Acoustic room response - measurement and interpretation

Larry McFarren wrote:
> I'd like to get your suggestions on a good way to look at acoustic
> room response for purposed of equalization.
>
> I'm thinking that I can play some wideband tones, which get recorded
> into a PC via a microphone.
>
> Does anyone have any suggestions on the tones that would be good to
> use?
>
> Once the data is recorded in my PC, what type of processing needs to
> take place to 'characterize' the room? (taking the inverse FT, etc.?)
> If someone has a idea on a procedure for this I'd like to hear what
> you have to say.
>
> Thx.

Larry,
I propose a completely different approach. What others and yourself wrote
will give you the room response plus the loudspeaker response and no way to
separate both(or very difficult and unprecise).
Why don't you directly measure the room response?
First you need a stimulus, can be a gunshot, handclap or best a bursting
balloon. The position should be in the place your usual sound source is
located.
Then you put your microphones in the place you want to know the response to
be valid. A good way is to use a sphere with two tiny electret mikes built
in. You will get the direct sound plus the room response. subtract the
direct sound by replacing the beginning with silence until the first
reflection arrives, If the pulse was longer, replace it with another source,
or subtract a previously recorded sample in free air.
-- 
ciao Ban
Bordighera, Italy

Larry McFarren wrote:
> In article <c0gii7$17cg55$1@ID-210375.news.uni-berlin.de>, 
> goldentully@hotmail.com says...
> 
>>What is a "wideband tone"?  The standard ways are an impulse, swept
>>sinusoids, and pink/white noise.  I believe that pink noise is the most
>>commonly used method.
>>
>>Once the data is in the PC, an FFT will give you the room's frequency
>>response.
>>
>>A common method is to use a real-time spectrum analyzer that basically
>>displays a live FFT.  While running noise through the speakers, an equalizer
>>is adjusted in order to make the live FFT response flat.  It's important to
>>use a microphone with a flat frequency response (generally one designed
>>specifically for this purpose) or you will be compensating for that as well.
>>
>>
> 
> 
> Thanks for the feedback.
> 
> Yes, by wideband tone, I was referring to an impulse (wide bandwidth).
> 
> OK, so let's say that I play pink noise out my speakers (left and right at 
> separate times) and collect the data with my PC. I'm not looking to use 
> external eq or spectrum analyzers. This is the functionality that I'd like to 
> implement myself. Also, the microphone calibration is something that I can 
> take care of.
> 
> Assuming that I'm able to record the data into my PC with a fairly high degree 
> of accuracy (mic. calibrated correctly - freq. response) and high SNR, I'm 
> searching for a procedure to analyze the data.
> 
I would like to add that I have been using Maximum Length Sequence
(MLS) to measure room impulse responses (RIR) and loudspeaker impulse
responses and so far it has given a very good performance. Regarding
the flat response of the measuring microphone I want to point out that
the loudspeakers you use to measure the RIR also have to have a flat
response or relatively flat in the audio frequency.

> So, you've mentioned taking a FFT. OK, I believe that I should be able to do 
> this with MATLAB. Any suggestions on the size of the FFT? Any other issues to 
> contend with regarding the FFT? Sorry, I am new to DSP and really need things 
> spelled out at this point.
> 
When doing an FFT analysis I would said that the size of the FFT
depends on the length of your RIR. I usually use an FFT of 2 to 3 times
the length of the RIR to minimize the effect of aliasing since, I
assume that what do you want is to come back to time domain with an
inverse filter (FIR I guess) that compensates for the effect of the
room.

> Now, once I look at the FFT and determine where my peaks are in the freq. 
> response, I will want to reduce their amplitude in order to get a flatter 
> freq. response. I won't be handling the dips in the freq. response just yet. 
> For now, I just want to smooth out the peaks. So how do I go from analyzing 
> the peaks in the FFT to creating a filter, perhaps a FIR filter?

