• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Motu M4, spurious tones at 44.1 kHz, excellent performance at 48 kHz

capslock

Senior Member
Joined
Jun 19, 2020
Messages
316
Likes
145
This is an M4 recently bought off Amazon Warehousedeals. It has firmware 2.0 and the additional screws on the combi connectors, so I suspect it is v3 but haven't opened it up.

Performance in loopback is better than what Amir measured (I suspect on v1), however, I get strange spurious tones such as twin peaks at 2.9 and 3.0 kHz instead of the pure third harmonic I'm getting at 48 kHz sampling rate. Also note the 4.9 kHz tone. This happens at both 44.1 and 88.2 kHz but not at 48, 96 and 192 kHz. Does the Motu have only a multiple of 48 kHz as a clock and use a sample rate converter or analog PPL to generate 44.1 and multiples?

Interestingly, the THD and THD+N numbers do not change significantly. I'm even getting slightly higher THD with 48 kHz, e.g. 0.0020 instead of 0.0018%. Some of the energy of the harmonics ends up away from the harmonic frequency, but ARTA should account for everything that is not at the fundamental.

This was measured with a 6.3 mm symmetric cable from Line Out 3 to Line In 3 on the left channel. Setting 0 dB as a generator level resulted in a wide floor of spuriae at -100 dB. Setting -1 dB results in -3.9 dB being displayed in ARTA in the dB fs setting. Not sure why this is, so far I thought dB fs ingnores the calibration.

Setting generator level at -3 dB in ARTA results in the peak sitting at -5.9 dB. THD+N increases from 0.0005% to 0.0006 but THD improves from 0.00018 to 0.00013, with the 2nd harmonic at roughly -124 dB from the fundamental and 3rd at -120 dB.
 

Attachments

  • L - 1 dB 44,1 kHz.PNG
    L - 1 dB 44,1 kHz.PNG
    190.5 KB · Views: 128
  • L -1db.PNG
    L -1db.PNG
    190.6 KB · Views: 124
OP
C

capslock

Senior Member
Joined
Jun 19, 2020
Messages
316
Likes
145
Sorry, must have saved the same image twice. Will redo and post the 48 kHz spectrum tonight. It does look clean.
 

AnalogSteph

Major Contributor
Joined
Nov 6, 2018
Messages
3,334
Likes
3,278
Location
.de
I would rather presume that all the clock generation for I²S is handled by the MCU, which seems to be an NXP LPC5528. Why this would create such issues, I don't know. Has to be some sort of aliasing.

Some other things I would still try to narrow things down:
a) 12 kHz, for a better look at LF jitter
b) split sample rates for ADC and DAC (if possible), i.e. a 44.1k multiple for the DAC and 48k multiple for the ADC and vice versa. There should be a pattern emerging. My bets would be on the ADC side, which is generally the more critical one. If this test is not possible, perhaps you've got an external source floating around that'll do the job.
 

Tangband

Major Contributor
Joined
Sep 3, 2019
Messages
2,994
Likes
2,789
Location
Sweden
Sorry, must have saved the same image twice. Will redo and post the 48 kHz spectrum tonight. It does look clean.
This is interesting - If your testing is correctly done. One can see the same (very small ) differences in usb-spdif converters where there is only one master clock inside. In such case, there is slightly worse performance with the sample frequency thats not perfect for the clock frequency. At - 120 dB and 44,1 kHz , its still excellent.

If all this is correct, then the result points to a slightly better recording quality If 48 or 96 kHz is used.
Here you can see the (very) small difference between 44,1 and 48 kHz conversion in SMSL po100 pro, maybe indicating that it has only one crystal clock optimised for one frequency sampling. ( Vintageflankers measurements )

37DBEF6E-3D36-41A1-B9B4-1708503C0D48.jpeg
 
Last edited:
OP
C

capslock

Senior Member
Joined
Jun 19, 2020
Messages
316
Likes
145
Ok, here goes. I rechecked the calibration. Setting 0 dB output in ARTA results in harmonics up to -85 dB, so either output or input is clipping slightly. - 1 dB is 4.495 V rms symmetrically at 1 kHz or 4.513 V according to my UNI-T UT139C True RMS meter. Interestingly, it is not truly symmetric. I measured 2.289 V from Tip to Sleeve and 2.223 V from Tip2 to sleeve as opposed to 4.495 V Tip to Tip2 (at 1 kHz). Same thing at lower frequencies. This unbalance I have never seen mentioned in reviews here or elsewhere. I wonder if channel differences can in part be explained by how well the outputs or inputs are balanced.

Consequently, -2 dB is 4.022 V RMS Tip to Tip2. I did the whole output and input calibration thing, but the dB FS scale was not changes (well, that's actually good). I still don't understand why it displays -1 dB according to its generator as -3.9 dB in its spectrum analyzer, but then, this is the first time I have used a symmetric soundcard with ARTA while actually paying attention to such details.

