um that TI DSP paper is way dated - current desktop PC processors run rings around that era's dedicated DSP chips
my nearly 6 year old desktop has Quad Core2, 2.4 GHz, 8 Mbyte L2 cache
http://www.nasoftware.co.uk/home/attachments/018_PPC_Intel_comparison_whitepaper.pdf
gives more recent Intel PC CPU DSP times, it is even a few years old - points to the current generation's PC CPU's doubling of DSP instruction bit width...
...got to buy me a new PC - its a shame only embarassingly garish gaming PCs use the better CPUs today off the shelf
my nearly 6 year old desktop has Quad Core2, 2.4 GHz, 8 Mbyte L2 cache
http://www.nasoftware.co.uk/home/attachments/018_PPC_Intel_comparison_whitepaper.pdf
gives more recent Intel PC CPU DSP times, it is even a few years old - points to the current generation's PC CPU's doubling of DSP instruction bit width...
...got to buy me a new PC - its a shame only embarassingly garish gaming PCs use the better CPUs today off the shelf
Last edited:
John,
The problem may be that those amplifiers sounded good for other reasons, not related to PIM. If you had measured their PIM, you might have found that they did not have any lower PIM than amplifiers with more NFB.
Barrie's paper did not really add anything to my paper on TIM, and did not contradict it. Barrie's paper just did not show explicitly that the PIM was dependent on movement of the closed loop gain pole, and he did not show experimental result from a real amplifier with and without large amounts of feedback.
My paper is here, and its explanations are true to this day:
http://www.cordellaudio.com/papers/phase_intermodulation_distortion.pdf
Any nonlinearity that moves the closed loop bandwidth around as a function of signal will generate PIM. When it involves NFB, usually it is amplitude modulation of the open loop gain that moves around the closed loop pole, thus effectively resulting and an amplitude-to-phase conversion (good old differential phase). The open loop bandwidth and the LF open-loop gain have virtually nothing to do with the magnitude of this effect.
Even in an amplifier with NO NFB, there are mechanisms that create signal-dependent bandwidth and phase.
PIM exists, it may be audible. It is virtually impossible for PIM to exist in the absence of other distortions, like THD20 and 19+20KHz CCIF IM.
Cheers,
Bob
They certainly are, depends what you want to explore. Large FFT highly averaged is nothing to call home about in case you need to examine non-periodic phenomena with random occurence.
SY - No. It does not help much. You have to catch it in time domain, single-shot the best. On contrary, you will not seek for THD 0.001% with scope. Frequency and time domains ARE dual, but we have the instruments that we use in time domain or frequency domain according to our needs.
Edit: Richard was faster.
Edit: Richard was faster.
Once you've captured it in time domain (and the techniques for doing that are well known), it's trivial to convert to the frequency domain. And the reverse as well, of course.
And incorrect as well. Transformations from time to frequency domain do not have to be limited to harmonically related phenomena nor are they limited to periodic signals. That's been explained in this thread alone at least a dozen times.
And incorrect as well. Transformations from time to frequency domain do not have to be limited to harmonically related phenomena nor are they limited to periodic signals. That's been explained in this thread alone at least a dozen times.
One point has been thoroughly made, over and over... the math of the FFT in all directions is correct. Thats a comfort to know [tongue in cheek]. Now can we get to what we actually have to test with? Amps (power) could use a transient stimulous and capture it and FFT it and all. Dont bother to tell me yet again that it can all be done with sine waves or deconvolved back and forth and broken down etc. Or that its done all the time on a super computer. Thats all fine and good for the FFT math and computer SIM guys doing R&D on a grand scale.
What can be done with a PC that is real time using 3D FFT's on a PC that is at least 10% accurate (+/- 1dB) to -100dB (or below)?
A transient waveform [and capture?] is the test stimulous for exciting a power amp.
Thx-RNMarsh
What can be done with a PC that is real time using 3D FFT's on a PC that is at least 10% accurate (+/- 1dB) to -100dB (or below)?
A transient waveform [and capture?] is the test stimulous for exciting a power amp.
Thx-RNMarsh
Last edited:
It's 2013. We have marvelous things like impulses and MLS, even in cheap software. And way better than 1dB resolution at -100dB or below. No supercomputers needed.
Guess what? Amps act exactly like you'd expect them to, at least amps designed by real engineers (as opposed to former hifi salesmen, Japanese gurus, and 1970s era fossils with "philosophies").
I've never used one of your amps. Odd that you would think that I was talking about them. What's the part of their performance that you think would show problems in a simple impulse test?
BTW, the amp impulse response is routinely taken during speaker measurements. So it isn't like this is an exotic measurement or that normal amplifiers show odd behavior.
Now FFT has come up again, it reminds me of the last time it surfaced on this thread.
On the issue of the perfect reversibility of the transform between time and frequency domains, I many posts ago made the rather contrived joke that this was only true if one would have the mathematical formula for random noise. I still can see it flying over all heads. The point was of course that the statement of perfect reversibility is only true for discrete FFT. This may have some relevance for audio measurements, since it requires sampling of the measured signal, with an unavoidable associated loss of information. Also, a discrete FFT only produces data, to get information out requires trade-offs.
I mention it again because of what Pavel and Dick are posting. Thought experiment: what is the FFT of a Kaiser Wilhelm sine? With this I mean a sine with a spike on the upper-half (not to be confused with its mirror image, the wineglass sine).
Edit: I remember Scott posting at the time of the last discussion on some measurements that were developped to, if I remember correctly, identify a peculiar kind of one sided crossover distortion. Pretty much a Kaiser Wilhelm situation if my memory doesn't play tricks on me, but I didn't have time to read it with sufficient attention at the time. Could you please post again?
On the issue of the perfect reversibility of the transform between time and frequency domains, I many posts ago made the rather contrived joke that this was only true if one would have the mathematical formula for random noise. I still can see it flying over all heads. The point was of course that the statement of perfect reversibility is only true for discrete FFT. This may have some relevance for audio measurements, since it requires sampling of the measured signal, with an unavoidable associated loss of information. Also, a discrete FFT only produces data, to get information out requires trade-offs.
I mention it again because of what Pavel and Dick are posting. Thought experiment: what is the FFT of a Kaiser Wilhelm sine? With this I mean a sine with a spike on the upper-half (not to be confused with its mirror image, the wineglass sine).
Edit: I remember Scott posting at the time of the last discussion on some measurements that were developped to, if I remember correctly, identify a peculiar kind of one sided crossover distortion. Pretty much a Kaiser Wilhelm situation if my memory doesn't play tricks on me, but I didn't have time to read it with sufficient attention at the time. Could you please post again?
Last edited:
A decent PC, something basic in a sound card (I use an M-Audio 192), basic software like AudioTester (you can get fancier signal generation capability if you want to create oddball stimuli, then use the FT analysis in Audiotester to measure the output). A simple analog interface box to get levels and impedances correct for what you're measuring. Really, that's all you need. This has enormously better capability than the ultra-expensive stuff I used to work with when I was at Nicolet.
- Status
- Not open for further replies.