I love the comments. E. Brad Meyer's listening tests have been "totally" debunked, link please? But then he offers his own listening tests where 192/24 files are down sampled to 44.1/16 and then subjected to another sample rate conversion/dithering back up to 192/24 just to "prevent cheating". The irony of it all, you call this science?
Then there's the evolutionary survival characteristics that make us sensitive to the Gibbs effect in audio. You can't make this stuff up.
Sometimes though I DO think they make it up. He doesn't sound THAT slow in the head
Jan
Thank you jcx. This link is good
http://www.analog.com/media/en/training-seminars/tutorials/MT-003.pdf
I have to do a survey.
Swimming in the web sea, I found this (very interesting book)
http://www.amazon.com/Methodology-Digital-Calibration-Circuits-Systems/dp/1402042523
Thank you Scott for explaining.
Re dither exercise, I don’t know how to write theTPDF dither equation for to enter it in Excel.
Here is a math-less introductory paper about dither in audio:
http://www.users.qwest.net/~volt42/cadenzarecording/DitherExplained.pdf
George
Very easy add two uniformly distributed random numbers and scale and offset the result. If your random returns a 0 to 1 value just add two of them and -1. the result. I got the same numbers as your table within fractions of a dB from my numerical sim.
Last edited:
Sometimes though I DO think they make it up. He doesn't sound THAT slow in the head
Jan
Yes I gather they make up things based on what feels right hoping some future researcher will verify it.
or you can subtract your previous random number from the current to get a frequency shaped "high pass" TPDF noise for dithering - page 14 of http://downloads.izotope.com/guides/izotope-dithering-with-ozone.pdf I linked earlier
Diff current outputs i can use with a clear conscience. An added inverter.... no.
THx-RNMarsh
the digital theories are nice to know... the basics. necessary, in fact, to circuit design of ADC/DAC et al. As well as knowing SS Physics, IC mfr processes etc.
BUT, I am most interested in the Application of the IC's and final results matters a lot more in that case. Thus, we need to know the Standards to which we can do intelligent comparisons. The Standard for Dynamic range is NOT the arbitrary 0FS to random noise floor level.... nor a theoretical level based on bits and structure. It is based on a measurement.
The three Standards bodies I brought up (AES, IASA, EIAJ) all have a practical app oriented definition based on spurious signal artifacts included and defined levels - at least for consumer electronics apps. It is a lot closer to what you have for an IC but also for the audio system which we listen to.
Again, it is the peak levels of the THD/noise/spurious signal levels present that is the noise floor. here is an example of the range including harmonics and other spurious signals at the systems' output. Of course, it includes more than the IC, alone when we measure the product.
View attachment SFdynamic range.pdf
Numbers, we use in audio, instead of the mfr 0FS number could be normalized to these max levels:
And tiny bit where and why we are into 24bit now: Based upon improved knowledge of human hearing;
View attachment The Relationship of Data Word Size to Dynamic Range and Signal Quality in Digital Audio Processi.pdf
THx-RNMarsh
BUT, I am most interested in the Application of the IC's and final results matters a lot more in that case. Thus, we need to know the Standards to which we can do intelligent comparisons. The Standard for Dynamic range is NOT the arbitrary 0FS to random noise floor level.... nor a theoretical level based on bits and structure. It is based on a measurement.
The three Standards bodies I brought up (AES, IASA, EIAJ) all have a practical app oriented definition based on spurious signal artifacts included and defined levels - at least for consumer electronics apps. It is a lot closer to what you have for an IC but also for the audio system which we listen to.
Again, it is the peak levels of the THD/noise/spurious signal levels present that is the noise floor. here is an example of the range including harmonics and other spurious signals at the systems' output. Of course, it includes more than the IC, alone when we measure the product.
View attachment SFdynamic range.pdf
Numbers, we use in audio, instead of the mfr 0FS number could be normalized to these max levels:
And tiny bit where and why we are into 24bit now: Based upon improved knowledge of human hearing;
View attachment The Relationship of Data Word Size to Dynamic Range and Signal Quality in Digital Audio Processi.pdf
THx-RNMarsh
Last edited:
Where did you get that chart? That -10 dBV consumer reference level is rather antiquated, don't you think? Red book CD gave us 2 VRMS, which is higher than the 1.23 VRMS pro level. And where does that -86 dB noise floor figure come from? Magnetic tape or something?
se
Scott
I am happy that my simplistic formula output results (I had rounded them to a single decimal point) compared favorably to your involved numerical simulation.
What I did for to calculate the undithered DR was to divide the Full Scale Output (V) by the LSB (V), result expressed in dB
What I did for to calculate the dithered (1 LSB) DR was to perform the same division but this time I had first reduced the DAC’s number of bits by ½ bit, (e.g 23.5 bits instead of 24 bits used for the undithered case)
Here is the small .xls sheet ( enter Bit no into cell P10, read the DR from P12):
https://www.dropbox.com/s/g6m8p215xj6rll4/Some%20useful%20formulae.xls?dl=0
Thank you and jcx for the help.
George
Last edited:
Where did you get that chart? That -10 dBV consumer reference level is rather antiquated, don't you think? Red book CD gave us 2 VRMS, which is higher than the 1.23 VRMS pro level. And where does that -86 dB noise floor figure come from? Magnetic tape or something?
se
It just popped up somewhere to make a point only....... looks for analog days...... use the CD standard of today: Normalize to 2v. The example IC Headline spec given a few pages ago had a spec based on 3v (0FS).
I think the approx -86 is similar to others comments that 16 bit in practice measures closer to -90dB. Normalize to one level and Not to IC mfr 0FS level. use the audio standards and not the Headlined numbers used for marketing ADC/DAC IC's.
-RM
Last edited:
Some good info about listening testing and results are from F.Toole (retired) and Sean Olive. Sean Olive has his own blog called Audio Musings. Latest is "Some New Evidence that generation Y May Prefer Accurate sound reproduction". but he is a good person to ask about hearing/listening/testing and what is and isnt audible re. bits etal.
THx-RNmarsh
THx-RNmarsh
yep.
Do the conversions needed to normalize. Then, measure all the artifacts combined as peak levels.... to get your dynamic range which to compare against established range needed per the researchers. You are looking for (and measuring) the Spurious Free dynamic range. Talk to Sean Olive if you need help with the latest info on listening/hearing for dynamic range required. But, for a number--- 118dB has been established a few years ago.
IMO, Dithering doesnt help much when you use the Standards and hearing models for dynamic range required.
THx-RNMarsh
Last edited:
Apparently Technics went to some effort at the time to precisely measure the phase alignments, the PR phrase was Linear Phase - so on axis was probably close to correct.
Yes, experiments bear doing, like very deliberately positioning drivers such that physical anomalies do their best to create acoustic artifacts, and see what is heard. Unfortunately, those speakers have been idle for a while, are not set up to go - a bit of work to restore a functioning system ... down the track, perhaps ...
- Status
- Not open for further replies.