Why is there divergent opinions here from gearslutz?

[Nickerz]

When I'm looking for an ADC I see the RME ADI-2 PRO repped pretty hard, but it's a pretty middling lowing performer at the gearslutz loopback tests. The Apollo x16 did poorly as well, assuming that was poorly measured, the x4 performs on par with units that were released nearly 20 years ago like old Motus. And little mention of the TI units here which are at mastering level appearantly?

Long story short, I'm trying to find the most cost effective ADC and having trouble figuring my way through this, because what should be pretty straight forward doesn't seem to be so. I'm not looking for THE ABSOLUTE BEST, but instead the best value proposition given that I DO NOT NEED MIC PRES FROM MY INTERFACE and I don't need many channels.

My understanding so far is this.

For me, a TI kit seems to be basically on par with a Blue 245.

 

[pozz]

That GS thread features a deficient null testing procedure that doesn't take into account filter behaviour, particularly phase, at the highest frequencies.

By TI units you're referring to the chips they use? Chips aren't as important as the whole package, which is mostly analog design.

For ADCs, you need to know the max level you're going to be sending into the device, and choose based on that as well as THD, SNR etc.

 

[Blumlein 88]

Null testing turns out not to be quite so simple as hoped for even when done correctly.

pkane is a member here and wrote Deltawave for null testing.
https://deltaw.org

First off the raw number his software gets is effectively the same as what GS is getting in that thread. Paul's software can do more, for instance one option is to EQ and adjust phase of the comparison file. When you do this the nulls get much deeper for most gear. Now that doesn't mean the gear doesn't muck up phase and level a bit, but it indicates that is all that is happening. Usually this effects the frequency extremes. As we are very insensitive to phase above 2500 hz, the fact a bit of gear is off with phase at the treble end ruins the null, but it would sound quite fine to us. Paul has added a PK metric that does some response weighting and sensitivity at different frequencies into account. It appears to give a much better number for comparing which DAC is really good to our hearing.

You can download his software and download the test files from Gearslutz and see what results you get on the RME with his metric.

 

[mansr]

By TI units you're referring to the chips they use?

Probably the PCM4222 demo board: https://www.ti.com/tool/PCM4222EVM

 

[manisandher]

I can get nulls >80dB with my RME ADI-2 Pro FS R using DeltaWave. That'd put it pretty much at the top of the GS list. But it gets a null of only 53dB using the GS approach. The RME into my Prism AD124 gets an even deeper null when DeltaWave corrects for clock drift.

I wouldn't go solely by the GS list, but still think it's a useful reference.

Mani.

 

[Nickerz]

That GS thread features a deficient null testing procedure that doesn't take into account filter behaviour, particularly phase, at the highest frequencies.

By TI units you're referring to the chips they use? Chips aren't as important as the whole package, which is mostly analog design.

For ADCs, you need to know the max level you're going to be sending into the device, and choose based on that as well as THD, SNR etc.

So I'm just trying to run an ISA110LE into my DAW in sort of the best bang for buck with the best conversion since I'm going to use real mic pres I don't need to worry about their pres. It looks like it's something like either the 828mkii or the PCM4222.

 

[Nickerz]

Null testing turns out not to be quite so simple as hoped for even when done correctly.

pkane is a member here and wrote Deltawave for null testing.
https://deltaw.org

First off the raw number his software gets is effectively the same as what GS is getting in that thread. Paul's software can do more, for instance one option is to EQ and adjust phase of the comparison file. When you do this the nulls get much deeper for most gear. Now that doesn't mean the gear doesn't muck up phase and level a bit, but it indicates that is all that is happening. Usually this effects the frequency extremes. As we are very insensitive to phase above 2500 hz, the fact a bit of gear is off with phase at the treble end ruins the null, but it would sound quite fine to us. Paul has added a PK metric that does some response weighting and sensitivity at different frequencies into account. It appears to give a much better number for comparing which DAC is really good to our hearing.

You can download his software and download the test files from Gearslutz and see what results you get on the RME with his metric.

Another way to think about this might be something like.... are there interfaces that are topping both? Would the TI kit be a nobrainer?

 

[pkane]

When I'm looking for an ADC I see the RME ADI-2 PRO repped pretty hard, but it's a pretty middling lowing performer at the gearslutz loopback tests. The Apollo x16 did poorly as well, assuming that was poorly measured, the x4 performs on par with units that were released nearly 20 years ago like old Motus. And little mention of the TI units here which are at mastering level appearantly?

Long story short, I'm trying to find the most cost effective ADC and having trouble figuring my way through this, because what should be pretty straight forward doesn't seem to be so. I'm not looking for THE ABSOLUTE BEST, but instead the best value proposition given that I DO NOT NEED MIC PRES FROM MY INTERFACE and I don't need many channels.

