More to a measurement then the transfer function?

D

David Sturzenbecher

Junior
Jan 15, 2011
345
7
18
Hello,
I have been thinking quite a lot about transfer function measurements, specifically those capturing the crossover, EQ, and FIR processes taking place inside of a DSP. If I am viewing a transfer function (Magnitude and Phase) of a DSP process, does that show me everything that is going on? I know some manufactures implement horn reflection compensation through FIR processes. How would those manifest themselves inside a measured transfer function? Would you see a secondary arrival in the impulse response and need to window to that?

-Thanks
Dave
 
Re: More to a measurement then the transfer function?

David Sturzenbecher said:
Hello,
I have been thinking quite a lot about transfer function measurements, specifically those capturing the crossover, EQ, and FIR processes taking place inside of a DSP. If I am viewing a transfer function (Magnitude and Phase) of a DSP process, does that show me everything that is going on? I know some manufactures implement horn reflection compensation through FIR processes. How would those manifest themselves inside a measured transfer function? Would you see a secondary arrival in the impulse response and need to window to that?

-Thanks
Dave
Click to expand...
The transfer function gives some information, but no where near the "whole picture".

With a normal loudspeakers with separated drivers, changing the crossover point will change the polar pattern etc.

When looking at just a DSP, how would you get a second arrival-unless you are doing multipath and have a different delay in the paths?

A second arrival may or may not appear on the transfer function.

It will depend on the size of the time window you are using the amount of time between the 1st and 2nd signal arrivals.

If the 2nd arrival fits in the time window, then you will see it as a classic combfilter and a series of notches in the response that are equally spaced in FREQ, On a linear scale they will look like the teeth on a comb (hence the name), but on a linear scale they will get closer and closer together as the freq increases

So-as usual-the answer is "it depends".
 
Re: More to a measurement then the transfer function?

"Horn correction" type processing will not be happening so late as to be something you'd look at with impulse response. If anything you'd be looking for oddities in the phase trace of the transfer function. And of course all that fancy processing does come at a cost of overall latency. Your measurement mic might be 10 milliseconds from the speaker but the measured delay might add several more mS to that because of the latency.
 
Re: More to a measurement then the transfer function?

Putting on the scientist hat that basically specializes in interpreting measurements, I would say any measurement is only as good as the question the measurement was designed to answer.

Many people with a basic technical background are familiar with the idea of accuracy and precision of a measurement. The next step in statistics is the validity and reliability of the measurement.

What many people are not as familiar with ( and I sometimes struggle with this concept even with my college students) is that you can and should apply the same thought process to the entire experimental design.

In this case, validity comes down to: can the measurement process provide a definitive answer to the research question (hypothesis) being tested.

Applying that to this thread, designing a speaker, tuning a speaker, tuning a system, tuning a system in a particular space, checking the performance of a system, or running a show, all lead me to pose different questions, therefore; the transfer function gets set up and used in different ways.

Sent from my XT1254
 
  • Love
Reactions: Danh QVS
Re: More to a measurement then the transfer function?

Jay Barracato said:
Putting on the scientist hat that basically specializes in interpreting measurements, I would say any measurement is only as good as the question the measurement was designed to answer.

Many people with a basic technical background are familiar with the idea of accuracy and precision of a measurement. The next step in statistics is the validity and reliability of the measurement.

What many people are not as familiar with ( and I sometimes struggle with this concept even with my college students) is that you can and should apply the same thought process to the entire experimental design.

In this case, validity comes down to: can the measurement process provide a definitive answer to the research question (hypothesis) being tested.

Applying that to this thread, designing a speaker, tuning a speaker, tuning a system, tuning a system in a particular space, checking the performance of a system, or running a show, all lead me to pose different questions, therefore; the transfer function gets set up and used in different ways.

Sent from my XT1254
Click to expand...

I guess the main purpose of my question has to do with extracting presets from one piece of gear and putting those presets into another piece. Many times I run into situations where I am told I "have to use a specific piece of gear to use our presets" and many times that piece of gear does not fit into my over system design goal. Situations that the "blessed" manufacturers DSP only has analog I/O, or you need to use this amplifier that has onboard DSP (with our preset) but doesn't carry my specific networked audio protocol. If take a transfer function of the blessed manufacturers product, and match it (phase/magnitude) with the products used in my system design...am I missing anything? I realize that this only applies to the time invariant aspects, and any inherent limiter settings will need to be addressed by other means.
 
Re: More to a measurement then the transfer function?

David Sturzenbecher said:
Hello,
I have been thinking quite a lot about transfer function measurements, specifically those capturing the crossover, EQ, and FIR processes taking place inside of a DSP. If I am viewing a transfer function (Magnitude and Phase) of a DSP process, does that show me everything that is going on? I know some manufactures implement horn reflection compensation through FIR processes. How would those manifest themselves inside a measured transfer function? Would you see a secondary arrival in the impulse response and need to window to that?

