Are you claiming that the A/B difference of with/without ferrite bead is similar to oboe vs. violin? It seems that you can't make it even in the shallow end so better stop here.
I am claiming that some sounds are transferable from short term auditory memory to long term memory. In that case the A/B time can be extended.
If sounds were not found to be easily transferred to long term memory then we could use other methodology to allow for fast A/B switching. In the case of the ferrite bead, the sound differences were sufficiently memorable for longer time interval A/B switching to suffice.
Also, there should be no need to resort to ad hominin personal attacks. Its always you that starts the bickering.
You can have the last word.
If sounds were not found to be easily transferred to long term memory then we could use other methodology to allow for fast A/B switching. In the case of the ferrite bead, the sound differences were sufficiently memorable for longer time interval A/B switching to suffice.
Also, there should be no need to resort to ad hominin personal attacks. Its always you that starts the bickering.
You can have the last word.
Last edited:
Oh my…
About the thermistor, I would want to have plenty of capacity in that application. Unfortunately I am not the best person to consult regarding switching supplies.
About the thermistor, I would want to have plenty of capacity in that application. Unfortunately I am not the best person to consult regarding switching supplies.
That's possible. But are we talking about DAC jitter? I thought we were talking about jitter on the clock to the DAC. If the DAC has a built-in jitter cleaner or clock/data recovery you'll never see any benefit from providing an ultra-clean clock. The clock just has to be clean enough that the jitter cleaner remains locked. What you see at the output of the jitter cleaner (thus, the internal clock to the DAC) is the phase noise of the VCO within the jitter cleaner. If you have actual data that show the opposite I'm all ears. Just drop a DOI or link to the AES paper.
Actually more like a few, maybe up to 10-15, seconds at the most. That's why it's important to be able to switch from A to B quickly.
Sure. When you're talking about identifying a violin from a saxophone, that's certainly true. But if you're walking about identifying a minuscule detail difference - as is likely with most audio tweaks - I highly doubt you'll have any exact memory of the detail if you go more than 10-15 seconds between trials.
Let's take a different example: Let's say you're able to tell the difference between two shades of red. One has RGB of 255,0,0 the other 253,0,0. That's a noticeable but subtle difference. Do you think if I showed you one of the two reds the next day that you could correctly determine whether it was the "255 red" or the "253 red", say, 19 times out of 20? The difference you're trying to hear in audio is likely even smaller than this.
Tom
I think you may want to look at how human memory systems work. At least I'm not aware of any direct link between the audiotory working memory and long-term memory. Long-term memory contains episodic memories, not semantic ones. The other aspect of this is that long-term memory is notoriously unreliable. You can have a look at Elizabeth Loftus research on eyewitness recall, for example. She built her career on this starting in the mid-1970s. There are tons of other examples, but that one is one of the better ones.
But if on the other hand you do have solid evidence for this I'm all ears.
Tom
Tom,
I am well aware of all the textbook and AES paper claims you are making. In fact you left out a lot details, overstated some claims, etc. I used to believe that stuff at that level too, but gradually I found outs reality is more complicated than your simplistic explanations recognize. Its taken me years to find out all this stuff, and so far as I can tell you are just starting to poke around it it.
For example, you talk about 'jitter elimination' which for the cases you mentioned doesn't exist. As an EE you should try to get it right. Most of what actually exists is 'jitter attenuation.' There is a difference, and if you read ESS datasheets until you understand what DPLL_BANDWIDTH does, then maybe you will notice a little note ESS put on that page that recommends to set it to the lowest stable value. When you know why, come back here and explain it. At that point maybe you will start to understand why jitter elimination is a marketing term, not an engineering one.
After that, maybe over time you will start to figure why Bruno Putzeys didn't use an off the shelf ASRC for Mola Mola dac. It was because in his mind they weren't good enough for his dac.
There isn't enough time to educate you about every last thing you got wrong. Certainly not by writing posts in a forum.
Good day, sir.
I am well aware of all the textbook and AES paper claims you are making. In fact you left out a lot details, overstated some claims, etc. I used to believe that stuff at that level too, but gradually I found outs reality is more complicated than your simplistic explanations recognize. Its taken me years to find out all this stuff, and so far as I can tell you are just starting to poke around it it.
For example, you talk about 'jitter elimination' which for the cases you mentioned doesn't exist. As an EE you should try to get it right. Most of what actually exists is 'jitter attenuation.' There is a difference, and if you read ESS datasheets until you understand what DPLL_BANDWIDTH does, then maybe you will notice a little note ESS put on that page that recommends to set it to the lowest stable value. When you know why, come back here and explain it. At that point maybe you will start to understand why jitter elimination is a marketing term, not an engineering one.