I think is a good idea not to compensate for dips because that
means your inverse filter will have strong peaks, which is not good for
the sake of your loudspeakers, and also because dips are not as audible
as peaks. Ok, when you have your spectrum a simple method to smooth it
is by using an average sliding windows. Now, lets say your smoothed FFT
is H (in complex numbers) then you will need to create invH=1/H
(equalizer filter) but take care that invH could be unstable. To avoid
this you can obtain a minimum phase version of H (say Hmin), which has
the same magnitude but different phase. Therefore you can obtain a
stable inverse FIR filter by applying the inverse FFT to 1/Hmin.

Of course there are some more method to compute inverse filters.

> 
> Thanks again. Please continue your feedback.
> 
> 
> 
> 

Regards,

Pablo

The time to setup a standing wave is at least twice the propagation delay of
the
wave in a room (one reflection). Thus, if a room had a dimension of 100
feet,
a reflection could take (up to) 200 feet to travel (depending on the
position of
the speaker, walls, and microphone). If we were to truncate the measurement
prior to the return of the reflection, it seems the measurement would not be
right.

The speed of sound, IIRC, is about 1150 feet per second.So a 100-foot room
would seem
to require at least 170 milliseconds before the reflection arrives at the
microphone.
Thus, if a swept (stepped?) sine measurement were to occur faster than this,
it seems
that the results would not be correct.

Perhaps there is some technique I am not familiar with where the sine is
swept continuously,
and that prior received energy from a previous time (and thus different
freuqency) can be used to
modify the results?

    -- Tom


"Bob Cain" <arcane@arcanemethods.com> wrote in message
news:c0higa08il@enews2.newsguy.com...
> TOM wrote:
>
> > Swept sine measurements need to be done slowly. Because cancellation is
due
> > to reflection and acoustic delay, the measurement needs to be done
slowly
> > enough to allow the standing wave to develop at each measurement
frequency.
> >
>
> Not true, I'm afraid.  They don't need to develop in order
> to predict from a measurement that they will if sustained.
>
> If that were true the impulse response method wouldn't work
> for other reasons than poor SNR.
>
>
> Bob
>
> -- 
>
> "Things should be described as simply as possible, but no
> simpler."
>
>                                               A. Einstein

TOM wrote:

> The time to setup a standing wave is at least twice the propagation delay of
> the
> wave in a room (one reflection). Thus, if a room had a dimension of 100
> feet,
> a reflection could take (up to) 200 feet to travel (depending on the
> position of
> the speaker, walls, and microphone). If we were to truncate the measurement
> prior to the return of the reflection, it seems the measurement would not be
> right.

Right.

> The speed of sound, IIRC, is about 1150 feet per second.So a 100-foot room
> would seem
> to require at least 170 milliseconds before the reflection arrives at the
> microphone.
> Thus, if a swept (stepped?) sine measurement were to occur faster than this,
> it seems
> that the results would not be correct.

Wrong. You need to continue measurements after the stimulus ends ot 
repeats, but you don't need to keep the stimulus substantially unchanged 
during the time of one room delay. It is conceptually simpler to do so, 
but not required.

> Perhaps there is some technique I am not familiar with where the sine is
> swept continuously,
> and that prior received energy from a previous time (and thus different
> freuqency) can be used to
> modify the results?

Essentially, correlation.

>     -- Tom

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

In article <wEPWb.21467$tD.2040@newssvr24.news.prodigy.com>,
Larry McFarren <invalid@address.com> wrote:
>I'd like to get your suggestions on a good way to look at acoustic room 
>response for purposed of equalization.
>
>I'm thinking that I can play some wideband tones, which get recorded into
>a PC via a microphone.

If you just use one microphone, but want the response of the entire room,
you might need to random walk the mic around the room, perhaps even
with a GPS location recorder attached to weight the response depending
on mic location (e.g. maybe less for the "cheap seats"?).

Otherwise there is the risk that the choosen equalization might make
one spot in the room (the mic location) sound better at the cost of
making the bulk of the room (including even those relatively near to
the mic) sound worse.


IMHO. YMMV.
-- 
Ron Nicholson   rhn AT nicholson DOT com   http://www.nicholson.com/rhn/ 
#include <canonical.disclaimer>        // only my own opinions, etc.