My understanding so far is this.

For me, a TI kit seems to be basically on par with a Blue 245.

Here's the result for ADI-2 Pro when using DeltaWave set to correct for phase differences (which are not large):

When not adjusting for phase differences, the result is this (which is exactly the same as the Gearslutz value):

Now, here's the PK Metric result (error signal adjusted for audio frequency and masking perception):

 

[Blumlein 88]

Another way to think about this might be something like.... are there interfaces that are topping both? Would the TI kit be a nobrainer?

Well, if you took the time you could get the entire list of files from the GS thread and run them in Deltawave to see.

I do think the TI kit is probably a no brainer if it has the features you need.

That GS list I would say the top tier of results is from -54 db to -60 db for ADC/DAC doing loopbacks. There is one exception (the Eventide). I have a Zen Tour which ranks a little higher than the RME. I know what hurts it is there being a very slight ripple in the treble response from the filter in use. If you fix that with Paul's software it drops way down just like Paul's example above. I'm pretty sure all the top tier will do that.

I'm surprised the UA devices rank so low. Maybe someone needs to download the files and see why.

 

[Blumlein 88]

@pkane been meaning to ask you this for some time.

I've always noted the very best results are for separate converters with the ADC being the master clock. It has always made me think maybe the software doesn't quite fix timing and alignment perfectly. And this kind of clocking is better. My theory being when in loopback, you have the same clock doing DAC and ADC (or I think you do maybe one slaves to the other internally), but there is a time delay from going out over the analog wire and back to ADC. With the ADC the master it sends the clock over wire which times the DAC and any delays are closer so the timing lines up more perfectly than even in an interface loopback. The delay in cabling for the clock being similar to the delay in signal going from DAC to ADC.

Anyway, do you have any thoughts about why the results of the good units using an ADC clock for the master are so much better? An SPL Madison is -48 db in loopback, but -79 db feeding a Focusrite blue 245 with the 245 as master clock. It is -74 db when feeding a TI evaluation board with the TI the master clock.

 

[pkane]

@pkane been meaning to ask you this for some time.

I've always noted the very best results are for separate converters with the ADC being the master clock. It has always made me think maybe the software doesn't quite fix timing and alignment perfectly. And this kind of clocking is better. My theory being when in loopback, you have the same clock doing DAC and ADC (or I think you do maybe one slaves to the other internally), but there is a time delay from going out over the analog wire and back to ADC. With the ADC the master it sends the clock over wire which times the DAC and any delays are closer so the timing lines up more perfectly than even in an interface loopback. The delay in cabling for the clock being similar to the delay in signal going from DAC to ADC.

Anyway, do you have any thoughts about why the results of the good units using an ADC clock for the master are so much better?

It's not just clock drift that causes this, but also clock variations, such as simple periodic jitter (e.g., due to 60Hz mains frequency) that are different between the two converters. When both converters are driven from the same clock, these are automatically sync'ed and there are no timing errors. This is also why DeltaWave phase EQ can improve the result so much over the Gearslutz RMS value.

 

[mansr]

My theory being when in loopback, you have the same clock doing DAC and ADC (or I think you do maybe one slaves to the other internally), but there is a time delay from going out over the analog wire and back to ADC. With the ADC the master it sends the clock over wire which times the DAC and any delays are closer so the timing lines up more perfectly than even in an interface loopback. The delay in cabling for the clock being similar to the delay in signal going from DAC to ADC.

The signal propagation times involved here are insignificant compared to the sample interval, so I doubt that's the explanation.

 

[pkane]

Well, if you took the time you could get the entire list of files from the GS thread and run them in Deltawave to see.

I do think the TI kit is probably a no brainer if it has the features you need.

That GS list I would say the top tier of results is from -54 db to -60 db for ADC/DAC doing loopbacks. There is one exception (the Eventide). I have a Zen Tour which ranks a little higher than the RME. I know what hurts it is there being a very slight ripple in the treble response from the filter in use. If you fix that with Paul's software it drops way down just like Paul's example above. I'm pretty sure all the top tier will do that.

I'm surprised the UA devices rank so low. Maybe someone needs to download the files and see why.

Here are a few of UAD results comparison (GS=Gearslutz, DW=DeltaWave):

I wrote a short script to download files from GearSlutz and remeasure them using DW. It's not perfect, but works for about 125 of the converters

 

[Nickerz]

Here are a few of UAD results comparison (GS=Gearslutz, DW=DeltaWave):
View attachment 118275

I wrote a short script to download files from GearSlutz and remeasure them using DW. It's not perfect, but works for about 125 of the converters

So from what I gather the DW metric is a sort of corrected figure. So things can get better, but not generally worse. So one should expect the TI numbers to be even higher then I assume?

