong J-Test, 24 bit instead of 16bit DOH! Nothing of concern here.


Overall the performance of the DAC section appears to be pretty average. It has a high noise floor and poor linearity in comparison to many cheap DACs

That crossed my mind, but I didn't say anything.

 

[March Audio]

That crossed my mind, but I didn't say anything.

Always suggest stuff, its helpful, no offence is taken as I am capable of being quite dumb! .

 

[Blumlein 88]

The shape of the skirt is affected by the window function, its not really the effect of not being centre bin, although obviously this is also affected by the window type and its leakage. That will create a centre amplitude difference, but the ultimate spread should be the very similar. The leakage is there whether you are centre bin or not. It shouldnt go from no leakage to significant leakage. I had been using Kaiser windows on ARTA, but the QA doesnt have this option. So I am a bit surprised by your audacity result there, I will test it out.

I have been using Hann window in all the measurement apart from linearity. (flat top), so it was really a comparison to the SMSL which had a thinner skirt indicating higher close in phase noise on the miniDSP.

https://en.wikipedia.org/wiki/Window_function

I use the Hann window as well. It definitely is from not being centered in the FFT bin close in. The one I posted was 100 ppm fast vs dead on. Digitally generated no other confusing factors. One spreads the base the other doesn't. Now actual jitter will broaden the band as well. Until you synch up the speed you can't be sure which is which.

Here are some shots of dead perfect speed , 83 parts per trillion fast, and 100 ppm fast. As you can see close in the 100 ppm is visibly different well above noise floors. Very close in which is the second image shows close enough you can see how it sits off center in the FFT windows. These are 32 K ffts. Frequency is along the bottom in khz.

The Jtest is capable of much more precision than is usually obtained for close in jitter. You have to synch up the speeds however. Now I doubt such close in jitter is an audible issue. These little off center in the FFT bin results fall away by 1000 hz each side of the 12 khz tone. So super precise synching is not needed to see other kinds of spurious jitter further out. For most purposes that is all we care about. If you had a perfectly centered Jtest tone and it has a base that spreads out then it is due to close in phase noise and not a speed difference. Often in Wave Spectra I will just look really close in and see if the tone is off center of the bin or not for a quick check.

 

[March Audio]

Sorry but there is something wrong with your results there. You are going from no leakage (it literally looks to be 1 bin wide in blue) to below -180dB, to significant leakage. My understanding is thats not how the Hann window behaves.

40 bins is 58.4Hz in this example.

I have generated a 12kHz tone (GREEN) and varied around that frequency (BLUE). Hann and 32K FFT with 1.46Hz bin width. These are the minimum and maximum found, about 0.8 Hz difference, which are pretty much what I would expect. I wouldnt expect a DAC to be 0.8 Hz out off the top of my head but I would need to check that. I will try this with a different ADC and ARTA later.

 

[Blumlein 88]

Sorry but there is something wrong with your results there. You are going from no leakage (it literally looks to be 1 bin wide in blue) to below -180dB, to significant leakage.

40 bins is 58.4Hz in this eample.

I have generated a 12kHz tone (GREEN) and varied around that frequency (BLUE). Hann and 32K FFT with 1.46Hz bin width. These are the minimum and maximum found, about 0.8 Hz difference, which are pretty much what I would expect. I wouldnt expect a DAC to be 0.8 Hz out off the top of my head but I would need to check that. I will try this with a different ADC and ARTA later.

View attachment 11494

View attachment 11493

I am not sure your results disagree with mine. To pick a point of reference. Look at where the graph crosses at -120 db. Both in Audacity and in Wave Spectra the crossing is about 75 hz either size of 12 khz. In your FFTs it looks to be maybe 60 hz either side. Not so different. Try using 12001.2 hz which is 100 ppm offspeed. The better DACs do seem to keep within 50-65 ppm difference. I have run across some that are more than 200 ppm off in speed. My Antelope and Forte are 15 ppm off from each other. The 18i20 is 77 ppm off from the Antelope.

A quick check to see how far off they are is to look at the recorded file in Audacity. The top track is 12 khz the bottom track is 100 ppm fast (would look the same 100 ppm slow). You could look at a recorded file of the MiniDSP and see how the speed looks on a 12 khz wave. And again none of this matters except for close in jitter noise.

 

[March Audio]

I am not sure your results disagree with mine. To pick a point of reference. Look at where the graph crosses at -120 db. Both in Audacity and in Wave Spectra the crossing is about 75 hz either size of 12 khz. In your FFTs it looks to be maybe 60 hz either side. Not so different. Try using 12001.2 hz which is 100 ppm offspeed. The better DACs do seem to keep within 50-65 ppm difference. I have run across some that are more than 200 ppm off in speed. My Antelope and Forte are 15 ppm off from each other. The 18i20 is 77 ppm off from the Antelope.

A quick check to see how far off they are is to look at the recorded file in Audacity. The top track is 12 khz the bottom track is 100 ppm fast (would look the same 100 ppm slow). You could look at a recorded file of the MiniDSP and see how the speed looks on a 12 khz wave. And again none of this matters except for close in jitter noise.

View attachment 11497

Well they sort of do because a hann window wont have zero leakage down to and beyond -180dB Your variation in skirt size is significantly different to mine. Its interesting thats for sure and would be nice to bottom out

 

[Rodney Gold]

The dsp functionality of the 2x4hd is its reason for being..as interesting as the measurements may be.

 

[Thomas savage]

Great work @BE718 , I’ve taken to scrolling down the measurements while keeping a eye out for a ‘ WTF ‘

2 WTF’s means this device is beyond my consideration.

 

[stunta]

The dsp functionality of the 2x4hd is its reason for being..as interesting as the measurements may be.

But if its messing with my bits, I don't want it. No one messes with my bits, yo!

 

[Blumlein 88]

@BE718
Well I get the same results with Audacity and Wave Spectra. Here are some measures of an actual device. The cyan line is an 18i20 which is running 77 ppm fast. The gold line is the same signal once the speed has been corrected within about 10 ppb more or less. The second image has a pink line which is a synthetically created file 77 ppm fast. Hann window, 32 K fft. Again it only matters close in. In the second image you see how they all come together 200 hz either side of the 12 khz tone. 77 ppm fast btw is offset by .924 hz. You see the point at which -120 db crosses the graph is very close to where your .8 hz offset occurred at roughly 75 or 80 hz left and right.

Now trying to get back on topic. I originally brought all of this up because you showed the SMSL as having less close in phase noise than the miniDSP. Yet the amount of the difference or most of it could be nothing more than speed differences in the clocks of the two DACs vs the clock in the QA. Or it could be a real difference in close in timing.

If you still have both devices an easy quick check to see if that is some of it would be to look at the 12 khz waveform in Audacity to see if one is smooth and one lumpy which would indicate such a speed difference.