• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Test and Measurement Procedures and Standards

March Audio

Major Contributor
Manufacturer
Joined
Mar 1, 2016
Messages
6,382
Likes
8,809
Location
Albany Western Australia
#1
Hi all,

Whilst most of the technical test and and measurements are performed by Amir in this forum, a number of members including myself contribute measurements from time to time. I will probably be contributing more to this testing over the next year as I have more free time available (Yay! :) ) To help make this data more consistent, comparable and easy for the readership to understand, we really should define some standards that we work to. I.E. type of test and purpose, acquisition and processing parameters, graph display parameters etc.

Also, Amir is the only one of us (that I know of) that has consistent access to what I consider "professional" test and measurement equipment. I do on occasions. Having said this some audio ADCs have excellent performance which puts them close to something like an AP and actually quite suitable for this sort of testing. Problem is we dont know how they compare in absolute terms to Amirs AP unit. So I suggest, for those that want to contribute to measurements, we could pass around a test unit (such as a dac) measure it consistently and see how they compare to validate our "domestic" measurement systems.

As DACs are currently the most popular measured device, shall we start with some suggestions for tests? We can move on to specific parameters afterwards.

My 2 pennies. Please add/remove/discuss

  1. Frequency response - 44, 48, 88 & 96 kHz replay bandwidth. Extend acquisition bandwidth beyond replay to show filter roll off and out of band noise
  2. No signal noise floor
  3. 1kHz tone @ 0dB, -1dB, -60dB FS. THD + Noise
  4. 48kHz J test. USB and spdif input (if available)
  5. Intermodulation 19kHz + 20kHz ???
  6. Small signal time waveform 16 bit and 24 bit @ -90 something dB

Thoughts?
 

Wombat

Major Contributor
Joined
Nov 5, 2017
Messages
6,568
Likes
5,862
Location
Australia
#2
Hi all,

Whilst most of the technical test and and measurements are performed by Amir in this forum, a number of members including myself contribute measurements from time to time. I will probably be contributing more to this testing over the next year as I have more free time available (Yay! :) ) To help make this data more consistent, comparable and easy for the readership to understand, we really should define some standards that we work to. I.E. type of test and purpose, acquisition and processing parameters, graph display parameters etc.

A good idea.

Also, Amir is the only one of us (that I know of) that has consistent access to what I consider "professional" test and measurement equipment. I do on occasions. Having said this some audio ADCs have excellent performance which puts them close to something like an AP and actually quite suitable for this sort of testing. Problem is we dont know how they compare in absolute terms to Amirs AP unit. So I suggest, for those that want to contribute to measurements, we could pass around a test unit (such as a dac) measure it consistently and see how they compare to validate our "domestic" measurement systems.

As DACs are currently the most popular measured device, shall we start with some suggestions for tests? We can move on to specific parameters afterwards.

My 2 pennies. Please add/remove/discuss

  1. Frequency response - 44, 48, 88 & 96 kHz replay bandwidth. Extend acquisition bandwidth beyond replay to show filter roll off and out of band noise
  2. No signal noise floor
  3. 1kHz tone @ 0dB, -1dB, -60dB FS. THD + Noise
  4. 48kHz J test. USB and spdif input (if available)
  5. Intermodulation 19kHz + 20kHz ???
  6. Small signal time waveform 16 bit and 24 bit @ -90 something dB

Thoughts?
 

DonH56

Major Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
4,789
Likes
7,880
Location
Monument, CO
#3
In addition to the specific tests targeting these devices IEEE Standard 1241 has a lot of useful info about testing ADCs that is also useful for DACs. I imagine there is a DAC standard but I do not have it (I was involved with 1241 and was designing mostly ADCs; I used the same methodology to test DACs). There is also the earlier transient waveform recorder Standard 1057.

To your list I would expand the IMD to check across the bandwidth, e.g. maybe 100/110 Hz, 1/1.1 KHz, and 10/11 kHz. I would also add some square wave and/or step response tests, and also liked the nasty TIM or wideband test signals combining e.g. a 100 Hz square wave with a 10 kHz sine wave just for grins.

Edit: Looks like IEEE Standard 1658 is for DACs... I was not involved with that one and do not have it on file. Guessing it is similar to 1241 but am too cheap to buy it.
 
Last edited:

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
30,949
Likes
95,904
Location
Seattle Area
#4
Great suggestion. One thing I have been doing is to make sure reader's eyes don't gloss over the measurements and skip over all of them. I find this to be a problem with many other measurement based reviews. So much stuff is shown that is really not separating the good from bad. One such thing is frequency response. Until we get to test some with tube buffer stage and such, I am not seeing the value of constantly showing near ruler flat responses that vary 0.1 db and such.

