[araragikoyomi]

Hi guys. So recently I checked an RMAA test for the performance of a dac and I noticed that its stereo crosstalk value was around -57dB and RMAA rated it as "Average". So, what is the worst value of stereo crosstalk that is considered "acceptable"? Does stereo crosstalk degrade audio quality by a ton? Also, if the stereo crosstalk of a dac is say, 60 dB, the stereo crosstalk for the amp connected to it is 80dB and the stereo crosstalk for the headphone is 90dB, how do you know what is the final value of the stereo crosstalk?

 

[jcx]

Quote:

I assume Nick knows most of this but for general education purposes:
 
crosstalk is a easily measured # that apparenty has little audible significance at levels easily achieved by DACs: world class phono cartridges often just have "better than 30 dB" spec for crosstalk, even closed back circumaural cans have leakage at -50-60 dB levels, electronics can easily reach/exceed 100 dB channel isolation/crosstalk - TRS common gnd contact ~3-10 mOhm series R gives bigger crosstalk with low Z cans
 
channel imbalance in DAC chips can be ~1% internally, few audio equipment use better than 1% tolerance resistors anywhere in the signal chain - so yes this number could be trimmed but it will normally be as good as any other audio equipement - touted HD650 driver matching is +/-1 dB ~ 10%
 
phase error is a "time domain" error - analog anti-imaging filters are more variable than the DAC's internal digital filtering and the analog filter design/implementaion is easier with oversampling DACs - there are claims of "linear phase" FIR digital filter pre-ringing being a cause of audible differences between higher sample rate systems and RedBook - but convincing, replicated DBT evidence is lacking
 
in loudspeaker crossovers Linkwitz-Riley LR4 with driver phase inversion are widely accepted and have ms group delay nonlinearity - as long as phase "errors" are small, matched in L,R channels it is another fairly low audibility isssue - in DACs the R/L differential phase errors can be sub us

 
 
Quote:

Originally Posted by jcx /img/forum/go_quote.gif

once again I'll point out that the crosstalk # is BS - to get to this level the measurement has to have been made on the pcb and/or without a headphone load - not at the headphone drivers, outside the amp where it counts

3-pin TRS common gnd connection to even 300 Ohm cans will not exceed ~80-90 dB separation due to the 10 mOhm common L/R gnd contact resistance

Neutrik 1/4" TRS Ag contact:
"
Electrical
Contact resistance < 6 mΩ (initial)
Dielectric strength 1 kVdc
Insulation resistance ≥ 1 GΩ @ 500 V dc
Rated current per contact 10 A
"
I have measured >10 mOhm common gnd + contact R with the Neutrik and <2" of wire to the panel

with 32 Ohm cans the the crosstalk wouldn't be better than ~70 dB

 
 
Quote:

Originally Posted by jcx /img/forum/go_quote.gif

no, I'm claiming a 20-40 dB improvement on -60 dB linear crosstalk is not audibly meaningful - open cans have higher natural air path crosstalk - coiled headphone cable has higher crosstalk - no live event recording can have less than that coupling between microphones

only totally synthetic/processed studio tracks can have huge R/L channel separation # - even there "stereo image" is painted on by the mastering engineer - nearly universally by pan-potting - mixing the individual instrument stubs in linear ratios to R/L channels

high end audiophile phono carts typically give up at 30-35 dB channel separation specs - and we All Know Vinyl is Superior no?

 

[jaddie]

The FCC minimum channel separation/crosstalk spec for FM Stereo used to be 29.7dB...yes, that's right, 29.7.  It had to do with how the signal was generated and handled, but 30 - 40dB wasn't hard to achieve, and 50dB wasn't uncommon.  
 
The bulk of what is perceived as stereo separation happens above 20dB with decreasing detectable improvements above 30dB or so. It's almost impossible to detect separation improvements above 40dB.  Localization of a phantom image depends less on channel separation and much more on relative intensity and inter-aural time delay of the sound, and human hearing response at different angles. 
 
So....long answer...separation above 40dB doesn't improve sound quality, below 40dB it slowly degrades, the final separation is equal to the device with the least separation in the system. Once degraded by a device, no device following it can restore separation.

 

[araragikoyomi]

Thanks for the answers I guess I won't have to worry about stereo crosstalk unless it's greater (closer to 0) than -40dB. So, does that mean that a DAC with ~ -45dB can still be considered "high-end"?

