Originally Posted by rock_bottom
What I'm saying is the measurement is invalid.
How many invalid measurements do you all run with in this forum? How often do you have access to anything but that THD number? If there has been one point I have been trying to get you all to agree is to not go by superficial claims. You are so quick to dig in here to prove THD wrong, but boy, have the cause be the other side, and you get stubborn as they come. No amount of what these things really mean makes a difference.
But running with it, I simply played the game in reverse at you. Prove to me that 0.2% distortion is audible. Where is the study that shows that with this kind of distortion spectrum? Now that the tables are reversed, you are crying foul. You want your analysis of the measurement to stand alone. That is what I have been trying to do in this thread but folks like you wouldn't have any of it.
Putting all of this aside, isn't it amazing that jitter can be peak as high as -58db? Remember, if the "Golden Ear" designers hadn't objected, this would have been a device in the market that people would have used to make USB DACs. The device used a PLL. Used buffering. Used local clock -- all the things that Arny said in the first post would assure us good performance. But it didn't. Did it?
Actually, you have rewritten the guy's story in terms of audiophile propaganda, by ignoring what's important (the PLL design and the measurements that showed what its problems were) and instead focusing on irrelevant stuff that just happens to agree with audiophile dogma.
You have to be specific with these kinds of attacks. Unlike you, I quoted exactly what the guy said. I am not implying it. He thought his device worked. Then the "Golden ears" objected and the rest of his analysis came later.
Your argument is that they could have found the problem using the measurements. This is true. But they performed three sets of measurements that did not: 1) THD 2) FFT at their standard resolution they use and 3) FFT of the noise floor. None were revealing. But the ears of someone was and without him, they would have shipped the product anyway. And with you all chanting that jitter doesn't matter, all would have been well.
Where does it say the listeners were trained? It does not.
It does in black and white. It says they used "Golden Ear" testers. I explained earlier in this thread that Golden Ear is a layman misnomer for Trained Listeners. In a DAC company, these people would be trained to hear artifacts of DACs more reliably than average Joe. I know, I played the same role at Microsoft for perceptual audio. Everyone called me "Golden Ear" but I never called myself that. I called myself a Trained Listener simply because that is how I got there. I trained myself to hear digital artifacts that at one point, I could not hear.
Of course not. Properly controlled listening tests are way too cumbersome and time-consuming, and uncontrolled listening tests are worthless.
This is where real world experience teaches you something that textbook does not. In real world, when we run extensive blind listening tests, we also include our Trained listeners in there. We know that in every one of those tests, their scores are far more reliable than the general public. Here is the example of research into this area by Harman/Sean Olive:http://seanolive.blogspot.com/2008/1...mances-of.html"The above graph plots the mean loudspeaker F-statistics for 4 groups of untrained listeners categorized according to their occupations. The performance scores of the untrained groups are scaled relative to the mean scores of the trained listener in order to facilitate comparisons between trained and untrained listeners. The trained listeners clearly performed better than any of the untrained groups, by quite a large margin. The relative performance of the untrained groups, from best to worst, were the audio retailers (35%), the audio reviewers (20%), the audio marketing-sales group (10%), and the college students (4%)."
Our experience completely matches that of Harman in countless other tests we have run. That result then, gives us confidence that trained listeners are far more right than wrong. Therefore, we can rely on their opinion with high confidence even their vote is the only one we are using, or that they are performing the test sighted. Remember, usually Trained listeners are dedicated to the job. If they are wrong often, they lose their job!
Yes, there is a risk that trained listeners could be wrong. So we do occasionally augment their work with large scale tests. But per above, I have yet to find a trained listener to be anything but highly reliable.
For you to be right then, and consider a test performed by a trained listener to be worthless, would be betting on the 5 to 10% chance they could be wrong. Those are not wise odds I would bet on
. But I realize in a forum discussion, and for people who haven't lived through the design of digital audio, that might be the way they live
Good engineering is a game of optimization. Take infinite factors and boil it into a successful product in the shortest amount of time and with least development and material expense. By definition then, you must take calculated shortcuts. The USB designer above didn't design 5 chips in parallel with 5 different DAC cores to see which is better. He picked one and ran with it. That is a shortcut. Sometimes it doesn't pay. But if you are good, it pays far more than it doesn't.
Measurements are a much better way to do this. But they must be valid measurements, in this case ensuring the frequency resolution is sufficient.
Which I have shown extensively in this thread but none of you will accept if unless I showed "audibility." Now that tables are turned, measurements all of a sudden are the gold standard? Here we have a wonderful combination of both measurements and audibility agreeing with each other. You would think there would be some celebration. But since it proves your point wrong, all of a sudden measurements are the main metric.
or the author may just be pandering to folks like yourself and John Atkinson.
There is no pandering. Knodoh-san is as genuine as it gets. You want to discuss things objectively, stick to facts and not innuendo. The man simply explains a journey he took in designing a new USB DAC. It is not like he goes on for miles about Golden Ears. If anything, he shows how upset he first was that they called his product into question until he discovered they were right.
It is because he didn't have a fundamental understanding of the problem.
Who says any of you did before coming into this thread? I could easily see light bulbs lighting up as I pointed to Paul's site with measurements of jitter. Did you know that prior to coming here? Where you not the one who is all surprised at how high jitter was in the products you were looking at?
And what "fundamentals?" Were you personally aware of the jitter induced by how the operating system and USB bus operate? Really? I mentioned that very issue on the first page. How come you were not giving me high-five for knowing that fundamental? Remember my constant emphasis of asynchronous USB which decouples USB packet timing from S/PDIF output?
Fact is that you didn't know what you claim the IC designer with a PhD had to discover the hard way. So please don't raise your status above the people who know these things from real life.
Again, we do not know if they were trained or not. Some of the story comes across as pandering to the audiophile community, so it's no wonder you ate it up. I was hoping that the essential part of it would come across, but you've chosen to emphasize that which is largely irrelevant.
You can't declare things. I showed exactly what the man wrote. You threw out the whole story and focused on the last paragraph. But if anyone reads the whole story, they realize that he is all about telling the whole journey. And that journey was one of discovery. That these things are not as simple as they look like in textbooks and forum postings from people not schooled in the exact science. And that real world audio engineering is different and more challenging than pure engineering.