Lowering the output level will decrease THD while increasing THD+N. THD is largely flat between -7 and -12 dB, but of course, THD+N degrades, so - 7 dB is a sweet spot. It has the added advantage of displaying the fundamental at - 9.9 dB FS, so I can simply add 10 dB to the harmonics :)

I'm attaching measurements of a 1 kHz sine at 44.1 kHz fs at - 1, -2, -3 and -7 dB, and an 11 kHz sine at -7 db at 44.1 and 48 kHz. No jitter effects are visible, but the 48 kHz plot is absolutely free of spuriae. For some reason, FFT length switched when I changed sample rate, so I am including both 64 and 128 k FFT length for the 48 kHz measurement - no difference that I can see.
 

Attachments

  • L 3 3 -1 dB 44_1.PNG
    L 3 3 -1 dB 44_1.PNG
    195.9 KB · Views: 68
  • L - 3 dB 44,1 kHz.PNG
    L - 3 dB 44,1 kHz.PNG
    184.5 KB · Views: 70
  • L - 2 dB 44,1 kHz.PNG
    L - 2 dB 44,1 kHz.PNG
    179.8 KB · Views: 72
  • L - 7 dB 11 kHz 44,1 kHz.PNG
    L - 7 dB 11 kHz 44,1 kHz.PNG
    125.6 KB · Views: 73
  • L - 7 dB 11 kHz 48 kHz 64 k FFT.PNG
    L - 7 dB 11 kHz 48 kHz 64 k FFT.PNG
    155.8 KB · Views: 66
  • L - 7 dB 11 kHz 48 kHz.PNG
    L - 7 dB 11 kHz 48 kHz.PNG
    122.8 KB · Views: 67
  • L - 7 dB 44,1 kHz.PNG
    L - 7 dB 44,1 kHz.PNG
    172.4 KB · Views: 69
OP
C

capslock

Senior Member
Joined
Jun 19, 2020
Messages
316
Likes
145
Here's 0 dB at 44.1 and 48 kHz sampling rate. 44.1 has all kinds of peaks up to -85 dB - at 48 kHz they don't reach beyond -100. Must be some sort of overflow problem. Why it would be so much worse at 44.1 kHz is anyone's guess.
 

Attachments

  • L -0 dB 1 kHz 44_1 kHz 128 k FFT.PNG
    L -0 dB 1 kHz 44_1 kHz 128 k FFT.PNG
    206.3 KB · Views: 70
  • L -0 dB 1 kHz 48 kHz 128 k FFT.PNG
    L -0 dB 1 kHz 48 kHz 128 k FFT.PNG
    214.7 KB · Views: 65
Last edited:
OP
C

capslock

Senior Member
Joined
Jun 19, 2020
Messages
316
Likes
145
Here's 1 kHz output at 48 kHz.
 

Attachments

  • L -1 dB 1 kHz 48 kHz 128 k FFT.PNG
    L -1 dB 1 kHz 48 kHz 128 k FFT.PNG
    214.7 KB · Views: 64
  • L -2 dB 1 kHz 48 kHz 128 k FFT.PNG
    L -2 dB 1 kHz 48 kHz 128 k FFT.PNG
    202.6 KB · Views: 66
  • L -3 dB 1 kHz 48 kHz 128 k FFT.PNG
    L -3 dB 1 kHz 48 kHz 128 k FFT.PNG
    198.8 KB · Views: 63
  • L -4 dB 1 kHz 48 kHz 128 k FFT.PNG
    L -4 dB 1 kHz 48 kHz 128 k FFT.PNG
    210.7 KB · Views: 48
  • L -7 dB 1 kHz 48 kHz 128 k FFT.PNG
    L -7 dB 1 kHz 48 kHz 128 k FFT.PNG
    208.9 KB · Views: 56
OP
C

capslock

Senior Member
Joined
Jun 19, 2020
Messages
316
Likes
145
If you thought that was pretty good, it gets better. I tried all permutations of R and L out (4 resp. 3) on L in (3) and R in (4). Turns out R (4) out into R in (4) is the best. This loopback performance is more than adequate to test most amplifiers while more convenient than having to use separte DAC and ADC units.

Note that for 4th harmonic, L out into R in was the best. In general, individual harmonics were pretty reproducible but I got changes by 3 dB on repeats occasionally.

PS: regarding Steph's suggestion to run output and input at different sample rates, Arta cannot do this. I'm not even sure the Moto could do this with a different software. I do have a Topping DM-7, so this is what I could try.
 

Attachments

  • L -7 dB 1 kHz 48 kHz 128 k FFT R 4 out L 3 in.PNG
    L -7 dB 1 kHz 48 kHz 128 k FFT R 4 out L 3 in.PNG
    202 KB · Views: 51
  • R -7 dB 1 kHz 48 kHz 128 k FFT L 3 out R 4 in.PNG
    R -7 dB 1 kHz 48 kHz 128 k FFT L 3 out R 4 in.PNG
    148.2 KB · Views: 54
  • R -7 dB 1 kHz 48 kHz 128 k FFT R 4 out R 4 in.PNG
    R -7 dB 1 kHz 48 kHz 128 k FFT R 4 out R 4 in.PNG
    212.5 KB · Views: 55
Last edited:

Tangband

Major Contributor
Joined
Sep 3, 2019
Messages
2,994
Likes
2,789
Location
Sweden
Ok - Thanks for the results. Im an amateur, but it might look like those options to explain this:

1. You Motu has only one clock at xmos . The dac has another clock . The Xmos 216 clock is a 24 MHz clock that seems to be native to 48 , 96 kHz .

