Thanks for reply, yes harder than previous indeed and still is for me even i now circuit route.
Don't know if it is my fault, i just did sim as schematic at post #82 without putting more load to sim next device. Now just tried same sim with 3 times 10Kohm in parallel at output (3xVSSA). No clipping, 2,0V RMS output falls to 1,95V RMS, distortion changes very little in fourth digit, phase/FR change in lows beause of output caps highpass C-R.
Why ignoring the fact that many people (including qualified designers) have believed that measurements have *no* direct correlation with perception?
What is needed here is an open mind. An understanding that if our IQ is 150, there could be other people with IQ of 160, 170, or even 300. An understanding that if we cannot do 5/5, there still could be other people who can do 100/100 or 200/200 (basically, no difference between 10/10 and 200/200, only physical endurance).
There are too many things in the listening perception that most designers have no idea about (when correlating it with numbers). Be open minded, it is not too difficult to understand.
Have you tried my "trick" (vocal vs sonic, and 0:20:00) to differentiate them?
If you need more "trick", I still have many in my dora emon bag 😀
That's my experience too. But something like relative comparison of FFT works fine.
You change the value of the resistors, I will ABX it and mention the difference in SONIC. I believe you will find out the clue by paying attention to current in the circuit. I'm not going to discuss anything about this. You find out by yourself.
ADD:
BTW, have you read my post regarding power opamp analogy? Lower THD (with passive load) may have compromise in drive-ability (basically a current issue, perceived as sonic superiority). So buffer is indeed mandatory, so as to reduce the effect of difference in amp and speakers.
Last edited:
Sorry, BYRTT, never got back to you on this ... Exchange HDD, is this a software utility, I can't see anything obvious via Google?
Hi Frank it is hardware so use
(HDD=Hard Disk Drive). You can have a HDD for every OS that match your platform (OS=Operating System). Think it is most easy on desktops, but laptops too and some have very easy solution to plug in/out. Example is my dedicated audio PC is running Win7 64bit on SSD (Solid State Disk). Then when i want to run my older DAW ProTools 7.3.1 whitch need WinXP 32bit i exchange the SSD with a HDD where XP onboard. This is the boot drives with OS, further the system have another HDD for data where i put audio files and these are available to whatever OS i start. You welcome to ask, then i simulate "Google" 😀.
Interesting to see it simmed in another program. Those distortion figures are orders of magnitude higher and probably far more realistic.
Hi Frank, I can make you sure that without measurements you will not be able to optimize your system and get rid of distortions you often speak about. 1st, you need to optimize DAC output level to get minimal distortion. Then, you may try to explore how different playback SW's affect the signal. Without that made, you will remain in a trial/error loop 🙂.
Attachments
Pavel, thanks for your thoughts ... but I don't think it's worth persisting with optimising onboard sound beyond a certain point, as you say the real solution is to get a decent USB unit, which will be the aim down the track. The main thing is that the current setup is good enough to separate these test files, using the SW which best transcends the hardware limitations.
The real downside of cheaper hardware is that the warmup period is so long, to get the level of resolution that's good enough. Ideally, more competent, and pricey, circuitry doesn't have these issues to the same degree - but I'm not really sure that is the case, 😕 ...
The real downside of cheaper hardware is that the warmup period is so long, to get the level of resolution that's good enough. Ideally, more competent, and pricey, circuitry doesn't have these issues to the same degree - but I'm not really sure that is the case, 😕 ...
Last edited:
So continuing the theme, I have some files using the same basic circuit in post #82 CIRCUIT. Differences this time are the use of the LM4562 opamp throughout and the reduction of R4 and R12 to 3k3 and R6 and R8 to 100 ohm.
Files... we have three for this test. One is the input and output buffers only. The second is the complete chain. The third is with the middle two LM4562's configured for 100% feedback (pins 1 and 2, 6 and 7 linked) and removal of the 100 and 680 ohms.
I'll ask you all before posting them.
Do you want to know which is the I/O buffer file so you can compare the other two against that or do you want them all annonymous ?
Or do you just want the 100% feedback vs the noise gain modified one ?
Files... we have three for this test. One is the input and output buffers only. The second is the complete chain. The third is with the middle two LM4562's configured for 100% feedback (pins 1 and 2, 6 and 7 linked) and removal of the 100 and 680 ohms.