Another one is the dual 19+20 Khz intermodulation test. The signals are not realistic in that we don't have full amplitude responses like that in music at such high frequencies. Is there something unique that this brings out that say, 1 Khz harmonic distortion does not?

How about 60 Hz harmonic distortion test that stereophile runs? I am not sure of the value of that either.

I think the most important thing we need to do is to make sure our graphs use similar settings as you suggested. I have even tried to standardize such in my own measurements with reference always being yellow and the device under test red for example. If we can go to this length a quick visual comparison would tell the whole story instead of having to read text and become familiar with every test setting.

So do you want to buy a cheap DAC to test and then send it to me and we post the two measurements and see where we go?

Again, very good idea to do this as there is so much product out there that we need to extended the labor force to cover a lot of it.
 

DonH56

Major Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
4,789
Likes
7,880
Location
Monument, CO
#5
One of the nastiest tests I have had to perform in the past was the NPR (noise power ratio) test but it might be similar to the J test? The NPR test essentially runs wideband noise or a multitude of signals into the input with a (very) narrow band notched (or one tone missing). See how much the converter fills in the notch. Really stresses the analog input/output of ADCs/DACs.

Keeping the look and feel of the results similar right down to the choice of colors and position and scale of plots is so very important and yet so easy to botch. I take it for granted that I have to do it so failed to complement Amir; thanks Amir!

PSRR is a good one but with linear and SMPS supplies you have to separate supply-related harmonics from AC line leakage. What I have done in the past is to tabulate key frequencies for each test including those that should be consistent, like 60/120/180 Hz, clock (sub)harmonics, and such plus the primary signal component and harmonically-related terms (i.e. where they fold into baseband -- alias -- for an ADC, and where they get multiplied up -- image -- at the output of a DAC). I might thus list 60, 120, 180, 1000, 2000, 3000, 4k, 5k-10k when testing a 1 kHz tone. THD used to include ten harmonics by default, no idea if that standard is used today.

SMPS usually have a fundamental'ish frequency but modulate around it so there is a more of a spread than a single tone and little bursts of noise that can reach well into MHz region. I recently helped debug an issue where a 500 kHz or so SMPS was causing noise up around 200~300 MHz. Normally no big deal, but when microvolts matter...
 
Last edited:

Blumlein 88

Major Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
11,108
Likes
16,144
#6
I for one would ditch the 1 khz THD. While the twin tone IMD test is unrealistic it is stressful in a number of ways to find when something is non-linear. For instance I have seen a few devices look alright with a one tone THD, but behave horribly with the IMD. I have not seen the reverse. If IMD is good, the THD is better. There is the fact that almost everyone uses THD and people understand it. Yet like frequency response, THD is boringly low enough with nearly all gear why post it unless you find one that is sub-standard.

I do like the Jurgen test as JA at Stereophile calls it. Superimpose silent noise floor with - 4db white noise at 48 khz sample rates, but let the ADC record at 192 khz. It shows the shape of the filter and how sharp or not it is. JA includes a 19.1 khz high level tone so you can see how it leaks (actually images) above the Nyquist frequency though that does make it look busy.

So if Amir wants to keep people's eyes from glazing over I think you need to limit it to three graphs and add any other substandard results you may have turned up.

I'll assume Amir wishes to keep the quarter sample rate tone or Jtest.

I would have an 18+19 or 19+20 khz twin tone IMD graph. I prefer 18+19 because some DAC filters are a bit soft at 20 khz and corrupt the result.

I would combine a -60 db 1 khz tone (for showing the noise floor) and a -4db white noise recorded at 192 khz during 48khz playback. A variation on the Jurgen test.

In addition an area that does vary quite a bit among DACs is the noise floor. I would list what the RMS noise level is while playing a - 60 db 1khz tone with the tone notched out.


Now I will also say, just for myself, I would like to see any and all tests that show something useful. I do understand however that most people skip pages like that.
 

Blumlein 88

Major Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
11,108
Likes
16,144
#7
It also occurs to me the size FFT used should be standardized. Amir uses 16 k FFT's I would prefer a little more, but we would do well to agree on using the same size. Further I think they should be scaled for the sample rate. 16 K for 48 khz rates, 32 K for 96 khz rates and 64 K for 192 khz rates so that the bin sizes and noise floors look the same across the different sampling rates.
 

