Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Mark Wilkinson]

I'm trying to understand why Smaart appears to have so much trouble finding the delay for bandwidth limited signals.

Always known the difficulty in finding delay for subs, but I've thought that was mostly about long wavelengths requiring long measurement windows, and problems with reflections making their way into the mic.

But I've been comparing x-over filters solely at line level, ie no mic no reflections, and smaart still can't find delay with default settings.

Example:  Reference channel.  Smaart generated Pink noise run through X-32 matrix out:  24db HP LR @100hz and 24db LP LR @650hz.  (A bandpass for Peter Morris's DIY60 mid section)
                Measurement.        Same pink noise, but from unfiltered X-32 matrix out, thru QSC PLD amp output: with the same bandpass filter settings on the PLD as on the reference channel. (amp output gain set to -26db against +26 input gain)

So, it should be two identical signals (if filters are constructed the same by the manufacturers) other than delay, and a little minor unfiltered amp freq/phase response variation.
And filters do look to be constructed the same, when I leave the delay finder set to the delay that smaart finds with unfiltered signal (0.77ms). In fact, any filter I set up equally on both the X-32 reference and PLD measurement, show flat response and phase, using 0.77ms delay.

To me, it would seem two line level signals with no mic involved would make for a pretty clean find, but Smaart cannot find the delay on the bandpassed signal. 
Do I need to adjust windowing or sample size parameters?

I guess I should be asking how the delay finder works....
My understanding is that it looks for the freq of highest energy and calculates delay there???
Is this a representation of the term Group Delay??
My understanding of Group Delay is that it is an average taken from some chosen slice within the measured bandwidth???  In the case of how smaart works, with highest energy being the chosen slice?
If this is so, how do i change the chosen slice / can smaart do it?

Many thanks !!!


   

[Merlijn van Veen]

I'm trying to understand why Smaart appears to have so much trouble finding the delay for bandwidth limited signals.

Always known the difficulty in finding delay for subs, but I've thought that was mostly about long wavelengths requiring long measurement windows, and problems with reflections making their way into the mic.

But I've been comparing x-over filters solely at line level, ie no mic no reflections, and smaart still can't find delay with default settings.

Example:  Reference channel.  Smaart generated Pink noise run through X-32 matrix out:  24db HP LR @100hz and 24db LP LR @650hz.  (A bandpass for Peter Morris's DIY60 mid section)
                Measurement.        Same pink noise, but from unfiltered X-32 matrix out, thru QSC PLD amp output: with the same bandpass filter settings on the PLD as on the reference channel. (amp output gain set to -26db against +26 input gain)

So, it should be two identical signals (if filters are constructed the same by the manufacturers) other than delay, and a little minor unfiltered amp freq/phase response variation.
And filters do look to be constructed the same, when I leave the delay finder set to the delay that smaart finds with unfiltered signal (0.77ms). In fact, any filter I set up equally on both the X-32 reference and PLD measurement, show flat response and phase, using 0.77ms delay.

To me, it would seem two line level signals with no mic involved would make for a pretty clean find, but Smaart cannot find the delay on the bandpassed signal. 
Do I need to adjust windowing or sample size parameters?

I guess I should be asking how the delay finder works....
My understanding is that it looks for the freq of highest energy and calculates delay there???
Is this a representation of the term Group Delay??
My understanding of Group Delay is that it is an average taken from some chosen slice within the measured bandwidth???  In the case of how smaart works, with highest energy being the chosen slice?
If this is so, how do i change the chosen slice / can smaart do it?

Many thanks !!!

Hi Mark,


Good question! What most people do not realize is that our dual-channel FFT analyzers are linear by nature. The pseudo-logarithmic response we see on our screens is composed of multiple bands of linear resolution.

The band limited response up to 650 Hz occupies approximately 50% of a logarithmic frequency scale BUT on a linear scale it's about 2.7% of all available data for the analyzer to work with.

Let's assume your Smaart system runs at its default sample rate of 48 kHz. The usable bandwidth will be 24 kHz (Nyquist). So out of your signal generator comes 24k kHz of unfiltered pink noise data. Your measurement input on the other hand sees only 650 Hz of filtered pink noise data returned.