2. The xmos does SRC calculations in your Motu when you do recordings in 44,1 kHz or 88,2 kHz - This will give slightly inferior result compared to recording music at 48 or 96 kHz.
A Xmos 216 seems to be able to do SRC with -120 SINAD range.

or…

3. Theres something wrong with your measurements .

4. Im wrong - the slightly worse results at 44,1 and 88,2 kHz depends on other things.

Can some of you digital experts comment on this ?

teardown of Motu m4 here:


A 24MHz TCXO sits at the xmos 216, maybe optimised for better results at 48, 96 kHz.

69860627-76AC-4565-9D59-29C5A36FA4A5.jpeg
 
Last edited:

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
You really need another interface to learn much more on the clocks. Could be lots of things. Try running a sweep and showing the spectragram view of that sweep. It might let you see if the filter is acting oddly in one rate versus another. Run maybe a -3 db 20hz-20khz sweep. Good thing is other than 0 dbFS all these are well below being audible.

Here is one viewed in Audacity using Spectrogram view. The light faint blue lines are around -120 dbFS. You can see aliasing from some of the harmonic distortion. etc. etc.
index.php
 
OP
C

capslock

Senior Member
Joined
Jun 19, 2020
Messages
316
Likes
145
Tangband, the newer revision has no XMOS but rather an Atmel controller.

In the picture you linked, which is of an older revision, you can see that the ESS DAC has its own crystal, but the AKM ADC doesn't. The newer revision has an ESS ADC that sports its own crystal. My understanding is that all ESS convertors do their own sample rate conversion, operating at a native frequency that is neither a multiple of 44.1 or 48 kHz.
 

Tangband

Major Contributor
Joined
Sep 3, 2019
Messages
2,994
Likes
2,789
Location
Sweden
Tangband, the newer revision has no XMOS but rather an Atmel controller.

In the picture you linked, which is of an older revision, you can see that the ESS DAC has its own crystal, but the AKM ADC doesn't. The newer revision has an ESS ADC that sports its own crystal. My understanding is that all ESS convertors do their own sample rate conversion, operating at a native frequency that is neither a multiple of 44.1 or 48 kHz.
Aha, ok.
Anyway, This Motu m4 seems to be one of the best options for the money. The correct way to do recordings anyway, is to always use 24 bit 96 kHz when using microphones and then do the later things ( SRC, effects, normalisation ) in the computer . Newest versions of audacity has sota SRC .
 
OP
C

capslock

Senior Member
Joined
Jun 19, 2020
Messages
316
Likes
145
Yes, but the mic in (channels 1 and 2) stinks in comparison to line in (channels 3 and 4), even at minimum gain. More to follow.
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
Yes, but the mic in (channels 1 and 2) stinks in comparison to line in (channels 3 and 4), even at minimum gain. More to follow.
I'd be interested in what ways they stink. Measures and reviews by others don't show it to be poor in that area.
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,634
Likes
1,692

AnalogSteph

Major Contributor
Joined
Nov 6, 2018
Messages
3,334
Likes
3,278
Location
.de
Interesting. Atmel indeed, and the same ARM Cortex-M7 job as in the Ultralite Mk5 no less. Does that mean that we should expect the same kind of ultralow roundtrip latency then?

BTW, I feel kind of sorry for the guys at iXBT - the rest of the world uses fancy audio analyzers or at least a Cosmos ADC + APU, and here they are doing loopback testing with RMAA like in the olden days. (Which, don't get me wrong, you can still learn plenty from, but it takes a bit more effort to coax out the equipment's actual limits.)

Their 24/44 test does not show any anharmonics though. I would try to repeat that with RMAA, and if the issue persists, this particular M4 may be kind of a dud. That sort of thing occasionally happens at MOTU... the M2 that German c't magazine recently included in an audio interface roundup had a noisy input channel, too. I guess soldering quality occasionally leaves something to be desired.
 
OP
C

capslock

Senior Member
Joined
Jun 19, 2020
Messages
316
Likes
145
Talking about duds, I had another M4 from a recommerce seller on fleabay. They sell stuff that customers have returned for refund and that the big retailers don't want to bother with for testing. Supposedly, it had a few scratches and was missing a few screws on the front plate. It turned out to be missing all screws from front and back. The line out seemed ok, it was even a little hotter than above unit with the same kind of asymmetry (2.489 resp. 2.462 V from Tip 1 to Sleeve, 2.389 resp 2.391 from Tip 2 to Sleeve - left resp. right).

Strangely, it was missing about 7 dB on the line inputs, so it couldn't be overdriven with 0 dB output, but it also had higher distortion for the same displayed level. Even more strangely, turning up the gain on the mic inputs would raise the noise floor by up to 25 dB. Needless to say I returned it (before even figuring out if there was a 44.1 kHz issue). The PCB showed no signs of burned up components or hand soldering.
 
Top Bottom