Blumlein 88

Major Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
11,108
Likes
16,144
#8
One other thought, I find slow sweeps to be interesting. They uncover some unexpected glitches in DAC noise floors. But they really only make sense if you watch a video of the sweep.

As an example I've one DAC that has an idle tone at 31 khz. It is there at all sample rates. All signals reflect in both directions around it. The level of the imaging/aliasing are low, but you see them move the noise floor. So an 18 khz tone results in a low level reflection at 13 khz and 44 khz even at 48 khz sample rates. Without watching a sweep I don't know if you would ever see what is going on with odd tones that show up on the noise floor of various tests.
 

DonH56

Major Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
4,789
Likes
7,880
Location
Monument, CO
#9
There is a direct theoretical correlation between IMD and HD with IMD tracking higher than HD (e.g. IMD3 products are about 9.54 dB higher than HD3 components). That tracks well in practice unless the distortion is dominated by other nonlinear effects. By and large I would prefer to see some IMD tests rather than THD but as you say THD is what folk know and understand (even if they do not understand its limitations).

16k FFTs are fast and easy but way smaller than what I would normally use. I do not know if Amir is following typical industry conventions for audio tests, however, and my work may have more stringent demands for things like frequency resolution and the FFT's noise floor.

Histograms can provide interesting data as well, and you can continue to collect data and bin it on the fly so records (arrays, file sizes) do not get huge. For long-term (hours, days, weeks) that is a good way to do it. You can fit the histogram to an ideal sinusoid easily enough and it will show both short- and long-term problems.
 
OP
March Audio

March Audio

Major Contributor
Manufacturer
Joined
Mar 1, 2016
Messages
6,382
Likes
8,809
Location
Albany Western Australia
Thread Starter #11
Brilliant, thanks guys for your responses, all good stuff (as always :) )


Amirs comments:

Great suggestion. One thing I have been doing is to make sure reader's eyes don't gloss over the measurements and skip over all of them. I find this to be a problem with many other measurement based reviews. So much stuff is shown that is really not separating the good from bad. One such thing is frequency response. Until we get to test some with tube buffer stage and such, I am not seeing the value of constantly showing near ruler flat responses that vary 0.1 db and such.
I absolutely agree there is a balance to strike between going overboard with lots of random test that are meaningless to the vast majority of readers, and providing enough pertinent and detailed technical information. I dont think we should dumb down too much, I would rather try and explain it. This is audio science review after all. We definitely need a sticky that describes the purpose of each test and maybe shows a good/bad example of data.
Frequency response - Yes I tend to agree, most dacs I have seen are consistently flat. As an aside whilst developing my speakers I did find that small tilts ( maybe .1/0.2 dB/octave) in frequency response over a wide range are audible, but we are talking when spread over many octaves. Anyway, I would like to see the filter and out of band noise levels, so Perhaps Blumleins suggestion of the Jurgen test is more useful?


Another one is the dual 19+20 Khz intermodulation test. The signals are not realistic in that we don't have full amplitude responses like that in music at such high frequencies. Is there something unique that this brings out that say, 1 Khz harmonic distortion does not?
I think I agree here, are we really telling ourselves anything useful with the test? Thoughts Don?

How about 60 Hz harmonic distortion test that stereophile runs? I am not sure of the value of that either.
Im afraid I'm not familiar with this test or its purpose.

I think the most important thing we need to do is to make sure our graphs use similar settings as you suggested. I have even tried to standardize such in my own measurements with reference always being yellow and the device under test red for example. If we can go to this length a quick visual comparison would tell the whole story instead of having to read text and become familiar with every test setting.
100% agree, for the rest of us not using AP software we should try and standardise what analysis software to use. I tend to use ARTA and Adobe audition, but there probably better alternatives.

So do you want to buy a cheap DAC to test and then send it to me and we post the two measurements and see where we go?
Yep, I will try and sort this out. Need to give a bit of thought to it though to minimize variables. For example bus powered USB could be different on different PCs (OMG thats dangerously close to audiophile talk. :eek: ). Also we might want a DAC thats not actually too good so we have some spuria to compare :) .

Again, very good idea to do this as there is so much product out there that we need to extended the labor force to cover a lot of it.
Yes, no idea how you find the time :)



Dons Comments:

One of the nastiest tests I have had to perform in the past was the NPR (noise power ratio) test but it might be similar to the J test? The NPR test essentially runs wideband noise or a multitude of signals into the input with a (very) narrow band notched (or one tone missing). See how much the converter fills in the notch. Really stresses the analog input/output of ADCs/DACs.
Sounds like a good test to me, but as you say we could probably cover it with J test.

