The Ultimate Converter DA/AD Loopback Shootout Thread!

[CD_Audio]

Originally Posted by panatrope

As I said, the sound cards have a filter in them in front of the A-D and after the D-A. The filters may be better ...

when a filter is better you get a lot of quality,
otherwise the bass frequency ... vaporize ...

would also speak about the analog slew rate
but then we go off topic.

[nms]

good morning gents..

@pantrope - It states in the AES paper that it works by upsampling the audio x4 and then aligning the two audio files at that high resolution. Correlation depth is different than RMS because the algorithm is designed to be less swayed by uninteresting differences (why it's not able to be thrown by noise floors). The difference file measurements in RMS are of course based on whatever is left in the difference file, noise floor and all.

I don't have time to get into it all further but feel free to try testing on your own, researching, or reading through the thread. It's good if we can keep the thread less cluttered and not waste too much time thinking aloud or going in circles covering the same stuff.

[CD_Audio]

Originally Posted by nms

good morning gents..

.

good morning
don't upload your square wave test ?

.....

EL8X Distressor

[Antoine Fabi]

nms

How does the FF800 perform ?

I hear a more solid bass, but i'm curious bout the facts

[ssiicckkoo]

I feel like the difference files you posted way back were the most revealing. Is there a chance that we could get the same difference files uploaded for the rest of the tested converters?

BTW, thanks for posting the ERGO results! Do the benefits of the room correction outweigh the transparency loss in conversion for you?

[nms]

Not anymore. I bought a Lavry DA10 within a couple days of going over those results. The sawtooth and sine wave results looked worse than any unit and I was expecting it to place last in the rankings. I was surprised it placed even that high. My Ergo has only worked well with my VXT6's. On my Focal Twins they try to force a 9db boost at 32hz which makes the EQ curve unusable by my standard. Using my measurement mic I realized there was more 32hz energy in the room if I move the mic slightly further from the mix position by 6", so I'm going to try a calibration from there just to see how much that reduces the problem. I think the Lavry will give me a better result though and I bet I can create a better low end EQ curve for L & R channels myself and save it as an EQ I can insert at end of my chain in my daw for now.
With better monitors and a more well treated room I feel the ratio of harm to good has shifted. I'll still keep it for future use as it'd be good to take for a portable setup touring and or as part of a temporary minimal setup where you don't have a proper room.

I'll try to make time to compile the full collection of difference files today and upload them.

[ssiicckkoo]

I think I'll be going with IK's ARC System to compliment my Dynaudio BM5a MKIIs. The ERGO scares me now.

I am leaning towards the Konnekt 48 w/remote over anything Motu though. I just have a hard time trusting a company that would take two (or more) steps back with a MKIII after such an arguably superior MKII. I really want to, but I can't come up with any GOOD excuses for them. And the TC interface fairs well in the race as far as I can tell.

[ipbrjt3]

This whole thread is so cool...waaay over my head, but extremely educational. I'm wondering if there are any tests for the 896 series in the works??

[Drumnbass411]

theres two 828mk2's on ebay!!!...go buy them nowwwww!!!

[nms]

Originally Posted by ssiicckkoo

I am leaning towards the Konnekt 48 w/remote over anything Motu though. I just have a hard time trusting a company that would take two (or more) steps back with a MKIII after such an arguably superior MKII. I really want to, but I can't come up with any GOOD excuses for them. And the TC interface fairs well in the race as far as I can tell.

the TC is a great choice. As for companies adding features to make the feature sheet pretty while losing transparency, same thing from the FF800 to UFX. They're a business though.. It's all about what sells and markets well and turns a profit. There's a bigger market for what they do than a unit with the same price, less features, but better sound.

@ipbrjt3 - I don't think any of the other Motus will test as cleanly but when someone can get me files for one I'll add it.

[panatrope]

Originally Posted by nms

good morning gents..

@pantrope - It states in the AES paper that it works by upsampling the audio x4 and then aligning the two audio files at that high resolution. Correlation depth is different than RMS because the algorithm is designed to be less swayed by uninteresting differences (why it's not able to be thrown by noise floors). The difference file measurements in RMS are of course based on whatever is left in the difference file, noise floor and all.

I don't have time to get into it all further but feel free to try testing on your own, researching, or reading through the thread. It's good if we can keep the thread less cluttered and not waste too much time thinking aloud or going in circles covering the same stuff.

