You’ve been provided what you require but didn’t spare a glance. 😉
Does measurement at post #344 look as a simulation to you? Excel data representation.
It is a measurement of real amplifier! There was enough explained at provided link.
Does measurement at post #344 look as a simulation to you? Excel data representation.
It is a measurement of real amplifier! There was enough explained at provided link.
That's exactly what I was talking about. Right in the text it makes the points I was trying to make: "Theories explain facts" and "Scientific knowledge is always tentative and subject to revision should new evidence come to light."
It is not enough for a theory to explain facts. The real test is in being able to predict facts!
Jan
Jan
To my earlier post about TPC/TMC and the loop response without compensation, here is a loop gain plot without compensation. With TMC/TPC, you exploit the high gains available while the TIS/VAS is operating in the gm regime - that's the flat top part of the plot below. When you add a comp capacitor around the TIS/VAS, the amplifier operating regime becomes that of an integrator, so you get the dominant pole splitting shape most opamp open loop plots display (here I refer to internally dominant pole compensated opamps like an LM4562 or OPA1641/2/4)
Here is the same amplifier, but with dominant pole compensation.
Here is the same amplifier, but with dominant pole compensation.
This also goes for loudspeaker measurements as well.
All these things are tools.
Tools that are able to predict certain behavior beforehand to a certain degree of precision.
Why?
Because beforehand you want to get some kind of idea of possible potential problems.
Which doesn't only help with designing and developing things.
It's also a good workflow in finding potential problems and issues later down the road.
Key words here are "tool" as well as "a certain degree of precision".
Some people still seem to be mistaken that it's a goal instead, or get hung up by the fact that it's never completely exact.
That's just not the right way of using these tools to begin with.
Yeah... mine too when as a member of the Hare Krishna wandering airports I switched from a Gillette... but he insisted I give it back.
In a presentation of observation from the time domain (in your call to publish something from that perspective) it seems instructive to begin from a zero time state in how a transient such as a step function behaves from time zero t0 to t1.
If we assert that all devices have some form of non-linear transfer function it can be mathematically concluded that a derivative of any point on the slope has a finite slope describing the transfer gain at that point. Imagine any two points on the transfer function and a curve is drawn between them. As you bring them together the curve tends to straighten. It is a straight line just before the two points touch and you lose the line. The derivative just tells us what the slope of the line is before touching.
What I am getting at is that if one maintains the peak to peak input swing within these two minimalist points before they touch there is absolute linearity. This leads to a general principle that linearity improves as peak to peak input signal amplitudes decrease regardless of how those signals diminish. It is this principle that has led to recent examinations of I/V conversions from DAC input stimulation, magnifying input differentials by 100x as Walter Jung did in his AD811 I/V of years ago
Last edited:
I read that as saying that when you decrease signal levels towards zero, distortion goes to zero.
What you describe is the .ac small signal simulation in a circuit simulator, linearize the device around the bias point to get the frequency domain response.
But it is totally useless (and really nonsensical) for transient/large signal analysis.
Jan
What you describe is the .ac small signal simulation in a circuit simulator, linearize the device around the bias point to get the frequency domain response.
But it is totally useless (and really nonsensical) for transient/large signal analysis.
Jan
What about for verification after prediction? Don't people tend to rely most on what they know how to measure, and tend to ignore things that can be also problems but are (possibly much) harder to measure?
For example, Bob Cordell's book mentions a lot of different types of distortion and noise mechanisms. They include nonlinear and linear distortions, white noise, colored noise, correlated noise, etc.
Seems to me the more loop gain there is, the more opportunity there could be for some of the less commonly measured noise problems. EMI/RFI noise getting in through the speaker terminals, AC power, etc.
How do people here like to measure for that stuff? And how do you know when its really inaudible if there is no published threshold for the average ear?
Last edited:
Do a subtractive test with music where the signal is subtracted while anything non-linear produced by the amplifier (various kinds of distortion and noise) stays at its normal volume, and notice that while you very accurately trim the amplitude and phase correction networks to get the best cancellation of the music, the residue gradually drops to 60 dB below the normal signal level, but still sounds like music rather than distortion. Music with the equalization way off, but still music.
Last edited:
There was a guy working on a subtractive test setup just for such purposes. Don't know how far he got with it. Part of it was to verify the test apparatus works as intended to detect known small differences.
Back here in my lab, just did an empirical shielding test with some discrete DSD dac boards. Without the shielding there is an easily audible difference. Subtractive test gear would be nice, but I think it might be a bit complicated to have have two complex and expensive dacs with different shielding in multiple chassis synchronized with each other without introducing ground loops. Besides, its so easy to hear on ESL speakers its would be pretty much just a measurement to show off to others.
Back here in my lab, just did an empirical shielding test with some discrete DSD dac boards. Without the shielding there is an easily audible difference. Subtractive test gear would be nice, but I think it might be a bit complicated to have have two complex and expensive dacs with different shielding in multiple chassis synchronized with each other without introducing ground loops. Besides, its so easy to hear on ESL speakers its would be pretty much just a measurement to show off to others.
Last edited:
The cancellation is indeed incomplete.
I gradually attenuated the signal by up to 60 dB without attenuating the distortion, and it still sounded undistorted to me. That tells me that the amplifier's non-linear distortion on music is low enough for me by a very large margin - as it should be, of course.
It doesn't say anything about whether the amplitude and phase responses are good enough, as I corrected for those.
I gradually attenuated the signal by up to 60 dB without attenuating the distortion, and it still sounded undistorted to me. That tells me that the amplifier's non-linear distortion on music is low enough for me by a very large margin - as it should be, of course.
It doesn't say anything about whether the amplitude and phase responses are good enough, as I corrected for those.
There would be a delay between the input and output signals, so a scaled-down version of the output signal will not precisely overlay the input signal, since the two signals are not aligned in time due to the propagation delay. Summing the two signals will always produce some residue.
I corrected for the amplifier's phase shift, but the correction was imperfect. The amplifier has built-in high- and low-pass filters, second order Butterworth at 1 Hz and 135 kHz, the phase correction network was a passive RC network, first order for the low and second order for the high frequencies if I remember well.
I have used subtraction with a constant time delay to compute the distortion introduced by MP3 encoding/decoding. The results were a 128 kbps MP3 introduces 10-15% distortion. A 256 kbps MP3 introduces 1-3% distortion.
Ed
Ed
Do you mean the residual is on average 1-3% of peak or RMS level of the original, or how is it calculated?
Obviously its a different type of distortion than a more or less stationary curved transfer function in an amplifier produces.
Also, seems to me that thresholds of audibility for MP3 type distortion could different be from thresholds for HD nonlinearity.