Keeping the look and feel of the results similar right down to the choice of colors and position and scale of plots is so very important and yet so easy to botch. I take it for granted that I have to do it so failed to complement Amir; thanks Amir!
Ditto :)

PSRR is a good one but with linear and SMPS supplies you have to separate supply-related harmonics from AC line leakage. What I have done in the past is to tabulate key frequencies for each test including those that should be consistent, like 60/120/180 Hz, clock (sub)harmonics, and such plus the primary signal component and harmonically-related terms (i.e. where they fold into baseband -- alias -- for an ADC, and where they get multiplied up -- image -- at the output of a DAC). I might thus list 60, 120, 180, 1000, 2000, 3000, 4k, 5k-10k when testing a 1 kHz tone. THD used to include ten harmonics by default, no idea if that standard is used today.
Yes it is very relevant but with todays bus powered and SMPS would it be difficult to establish what spuria are related to the supplies? Changing supply could give the game away if there is an issue. Perhaps if odd spuria are noted in other plots then put extra effort in to identify rather than have it as a normal test?

SMPS usually have a fundamental'ish frequency but modulate around it so there is a more of a spread than a single tone and little bursts of noise that can reach well into MHz region. I recently helped debug an issue where a 500 kHz or so SMPS was causing noise up around 200~300 MHz. Normally no big deal, but when microvolts matter...
As above


Blumleins Comments:

I for one would ditch the 1 khz THD. While the twin tone IMD test is unrealistic it is stressful in a number of ways to find when something is non-linear. For instance I have seen a few devices look alright with a one tone THD, but behave horribly with the IMD. I have not seen the reverse. If IMD is good, the THD is better. There is the fact that almost everyone uses THD and people understand it. Yet like frequency response, THD is boringly low enough with nearly all gear why post it unless you find one that is sub-standard.
My experience is that THD on some DACS (and ADCs for that matter) can be problematic when you approach 0dB. Hence my desire to see it at 0dB and -1dB to see any problems. The THD + Noise figure is good for quick comparisons of general signal cleanliness and people are familiar with it. The -60dB plot is useful as it shouldnt "stress" the measuring ADC and will show the base noise floor which we could remove as a separate test.

I do like the Jurgen test as JA at Stereophile calls it. Superimpose silent noise floor with - 4db white noise at 48 khz sample rates, but let the ADC record at 192 khz. It shows the shape of the filter and how sharp or not it is. JA includes a 19.1 khz high level tone so you can see how it leaks (actually images) above the Nyquist frequency though that does make it look busy.
Agreed, I like :) .

So if Amir wants to keep people's eyes from glazing over I think you need to limit it to three graphs and add any other substandard results you may have turned up.
...mmmm......I'm too geeky, I want all the data......., no seriously see my comments above.

I'll assume Amir wishes to keep the quarter sample rate tone or Jtest.
I think its useful

I would have an 18+19 or 19+20 khz twin tone IMD graph. I prefer 18+19 because some DAC filters are a bit soft at 20 khz and corrupt the result.
I'm not sure, see above, needs a bit more discussion

I would combine a -60 db 1 khz tone (for showing the noise floor) and a -4db white noise recorded at 192 khz during 48khz playback. A variation on the Jurgen test.
Yep

In addition an area that does vary quite a bit among DACs is the noise floor. I would list what the RMS noise level is while playing a - 60 db 1khz tone with the tone notched out.
Yep


Now I will also say, just for myself, I would like to see any and all tests that show something useful. I do understand however that most people skip pages like that.
Yes, see my previous comments

Also

16k FFTs are fast and easy but way smaller than what I would normally use. I do not know if Amir is following typical industry conventions for audio tests, however, and my work may have more stringent demands for things like frequency resolution and the FFT's noise floor.

It also occurs to me the size FFT used should be standardized. Amir uses 16 k FFT's I would prefer a little more, but we would do well to agree on using the same size. Further I think they should be scaled for the sample rate. 16 K for 48 khz rates, 32 K for 96 khz rates and 64 K for 192 khz rates so that the bin sizes and noise floors look the same across the different sampling rates.


I agree here. As we looking at static data, is there any reason to not use large FFTs? Also I would use averaging to reduce random noise. If we agree we need to think about its parameters.

Oh, yes, impulse response?
 