It is actually good to discuss the measurement method, because then the conclusions that you make can be justified. As the time and response alignment is crucial to the result, then maybe we need to understand it. For example, at orignal sampling rate the top frequency of half the sampling rate (Nyquist frequency) can only be aligned to an average phase accuracy of 90 degrees, and a quarter of the sampling rate to an average of 45 degrees. I postulate that the top octave of the converted signal is probably the most revealing in terms of convertor performance, especially in terms of Delta Sigma processes which most modern convertors use. Assuming the upsampling is accurate - an exact multiple is synchronous and usually is - the resolution is four times better but still not exact. (See the paper for the analysis of depth of null versus amplitude and phase difference.) So maybe timing is important and maybe a 64x upsampling might be capable of better accuracy. I don't know - I didn't write the program.

Likewise where correlation differs from difference RMS depends on the method used for correlation. Do you understand how correlation difference is calculated in this program, or do you take the author at his word that it is "better"? And the assertion that "uninteresting" differences are discarded (I assume meaning not relevant to the assesment process). Short of seeing the source code for the program, I would have to decompile it to work out what exactly was going on.

With modern conversion processes, differences are very fine. And while a null method can be very good in such cases, one has to be very careful about preserving the nature of the subject and reference signals. It would be intersting if the original author might comment about his creation being used for this particular task - and the impact of multiple passes being used to 'amplify' the differences.

As this seems to have elements of an RME vs MOTU shootout, maybe two devices should be tested in a synchronised manner (ie., fed and clocked by the same digital input, and the results recorded pairwise in that situation). Curiousity also suggests that it might also be interesting to swap the loopback ie., RME analog out to MOTU analog in and vice versa, and maybe even the other two permutations.

Finally, it might be more revealing to devise a synthetic test signal (pseudo-random sequence with defined characeristics), which could be more readily correlated or convolved with the resulting returned signal to more accurately characterise the round trip performance. What do ADC/DAC chipmakers use to examine the dynamic performance of their chips (the ones they don't put in the spec sheets)?

My observation of your current analysis of the results remains 'noted with interest' until there is some explanation of why the results are what they are, and why the ranking based on "correlation depth differs from RMS" . Then the results become more reliable, and useful to readers of this thread who appear to be using its contents to make purchasing decisions.

[didier.brest]

Time alignment is performed in the frequency domain where a time delay t is equivalent to a frequency linear phase shift applied by multiplying each complex value of the Fourier transform at a frequency f by the complex number exp(-j2.pi.ft). The time alignment is done within a few nanoseconds and the gain correction within 0.001 dB.


I compared a 44 kHz recording with the same recording from which I removed the first sample. I got the the following result :
parameters: 22,68usec, 0,000dB (L), 0,000dB (R)..Corr Depth: 300,0 dB (L), 300,0 dB (R)

[CD_Audio]

@NMS
in my FF800 the model of opamp that governs all the analog bus are JRC 4580,
which model of opamp are used in the MOTU ?

Bonis

add

The FF800 have a internal switching power supply...and the MOTU power supply is ?

[gordbort]

Originally Posted by panatrope

As this seems to have elements of an RME vs MOTU shootout, maybe two devices should be tested in a synchronised manner (ie., fed and clocked by the same digital input, and the results recorded pairwise in that situation). Curiousity also suggests that it might also be interesting to swap the loopback ie., RME analog out to MOTU analog in and vice versa, and maybe even the other two permutations.

I'm working on it. I've got a UFX, 828mkII, and 2408mk3. I'm planning on getting to it today.

[Antoine Fabi]

Originally Posted by didier.brest

Time alignment is performed in the frequency domain where a time delay t is equivalent to a frequency linear phase shift applied by multiplying each complex value of the Fourier transform at a frequency f by the complex number exp(-j2.pi.ft). The time alignment is done within a few nanoseconds and the gain correction within 0.001 dB.


I compared a 44 kHz recording with the same recording from which I removed the first sample. I got the the following result :
parameters: 22,68usec, 0,000dB (L), 0,000dB (R)..Corr Depth: 300,0 dB (L), 300,0 dB (R)

Oh!

Thanks a lot for the info Didier!

That proves that the method is reliable.

Cheers!