650 / 24,000 = 0.027

So of the 100% of data send away only 2.7% is returned. The amplitude of the impulse response will drop to 20*log(0.027) = -31 dB and is stretched out over time (group delay) by the 360° of phase shift introduced by the 4th order LR filter.

In other words there's insufficient data for the analyzer to work with even though on a logarithmic scale is occupies half the screen.

Subwoofers are even worse. Let's assume 100 Hz operational bandwidth of the subwoofer. That's only 100 / 24,000 = 0.004% = 4 ‰ of data returned and amplitude drops to 20*log(0.004) = -48 dB!!!

Wavelength and measurement windows have nothing to do with it. Reflections will contaminate and time smear the returned measurement signal even more making the job for the analyzer even harder.

Check out this excellent Meyer Sound webinar by Mauricio "Magu" Ramírez.


Regards,


Merlijn van Veen

www.merlijnvanveen.nl

[Ivan Beaver]

Here is a "simple" way of thinking about it.

Draw a line on a road.

It is pretty easy to say when a motorcycle crosses the line-because it is small.  Like a high freq signal.

But "when" does a tractor trailer "cross" the line"?  When the nose touches it?  When it is in the middle"  Or the end?

The truck is larger (like a low freq waveform).

Or look at different waves on an oscope.  For a given amplitude and fixed time setting, the higher freq are MUCH easier to see where they "arrive" vs lower freq.

There are a lot of "assumptions" and data points the writers of the programs must enter.

Not all situations fit within the "easy/simple" definitions.   

Just like power capacity of a loudspeaker and power output of an amplifier.

It is pretty easy for a simple fixed test condition-but that condition never exists in the real world.

And if you want to go down the road of "what is the real world", exactly what music or test signal would no choose that could describe all forms of music.

There isn't one.

The nice thing about standards is that there are so many to choose from-----------

[Doug Fowler]

Hi Mark,


Good question! What most people do not realize is that our dual-channel FFT analyzers are linear by nature. The pseudo-logarithmic response we see on our screens is composed of multiple bands of linear resolution.

The band limited response up to 650 Hz occupies approximately 50% of a logarithmic frequency scale BUT on a linear scale it's about 2.7% of all available data for the analyzer to work with.

Let's assume your Smaart system runs at its default sample rate of 48 kHz. The usable bandwidth will be 24 kHz (Nyquist). So out of your signal generator comes 24k kHz of unfiltered pink noise data. Your measurement input on the other hand sees only 650 Hz of filtered pink noise data returned.

650 / 24,000 = 0.027

So of the 100% of data send away only 2.7% is returned. The amplitude of the impulse response will drop to 20*log(0.027) = -31 dB and is stretched out over time (group delay) by the 360° of phase shift introduced by the 4th order LR filter.

In other words there's insufficient data for the analyzer to work with even though on a logarithmic scale is occupies half the screen.

Subwoofers are even worse. Let's assume 100 Hz operational bandwidth of the subwoofer. That's only 100 / 24,000 = 0.004% = 4 ‰ of data returned and amplitude drops to 20*log(0.004) = -48 dB!!!

Wavelength and measurement windows have nothing to do with it. Reflections will contaminate and time smear the returned measurement signal even more making the job for the analyzer even harder.

Check out this excellent Meyer Sound webinar by Mauricio "Magu" Ramírez.


Regards,


Merlijn van Veen

www.merlijnvanveen.nl

Well done. 

[Mark Wilkinson]

Hi Mark,


Good question! What most people do not realize is that our dual-channel FFT analyzers are linear by nature. The pseudo-logarithmic response we see on our screens is composed of multiple bands of linear resolution.

The band limited response up to 650 Hz occupies approximately 50% of a logarithmic frequency scale BUT on a linear scale it's about 2.7% of all available data for the analyzer to work with.

Let's assume your Smaart system runs at its default sample rate of 48 kHz. The usable bandwidth will be 24 kHz (Nyquist). So out of your signal generator comes 24k kHz of unfiltered pink noise data. Your measurement input on the other hand sees only 650 Hz of filtered pink noise data returned.

650 / 24,000 = 0.027

So of the 100% of data send away only 2.7% is returned. The amplitude of the impulse response will drop to 20*log(0.027) = -31 dB and is stretched out over time (group delay) by the 360° of phase shift introduced by the 4th order LR filter.