-Thanks
Dave
Click to expand...

Hi Dave, I really am far to much of a noob when it comes to measurement, but i have been hitting it pretty hard lately...asking myself the same questions you pose...

Here's as much as I've been able to gather....please correct !

Starting with a snip from one of Dave Gunness' patents http://www.fulcrum-acoustic.com/ass...ng-dsp-filters-improve-transient-response.pdf

Dave G patent snip.JPG

If I understand the patent, blocks 40 and 41 are about finding driver resonances, horn reflections, etc,... that are NOT minimum-phase (IIR) fixable.
It appears Dave calls these driver specific fixes LMA's (Loudspeaker Mechanism Algorithms.)
I don't see how a transfer function can measure this, hopefully somebody can chime in.
I think an impulse response, along with a time sensitive spectrograph or something similar, could show it....

Blocks 42-45 seem to be about more well traveled technique. First use minimum-phase eq, either IIR or FIR replicated IIR.
Then, instead of IIR crossovers, use linear-phase crossovers. Then, keep iterating till your're happy.
I think Dave calls these steps LSA's (Loudspeaker System Algorithms).

To me, 42-45 are what we've been doing all along,....... only as suggested by the patent, crossover is now linear-phase via FIR;
and eq is still done via minimum-phase, but implemented with either FIR or IIR.
I think this tuning is the domain of the transfer function.

I've been having a lot of luck with 42-45, measuring with transfer function, and then generating FIR for both linear-phase crossover, and minimum phase eq.

What I Really want to learn, is how to measure for blocks 40-41, and then how to insert them into FIR generation.
I think that is what you're asking too... right?

Thx, Mark
 
Last edited:
Re: More to a measurement then the transfer function?

Mark Wilkinson said:
What I Really want to learn, is how to measure for blocks 40-41, and then how to insert them into FIR generation.
I think that is what you're asking too... right?

Thx, Mark
Click to expand...

Most of the real research in this area is in developing a model which makes sense for specific observed non-linear (time varying) behavior. Years ago, there was a lot of excitement around higher-order voltera impulse responses, and there are some papers in AES journals and conferences on the subject. I've not really been following it from there, where the bleeding edge is. It seems like looking into research in other similar fields might get you more information. There is nothing to measure if you don't have a working model. Certainly linear transfer functions are not it.

I put a number of references in my AES presentation which would be a good starting place.
http://www.cascadeacoustic.com/files/research/AES-PNW_Audio_Band_Test_and_Measurement.pdf
You can send me a PM if you need help finding anything.
 
Last edited:
Re: More to a measurement then the transfer function?

Mark DeArman said:
Most of the real research in this area is in developing a model which makes sense for specific observed non-linear (time varying) behavior. Years ago, there was a lot of excitement around higher-order voltera impulse responses, and there are some papers in AES journals and conferences on the subject. I've not really been following it from there, where the bleeding edge is. It seems like looking into research in other similar fields might get you more information. There is nothing to measure if you don't have a working model. Certainly linear transfer functions are not it.

I put a number of references in my AES presentation which would be a good starting place.
http://www.cascadeacoustic.com/files/research/AES-PNW_Audio_Band_Test_and_Measurement.pdf
You can send me a PM if you need help finding anything.
Click to expand...

Thank you Mark,
your paper looks spot on for what i seem to need to learn to keep moving on...
...will study...
 
Re: More to a measurement then the transfer function?

Mark Wilkinson said:
Thank you Mark,
your paper looks spot on for what i seem to need to learn to keep moving on...
...will study...
Click to expand...

Well the slides are pretty much bullet points, I wish I had time to write a real paper. If you are interested in the application side of things, I would check out all of Klippel's and Steve Temme's publications from the past 10 years. They are really approachable. I have built a number of specialized apparatus and software instruments for doing transducer measurements based on Klippel's work and also some Japanese fellows (name alludes me I'll see if I can dig it up some time). I really like and appreciate Klippel's work because of the significant concentration on modeling. Steve Temme's company Listen, Inc. produces Sound Check, what I consider the best of piece of loudspeaker measurement software available. Many of his papers are available on their site. Antonin Novak, one of the references from my slides, has really picked up doing the application research for Swept Sine/Matched filtering applications in audio. He has a number of more recent pubs in AES.
 
Last edited:
Re: More to a measurement then the transfer function?

Mark DeArman said:
Well the slides are pretty much bullet points, I wish I had time to write a real paper. If you are interested in the application side of things, I would check out all of Klippel's and Steve Temme's publications from the past 10 years. They are really approachable. I have built a number of specialized apparatus and software instruments for doing transducer measurements based on Klippel's work and also some Japanese fellows (name alludes me I'll see if I can dig it up some time). I really like and appreciate Klippel's work because of the significant concentration on modeling. Steve Temme's company Listen, Inc. produces Sound Check, what I consider the best of piece of loudspeaker measurement software available. Many of his papers are available on their site. Antonin Novak, one of the references from my slides, has really picked up doing the application research for Swept Sine/Matched filtering applications in audio. He has a number of more recent pubs in AES.
Click to expand...

