Anechoic type frequency response plots possible?

[natureguyusa]

I would like to obtain a frequency response scan of a speaker that excludes room reflections as if the speaker was in an anechoic chamber. I would like to know if REW can do these types of scans using some kind of impulse response analysis or time delay spectrometry techniques. I am trying to test individual speakers and I am getting a lot of room reflections issues that is making it difficult to acurrately obtain real data. If REW can do this, great, give me a path of how to set the program up to do so, please. I can not find a way by reading the instructions as written. Since REW can do so many things, I was hoping it would have a way to do this. Otherwise, I might be stuck buying into programs in the $1000 range.

 

[Matthew J Poes]

I would like to obtain a frequency response scan of a speaker that excludes room reflections as if the speaker was in an anechoic chamber. I would like to know if REW can do these types of scans using some kind of impulse response analysis or time delay spectrometry techniques. I am trying to test individual speakers and I am getting a lot of room reflections issues that is making it difficult to acurrately obtain real data. If REW can do this, great, give me a path of how to set the program up to do so, please. I can not find a way by reading the instructions as written. Since REW can do so many things, I was hoping it would have a way to do this. Otherwise, I might be stuck buying into programs in the $1000 range.

You are probably talking about windowing. It is possible to add windows in REW. However it is not a free lunch. First, you have to take measurements that have the minimum of early room interactions, because you can’t window those out without loosing too much resolution. Then you can investigate the impulse and find the first big reflection and set the window to cut off after that point. In my living room I can obtian about 15-17hz resolution that is close to accurate.

Otherwise you need to measure outside.

When you use this technique the results are only accurate down to about 300-500hz or So. You can extend this by absorbing the floor and ceiling bounce, which in most rooms is the first major early reflection, and generally clearing the room of reflective surfaces.

I did just this yesterday and am attaching the results here:

This has been confirmed against both outdoor and NRC anechoic results and is generally correct.

Now as for more sophisticated ways of doing this using actual scanning. If you know of software for $1000 that can do that, let me know! As far as I know, only folks like Klippel have commercialized this. It’s know as sound field expansion and uses a transformation of spherical harmonics. It’s a complex mathematical analysis of data that REW cannot do. Matlab can do it and some Matlab functions for this exists. None of this is $1000 or cheap.

Earl Geddes has developed a DIY system for this and has shared how he does it on DIY audio. Someone over there was trying to recreate this. However as far as I know it was never released in a shareable format.

I have ultimately decided it’s easier to take the speakers outside and measure on a tall stand. I’ve also looked into building a small anechoic chamber in a shed. All of this has seemed easier for me than trying to recreate what Klippel is doing.

In the above measurements, I’ll be using that data to create a hemispherical response for that speaker. I will use a software program to do something similar. It interpolates the missing polar data giving me 1 degree or better resolution even though the source data is 5 degree or wider. I will be using nearfield measurements of the loudspeaker taken on the floor (groundplain) and splicing together. This should make it accurate to below 100hz within reason. However what I’m doing won’t be quite on par with Klippel or Geddes.

 

[natureguyusa]

You are probably talking about windowing. It is possible to add windows in REW. However it is not a free lunch. First, you have to take measurements that have the minimum of early room interactions, because you can’t window those out without loosing too much resolution. Then you can investigate the impulse and find the first big reflection and set the window to cut off after that point. In my living room I can obtian about 15-17hz resolution that is close to accurate.

Otherwise you need to measure outside.

When you use this technique the results are only accurate down to about 300-500hz or So. You can extend this by absorbing the floor and ceiling bounce, which in most rooms is the first major early reflection, and generally clearing the room of reflective surfaces.

I did just this yesterday and am attaching the results here:
View attachment 12671
This has been confirmed against both outdoor and NRC anechoic results and is generally correct.

Now as for more sophisticated ways of doing this using actual scanning. If you know of software for $1000 that can do that, let me know! As far as I know, only folks like Klippel have commercialized this. It’s know as sound field expansion and uses a transformation of spherical harmonics. It’s a complex mathematical analysis of data that REW cannot do. Matlab can do it and some Matlab functions for this exists. None of this is $1000 or cheap.

Earl Geddes has developed a DIY system for this and has shared how he does it on DIY audio. Someone over there was trying to recreate this. However as far as I know it was never released in a shareable format.

I have ultimately decided it’s easier to take the speakers outside and measure on a tall stand. I’ve also looked into building a small anechoic chamber in a shed. All of this has seemed easier for me than trying to recreate what Klippel is doing.

In the above measurements, I’ll be using that data to create a hemispherical response for that speaker. I will use a software program to do something similar. It interpolates the missing polar data giving me 1 degree or better resolution even though the source data is 5 degree or wider. I will be using nearfield measurements of the loudspeaker taken on the floor (groundplain) and splicing together. This should make it accurate to below 100hz within reason. However what I’m doing won’t be quite on par with Klippel or Geddes.