Antoine

[rmedek]

For what it's worth…

I own the Traveler tested here. The last project I worked on I recorded vocals using a U67 through its onboard preamp, and a KSM44 through a MA5 through various conversion. The U67 vocals sound amazing compared to the KSM44, and the MA5 tracks sound amazing compared to the KSM44 tracks through other pres.

Which is to say I think the MOTU preamps are nothing to write home about, but are perfectly serviceable and work well. They do get a little noisy when you push the gain, I never had any luck using them with ribbons.

Just saying I don't think anyone with a MOTU interface should be thinking, "dang now I NEED to buy preamps because MOTU pres put a cloud of haze blah bla…"

[panatrope]

Originally Posted by didier.brest

Time alignment is performed in the frequency domain where a time delay t is equivalent to a frequency linear phase shift applied by multiplying each complex value of the Fourier transform at a frequency f by the complex number exp(-j2.pi.ft). The time alignment is done within a few nanoseconds and the gain correction within 0.001 dB.


I compared a 44 kHz recording with the same recording from which I removed the first sample. I got the the following result :
parameters: 22,68usec, 0,000dB (L), 0,000dB (R)..Corr Depth: 300,0 dB (L), 300,0 dB (R)

Excellent calibration of time alignment, Didier, in the first instance. However, integer number of samples would be expected to be exact. Non-integer sample number differences are perhaps more meaningful. If you can afford the time and effort, maybe carefully sample rate convert the original file to 96kHz, remove say the first three sample, re-convert to 44K1, and see if the resulting offset is reported as 31.25us. Of course the corr and diff measurements will now re[present the accuracy of the SRC, but that's another story ....

In terms of amplitude resolution, maybe it would be interesting to compare a 24-bit file with the same file truncated to 16 bits (or any smaller resolution that can be implemented). The level difference should be equal to the theoretically expected quantisation distortion, and the correlation depth results would become more meaningful as regards magnitude of impairment. (Using controlled amounts of dither instead of straight truncation would add to the calibration).

There is still some question over the multiple pass methodology. If the program is capable of precision results, why not rely on the results of a single pass.

All very interesting.

[nms]

The reason we did 10 pass was to magnify the effects so we could listen or visually spot them more easily. As well, if there are any glitches 10 passes gives more to average from. The Diffmaker analysis scales in an expected manner though whether testing the 1 or 10 passes. We've done single pass testing in another thread.

The reason I didn't want to get into several posts about how Diffmaker works and whether it does a good job is because anyone can visit the Diffmaker website and read the AES paper. It's generally not disputed that the program does what it does well. I still need to catch up on result content for the thread plus have new results on the way from a couple people and meanwhile work in my studio which takes priority of course... So reading over the thread or visiting the Diffmaker site for answers is reccomended!

I really have no idea about the Motu preamps, but that's one thing I'd assume only got better as newer units were made just as I'd assume the same with UFX. I think that has taken priority for RME and Motu where sound improvement is concerned rather than the conversion itself.

One point I want to make though where the old Motus are concerned, for tracking purposes I highly recommend clocking it from a better clock like a Mytek clock output or Antoine in your case the BLA MC2. I don't find any measurable improvements from external clocking when it comes to loopbacks, but in tracking there is definite improvement.

[gordbort]

I've been working on the MOTU/RME tests. As a sanity check, I've first done UFX by itself and 2408 by itself. I did the 10 passes and then ran just the end music through AudioDiffmaker to see what happened. I got a couple of dB worse on Corr Depth for each, but they were in the ballpark of the previous results reported in this thread.

To nms: do you have a place to upload the files or do I have to find my own server?

[nms]

@gord - if checking stuff yourself you want to run the tests with stereo selected, clock drift correction enabled, & uncheck the box for limiting to only the first few secs. Upload to somewhere like sendspace.com

[CD_Audio]

Originally Posted by nms

The reason we did 10 pass was to magnify the effects so we could listen or visually spot them more easily

The test has very little meaning in the real world.

[nms]

Sure it does. It's just a way of magnifying the effects so it's easier to spot some things, but you do have to remember it's from 10 passes and keep that in perspective. The converters all stay on the same relative playing field. You forget I take 1,4, and 10 pass recordings from everyone for the precise reason that it gives a chance to also measure or listen to results that more closely simulate tracking or a file that's been passed through a few extra times during mixing.

