| A/D - D/A accuracy tests [message #347676] |
Sun, 08 June 2008 08:25  |
 |
 Matt_G
Messages: 472
Registered: July 2004
Location: Brisbane Australia |
Active Member |
|
|
Ok, I've got some Lynx Aurora's in on demo at the moment & I've been doing the usual listening tests but to me conversion should be about 'accuracy'. Quality conversion should represent the source signal it's playing back or capturing as closely as possible. Therefore you would need someway of measuring the source against the conversion capture. I'm only concerned with the whole round trip not half at this point.
I've taken a clients 24bit - 96kHz mix, fed it out of the Aurora's D/A & then captured straight back into the A/D (set to internal sync) & performed a null test against the the source file. I'm getting cancellation down to -38dbfs peak as a result, the peak levels from source to capture match & the RMS levels are only 0.1db difference so I think this is a fairly good result.
As a comparison I tried the same test on my Cranesong HEDD 192 which has the current D/A board with the older A/D board. I turned off the process switch, set the clock source of the HEDD as master for Pro Tools & repeated the test. Amazingly the results were worse by around 10db, also the peak levels were down -0.6db but the RMS levels were only 0.1db difference.
I would be curious to know what sort of measurements a round trip of Lavry Gold would give by comparison.
Has anyone done this sort of test with their converters or is my test system flawed?
Matt
Matthew Gray Mastering
Brisbane Australia
|
|
|
| Re: A/D - D/A accuracy tests [message #347704 is a reply to message #347676 ] |
Sun, 08 June 2008 10:43  |
Dave Davis
Messages: 375
Registered: December 2005
Location: Cincinnati, Ohio |
Active Member |
|
|
No it's actually a pretty fair test. Using digital sources you can repeat as close to identically as possible is better for measurement than analog sources that are "deterministic" and can be different every playback. The really difficult thing to do is level-matching during the round trip. Analog gain settings of the converter can greatly affect the character of conversion, especially as your peaks approach 0dBFS.
We've done similar tests of our Digi 888/24 and HD192 and Metric Halo and Sonic interfaces, and discovered many interesting things. I'm at home, so the results are not exact but off the top of my head, the Digi HD192s and my 2882 were nearly identical nulling to about -36, the MH/Sonic stuff, in the -40's or better ( recall it being at least 6 dB better than Digi).
I'm not surprised the HEDD is so different. It's function is to add euphonic color, after all! 
Don't forget that the character of the residue is at least as relevant as the magnitude. In other words, if most of the error is hash concentrated in the extremes, high or low, they're of less consequence than more musical broadband errors. In other words, the more music you can hear in the residue, the worse the performance, and one should consider A-weighting results to better scale the audibility of the degradation. So it's not just numbers... I'd expect the HEDD's residues are generative (added, new content, not more or less of what's there) because that's it's nature.
-d-
Dave Davis
MUSIC|MEDIA|DESIGN
|
|
|
|
|
| Re: A/D - D/A accuracy tests [message #347726 is a reply to message #347725 ] |
Sun, 08 June 2008 12:10 |
Bruno Putzeys
Messages: 710
Registered: November 2006
Location: Rotselaar (Belgium) |
Gold Member |
|
|
Linear errors (i.e. gain, frequency&phase response, delay) will dominate the null test unless you first characterise the linear errors and roll these into a loooong FIR filter that you apply to the reference signal (such as to apply the exact same linear error).
Have more faith, Bambi. The only right way of dealing with headlights is staring them down.
|
|
|
| Re: A/D - D/A accuracy tests [message #347751 is a reply to message #347676 ] |
Sun, 08 June 2008 14:33 |
Dave Davis
Messages: 375
Registered: December 2005
Location: Cincinnati, Ohio |
Active Member |
|
|
Agreed, Bruno, but as a gross measure, this method provides a relative measure of fidelity to a source and, as you note, relative linearity vs source. And I think that's what Matt's after.
-d-
Dave Davis
MUSIC|MEDIA|DESIGN
|
|
|
| Re: A/D - D/A accuracy tests [message #347753 is a reply to message #347726 ] |
Sun, 08 June 2008 14:43 |
Kees de Visser
Messages: 79
Registered: April 2007
Location: France |
Member |
|
|
Don't forget that it's rather unlikely that the group delay will be an exact number of sample periods, so optimal nulling might require sub-sample accuracy.
There's an interesting free application for Windows users, DiffMaker (http://www.libinst.com/Audio%20DiffMaker.htm) , which can make tiny timing and level adjustments for better nulling. Highly recommended.
Kees de Visser
Galaxy Classics
|
|
|
|
|
|
|
| Re: A/D - D/A accuracy tests [message #347855 is a reply to message #347676 ] |
Mon, 09 June 2008 06:28 |
Dave Davis
Messages: 375
Registered: December 2005
Location: Cincinnati, Ohio |
Active Member |
|
|
Converters should indeed be all about accuracy in mastering. Euphonics are not the game in conversion (btw: noted that HEDD process was bypassed in Matt's test). The goal MUST be input=output. Anything else is a recipe for for tail-chasing, and excuse for the eventual "new and improved" reissue.
We're closing in on a generation of gear that has no sound of it's own. This is a very good thing, when coupled with sufficient resolution in the digital domain to effectively replicate the "infinite" resolution that analog provides.
Unless one is interested in using converters as processors (not unheard of), input=output is more than a truism, it's a grail.
-d-
Dave Davis
MUSIC|MEDIA|DESIGN
|
|
|
|
|
|
|
Members
Search
FAQ
Register
Login
Home