Thanks Matthew! I will definately look into windowing. I have no idea how to do that yet. Outdoor scanning is problematic for me. Here are some programs I found that are supposed to do quasi anechoic response: OpenTDS, a free program on Github, written in the Juice Framework. I haven't figured out what a Juice Framework is or how to get it to run yet, but it uses sound cards without needing external boxes. Goldline TEF Analyzer System $2100, with extra software modules available. Audiomatica Pocket CLIO for around $650 uses LogChirp impulse response analysis for anachoic frequency response. Audiomatica also has a $2100 unit that is better. AFMG Easera for $780 + $780 for the TEF/TDS software module. Smaart V8 from Rational Acoustics for $1000, but it isn't clear if they have a software module to do this. I spoke with a sales rep from Listen, and their SoundCheck software will definately do anechoic response on your desktop, but at a cost of around $11,000. It looks like Klippel runs in the same price range as Listen.

 

[Matthew J Poes]

None of those systems do what you think. You can’t just create anechoic from no where. Anechoic requires an anechoic chamber or free space environment. What they all do is offer different ways of doing windowing and splicing with nearfield measurements to create a semi-anechoic like measurement. It isn’t perfect but good enough.

Klippel is around $30,000 or so for a full package, which is needed or the 3D scanning to work.

I know a guy who uses EASE with TEF, he had to convert his garage into a hemispherical anechoic chamber to get anechoic results.

http://www.audiomatica.com/wp/wp-content/uploads/polar_measurements_and_ease-1.pdf

Notice that in Audiomaticas instructions for anechoic measurements, it says you need a suitable anechoic space.

This explains how to apply a window to the measurement in REW.

 

[Matthew J Poes]

@John Mulcahy can you explain a little REW works vs Time Delay Spectrometry? Does TDS offer any advantages in creating a quasi-anechoic measurement of a speaker can the techniques I’ve described.

 

[John Mulcahy]

No, TDS is essentially equivalent to windowing the impulse response. For more detail see Swen Müller, Paulo Massarani, "Transfer Function Measurement Using Sweeps", JAES Volume 49 Issue 6 pp. 443-471, June 2001. An extended version of the paper is available at CiteSeerX.

Did you mean to include a link in your previous post? ("This explains how to apply a window to the measurement in REW").

 

[Matthew J Poes]

I sure did! It was just a link to @John Mulcahy own material.

https://www.roomeqwizard.com/help/help_en-GB/html/impulseresponse.html

This might be another good example for a video.

 

[natureguyusa]

Ok, thank you for the infomation on windowing. I figured it out how to see the IR and adjust a window on a scan. It helped, but one issue that I see is that when I make a reference cal scan with a reference mic, I am unable to apply a window on the reference scan. Therefore, when I scan with a reference mic again, then apply the windowing, the scan is not flat, as it should be if based on the reference calibration scan. Is there a way to apply a window to the reference cal file so that subsequent scans, with the same window, are referenced on that file (and corrected relative to the reference file)? This is using a flat, referenced speaker as well.

 

[Matthew J Poes]

Ok, thank you for the infomation on windowing. I figured it out how to see the IR and adjust a window on a scan. It helped, but one issue that I see is that when I make a reference cal scan with a reference mic, I am unable to apply a window on the reference scan. Therefore, when I scan with a reference mic again, then apply the windowing, the scan is not flat, as it should be if based on the reference calibration scan. Is there a way to apply a window to the reference cal file so that subsequent scans, with the same window, are referenced on that file (and corrected relative to the reference file)? This is using a flat, referenced speaker as well.

Can you either describe a bit more what you are trying to do or show me an example? I am not sure what you mean by reference calibration scan. What are you calibrating? Reference to what? Once I know what you are trying to do I can probably explain if what you are trying to do is possible and how.

 

[John Mulcahy]

Do you mean you are running the soundcard calibration? What makes you say you can't apply a window to the result?

 

[natureguyusa]

Matthew, I will try to explain better. John, yes, I mean soundcard calibration. I can apply a window to the result, but I am unclear if applying a window to the result is changing the cal file operation of zeroing out the base system and therefore subsequent scans are more accurate scans of the decive under test. When I used the sound card calibration with a reference mic, the resulting rescan of that reference mic was nearly perfectly flat. When I applied a window to the results, I then start seeing boggles showing up in what should be a flat response which makes me think that the sound card cal file is not "windowed". I generally would like to test speaker builds and microphones. Currently, I am working on some microphone builds where I would like to critically find out the frequency response of them and see what the directivity looks like. I was hoping that in combination of sound card calibration and windowing would zero out the speaker and reference microphone response and room effects so that that I can run scans on a device under test to see it's frequency response much more accurately. Windowing sounds really promising to get rid of some room effects as I was previously getting problematic scans due to room reflections. Am I doing this right or are there better ways of getting REW set up to do what I am trying to do?

 

[John Mulcahy]

You are comparing two different things. If you want the response to be flat with a certain window, apply that window to the calibration measurement before you save it as a cal file. The cal file remains as you saved it, changing the window on your measurement doesn't change the cal file.

 

[natureguyusa]

Awesome John! That's what I was missing. I ran some mic scans and I am getting good, consistent measurements now. Wow, what a difference! Thank you very much!