After that, maybe over time you will start to figure why Bruno Putzeys didn't use an off the shelf ASRC for Mola Mola dac. It was because in his mind they weren't good enough for his dac.
There isn't enough time to educate you about every last thing you got wrong. Certainly not by writing posts in a forum.
Good day, sir.
Would you maintain such evidence proves it is not possible for any human to accurately recite 31,811 digits of pi from memory?
Do you remember the year you were born, or is your memory too unreliable?
Obviously there are many cases where memory can be pretty reliable.
Regarding memorizing a sound, go read hhoyt's post again about how he learned how hear small differences in order to pass ABX. How does a music student learn how to transcribe a symphony from listening? What you seem to be missing is that in both cases it takes practice. And its not the same thing at all as eye witness testimony of a multiple car auto accident.
EDIT: Regarding 'miniscule,' that is an assumption on your part. How do you know that the sound of familiar instruments did not change in ways obvious to a musician? How do you know the perceived width of the stereo illusion of soundstage did not become smaller? Do you know how accurately a human can localize the relative angular position of a sound source?
Last edited:
Those are both examples of semantic memory, not auditory memory. Your point?
No I don't. Do you? Do you have hard data on this? If you do I'm all ears. I've been wanting do a writeup on what humans can hear (based on scientific evidence) for years. This would be a great addition. My guess would be, "not very accurately". I'd be surprised if we could localize instruments in a symphony blindfolded to within ±10º.
You are correct that with practice we can become better at listening. Interestingly, Floyd Toole found no significant difference in the listening preferences between trained and untrained listeners. Basically, there's no reason for an audio company to hire a panel of experts to listen to their products. They can quite literally just pull people off the street. In a blind test, that is. You can find his paper here: Toole (2003).
Tom
Regarding listener abilities here is a chart from a study published in EIA-J 1979 (from Sound Reproduction by Toole, 2008).
Note especially rows 3, 4 and 6.
Note especially rows 3, 4 and 6.
"Localization accuracy is 1 degree for sources in front of the listener and 15 degrees for sources to the sides. Humans can discern interaural time differences of 10 microseconds or less.[9][10]"
https://en.wikipedia.org/wiki/Sound_localization
Regarding memorizing a sound, I have done it by copying a wave file to a DAW or other audio editor and looping the sound until I can recognize it in a blind test. One DBT ABX listening test posted in the forum by PMA was a hi res recording of violin music. I found a point in one file where a quickly executed finger vibrato had one vibrato cycle almost missing. It went by very fast. Eventually I had the sound memorized well enough I could count the number of vibrato cycles in real time. My brain had learned how to recognize the sound in what Professor and Nobel Laurate, Daniel Kahneman, called 'System 1.' If fact, retraining of System 1 processes is what CBT (cognitive behavioral therapy) is all about. All brain processing that is not directly observable by conscious awareness (System 2) is performed by the much more powerful System 1. Suggest reading Kahneman's book, Thinking Fast & Slow; it was National Science Foundation book of the year, and a NY Times best seller. Can't go wrong with that one.
One might also note the similarity between what I did and what hhoyt did to pass ABX with various types of sound reproduction aberrations: It takes some kind of training, even if self-training. https://www.diyaudio.com/community/threads/the-black-hole.349926/post-7310119
Also of possible interest, an overview of perceptual training techniques is included in an article about vision. The underlying principles of perceptual training are general though:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10027970/
In addition, the field of psychoacoustics has been largely supplanted by the field of Auditory Scene Analysis (the study of hearing in the presence of noise, where noise is used in the general sense of meaning any unwanted signal): https://www.frontiersin.org/articles/10.3389/fnins.2016.00524/full
Mark
Last edited:
I've never said that you can eliminate jitter. There will always be noise and that will result in jitter. What I have said that jitter cleaners produce an output that has the same level of jitter almost regardless of the jitter on the incoming clock. As you're probably aware, a PLL will act as a lowpass filter on the reference clock and as a highpass filter on its VCO. The cutoff frequency of these transfer functions is set by the PLL loop bandwidth. This means that if you push the PLL bandwidth low enough for the cleaner to do any work, all you'll be left with is the VCO noise. You have no control over that VCO as it's built into the jitter cleaner. At that point it makes no difference whether you use a fancy femto-second clock or a plain Jane clock with jitter in the pico-seconds as the reference clock.
You can argue that since the transfer function of a PLL is a lowpass filter from the reference input some of the reference noise will make it to the output. That's true. But that's easy to deal with by pushing the PLL loop bandwidth down. You'll then have to accept longer lock time, but I doubt that'll be an issue in audio. The whole point of the jitter cleaner is to make the VCO the dominant noise source in the system. The VCO is then designed such that the system performance (DAC in this case) meets spec.
