Audibility of Audio Power Amplifier Distortions
Reply #90 – 2014-11-30 21:06:46
As I promised Steven earlier, here are some additional quotes from the Meyer amplifier test. One of the key requirements for a correct listening test is critical test signals/music tracks. We need content that is revealing of the differences we like to find. This is stipulated strongly in ITU BS1116 and is standard practice in research/industry. It is abundantly easy to use tracks that don't bring out the difference and proceed to declare there is no audible difference. When a distortion is not uniformly apparent, finding the difference becomes a statistical effort. If I suck out 43 Hz out of the system, but you play content that has little in that frequency, then you won't hear the difference before and after I do that. I think we all agree that a system that has big hole at 43 Hz is broken. Yet we can trivially create results from double blind tests that say there is no problem at all. We may get lucky and throw music at the system that does have 43 Hz content, or the opposite. What are the odds? Well, just like gambling, we like to improve our odds. We do that by using trained listeners who know what to listen for and find them easier than average listeners. And we help them further with critical music segments that makes the job easier and less subject to errors/nocebo, etc. Let's put this knowledge to practice. Here are the results of the double blind tests that Meyer ran in his stereo review article: Look at the difference the type of signal makes. I have highlighted Pink Noise. See how that is revealing in all three test cases. Now look at the rest of the outcomes in test #2. All would be dismissed as not being good enough and equiv. declared in the two amplifiers. The job of any amplifier is to be the proverbial "wire with gain." In other words it only amplifies but doesn't change anything. The moment one amplifier sounds different than another with pink noise, the game is over. It matters not that a million other pieces of music don't show it. It matters that we violated the "wire with gain" as that statement does not make an exclusion of a full spectrum signal like pink noise. In this case we have objective proof that the pink noise data is correct. We know that the frequency response varied between the two amplifiers. So there was no need to throw music at the system. The job was done the moment pink noise showed revealing difference in amplifiers. Unfortunately the audio test field is littered with people who throw random tracks of music at a system and proceed to declare equivalence. No attempt is made at determining what the potential problems are, and what kind of content best exercises it. This is my #1 beef with many listening tests. I hope the above data shows why.