Align IR start feature

[Breeman]

Hi John,

I just want to confirm that I understand how the "Align IR start" feature works. I measured a loudspeaker at different listening positions that are close together and wanted to align the impulse responses before averaging them. However I seem to be having an issue where if I try to align two impulse responses for an example, I have to click on the Align IR start button repeatedly before the impulses align in what I deem to be correct, this is shown in picture C below. Picture A is before attempting to align and picture B is after the first click of the button. The impulses are also shifted to time zero which may or may not always be desirable. Is it possible that I could choose one impulse as a reference and align the other/s to the start time of the reference?

 

[Breeman]

While on this topic. Is it possible that we could have an extra parameter to define when attempting to align impulses like the ones below? These impulses won't align regardless of how many times I click the "Align IR start" button. I suspect the alignment feature works by measuring an impulse level at the start meaning the level of the small wiggles on the green impulse below will prevent proper alignment.

 

[John Mulcahy]

Strange looking responses with their multiple peaks. Can you attach the mdat file?

Align IR start is usually used with subs and the like, it aligns within one sample (21 us at 48 kHz sampling). Time align would be the norm for speaker measurements, but I doubt that would work reliably with those measurements. Was there more than one speaker playing? Or does the speaker have multiple tweeters?

 

[John Mulcahy]

The start time is usually defined as the first point where the response exceeds 10% of the peak.

 

[Breeman]

The start time is usually defined as the first point where the response exceeds 10% of the peak.

Just to make sure I understand. Is it 10% of the peak or 10% above zero?

 

[Breeman]

Strange looking responses with their multiple peaks. Can you attach the mdat file?

Align IR start is usually used with subs and the like, it aligns within one sample (21 us at 48 kHz sampling). Time align would be the norm for speaker measurements, but I doubt that would work reliably with those measurements. Was there more than one speaker playing? Or does the speaker have multiple tweeters?

They are looking strange because they were taken in the car.

 

[John Mulcahy]

Not sure I understand the difference? The peak is at 100%, so 10% is at 10%.

 

[Breeman]

Here's file you can use for the first post

 

[Breeman]

Not sure I understand the difference? The peak is at 100%, so 10% is at 10%.

Understood thanks. Can this value be changed to cater for the second scenario of the attached file? I'm still trying to figure out why the wiggles in front of the main impulse occur. They only occur on measurements of the opposite audio channel from where the microphone is located. In other words, they only occur if the microphone is located in the driver seat and I'm measuring loudspeakers located on the passenger side and vice versa. When measuring the loudspeakers closest to the microphone (like in the first attachment, Impulse tests.mdat), they aren't there.

 

[Breeman]

They are looking strange because they were taken in the car.

No multiple tweeters, just a normal 2-way passive loudspeaker per front channel. Each audio channel was measured one at a time

 

[John Mulcahy]

In such a highly reflective environment alignment will be very difficult. Best to just eyeball each response and use "Set t=0 at cursor" with your own judgement, that's what I did on the attached.

 

[Breeman]

In such a highly reflective environment alignment will be very difficult. Best to just eyeball each response and use "Set t=0 at cursor" with your own judgement, that's what I did on the attached.

Thanks for the suggestion. Too much eyeballing is exactly what I was trying to avoid because it can get tediuos when dealing with many measurements. I thought the issue could largely be mitigated by having a user configurable value for the impulse rise threshold. In this example, a threshold of say 20% will make the alignment work as intended, right? Please consider making it configurable between 10 and 20 or 30%.

 

[Breeman]

Hi John,

I just want to confirm that I understand how the "Align IR start" feature works. I measured a loudspeaker at different listening positions that are close together and wanted to align the impulse responses before averaging them. However I seem to be having an issue where if I try to align two impulse responses for an example, I have to click on the Align IR start button repeatedly before the impulses align in what I deem to be correct, this is shown in picture C below. Picture A is before attempting to align and picture B is after the first click of the button. The impulses are also shifted to time zero which may or may not always be desirable. Is it possible that I could choose one impulse as a reference and align the other/s to the start time of the reference?
View attachment 52680

View attachment 52681

View attachment 52682

We you able to see the changes with alignment with every click?

 

[John Mulcahy]

Yes, I have fixed the fractional-sample shifts that were occurring. The IR start feature tries to find where the IR begins, the nominal 10% level/1 % energy threshold is a starting point for that but it isn't simply a case of picking the 10% point. High noise levels ahead of the arrival at the measurement mic make the process very difficult, there isn't an easy fix. I tested higher thresholds, but they can lead to the start being placed after the initial part of the response (and so later than it should be) for measurements with lower noise levels and didn't produce alignments that approached the manual method since the span can be up to 1 sample period.

 

[Breeman]

Yes, I have fixed the fractional-sample shifts that were occurring. The IR start feature tries to find where the IR begins, the nominal 10% level/1 % energy threshold is a starting point for that but it isn't simply a case of picking the 10% point. High noise levels ahead of the arrival at the measurement mic make the process very difficult, there isn't an easy fix. I tested higher thresholds, but they can lead to the start being placed after the initial part of the response (and so later than it should be) for measurements with lower noise levels and didn't produce alignments that approached the manual method since the span can be up to 1 sample period.