Where does the 828mkii place on the DW scale?

And how could one assertain which is right or wrong in terms of measurement?

 

[pkane]

So from what I gather the DW metric is a sort of corrected figure. So things can get better, but not generally worse. So one should expect the TI numbers to be even higher then I assume?

Where does the 828mkii place on the DW scale?

And how could one assertain which is right or wrong in terms of measurement?

PK Metric is the same as the simple RMS error value, except corrected for audibility and perception of frequencies. For example, frequencies above 20KHz will be weighted a lot less than frequencies around 3KHz. This is using industry-standard equal loudness curves.

Yes, with PK Metric the results will be generally lower dBFS error values than with pure RMS metric, like those on Gearslutz.

As far as I know, TI evaluation board is just an ADC and doesn't contain a DAC, so I don't know how you'd get the result just for the TI board from the Gearslutz thread, considering both are required.

Here's the result for Motu 828mk2:

DW error adjusted for phase is -81.2dBFS, and DW PK Metric (perceptual error) is -88.8dBFS

 

[win]

For an adc, I would recommend the venu360.

Has 6 analog out channels, and three analog or digital xlr inputs.

 

[AnalogSteph]

As-is, the GS null testing algorithm seems to be so poor that you're probably better off studying the specs, along with datasheets of converter chips used (particularly for digital filter performance).

I mean c'mon, there's Dell Optiplex 755 onboard audio (front out - rear in, chip's an AD1984) sitting in the middle of pro and semi-pro interfaces! This is ridiculous. You know what the chip specs of an AD1984 are? DAC DR 96 dB(A), THD+N -86 dB, ADC DR 90 dB(A), THD+N -81 dB! That's like 89 dB(A) worth of dynamic range in loopback. Oh, and the EMU 1212M fares worse. The 1212M with AK5394 ADC and CS4398 DAC that RMAA says gets 110-114 dB(A) DR and 0.0009% (-101 dB) THD in loopback. Yeah right.

It really does help if your performance metrics aren't complete garbage, you know?

Let's hope @pkane can straighten this out, but I still feel these null tests aren't that informative compared to more traditional engineering metrics. You can at best tell that there is a difference, but if it's not obviously pure noise you'll never know what it's caused by (frequency response? phase response? distortion?).

OP, is there anything known about the maximum dynamic range you can get through one of these ISA 110 LEs? I did find the schematic for the input stage and was able to estimate its EIN from that (about -131 dBu unweighted, 150 ohm, max gain).

I suspect any interface or ADC that you can get a good honest 110 dB(A) and change of dynamic range into will generally do in practice (studio). If latency is not a concern, it could even be quite old... this requirement has not exactly made anti-alias filter performance any better. Chip wise, my favorites would be PCM4220/4222, AK5394(A) and AK5393 followed by CS5396, PCM4202/4204, AK5385A/B, CS5394 and AK4620A/B (if used with differential input), maybe even the old PCM1804... all of these can generally be trusted down to 44.1 kHz. If sample rate is going to be at least 48 kHz or even 96 kHz, CS5381 (higher-end) and CS5361 (solid midrange) are not at all slouches either. The CS4272 ADC is about on par with the latter, though whether an interface actually uses it to its full potential and supports the levels that may be required here is another matter. At 96 kHz AK5388 would also be a good option (it's quite a high-performance part with a so-so digital filter), or even the old AK4528 codec that the 828MkII uses (the MOTU seems to have 8 of them so must be using two ADC channels in parallel in order to get dynamic range over the 110 dB(A) mark).

 

[KSTR]

I can get nulls >80dB with my RME ADI-2 Pro FS R using DeltaWave. That'd put it pretty much at the top of the GS list. But it gets a null of only 53dB using the GS approach. The RME into my Prism AD124 gets an even deeper null when DeltaWave corrects for clock drift.

I wouldn't go solely by the GS list, but still think it's a useful reference.

Mani.

@Nickerz, the actual distortion-limited number for the ADI-2 Pro in loopback is more like 110dB++. You can listen to the residual distortion (already in the noise floor) here, note the 90dB boost on the file and that dynamic gain drifts / envelope modulation (which is irrelevant) still dominates the distortion.

 

[KSTR]

As-is, the GS null testing algorithm seems to be so poor that you're probably better off studying the specs, along with datasheets of converter chips used (particularly for digital filter performance)

It is really a pity when someone puts great efforts in producing tons of numbers which are all moot because the procedure to obtain them is thoroughly flawed. Those wrong and useless numbers are impressed into the internet forever, meh.

 

[Wombat]

Maybe we are not Slutz.