 

[jaddie]

Quote:

... So, does that mean that a DAC with ~ -45dB can still be considered "high-end"?

Well, that's a completely different question!  
 
"High End" means many things to many people.  But in general it means "exotic" in some way, could be price, build quality, or some unquantifiable aesthetic characteristic, real or imagined, that places that equipment into that class.  I don't think you can ever define "High End" by performance specs alone, though.  Some so-called high-end equipment has far from the best specifications.  
 
So, I don't know.  If it costs $100 and has great performance specs, it might be good,  but not considered high-end. If it costs $5000, and is in a cabinet made from African Grenadilla wood with quartz knobs and 24K gold plated contacts, but has questionable performance specs, it might be high end.  Or you might improve it with a generous application of a pre-emergent, inferral mixture of Lorsban with Atrazine in a tank mix.
 
Just not my call to say.

 

[jcx]

-57 dB crosstalk is likely not audible if linear, but certainly a technical warning sign - I don't know how to get that poor a number with modern monolithic DAC
 
even the mid '80's Phillips TDA1541 with "legendary" audiophile cred gives 80 dB channel separation spec for the chip
 
some tweak audiophile favored circuits can have next to no PSRR - maybe the channels talk that way
 
 
but it could be instrumentation or setup error as well

 

[mikeaj]

Probably instrumentation or setup, considering how most RMAA tests are run.  Not just DAC but ADC, maybe grounding issues.

 

[stv014]

Quote:

Probably instrumentation or setup, considering how most RMAA tests are run.  Not just DAC but ADC, maybe grounding issues.

 
RMAA also consistently reports the crosstalk to be better than it really is by 6 dB, so -57 dB crosstalk in RMAA is in fact -51 dB. Grounding is a common problem as well, due to the lack of differential inputs on sound cards, so the loopback tests often create ground loops. But it is also possible that a common ground wire had too high impedance in the test because of a bad/faulty cable, contact problems, etc. I have seen impedance as high as 5 Ω with some poor cables and connectors, and sound cards can have an input impedance of less than 5 kΩ.

 

[araragikoyomi]

Okay, maybe "high-end" was a wrong word. I meant something like a decent/entry level audiophile DAC. Yes, I know that ~57dB is low compared to many other DACs nowadays. But if i'm not wrong, the sansa clip (not sure if it's the plus) has a stereo crosstalk value of just around -44dB? Actually, I wasn't the one who carried out the RMAA test; I found the values online. I do not have the knowledge and equipment to carry out an RMAA test on my own yet

EDIT: Sorry, I searched again and the rmaa results for the clip was ~ -80dB without load and -51dB with a 16ohm load.

 

[jcx]

DAP do see low Z iem, 1/8" TRS have much higher common gnd contact R than the 1/4" in my example above - so its possible that the contact R, low Z load explains all of the crosstalk for the Clip

 

[jaddie]

Quote:

Okay, maybe "high-end" was a wrong word. I meant something like a decent/entry level audiophile DAC. Yes, I know that ~57dB is low compared to many other DACs nowadays. But if i'm not wrong, the sansa clip (not sure if it's the plus) has a stereo crosstalk value of just around -44dB? Actually, I wasn't the one who carried out the RMAA test; I found the values online. I do not have the knowledge and equipment to carry out an RMAA test on my own yet

EDIT: Sorry, I searched again and the rmaa results for the clip was ~ -80dB without load and -51dB with a 16ohm load.

As jcx said, could be load related high ground resistance.  Or, as others have said, could be a problem with the measurement setup.  The mistake here is trying to equate sound quality with a crosstalk figure.  Of all the performance parameters that would seem out of spec, crosstalk has the least impact on sound quality, as i said, until you get below 40dB or less.  How about noise and distortion, both intermod and THD?  Linearity? Flat response should be pretty easy these days, but what does it do under a load, and are we driving headphones with it?  If so, how about the output impedance?  Point is, I'd let go of the crosstalk as a figure of merit, and look at the entire picture, along with how it's made.  A good part of the value in an equipment purchase is physical design and life expectancy.  

 

[araragikoyomi]

Thanks for all your replies, I now understand a lot more about stereo crosstalk And it looks like I won't have to worry too much about stereo crosstalk in the future.