Last edited:

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
30,949
Likes
95,904
Location
Seattle Area
#12
It also occurs to me the size FFT used should be standardized. Amir uses 16 k FFT's I would prefer a little more, but we would do well to agree on using the same size.
Mine are 32 K actually and is the limit of what I can without pulling the samples into the computer. So if that is good with everyone, we should use that.

BTW, there will be times where more taps would be useful and those could continue to exist and be posted.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
30,949
Likes
95,904
Location
Seattle Area
#13
16k FFTs are fast and easy but way smaller than what I would normally use. I do not know if Amir is following typical industry conventions for audio tests, however, and my work may have more stringent demands for things like frequency resolution and the FFT's noise floor.
I am limited to 32K ones in the AP I have. Fixing that will cost me $28,000 for a new unit so I am not going there :).

This has actually proven to be a benefit. I go to other sites where they use PC analysis and crank this up to 2 million or whatever. Once there, all kinds of little spikes show up at say -150 db that gets in the way of seeing the larger picture. Something that may show "no difference" in mine, will show different versions of "grass" in ultra-high-resolution FFTs. Since what I have is already revealing well below threshold of hearing, I am think it is a good thing to maintain at least at baseline level.
 

DonH56

Major Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
4,789
Likes
7,880
Location
Monument, CO
#14
S'ok with me, cheapskate. ;) :D

And yes, higher resolution can be a blessing and/or a curse.
 

Blumlein 88

Major Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
11,108
Likes
16,144
#15
I am limited to 32K ones in the AP I have. Fixing that will cost me $28,000 for a new unit so I am not going there :).

This has actually proven to be a benefit. I go to other sites where they use PC analysis and crank this up to 2 million or whatever. Once there, all kinds of little spikes show up at say -150 db that gets in the way of seeing the larger picture. Something that may show "no difference" in mine, will show different versions of "grass" in ultra-high-resolution FFTs. Since what I have is already revealing well below threshold of hearing, I am think it is a good thing to maintain at least at baseline level.
How much would the camper van bring on the market, think it'll go for $28k? Have to keep your priorities man. :)
 

Blumlein 88

Major Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
11,108
Likes
16,144
#16
I find these sort of spectrograms useful from Audacity.

Forte to Zen spectrogram .png


In this case it is a running 1k FFT of a 20 to 20khz sweep recorded at 192 khz though I am only graphing to 48 khz.

If a signal drops below -100 db it goes to the gray background (this is a setting you get to choose). You have the option of adding background gain which I did here. I added 10 db to the actual signal. The light blue lines above the red line are harmonics. In this case you see the 3rd and the 5th. Without the gain they were not visible so are between -100 and -110 dbFS. The horizontal line is a 31 khz idle tone which resides at -104 dbFS and is in the DAC.

So the only place I have seen these are over at DIY forums at times. Many people don't know what to make of them, but in one reasonably neat graph much can be shown once a person knows what is going on in the graph.

The range of these can be set as low as 300 db (or maybe more I stopped there). For instance I could alter this and you can see this ADC has some aliasing at about -140 dbFS.

Here is a slightly different view. 96 khz sample rate and a sweep to 40 khz. You see this DAC has low levels of harmonics, but there are lots of them. The 3rd, 5th, 6th, 7th, and on to the 11th are visible. You also see the harmonics as they reach nyquist reflect back downward for about 4 khz before filtering cuts them out of the picture. So this can show you much about how the filters work, but a different view than most are accustomed to seeing.

Zen loop spectrogram.png Click to see full size.
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,244
Likes
1,077
#17
http://archimago.blogspot.hk/2017/06/measurements-audioquest-dragonfly-black.html
I think the Juergen inspired "digital filter composite" is important:

[1] The 0 and -4dBFS white noise can reveal filter steepness and intersample clipping.
[2] The 19+20kHz IMD test can reveal distortion in sensitive frequencies (e.g. 1kHz)
[3] The digital silence shows the differences between noise floor and distortion

The Dragonfly black for example showed pathetic results in this test.

Also I hope reviewers can show the reference noise floor of a known signal, for example a 16-bit flat dither like this one:
https://www.audiosciencereview.com/forum/index.php?threads/digital-distortion.2059/#post-55543
I was testing pure SPDIF connection without analog so the low noise is expected, but at least readers know the 16-bit dither is at around -133dB in my graph.