I spent seven years of my life designing clock jitter cleaners for National Semiconductor and TI. I have circuits in the LMK040xx, LMK048xx, LMK033xx series to name a few. I wouldn't consider myself to be an expert on these things, but I did learn from experts. I do know the fundamentals even if it's been a while since I last worked on them. I've spent more hours in front of an HP/Agilent/Keysight E5052B phase noise analyzer than I care to remember.
The LMK04000 has two PLLs. PLL1 is used to clean up the reference. PLL2 uses the output of PLL1 as its reference and provides a clean clock output from a built-in VCO. Significant effort went into designing the VCO to ensure that its noise was low enough for use in 4G/LTE cellphone systems. The VCO for PLL1 was a crystal oscillator. It's pretty easy to get good phase noise from an XO due to the extremely high Q of the crystal. A CMOS inverter is a good place to start.
Tom
Regarding PLL verses the PPLL used in ASRC, PLL is inferior at jitter attenuation. Please see the explanatory thread by werewolf: https://www.diyaudio.com/community/threads/asynchronous-sample-rate-conversion.28814/
BTW, FIFO buffering is more effective at jitter 'elimination' than either of the above. Each method has its pros and cons.
The ASRC built into ESS dacs is not of same quality as TI's SRC4392 or AKM's AK4137.
At best even ASRCs only attenuate jitter, and both PLLs and ASRCs can be affected by low level power supply noise. Some of this stuff you will not find in datasheets, just as you won't typically find excess noise specs in film resistor datasheets (although foil resistor datasheets may contain a spec for that). This brings up another very important issue: 'WYSIATI Bias'
https://www.shortform.com/blog/what-you-see-is-all-there-is/
BTW, FIFO buffering is more effective at jitter 'elimination' than either of the above. Each method has its pros and cons.
The ASRC built into ESS dacs is not of same quality as TI's SRC4392 or AKM's AK4137.
At best even ASRCs only attenuate jitter, and both PLLs and ASRCs can be affected by low level power supply noise. Some of this stuff you will not find in datasheets, just as you won't typically find excess noise specs in film resistor datasheets (although foil resistor datasheets may contain a spec for that). This brings up another very important issue: 'WYSIATI Bias'
https://www.shortform.com/blog/what-you-see-is-all-there-is/
And I'm sure you have great confidence in your ability to recognize the sound. Unfortunately, however, confidence does not mean that you actually did recognize the sound. Psychology is littered with examples of how someone will mis-remember with great confidence.
I appreciate your thoroughness, though.
Yeah. That's a pretty decent text. I found it a bit long and repetitive, though. Perhaps because I knew most of the material from Cognitive Psychology already. I may be wrong here, but I don't recall that book going into detail regarding human memory. Rather, it was about decision-making.
I base my understanding of human memory on a senior level class I took as part of my psychology degree. The class was taught entirely from current research papers, so unfortunately I can't point you to a text. If you have academic access so you can reach behind paywalls I can probably dig out some references for you though.
If you're interested in a more practical approach than Kahneman, I suggest reading Dan Arieli, "Predictably Irrational". He researches decision-making in the business context. His book is a very entertaining read.
Tom
True. For that specifically I would suggest the perceptual training literature. Recognition is in part a memory function, although recognition may only require something more or less like retention of a hash of some much more complex information. Regarding memorizing sound, how many people have memorized the essence of a song to the point they can't keep it from repeating playing back inside their head. They can tell you the lyrics, and if they can sing, perform the melody. Some can recognize the chord progressions by ear and memorize those too. A trained composer may memorize the musical form and structure. A percussionist may memorize the groove. Is human memory equal in detail to a CD recording? Not exactly, information is not represented and or encoded in the brain in the same way as on a CD. But, a lot of information may nonetheless be memorized.
Last edited:
*versus
I also see you've slid the goal posts to now cover ASRC. Whether the PLL is inferior depends on the PLL design, specifically its VCO. I've already covered why.
Anyway. It was fun to chat. I'll go grab some food and move on with my day.
Tom
You didn't see where I said I learned how to pass a blind test? Again much like what hhoyt did. ABX, dude. Please pay attention.
Its not that simple. ASRC allows for a lower loop corner frequency in practice. Add adaptive FIFO buffering and you are getting closer to Bruno Putzeys' solution. You think if he could get the same real world performance from a PLL alone he would have gone to all trouble he did? Again, you aren't getting into the practical details of making something work in a real world physical system.
There is no sliding, its part of that Benchmark patented 'jitter eliminator' technology marketing fluff you seem to have been reading. You think they could have done it with a PLL alone in actual practice?