do it like Edison - "Mary had a little lamb,,," each speaker set to play at the same level - the most intelligible at that low level is the "winner". Of course it can be done like looking at a low level sine wave with a DAC.
I'm probably thinking way too basic (certain of it!), but what about:
1) digitally convolve a low-level signal (sine wave at a particular frequency) with a high-level signal at another frequency
2) play the signal and record it
3) band-pass the recorded signal around the low-level frequency
4) time domain cross-correlate it with the original low-level signal
5) do it with another driver and then compare the 2 to see which has more low-level info.
6) screw around with phase unwrapping to understand modulation more.
1) digitally convolve a low-level signal (sine wave at a particular frequency) with a high-level signal at another frequency
2) play the signal and record it
3) band-pass the recorded signal around the low-level frequency
4) time domain cross-correlate it with the original low-level signal
5) do it with another driver and then compare the 2 to see which has more low-level info.
6) screw around with phase unwrapping to understand modulation more.
Just had a bit of a play around with this, and the results look good so far.
Here's a link to what I've done today, including recorded files.
There's the recorded noise and 3x recordings of noise + sweep, with the sweep level varying per recording.
The sweep levels went as follows:
R2 - Very obviously audible
R3 - Just about audible
R4 - Sweep -10dB from R3 - I couldn't hear it.
You can listen to each file in turn to see what the mic was picking up, and then you can use the "Only Noise" file to null the recorded pink noise in each case, which should result in just the sine sweep being audible.
There was some background noise, particularly once you've done the null on R4, but I think it's interesting that, in this case at least, the mic has recorded things that I couldn't hear.
Chris
Here's a link to what I've done today, including recorded files.
There's the recorded noise and 3x recordings of noise + sweep, with the sweep level varying per recording.
The sweep levels went as follows:
R2 - Very obviously audible
R3 - Just about audible
R4 - Sweep -10dB from R3 - I couldn't hear it.
You can listen to each file in turn to see what the mic was picking up, and then you can use the "Only Noise" file to null the recorded pink noise in each case, which should result in just the sine sweep being audible.
There was some background noise, particularly once you've done the null on R4, but I think it's interesting that, in this case at least, the mic has recorded things that I couldn't hear.
Chris
The discussion of a speaker's ability to reproduce low level detail and high level at the same time, prompted my post which was moved here.
Update with new content - follow the link in my previous post, but I've done some sweeps starting at a sensible level (the top curve clips at 86dB) and going down in 6dB increments.
The lowest-level sweep has a very low real SPL, ranging from 30 to 8dB over the range.
As the levels went lower, I used more and more sweeps to try to get REW to capture as much correlated data as possible, which appears to have helped.
In the penultimate sweep (labelled Almost Inaudible) I could hear the sweep over a small fraction of the range - presumably, Fletcher & Munson would have something to say about that.
The last sweep was pretty much inaudible to me - I couldn't reliably tell whether any sound was happening, so I was probably imagining things a little.
Chris
PS - link again for those that missed it: DDR Testing - Google Drive
The lowest-level sweep has a very low real SPL, ranging from 30 to 8dB over the range.
As the levels went lower, I used more and more sweeps to try to get REW to capture as much correlated data as possible, which appears to have helped.
In the penultimate sweep (labelled Almost Inaudible) I could hear the sweep over a small fraction of the range - presumably, Fletcher & Munson would have something to say about that.
The last sweep was pretty much inaudible to me - I couldn't reliably tell whether any sound was happening, so I was probably imagining things a little.
Chris
PS - link again for those that missed it: DDR Testing - Google Drive
Pano,
I'd recommend you download the files from my Google Drive link and play around in Audacity. I think you might be surprised.
Chris
Last edited:
I was just looking at them on Google and couldn't hear anything. Will try again.
You and Dave will remember my attempts from a few years back.
Low Level Detail: An experimental search and test.
I was not able to follow up on that. Might be time to try again.