Now take a look at Amir's Exasound E32 review:
https://www.audiosciencereview.com/...xasound-e32-dac-review-and-measurements.1990/
It shows something like -135dB

...and Blumlein 88's Focusrite 18i20 review:
https://www.audiosciencereview.com/forum/index.php?threads/focusrite-18i20-dac-measurements.2128/
Now it looks like -145dB

But I suppose the Exasound E32 actually has lower noise floor right? Those confusing FFT size, window type, overlapping etc made the graphs inconsistent and hard to read.
 

Blumlein 88

Major Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
11,108
Likes
16,144
#18
http://archimago.blogspot.hk/2017/06/measurements-audioquest-dragonfly-black.html
I think the Juergen inspired "digital filter composite" is important:

[1] The 0 and -4dBFS white noise can reveal filter steepness and intersample clipping.
[2] The 19+20kHz IMD test can reveal distortion in sensitive frequencies (e.g. 1kHz)
[3] The digital silence shows the differences between noise floor and distortion

The Dragonfly black for example showed pathetic results in this test.

Also I hope reviewers can show the reference noise floor of a known signal, for example a 16-bit flat dither like this one:
https://www.audiosciencereview.com/forum/index.php?threads/digital-distortion.2059/#post-55543
I was testing pure SPDIF connection without analog so the low noise is expected, but at least readers know the 16-bit dither is at around -133dB in my graph.

Now take a look at Amir's Exasound E32 review:
https://www.audiosciencereview.com/...xasound-e32-dac-review-and-measurements.1990/
It shows something like -135dB

...and Blumlein 88's Focusrite 18i20 review:
https://www.audiosciencereview.com/forum/index.php?threads/focusrite-18i20-dac-measurements.2128/
Now it looks like -145dB

But I suppose the Exasound E32 actually has lower noise floor right? Those confusing FFT size, window type, overlapping etc made the graphs inconsistent and hard to read.
Yes that is why we need to be on the same page for FFT size. As for 18i20 vs E32 Amir was using a 32k FFT and I was using 128k FFT. So the difference from this should be 6 db. Also at least in some of the graphs Amir had the E32 at +3.6 db so the 135 or so becomes 139dbFS. I had my signals around -1 db peak so something like the 6 or 7 db difference in results means the total noise floor was probably very close to the same for both devices.

You might notice where I had the -60 db tone I noted the effective noise floor was 106-107 db. That would be over the 24 khz band with the tone notched out. The E32 is near that or maybe a couple db lower. The Exasound measurements page looks much better until you realize they used a 1024k FFT and averaged 12 times. That may have lowered the noise floor 15 db. They also ran the unit off a 12 volt battery for these tests instead of the power supply. So taking all this into account Amir's result was maybe a few db less good than expected, but not far off Exasound results. Contrary to some expectations, the noise floor of the 32 bit DAC chip in the Exasound isn't much different than the 8 channel recording interface from Focusrite.
 
Last edited:

Joamlt

New Member
Joined
Jan 20, 2018
Messages
3
Likes
4
Location
Malaysia
#19
Seems like the discussion has died down a bit.
I'd like to contribute as well in measuring DAC's or headphone amps, however I do agree we need to standardize our results.
As I understand from the recent discussion, allow me to summarize and correct me if i'm wrong.

Testing items is listed below:

  1. Frequency response - 44, 48, 88 & 96 kHz replay bandwidth. Extend acquisition bandwidth beyond replay to show filter roll off and out of band noise (is 15Hz -> 24kHz enough?, is 88.1 really necessary? is 61 points slow sweep enough?)
  2. No signal noise floor (FFT's at 32K, low average,)
  3. 1kHz tone @ 0dB, -1dB, -60dB FS. THD + Noise (1 graph with overlapped results?)
  4. 48kHz J test. USB and spdif input (if available) (FFT's at 32K, low average)
  5. Small signal time waveform 16 bit and 24 bit @ -90 something dB (is 24 bit necessary?)
Could we standardize all tests to either 16bit 44.1kHz or 24bit 96kHz?
Based on Amir's test , can we also standardize the graphs to dBv vs freq? however we would need a reference voltage @ 0dB
 
OP
March Audio

March Audio

Major Contributor
Manufacturer
Joined
Mar 1, 2016
Messages
6,382
Likes
8,809
Location
Albany Western Australia
Thread Starter #20
Hi Joamit,

Yes, I have been waiting for my new QuantAsylum QA401 analyser to arrive (which it did today) before commenting further. Ill play a bit more with it and take a set of measurements of my explorer2 before sending it to Amir to correlate results with his AP.

If I get time tomorrow Ill put together another suggestion of tests based on peoples feedback and see if we can come to a concensus.
 
Top Bottom