Blind testing in audio is a useless, pathetic joke. Of course, that's just my opinion. But I know it's true.
I Don't Know It For a Fact...I Just Know It's True — Real Time with Bill Maher Blog
Last edited:
Db to gain is about 1/3 millionth ! so not worth bothering about.
Possibly more cross talk caused by pcb track routing and component layout.
Possibly more cross talk caused by pcb track routing and component layout.
Yes, "self-talk" (or coupling among nodes of the same circuit), is a problem that is often present,
but seldom recognized.
but seldom recognized.
I have got caught out a couple of times with op amp cross talk.
Ran a op amp output between the pads of an ICL7660 charge pump and 10KHz got added to audio signal nicely !
Ran a op amp output past 8MHz crystal oscillator and that got picked up too.
There is a difference between a layout problem and the internal specs of an IC.
You can always use RF design rules, ground plane, etc., to keep stray C to a minimum.
I used ground planes and more careful routing to get rid of the problem.
The most prone areas are input pins on op amps where the impedance is higher.
See crystal circuit in middle now surrounded with grounded copper pour.
An externally hosted image should be here but it was not working when we last tested it.
Last edited:
Depends what one is talking about. For simple measured single tone hearing sensitivity then, yes, 100dB is a lot.
For distortion (not the subject of this thread), then maybe not so much. For example, it is completely noncontroversial that some people can hear quantizing distortion if CDs are undithered. Since truncation occurs between the 16th and 17th bits, its distortion is down around -96dBFS. Getting down close to -100dB and certainly far lower than 40dB down.
Ray, With all due respect, I would disagree slightly. Blind testing can work reliably when used for gathering statistics provided experimenters are experts in sensory perceptual testing. Of course, I have yet to come across an EE with expertise in that field, except for perhaps Jakob2. For those interested in becoming acquainted with the field, a good reference for study is: "Sensory Discrimination Tests and Measurements - Sensometrics in Sensory Evaluation," by Jian Bi. https://www.wiley.com/en-us/Sensory...nsory+Evaluation,+2nd+Edition-p-9781118733530
That's how crosstalk is measured, but not heard, unless you play the top ten test tones over and over. It measures worse than it's heard in the wild because music is self-masking, dynamic, not a steady state sine wave.
There is a big disconnect here between measurement and audibility with actual music. We can, and regularly do, measure what is completely inaudible.
The scientific study of the audibility of anything requires the collection of a mass of subjective data in means apart from other sensory bias, then statistical analysis. To get that subjective data apart from bias requires controls. Careful double-blind testing is one of those controls. The analysis of good, clean subjective data results in information about what is actually audible and what is not. But it's very different from measured data, which often doesn't represent audibility well at all. To be very clear, this is NOT saying "we cannot measure what we hear", which is a false statement. "We can measure everything we hear" is currently also false, but the list of what cannot be measured and quantified is getting smaller.
But the correct measurement techniques must also be applied to get valid results. For example, many will say that 3% THD is audible and objectionable, but it has been shown to be otherwise in many circumstances because 3% THD doesn't represent a dynamic performance figure, it's steady state, fixed level. Actual music isn't steady state, fixed level, nor is it a single frequency, rather a highly variable spectrum that tends to self-mask harmonically related products generated by nonlinearities. THD figures are, therefore, meaningless in practice other than a general means of testing. It's almost impossible for device nonlinearities to produce a fixed amount of any sort of THD throughout the dynamic range.
Crosstalk with music is also a highly dynamic property that also tends to be self masking. It's been known for many decades that the required stereo channel separation for acceptable stereo presentation is actually quite low for two-channel stereo, and even lower for multi-channel stereo. The original stereo separation figures for FM multiplex (USA) was less than 30dB. Vinyl dips to 20dB at many frequencies, but doesn't suffer from lack of dimensionality. Matrix Quad systems presented believable 4-channel sound fields with as little as 3dB separation between adjacent channels, and rarely more than 10 or 12dB, though non-adjacent channel separation was somewhat better.
Stereo microphone pairs measure as little as 6dB separation at some frequencies, rarely more than 20dB at any frequency.
Lastly, check any HRTF for crosstalk. It's simply massive.
Crosstalk of better than 50dB is perfectly fine for any system.
Way off topic. Nothing at all to do with dual chip crosstalk.
Lets not forget that you posted, "Please see Ioannidis: Why Most Published Research Findings Are False"
How so? Xtalk is a ratio, like -100dB. That means that whatever is in one channel, xtalks to the other at 100dB below that. So the xtalk on the other channel varies just as the main signal in this channel does.
Jan