Thank John

 

[sm52]

John, is it possible to use the analysis of the behavior of the phase at the end of the range to determine t=0? You once wrote about it. When I manually move the phase response, I can find its position in time when some part of it is parallel to the horizontal axis. In fact, there is no flat phase characteristic, but it is visually clear that when the graph moves in one direction, the slope from the left bottom goes up to the right, and in the other from the left top to the right down. It means 'horizontal' between them. This is t=0. And, if t=0 falls between the first peak, which is less than 100%, and the second, whose level is 100%, then will accept that the first peak is t=0.
As for Breeman's concern, I think that if one measured driver is running during the measurement, and not two, then something in the measurement chain gives a lot of noise (interference).
I also took measurements in the car. Yes, there are always a lot of reflections in the car, but this is not the reason for Breeman's strange graphs.

 

[John Mulcahy]

John, is it possible to use the analysis of the behavior of the phase at the end of the range to determine t=0?

In some circumstances, when reflections are low and when the measurement extends a long way past the bandwidth of the system being measured and the signal-to-noise ratio is high enough at high frequencies. Flat phase is equivalent to saying group delay is zero, if the group delay in the region past the end of the system response is not zero (but flat) it will show the delay that needs to be removed. Most often the phase and group delay are too noisy to use reliably for that and/or the measurement bandwidth is not sufficient to reach the region where the phase truly becomes linear. Here is an article about it: Phase response and receive delay

 

[jtalden]

I'm still trying to figure out why the wiggles in front of the main impulse occur. They only occur on measurements of the opposite audio channel from where the microphone is located. In other words, they only occur if the microphone is located in the driver seat and I'm measuring loudspeakers located on the passenger side and vice versa.

I have a theory as to the source of the impulse wiggles.
The impulse wiggle spacing suggest a low level ~17 kHz sound is arriving prior to the sound from the more distant speaker being measured.
The closer speaker should have no signal and thus be silent, but some amplifiers have a little more crosstalk between channels at the very high frequencies. Thus, there may be low level high frequencies leaking into the nearer channel from the sweep. This is easy to check out by muting the nearer channel completely or just covering that tweeter during a measurement. If the wiggles are gone, it confirms the situation.
If this checks out and you want to measure the relative leakage, you could mute the more distant speaker, but still send the measurement signal there. The sound measured then would be only whatever is leaked into the nearer speaker. I would expect the response to be only at very high frequencies and maybe 30 dB or more down when compared to the normal measurement.

This doesn't impact any of the other discussion or actions here and I wouldn't expect this to be any kind of sound quality issue. It is just offered to ease your concern over the source of wiggles you are seeing.

 

[sm52]

John, thanks for the clarification. I read the article. The group delay has the last word. But I don't see how that can be applied. If the group delay changed from a change of t=0, it would be possible to move the scale until the group delay was at the end of the range = 0. But the GD does not change after measurement.

 

[John Mulcahy]

But the GD does not change after measurement.

? If you move t = 0 the group delay plot will shift along the time axis by the same amount.

 

[Breeman]

I have a theory as to the source of the impulse wiggles.
The impulse wiggle spacing suggest a low level ~17 kHz sound is arriving prior to the sound from the more distant speaker being measured.
The closer speaker should have no signal and thus be silent, but some amplifiers have a little more crosstalk between channels at the very high frequencies. Thus, there may be low level high frequencies leaking into the nearer channel from the sweep. This is easy to check out by muting the nearer channel completely or just covering that tweeter during a measurement. If the wiggles are gone, it confirms the situation.
If this checks out and you want to measure the relative leakage, you could mute the more distant speaker, but still send the measurement signal there. The sound measured then would be only whatever is leaked into the nearer speaker. I would expect the response to be only at very high frequencies and maybe 30 dB or more down when compared to the normal measurement.

This doesn't impact any of the other discussion or actions here and I wouldn't expect this to be any kind of sound quality issue. It is just offered to ease your concern over the source of wiggles you are seeing.

Interesting observation and thanks for your suggestion. I will definetely try it out and report back. If it turns out to be true, the leakage would be caused by amplifier crosstalk because I feed the sweep to only one channel at the time meaning no other audio channel is connected except of the loudspeaker I'm measuring. Just curious to know how you determined the frequency of the low level sound from the wiggle spacing?

 

[jtalden]

Evenly spaced impulse wiggles reflect a strong frequency resonance present near the highest frequencies measured. These wiggles were approximately equally spaced, so I measured the period of 10 wiggles and divided by 10 to estimate the period of one cycle. The inverse of that is the frequency. In this case it corresponded well to the highest frequency that was measured prior to the SPL roll-off. I also noted the wiggles started roughly 2 ms prior to the main peak. This all seemed appropriate if there was some sound leakage from the nearer speaker.
Amplifier crosstalk increases with frequency so if you measure for it there will be other high frequencies above the noise floor as well.

 

[Breeman]

Evenly spaced impulse wiggles reflect a strong frequency resonance present near the highest frequencies measured. These wiggles were approximately equally spaced, so I measured the period of 10 wiggles and divided by 10 to estimate the period of one cycle. The inverse of that is the frequency. In this case it corresponded well to the highest frequency that was measured prior to the SPL roll-off. I also noted the wiggles started roughly 2 ms prior to the main peak. This all seemed appropriate if there was some sound leakage from the nearer speaker.
Amplifier crosstalk increases with frequency so if you measure for it there will be other high frequencies above the noise floor as well.

Thanks jtalden. Any pointers on where I can read up more on this topic?

 

[jtalden]

Sorry, I don't.