You and Dave will remember my attempts from a few years back.
Low Level Detail: An experimental search and test.
I was not able to follow up on that. Might be time to try again.
Had a quick look, and the files seem to work fine if you download them. The preview function doesn't appear to be working, though.
Chris
Chris
Chris I processed your files, you got good extraction of the sweep. Here is what I found.
Noise:______RMS level -26dBFS
Region 2:__Sweep level -20dBFS
Region 3:__Sweep level -39dBFS
Region 4:__Sweep level -49dBFS
On regions 3 and 4 there is a good null on the noise, about 30dB reduction. Region 2 was not as good for some reason, about 16dB reduction. That might be influencing the measurement of the Region 2 sweep.
Thanks for doing this!
Noise:______RMS level -26dBFS
Region 2:__Sweep level -20dBFS
Region 3:__Sweep level -39dBFS
Region 4:__Sweep level -49dBFS
On regions 3 and 4 there is a good null on the noise, about 30dB reduction. Region 2 was not as good for some reason, about 16dB reduction. That might be influencing the measurement of the Region 2 sweep.
Thanks for doing this!
To look at that another way, if we count the noise as the 0 dB level:
Noise:______RMS level 0 dB
Region 2:__Sweep level +16 dB*
Region 3:__Sweep level -13 dB
Region 4:__Sweep level -23 dB
*noise null not good on this, may affect sweep measurement.
Noise:______RMS level 0 dB
Region 2:__Sweep level +16 dB*
Region 3:__Sweep level -13 dB
Region 4:__Sweep level -23 dB
*noise null not good on this, may affect sweep measurement.
Entirely possible there was other background noise happening during R2. Not sure.
Your dB references sound about right from my fader positions, but I didn't write those down.
Glad the sweep extraction is working, though.
Question is, though, what do we do with the extracted sweep?
I picked it fairly arbitrarily as a test signal, but I'm not certain it's the best one to use.
Chris
Your dB references sound about right from my fader positions, but I didn't write those down.
Glad the sweep extraction is working, though.
Question is, though, what do we do with the extracted sweep?
I picked it fairly arbitrarily as a test signal, but I'm not certain it's the best one to use.
Chris
The sweep and noise is a good place to start, then move on to other signals like music.
If you'd like to collaborate on this, I'd be happy too. I can make some precise files for the tests.
If you'd like to collaborate on this, I'd be happy too. I can make some precise files for the tests.
Yeah, let's do it.
So far, here's how it's working:
- Generate required signals in REW
- Load those on to SSD, putting test signal on Ch1, noise on Ch2, and Ch3-30 empty
- Connect SSD to mixing desk
- In mixing desk, read Ch1&2, record the mic input to Ch3. Means time alignment etc is all taken care of.
- To alter the level of the test signal, it's as easy as pushing a fader and re-recording.
To get at the individual files, I plug the SSD back into the laptop so I can browse.
... and I've just realised that method is slow and flawed. There's a much nicer way of doing it.
I'm going to make a diagram. Won't be long.
Chris
So far, here's how it's working:
- Generate required signals in REW
- Load those on to SSD, putting test signal on Ch1, noise on Ch2, and Ch3-30 empty
- Connect SSD to mixing desk
- In mixing desk, read Ch1&2, record the mic input to Ch3. Means time alignment etc is all taken care of.
- To alter the level of the test signal, it's as easy as pushing a fader and re-recording.
To get at the individual files, I plug the SSD back into the laptop so I can browse.
... and I've just realised that method is slow and flawed. There's a much nicer way of doing it.
I'm going to make a diagram. Won't be long.
Chris
Okay, here goes.
The idea here is quite neat, and mostly relies on the fact that it's a digital mixing desk which lets you do all sorts of interesting things with routing.
See below.
So the idea is that all the nulling etc can be done within the desk itself, and as a result it's possible to spit out a pink-noise-nulled test signal.