In other words there's insufficient data for the analyzer to work with even though on a logarithmic scale is occupies half the screen.

Subwoofers are even worse. Let's assume 100 Hz operational bandwidth of the subwoofer. That's only 100 / 24,000 = 0.004% = 4 ‰ of data returned and amplitude drops to 20*log(0.004) = -48 dB!!!

Wavelength and measurement windows have nothing to do with it. Reflections will contaminate and time smear the returned measurement signal even more making the job for the analyzer even harder.

Check out this excellent Meyer Sound webinar by Mauricio "Magu" Ramírez.


Regards,


Merlijn van Veen

www.merlijnvanveen.nl

Hi Merlijn,

Thanks for your helpful, clear reply! Very kind.
I also watched the webinar you linked.
The nature of FFT workings began to de-cloud some. (Want to watch the webinars on filters)

Then I got the idea to try to drop sample rate as low as Smaart (7) would allow, but found Smaart's MTW only supports 48000 or 44100.  So I started reading about MTW vs FPPO vs 16K FFT....and how Time Constants effect the measurements......

Now I'm as lost as ever haha.

Do you have any webinar or other links, besides the Smaart forum, that you would recommend to a beginner?

Thx again, and Best,   Mark

[Mark Wilkinson]

Here is a "simple" way of thinking about it.

Draw a line on a road.

It is pretty easy to say when a motorcycle crosses the line-because it is small.  Like a high freq signal.

But "when" does a tractor trailer "cross" the line"?  When the nose touches it?  When it is in the middle"  Or the end?

The truck is larger (like a low freq waveform).

Or look at different waves on an oscope.  For a given amplitude and fixed time setting, the higher freq are MUCH easier to see where they "arrive" vs lower freq.

There are a lot of "assumptions" and data points the writers of the programs must enter.

Not all situations fit within the "easy/simple" definitions.   

Just like power capacity of a loudspeaker and power output of an amplifier.

It is pretty easy for a simple fixed test condition-but that condition never exists in the real world.

And if you want to go down the road of "what is the real world", exactly what music or test signal would no choose that could describe all forms of music.

There isn't one.

The nice thing about standards is that there are so many to choose from-----------

Thx Ivan, I'm a big fan of keeping it simple   I never believe i really understand anything till it seems simple enough to explain to a middle school kid...

[Peter Morris]

I'm trying to understand why Smaart appears to have so much trouble finding the delay for bandwidth limited signals.

Always known the difficulty in finding delay for subs, but I've thought that was mostly about long wavelengths requiring long measurement windows, and problems with reflections making their way into the mic.

But I've been comparing x-over filters solely at line level, ie no mic no reflections, and smaart still can't find delay with default settings.

Example:  Reference channel.  Smaart generated Pink noise run through X-32 matrix out:  24db HP LR @100hz and 24db LP LR @650hz.  (A bandpass for Peter Morris's DIY60 mid section)
                Measurement.        Same pink noise, but from unfiltered X-32 matrix out, thru QSC PLD amp output: with the same bandpass filter settings on the PLD as on the reference channel. (amp output gain set to -26db against +26 input gain)

So, it should be two identical signals (if filters are constructed the same by the manufacturers) other than delay, and a little minor unfiltered amp freq/phase response variation.
And filters do look to be constructed the same, when I leave the delay finder set to the delay that smaart finds with unfiltered signal (0.77ms). In fact, any filter I set up equally on both the X-32 reference and PLD measurement, show flat response and phase, using 0.77ms delay.

To me, it would seem two line level signals with no mic involved would make for a pretty clean find, but Smaart cannot find the delay on the bandpassed signal. 
Do I need to adjust windowing or sample size parameters?

I guess I should be asking how the delay finder works....
My understanding is that it looks for the freq of highest energy and calculates delay there???
Is this a representation of the term Group Delay??
My understanding of Group Delay is that it is an average taken from some chosen slice within the measured bandwidth???  In the case of how smaart works, with highest energy being the chosen slice?
If this is so, how do i change the chosen slice / can smaart do it?

Many thanks !!!


