"phase unwrapping" can be a problem: http://www.ljmu.ac.uk/GERI/CEORG_Docs/OneDimensionalPhaseUnwrapping_Final.pdf
audio amplifier phase measurement issues are "one dimensional" and can be determined at fairly high S/N where it isn't a problem
Fourier based analysis are best at detecting new distortion frequency components in a (sparse) multitone test signal - if you keep the complex part phase information is available too where the S/N is good
by eyeball it is much harder to see small amplitude changes in large fourier components - although signal processing algorithms may do better
the worst case is detecting noise modulation - modulating mls "noise" is a good, secure broad band communication technique - without the original noise sequence you have a hard time finding the spread spectrum modulation
at this point some people want to start talking about chaos, stochastic resonance, below noise floor correlated signal, etc - but while human perceptual limits at the noise floor may not that well explored all you have to do is show a system where you can DBT these "night and day differences" - and if it can't be explained by "conventional" auditory thresholds, signal theory you can probably find some eager researchers out there willing to look
audio amplifier phase measurement issues are "one dimensional" and can be determined at fairly high S/N where it isn't a problem
Fourier based analysis are best at detecting new distortion frequency components in a (sparse) multitone test signal - if you keep the complex part phase information is available too where the S/N is good
by eyeball it is much harder to see small amplitude changes in large fourier components - although signal processing algorithms may do better
the worst case is detecting noise modulation - modulating mls "noise" is a good, secure broad band communication technique - without the original noise sequence you have a hard time finding the spread spectrum modulation
at this point some people want to start talking about chaos, stochastic resonance, below noise floor correlated signal, etc - but while human perceptual limits at the noise floor may not that well explored all you have to do is show a system where you can DBT these "night and day differences" - and if it can't be explained by "conventional" auditory thresholds, signal theory you can probably find some eager researchers out there willing to look
Sudden phase changes are a sign of a narrow resonance. Narrow resonances can sometimes be difficult to spot in an amplitude plot, either because the sweep is too fast (conventional spectrum analyser) or the resonance falls between points (FFT). Fortunately(?) the phase effects are broader than the amplitude effects so can be easier to see. A single resonance will usually shift the phase by 180 degrees. Phase is circular, graphs are linear, so you have to remember that +180 and -180 are the same thing.
Different dither settings in audio recording/mixdown softwares sound different too.
There has been mention of conduction quantization regarding QP - is this effectively causing a form of dithering affecting audio circuits ?...open question.
Eric.
Last edited:
Great!
The structural member cavity was resonating at that frequency. The phase shift could be considered to be more than 360 degrees as the signal hung around for milliseconds after being excited. My and others hearing perceived that as more energy or a sibilance problem.
An equalizer would not have fixed it! Completely removing the problem frequencies would have left a hole and fixed attenuation would always be there, but the problem only occurs when you have enough energy for long enough to store some to resonate.
So now the question becomes what would be better than an FFT to show these types of problems?
Why wouldn't the problem show up on an FFT analysis? Time and frequency domains are both accessible from any measurement system I'm aware of. No need to even take the FFT to spot a resonance like that.
It did show up but was not read correctly. No you don't need an FFT to see it, but if all you are using is an FFT you can miss it. That is what I though was the point.
You know when you only have a hammer everything looks like a nail.
So the question is what DSP analysis would show it? Could you still reconstruct the signal from the final data subject to sampling and other limits?
Nah,
Reconstructing the signal can be very handy when you are keeping records and want to check a new idea on your database. Sometimes after doing the same thing over and over something new pops out at you.
As to the second issue that presumes you know what you want to look at. For example capacitors can be tested for "soakage" but until you know to look for microphonics or resonances you would not find them. When an amplifier circuit warms up it may behave differently as it heats. Then you can look to see if load induced temperature variations also have the same effect.
Heyser promoted Time Energy & Frequency. It is nothing new and unless you look at all three you might miss something.
That was the point of this line. The next issue would be what are the limits of DSP analysis, but life is too short...
I think you are confusing two different things. The steady-state response of a system will shift by 180 degrees for each resonance. There will also be a transient response, which may last a long time if it is a high Q resonance. My guess is that you were measuring the steady-state response, but hearing the transient ringing. These two responses are two manifestations of the same transfer function, which arises from the forced and unforced solutions to the damped simple harmonic oscillator second-order differential equation. I realise I have to tread carefully, as any mention of transients may wake up trolls.
Thanks, in this case it was actually loosely coupled systems and a swept excitation. So if I understand the driving mechanism you need not get the perfect 180 degree shift. Which sort of raises the issue of a delayed response from the coupled system.
- Status
- Not open for further replies.