Thank you.
Don't recall having seen a clear statement of the whole conceptual problem under consideration.
For the sake of clarity, has it been stated how the amplitude modulation comes to be? It may have been produced by mechanical nonlinearity in a cymbal? Or, it is produced by sampling a frequency close to fs/2, where the close frequency exists only for a short time period, and without adequate ADC filtering? In other words, do we have a clear statement on the creation, or hypothesized creation, of amplitude modulation?
Last edited:
Okay. Does anyone think this discussion has come far enough that such a statement could be formulated?
Of course there are countermeasures, but to incorporate these one has first to recognize/accept the problem.
If the position is "can't be" or "doesn't matter" it gets difficult.
Claro. I wanted to comment on it as it seems that many shy away from toslink and may miss its good properties if done correct.
//
Really sounds like a very smart idea, never heard this before.Well excuse me, I did "dream" this up last night.
Excuse me if this has been shown before. I was trying to figure out a way to use the frequency domain to do down sampling with FFT's. This is a trick that does not work with power of 2 FFT's. If you take a file sampled at 192K and want to convert it to one sampled at 44.1k using power of two FFT's there is no way to have a common frequency per bin. Using an N X 44.1 FFT on one file and N X 192 FFT on the other all the bins have exactly the same spacing.
I tried the cymbal file truncated to 10sec taking the 10 X 192k file and using a real FFT I placed the bins from 0 Hz to 22050 Hz in a blank 10 X 44.1 sized file. The inverse FFT gives the resampled result which is almost exactly what the Audition resampler outputs. I have noticed that doing SRC back and forth in Audition or Cooledit builds up a very slight delay which I have not figured out. Creating MP3's seems to have the same problem which makes exact comparison harder.
Much less computing intensive as the normally used interpolation method.
I assume you placed -22050 Hz up to 22050 Hz in a blank file to create a real (as opposed to complex) signal.
Hans
I'm prepared to take a wild guess. Information being lost due to filtering before and after conversion, possibly leading to distortion of envelope attack resulting in ITD shifting and blurring of image etc.
Actually I used the real only FFT functions, they cut the size in half because the negative frequencies are always the complex conjugate of the positive ones.
I will repeat what I already said, perhaps a week ago, maybe two.
When a sine wave is modulated either by amplitude or frequency, the information in that modulation can cause a violation of nyquist if the upper sideband is high enough in frequency. It has already been clearly stated that a rapidly changing signal can violate nyquist.
As many instruments have a quickly changing envelope, the sideband products have to be considered when setting the sampling rate, as the sidebands above and below carry information about the envelope.
The choice of a 44.1 k rate does not leave much headroom to allow for envelope modulation sidebands, so we must consider this.
If I use two frequency generators, one putting out 17.5 khz, the second putting out 22.5 Khz, and add the two signals,I will see a 20Khz frequency being modulated at a 5k rate, the beat frequency.
If I then put that through a filter that is capable of removing just the 22.5K component, The filter will put out 17.5k. So, the filter did not create a new frequency out of thin air, it was always there.
That is specifically why I requested Hans investigate his filter output, as it was lower frequency than the input.
Note: as always, this statement can be easily tested for affirmation, or refutation. I still await Hans trying this as I asked him to.
jn
Last edited:
If you are getting a beat frequency then something is intermodulating, a nonlinear process. Still inevitable with sampling (which is similar to mixing).
However I think your question- does a modulated tone with sidebands beyond the passband get changed if one sideband gets removed by the antialiasing filter etc.
Next associated question would be whether this has any audible impact.
If I were to generate a modulated tone per your specs and pass it through a digital record/reproduce chain and look at the results would that add anything to the discussion?
However I think your question- does a modulated tone with sidebands beyond the passband get changed if one sideband gets removed by the antialiasing filter etc.
Next associated question would be whether this has any audible impact.
If I were to generate a modulated tone per your specs and pass it through a digital record/reproduce chain and look at the results would that add anything to the discussion?
Demian, Hans already did. That said, confirmation of this entirely unexpected result would be great.
I printed out two copies of Han's posting, star date 34414.
I then used my PSP* tools to sample one, and add it to the other, time shifted about 100 uSec.
IOW, I cut one, scotch taped it to the other.
Note that I drew a straight line through every green zero crossing and the first blue zero crossing.(to line them up.)
I circled on the blue waveform exactly where the green zero crossings occur on the blue.
ps. If one also takes the time to do so, figure out the time it takes for 3 negative going zero crossings on the green wave.
As close as I could guess, it is very close to 170 uSec for 3 full cycles, or 56.66 uSec per full cycle.
1/56.66 = .0176. So just by pencil paper, and good glasses, it looks like the green is 17.6 Khz. Close enough to 17.5Khz given the method used..
jn
*Paper Sheet Processing
I printed out two copies of Han's posting, star date 34414.
I then used my PSP* tools to sample one, and add it to the other, time shifted about 100 uSec.
IOW, I cut one, scotch taped it to the other.
Note that I drew a straight line through every green zero crossing and the first blue zero crossing.(to line them up.)
I circled on the blue waveform exactly where the green zero crossings occur on the blue.
ps. If one also takes the time to do so, figure out the time it takes for 3 negative going zero crossings on the green wave.
As close as I could guess, it is very close to 170 uSec for 3 full cycles, or 56.66 uSec per full cycle.
1/56.66 = .0176. So just by pencil paper, and good glasses, it looks like the green is 17.6 Khz. Close enough to 17.5Khz given the method used..
jn
*Paper Sheet Processing
Attachments
Last edited:
O.k when you only took the real values and their complex conjugates for the negative side, instead of of the complex values and their conplex conjugates, doesn’t this cause phase distortion in the time domain ?
Hans
You have to be careful about what you mean. If you add two frequencies then completely filter out one, then only the other one remains. The beat frequency observed in that case arises from the two frequencies summing constructively at times (thus creating a higher crest factor or peak summed amplitude), and then other other times they sum more destructively (creating a lower peak amplitude or crest factor).
The above is not technically modulation, since no nonlinear process has acted on the two frequencies to create other new frequencies. When modulation, such as mixing, takes place, then a nonlinear process such as the two frequencies multiplying together creates sum and difference frequencies. In that case new frequencies are created, not just a difference envelope.
However, as Audio1 (Demian) points out, sampling itself is a nonlinear process.
Last edited:
For clarity my example was for a continuous sine wave. BTW the cymbal file has several resonances near 3kHz I might be able to isolate them and look at the envelope before and after.
Yes, it is a very reputable manufacturer. The problem is with your OS or your anti-virus, I guess. Are you on Win or Mac?
But please bear in mind that this is a plug-in. You will still need a DAW, like Adobe, Wavelab or some freeware, to load it.
The FFT library does the book keeping, the results are identical but the two sided display of spectra is a nuisance for visualizing things. There is a speed up if it is known that the input is all real too.
- Status
- Not open for further replies.