If it was up to me we'd be testing these units by tracking every instrument separately from a higher quality DSD source, precisely matching levels, then comparing a resulting ITB mix of the tracks with an ITB mix of the originals. A guy like Warren from Zen Pro Audio could set that up, but I don't see that happening and being made public with conclusive & condemning evidence due to his business relationships. I imagine some companies would be pissed at me if I was in his position and ran this thread!

Rest assured though, we're nowhere near done testing. Stats measured from a new file will begin in the near future but this time instead of taking 1,4,10 pass I'm going to take 1 & 4 passes at 44 & 88khz which will reveal some significant evidence I think. I'll also add screenshots of the low end phasing measurements.

I'm sure some would find this all a bit excessive, but trying to push the boundaries of sound quality in music production & reproduction is a passion of mine. I know I'm far from being the only one with that goal and I find the amount of hearsay, overstated marketing, and poor testing being done to be pretty tragic and misleading.

[panatrope]

Originally Posted by gordbort

I've been working on the MOTU/RME tests. As a sanity check, I've first done UFX by itself and 2408 by itself. I did the 10 passes and then ran just the end music through AudioDiffmaker to see what happened. I got a couple of dB worse on Corr Depth for each, but they were in the ballpark of the previous results reported in this thread.

My particular reservation about MOTU is that they appear to no longer publish performance specifications for their products, in contrast to other manufacturers in the semi-pro/pro sphere. So any tests cannot be compared with the makers guaranteed performance, and if you think a piece of their equipment is not performing satisfactorily, it is solely their discretion as to whether they think it is faulty or not. If someone can point me in the direction of such specification, I will stand corrected.

No point in testing these devices if you can't test that they are performing to the manufacturer's standard.

(ps. The odd piece of their gear I have tested has performed quite reasonably.)

[nms]

Make no mistake.. no one in this thread is recommending the newer Motu products. I too find that odd about the secretive tech specs, though I wouldn't need them to be able to figure out if my interface is performing well. I'm sure people have great experiences with them, and still they did fare a lot better than many of the units.. but they're not for me and there are certainly cleaner sounding converters to be had as evident by the test results. Honestly, if you want to get top shelf conversion in an interface you need something like a Metric Halo ULN8 / LIO8 or a Symphony but of course no PC drivers (very lame). Otherwise just grab something with good drivers and connectivity and separate converters. I wonder how the new RME UCX will test?

It'd be awesome if the new UA Apollo had 2192 quality conversion but it sounds like they're going for something with cheaper quality and a bigger market.

[CD_Audio]

Originally Posted by panatrope

in the semi-pro/pro sphere. So any tests cannot be compared with the makers guaranteed performance,

Today, manufacturers have good initial designs, and build the final circuits with components of poor quality, to earn as much possible, rely on ignorance of the mass,
sell equipment at incredible high prices, if we analyze a **** is built with just enough quality Chinese components, even paying a thousand dollars, manufacturers should be compelled by law to write in the specific the model of main components composing it ,and not only the technical characteristics of input and output.

[nms]

Hugely agree. It's not about making the best sounding product, it's about making something that will pass most people's inspection especially when accompanied by some great marketing while keeping the price low enough that your market and profit are large enough.

I mean seriously.. some of these products like the pricing on pcie cards.... WTF? Is the required RME pcie card for a multiface worth $500? Massive rip off. I like the RME hdspe cards but I shudder to imagine what their profit margin and cost per unit is.

Things get seriously wacked out in the arena of interfaces and A/D converters and that's partly because of how hard it's been for people to easily know if they're getting something as great as it's made out to be.

[ronmac]

Originally Posted by nms

I mean seriously.. some of these products like the pricing on pcie cards.... WTF? Is the required RME pcie card for a multiface worth $500? Massive rip off. I like the RME hdspe cards but I shudder to imagine what their profit margin and cost per unit is.

Can't agree with your "Massive rip off" statement.

Providing a product that works well, continuously updating drivers and being extremely responsive to customers is something that costs.

Rock solid product is much more important than a point or two on chart or graph of specs.

[adrianww]

Originally Posted by nms

Hugely agree. It's not about making the best sounding product, it's about making something that will pass most people's inspection especially when accompanied by some great marketing while keeping the price low enough that your market and profit are large enough.