   

The delay finder works by taking the impulse response of the signal measured by microphone, and then measuring the delay to the tallest peak of the IR relative to the initial signal. The catch is the tallest peak does not always represent the start of the signal.

It’s important to understand that the output of a dual channel (Smaart) system is an impulse response (Time Domain).  This can be converted to your “normal” frequency and phase response (Frequency Domain) using a Fast Fourier Transform … or back to the impulse response by an Inverse Fourier Transform.

Once you have an appreciation of all of this you will see that the impulse response actually tells you a lot about a speaker’s behaviour. That’s why I kept referencing the  impulse response of the DIY Mid Hi in my forum posts - its almost perfect.
`
Group delay is technically the first derivative of phase Vs frequency. You could say it’s more or less the time the signal takes to emerge from the speaker as a function of frequency.

Picture is from Smaart's manual.

[Mark Wilkinson]

The delay finder works by taking the impulse response of the signal measured by microphone, and then measuring the delay to the tallest peak of the IR relative to the initial signal. The catch is the tallest peak does not always represent the start of the signal.

It’s important to understand that the output of a dual channel (Smaart) system is an impulse response (Time Domain).  This can be converted to your “normal” frequency and phase response (Frequency Domain) using a Fast Fourier Transform … or back to the impulse response by an Inverse Fourier Transform.

Once you have an appreciation of all of this you will see that the impulse response actually tells you a lot about a speaker’s behaviour. That’s why I kept referencing the  impulse response of the DIY Mid Hi in my forum posts - its almost perfect.
`
Group delay is technically the first derivative of phase Vs frequency. You could say it’s more or less the time the signal takes to emerge from the speaker as a function of frequency.

Picture is from Smaart's manual.

Hi Peter, thanks!

Your post was very timely....I spent last night clawing into Fourier transforms....and can now see what you are saying....that it all starts with the impulse response.  Clever dude, Fourier.

Where (what freq) is the first derivative of phase taken... for Group Delay?  The tallest peak, the "catch"?

[Peter Morris]

Hi Peter, thanks!

Your post was very timely....I spent last night clawing into Fourier transforms....and can now see what you are saying....that it all starts with the impulse response.  Clever dude, Fourier.

Where (what freq) is the first derivative of phase taken... for Group Delay?  The tallest peak, the "catch"?

Have a look at Charlie's paper regarding the tallest peak. (pgs. 2 - 9)

http://www.excelsior-audio.com/Publications/AES129_RH_Charlie_Hughes_Subwoofer_Alignment_with_a_Full-Range_System.pdf
https://www.youtube.com/watch?v=yoASUFGPWwg

Not sure I understand your phase / group delay question ... its the first derivative with respect to frequency ... i.e. more or less the slope of the phase trace at any given point/frequency.

FWIW I tend not use the delay finder that much, but try an overlap the phase traces - here is a picture of the DIY low to sub alignment. Its measured very close to the speaker, so the magnitude will not be accurate but the time should be.

Note: the coherence is for the high-pass signal

[Mark Wilkinson]

Have a look at Charlie's paper regarding the tallest peak. (pgs. 2 - 9)

http://www.excelsior-audio.com/Publications/AES129_RH_Charlie_Hughes_Subwoofer_Alignment_with_a_Full-Range_System.pdf
https://www.youtube.com/watch?v=yoASUFGPWwg

Not sure I understand your phase / group delay question ... its the first derivative with respect to frequency ... i.e. more or less the slope of the phase trace at any given point/frequency.

FWIW I tend not use the delay finder that much, but try an overlap the phase traces - here is a picture of the DIY low to sub alignment. Its measured very close to the speaker, so the magnitude will not be accurate but the time should be.

Note: the coherence is for the high-pass signal

Thx Peter,  the idea of aligning first arrivals via phase trace is getting more and more clear.
 
Phase alignment is what i have been trying to do all along, but i also want to understand how delay finder works, and what the terms such group delay, minimum phase, etc, all mean.

I can grasp that group delay is the derivative with respect to frequency, i was just have trouble picturing 'derivative at what frequency or frequency interval'.
Here's a pretty good link i found on that ...http://na.support.keysight.com/pxi/help/latest/Tutorials/Group_Delay6_5.htm

If you or others have further reading you've found that helps...much appreciated !!!!

[ProSoundWeb Community]