I'll ask you all before posting them.
Do you want to know which is the I/O buffer file so you can compare the other two against that or do you want them all annonymous ?
Or do you just want the 100% feedback vs the noise gain modified one ?
Jay, I am very open minded, and do believe any perceptible differences show up in measurements.
However, when differences become so small that it requires extended habituation of listening to one track that briefly trigger difference in perception when switching to comparison track, it reveals that the mind generally accepts both conditions to be the same in general case of extended listening to either test condition.
Mooly's case of before and after breakfast discrimination is example; after breakfast he may have been rushing. I experienced similar results when listening too much to both A and B; it seemed that my propensity to pick wrong match went up, leading to super high guessing percentage in results calculation. 1/7 if you force the result because you can really differentiate is much different than 3/7 because of really guessing. All statistically significant results are compelling. 7/7 shows refined aptitude for sure.
With well aligned tracks of A and B the subtracted tracks for aaa and bbb are on the order of 36dB difference. Lots of people seem to have difficulty discerning two tones at that degree of difference.
If aaa and bbb differ only in amplitude, difference is identical track of lower amplitude. If aaa and bbb differ only by noise content, then difference track reveals noise and it's relative level. If both aaa and bbb have only different noise then difference track would show this too.
In real case, phase differences are involved across spectrum that result in amplitude differences seen in display.
In previously posted example for bbb-aaa with improved alignment:
dips and peaks of reference spectra and difference spectra are extensively lined up above about 200Hz. Below 200Hz peaks and valleys show divergence. In this picture single 65k fft was used. In this case snapshot of spectra is of insufficient resolution. With 262k fft and averaging across entire waveform, resolution goes up and noise is reduced as well:
In the above overlays show reference spectrum, difference spectrum of original bbb-aaa, and bbb-aaa with improvement using 2 sample offset between aaa and bbb; aaa was delayed 2 samples.
For original aaa-bbb with 262k window averaged across track, the peak and dip alignment with reference extends down to about 70Hz. Peaking in difference tracks is seen; 60Hz line, 50Hz line with little side bands, and also lines at about 33Hz and 36Hz. For sample adjusted difference spectrum, sharp lines are visible immediately above 70Hz; with a 100Hz and 180Hz line also apparent. None of these details are apparent in spectra of aaa and bbb when viewed directly as overlays.
If aaa or bbb only are listened to in prepping for comparison, the brain burns in a snapshot memory, and sudden switching to compare triggers a difference response. Resolving spectral difference with fft seems to have similarity, in needing sufficient reference base of data.
Perhaps this is how 'SONIC' is different between various electronics; certainly this analysis works with speaker transducers and microphones.
With two similar drivers, when I use DSP to match spectra to within 0.1dB, audible differences are dramatically reduced to point of needing ABX to readily tell them apart.
Perhaps you stick with reference speaker, and have spent more time swapping electronics and critically listening, honing your abilities here.
Indeed my best scores where obtained by sticking with extended listening to either A or B; then when switching the sense of difference or no difference was immediate. If it wasn't I returned to extended listening, usually with a short break to clear my head. Listening must be relaxed, yet disciplined.
Your listening IQ is highly developed. Mine is highly functional; and will improve with exercise.
Mooly;
Audacity spectrum display is clunky. No overlay features or display zooming. With vinyl example, bigger fft or fewer data point by down sampling is require to see turntable havoc:
Nothing sounds quite like vinyl. Lots of super LF stuff to modulate brain wave activity. I think some of those LF patterns are addictive.
Thanks, that's a little lower distortion in simulation than the TLE2072. I suspect in practice with real parts it would be lower too.
These are the same set up but different music. Plenty of hi hats on here.
https://www.dropbox.com/s/7kqe76gyalyl8dy/L.wav
https://www.dropbox.com/s/4cpikhtxtatqf67/M.wav
https://www.dropbox.com/s/u5h4spybxajd4jt/N.wav
https://www.dropbox.com/s/7kqe76gyalyl8dy/L.wav
https://www.dropbox.com/s/4cpikhtxtatqf67/M.wav
https://www.dropbox.com/s/u5h4spybxajd4jt/N.wav
crisis here, those AAA,BBB were hard, these R,S,T/L,M,N even harder. See again tomorrow 
- Status
- Not open for further replies.