The thing is, "best sounding" will always be a subjective judgement for most, if not all, people. I know where you and others such as didier are coming from in terms of these transparency tests and I think it's great that so many people are putting their time and effort into trying to quantify this stuff. It's important and if I didn't think that I wouldn't have taken the time and trouble to test my LS56 and I wouldn't be planning on running the tests again more carefully just to confirm things.

But transparency is just one aspect of the whole picture. Something can be non-transparent, but sound pleasing nevertheless. Or the non-transparency can, for all practical purposes, be ignored since the effects are pretty much undetectable under normal conditions while other aspects of the interface (amount and type of I/O, driver stability, general compatibility with other equipment, etc.) are more important to the owner or user.

That's kind of the case for me. I'm not planning on getting rid of my LS56 since there isn't really anything on the market that offers all the same facilities as it does at the same price point. I'm quite prepared to accept some inaccuracy or colouration in the analog input and output stages, or even the conversion itself, in return for all the rest. Particularly since any colouration is slight enough (or benign enough) that it doesn't have any major impact on what I'm doing. (OK, if it made everything sound like it was being played back through an ancient Bakelite telephone on a long-distance call during a massive thunderstorm I might be thinking differently, but...)

Originally Posted by nms

I mean seriously.. some of these products like the pricing on pcie cards.... WTF? Is the required RME pcie card for a multiface worth $500? Massive rip off. I like the RME hdspe cards but I shudder to imagine what their profit margin and cost per unit is.

Ah, well here we basically just run up against good old economics. If people are prepared to pay $500 for an RME PCIe card, then that is indeed what it's worth. That's just how it goes. I know of industries where profit margins are measured in terms of a few per cent, while other industries run at 50, 60, 70 per cent or more. From the point of view of the end customer, how much profit the manufacturer (or distributor or retailer) makes on the deal shouldn't be a major part of the decision-making process. What is important is whether you feel that you've got value for your money and whether you end up thinking that something is worth what you paid for it.

Again, thinking of my own case with that LS56, sure I could wish that it came out better in this test and topped the table, but I still don't feel like I've been ripped off or that it doesn't give me great value for what I paid. So I'm more than happy enough with it.

Originally Posted by nms

Things get seriously wacked out in the arena of interfaces and A/D converters and that's partly because of how hard it's been for people to easily know if they're getting something as great as it's made out to be.

I kind of go back to my first point here. Someone could get one of the interfaces or converters that top the table here, thinking they've got the absolute bees knees in terms of what they were looking for, only to discover later that they absolutely hate it for some other reason(s). And, unfortunately, you can only make the final call on something when you've used it and lived with it for a while. There's always trade-offs.

Anyway...I still haven't retested the LS or tested the n12 yet as the company that was supposed to be supplying me with new cables failed to deliver. Once I've got some elsewhere, I'll see about running more tests.

[terence]

I always found my poor old Saffire LE more transparent and less processed sounding for recording (with decent preamps plugged into it) than my Steinberg/Yamaha n12, which is based around MR816 converters. This was against the grain of most other opinions, or what I wanted to believe myself having shelled out good money for it. The latter has now been delegated to live use. I feel slightly vindicated!

[nms]

Well in the case of RME the cost vs value is what's kept me from purchasing a single product of theirs and I do feel absolutely that some if not all of their stuff is overpriced somewhat in relation to the build cost, conversion quality & quality of components used. Let's be realistic here, look at the $500 pcie card for connecting the multiface.. Then look at the build, processing power, and components of say a video card or CPU & mobo that sells for $500 (or $250 for that matter). I almost picked up a multiface 2 a while ago, but didn't for that reason. A lot of people like to make an assumption that the more expensive a product is, the better it must be but it doesn't go like that. I like the multiface 2 and whole fireface line, but it hasn't been worth it to me to spend that on a unit that has converters I won't let near my audio and the only one of those with AES is the UFX.

In comparison, look how much cleaner that TC Konnekt comes up in testing and the amount of connectivity it has for what it costs. That's something easier for me to recommend. I don't like companies who sell their products for a more expensive price just because they can and appear to be doing nothing to push the sound quality to a level that's anywhere in the stratosphere of the best sounding converters out there. A company like Mytek is out there trying to make amazing sounding converters.. Not devoting themself to whatever they can market best while keeping costs down and profit margin up.