In this case, I'd be using REW to supply and record the test signal (sine sweep), but it could be anything. The test signal would come out in real-time, so could also be monitored via headphones.
As far as REW can see, it's just a speaker on the other end of the mixing desk that's playing a sine sweep. What REW doesn't know is that we're adding some pink noise to the test signal, and then nulling that pink noise with a previous recording.
The laptop would interface with the desk via USB, which is handy so long as I can get all the routing worked out. Plan B is to use a USB interface so I can patch things around in the analogue domain.
I'm pretty sure this will work, and having it all interface directly with REW makes analysis etc very straightforward.
Thoughts?
Chris
The idea here is quite neat, and mostly relies on the fact that it's a digital mixing desk which lets you do all sorts of interesting things with routing.
See below.
So the idea is that all the nulling etc can be done within the desk itself, and as a result it's possible to spit out a pink-noise-nulled test signal.
In this case, I'd be using REW to supply and record the test signal (sine sweep), but it could be anything. The test signal would come out in real-time, so could also be monitored via headphones.
As far as REW can see, it's just a speaker on the other end of the mixing desk that's playing a sine sweep. What REW doesn't know is that we're adding some pink noise to the test signal, and then nulling that pink noise with a previous recording.
The laptop would interface with the desk via USB, which is handy so long as I can get all the routing worked out. Plan B is to use a USB interface so I can patch things around in the analogue domain.
I'm pretty sure this will work, and having it all interface directly with REW makes analysis etc very straightforward.
Thoughts?
Chris
Attachments
I've made up a few signals of pink noise with an embedded sweep at differing levels below the noise. In the purely digital domain the sweep signal can be easily and cleanly pulled out of the noise even when the sweep is 50dB below the noise. Maybe more, I haven't gotten that far yet.
Using a digital desk would certainly speed up the process and is very handy
Unfortunately I don't have access to one during lock down. I have files that can be used by just about anyone with a soundcard. Maybe we can encourage some other people to test.
To proceed we should test how good a null we can get with pink noise doing a line level loop-back and then with the speaker. Your posted files showed a 30dB null which is good. At analog line level it could be much better. It's good to know what the limits of the test are.
The only problem I can see with the nulling of pink noise is that noise is random, so you want to be sure you are using exactly the same noise to null with, right down to bit level.
Using a digital desk would certainly speed up the process and is very handy
Unfortunately I don't have access to one during lock down. I have files that can be used by just about anyone with a soundcard. Maybe we can encourage some other people to test.To proceed we should test how good a null we can get with pink noise doing a line level loop-back and then with the speaker. Your posted files showed a 30dB null which is good. At analog line level it could be much better. It's good to know what the limits of the test are.
The only problem I can see with the nulling of pink noise is that noise is random, so you want to be sure you are using exactly the same noise to null with, right down to bit level.
Keeping it digital, I'd expect >80dB to be possible at 16-bit.
It's evening here, but I'll get back to you with how good the loop-back on the mixing desk is. In theory, it ought to be very good.
I'm getting around the random-ness of the noise by recording a minute or so of generated pink noise, and using that for all further tests.
Chris
It's evening here, but I'll get back to you with how good the loop-back on the mixing desk is. In theory, it ought to be very good.
I'm getting around the random-ness of the noise by recording a minute or so of generated pink noise, and using that for all further tests.
Chris
Looking forward to it Chris.
I can not getting work here chez moi. Audacity refuses to do an overdub using the same soundcard for in and out. The PreSonus DAW will do it, but there is something wrong with the recorded track. Can't get anything working that will properly play and record at the same time.
Measurement software like REW, HOLM, ARTA have no problem doing the in/out loop.
I can not getting work here chez moi. Audacity refuses to do an overdub using the same soundcard for in and out. The PreSonus DAW will do it, but there is something wrong with the recorded track. Can't get anything working that will properly play and record at the same time.
Measurement software like REW, HOLM, ARTA have no problem doing the in/out loop.