Another big thanks !! Hey, quick question....do you know of a relatively inexpensive accelerometer that could be used for identifying cabinet resonances ?
 
Re: More to a measurement then the transfer function?

Mark Wilkinson said:
Another big thanks !! Hey, quick question....do you know of a relatively inexpensive accelerometer that could be used for identifying cabinet resonances ?
Click to expand...

For speakers I use a laser displacement sensor now. I normally simulate the boxes and never have had a real issue although I have a pile of PCB Piezo stud mount ICP accelerometers I got when we worked on that quad-21in box. You can just check their website, they are for the most part pretty cheap, but you do need an ICP(IEPE) power supply. You can check eBay for the Endevco 2775A, which can normally be had for next to nothing ($20-50). It is a dual mode charge amp/ICP supply and will work for most any sensor you get your hands on. Charge coupled acceleromters are cheap on ebay but be aware they require special low capacitance wire normally with 10-32 connectors and need to be taped down so it doesn't move.
 
Last edited:
Re: More to a measurement then the transfer function?

Mark DeArman said:
For speakers I use a laser displacement sensor now. I normally simulate the boxes and never have had a real issue although I have a pile of PCB Piezo stud mount ICP accelerometers I got when we worked on that quad-21in box. You can just check their website, they are for the most part pretty cheap, but you do need an ICP(IEPE) power supply. You can check eBay for the Endevco 2775A, which can normally be had for next to nothing ($20-50). It is a dual mode charge amp/ICP supply and will work for most any sensor you get your hands on. Charge coupled acceleromters are cheap on ebay but be aware they require special low capacitance wire normally with 10-32 connectors and need to be taped down so it doesn't move.
Click to expand...

Thanks yet again Mark, you provide a wealth of information...
 
Re: More to a measurement then the transfer function?

Mark DeArman said:
For speakers I use a laser displacement sensor now. I normally simulate the boxes and never have had a real issue although I have a pile of PCB Piezo stud mount ICP accelerometers I got when we worked on that quad-21in box. You can just check their website, they are for the most part pretty cheap, but you do need an ICP(IEPE) power supply. You can check eBay for the Endevco 2775A, which can normally be had for next to nothing ($20-50). It is a dual mode charge amp/ICP supply and will work for most any sensor you get your hands on. Charge coupled acceleromters are cheap on ebay but be aware they require special low capacitance wire normally with 10-32 connectors and need to be taped down so it doesn't move.
Click to expand...

Many years ago I did some work on EAW's KF750 HF horn. I glued some small piezo transducer on it and found some resonance in the hard plastic horn flare.

Then I glued some rubber on the flare to deaden it and there was a noticeable improvement in sound quality.

I discussed this with EAW and they eventually sent me some new calcium load flares to test … these then became the standard horn.

@ Mark W … hence the rubber on the 60 DIY horn :-)
 
Re: More to a measurement then the transfer function?

Peter Morris said:
Many years ago I did some work on EAW's KF750 HF horn. I glued some small piezo transducer on it and found some resonance in the hard plastic horn flare.

Then I glued some rubber on the flare to deaden it and there was a noticeable improvement in sound quality.

I discussed this with EAW and they eventually sent me some new calcium load flares to test … these then became the standard horn.

@ Mark W … hence the rubber on the 60 DIY horn :-)
Click to expand...

Back in the old, old days when we were using Altec and JBL metal horns we would standardly coat them with auto undercoating to reduce resonances.
 
Re: More to a measurement then the transfer function?

Peter Morris said:
Many years ago I did some work on EAW's KF750 HF horn. I glued some small piezo transducer on it and found some resonance in the hard plastic horn flare.

Then I glued some rubber on the flare to deaden it and there was a noticeable improvement in sound quality.

I discussed this with EAW and they eventually sent me some new calcium load flares to test … these then became the standard horn.

@ Mark W … hence the rubber on the 60 DIY horn :-)
Click to expand...

Hi Peter, yep...I caught the rubber you added.... and exchange between Mark D and you, back when posted off-axis plots for the DIY60.
I've been hoping to measure stuff like that. You know.. trying to synchronize brain and ears :)

Your DIY 60/90 designs have turned into such great learning platforms....they respond both subtly and boldly...


I'm hoping to initiate a new thread with some phase questions, when it comes to phase relevancy.
Experiments with FIR and what feels like making phase and magnitude overlord....rePhase/minidsp ......are making me both believe in what the the transfer function shows, and at the same time question it too...



To Dave, OP, pls pardon if I've overstepped direction of your thread....I realize my first post was not relating to what you were asking about...
 
Top Bottom