Evaluating loudspeakers

If you don't have a lot of money to spend but you want to estimate the
frequency response of different loudspeaker setups what is the best
way to do it?

I bought a very nice microphone and built a box with foam on the
inside. My idea is to place the loudspeaker and the mic in the box and
measure the power level at different frequencies by playing a sine
tone at those frequencies.

Or is it better to have the microphone and loudspeaker in an open
space where the mic is 10 inches away from the loudspeaker?

I'm sure neither of the 2 methods above are valid for making such
measurements (if so please explain why), but how should I do then?

Also, should I use sine tones or is it better to use another test
signal like an "impulse", "white" noise or a chirp signal?

Hoping to get some good answers / advice

Thank you!

"Mauritz Jameson" <mjames2393@gmail.com> wrote:

> If you don't have a lot of money to spend but you want to estimate the
> frequency response of different loudspeaker setups what is the best
> way to do it?
>
> I bought a very nice microphone and built a box with foam on the
> inside.

Surprisingly, an electret capsule from RadioShack would do just as good.

> My idea is to place the loudspeaker and the mic in the box and
> measure the power level at different frequencies by playing a sine
> tone at those frequencies.
> Or is it better to have the microphone and loudspeaker in an open
> space where the mic is 10 inches away from the loudspeaker?

Without special anechoic room, the results are going to be wildly different 
from one location to another.
At 10 inches, the results would be very different also as they are 
determined by near effects such as axis misalignment.

> I'm sure neither of the 2 methods above are valid for making such
> measurements (if so please explain why), but how should I do then?

It would be difficult to get meaningful measurements without anechoic room.

> Also, should I use sine tones or is it better to use another test
> signal like an "impulse", "white" noise or a chirp signal?

There are softwares that measure frequency response using pseudo - noise 
signal. They claim that they could discriminate the reverberation from main 
response; as such they are less affected by echoes; at least in theory. 
Don't know how well does it work in practice.

Vladimir Vassilevsky
DSP and Mixed Signal Consultant
www.abvolt.com

There will be dozens of opinions on this by lots of very smart people, and there is no universally accepted method. But one thing that is generally agreed on is that the radiation pattern as a function of frequency is very important, so you can't just measure the on-axis response in an anechoic environment. Some people use a circular array of microphones around the speaker and sum the powers of all the microphones. 
Impulse testing has some drawbacks, as you need to average many results to get a decent to noise ratio, but one advantage is that you can chop off the room response assuming there is sufficient separation in time between the true impulse response and the start of the room response. 

Bob

On 10/9/12 11:31 PM, Robert Adams wrote:
> Impulse testing has some drawbacks, as you need to average many results to get a decent to noise ratio, but one advantage is that you can chop off the room response assuming there is sufficient separation in time between the true impulse response and the start of the room response.


On 10/9/12 11:18 PM, Vladimir Vassilevsky wrote:
>
> It would be difficult to get meaningful measurements without anechoic room.
>


"Mauritz Jameson"<mjames2393@gmail.com>  wrote:
>
>> Also, should I use sine tones or is it better to use another test
>> signal like an "impulse", "white" noise or a chirp signal?


back in the '70s and '80s a guy named Richard Heyser gained a lot of 
note and respect in the AES when he introduced what he called 
"Time-delay spectrometry" (TDS) as a means of measuring anechoic 
response in a non-anechoic environment.  but, as Bob points out, the 
impulse response of the loudspeaker has to be sufficiently separated 
from the reflections so you can separate them.  this must be the case no 
matter how you measure it.

TDS was essentially linear-frequency swept-frequency measurements.  a 
chirp.  and another thing i heard from Bob about 2 decades ago is that 
if you ever think you've invented something, better check the Bell 
Systems Journal first.  so the driving signal that Heyser was using is a 
chirp:

    x(t)  =  e^( j*pi*beta*t^2 )

          =  cos( pi*beta*t^2 )  +  j*sin( pi*beta*t^2 )


and if you do it right, you need to do both a cosine chirp and a sine 
chirp (and attach that result to j) and start at some negative frequency 
and pass through DC up to a positive limit.

the output is

    y(t)  =   h(t) (*) x(t)    where (*) means convolution

what Heyser suggested, before there were PCs, was to follow the result 
with a tracking band-pass filter so that room reflections (that are 
delayed in time so they don't have the same instantaneous frequency) are 
filtered out.  otherwise, you can examine this analytically as driven by 
a chirp.

Techron came out with a box (using the CP/M operating system) that did 
this called "TEF" (for Time-Energy-Frequency)

you can also try using Maximum Length Sequences (MLS, probably has other 
names) to get a total response with the room and, after examination, 
removing the reflections assuming they don't overlap.  i once wrote a 
quickie tutorial about MLS that is at the dspguru.com site somewhere.

both swept-sinusoid and MLS have a better crest factor (peak-to-rms 
ratio) than does an impulse.  if you do an impulse driving signal, 
you'll need to keep it low enough to not saturate anything and then run 
it many times and coherently average the output.

do you have some programmable DSP box with an A/D and D/A converters 
that you can use for this?  if you're doing this with a laptop and the 
built-in sound I/O, you better find out what the internal delay is to 
and from the converters.  you need to record the response synchronously 
with the driving signal, no matter how you do it.


-- 

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."

"robert bristow-johnson" <rbj@audioimagination.com> wrote in message 
news:k52t7i$kr4$1@dont-email.me...
> On 10/9/12 11:31 PM, Robert Adams wrote:

> you can also try using Maximum Length Sequences (MLS, probably has other 
> names) to get a total response with the room and, after examination, 
> removing the reflections assuming they don't overlap.

It is no problem to run white noise through the speakers and get a model of 
the room + speaker by LMS algorithm, for example.
With the speaker main resonance at tens of Hz, it is hard if possible to 
distinguish where the response of the speaker ends and where the response of 
the room starts.

VLV
 

>> what is the best way to do it?
go outdoors.

Careful with white noise at higher levels, it will burn the tweeters.

>> where the mic is 10 inches 

at such a short distance, you're still in the near-field (where your
"antenna" is physically large).

If you're evaluating "near-field" studio monitors, the manual will tell you
where to put your ear and where not. 

If it's about a line array (which is physically large by design), the
question of microphone placement and "room" acoustics that need to be taken
into account for meaningful results would be totally different.

This only as examples to rule out any single "correct" answer.

>If you don't have a lot of money to spend but you want to estimate the
>frequency response of different loudspeaker setups what is the best
>way to do it?

Do you want to estimate the frequency response of just the loudspeaker or
the loudspeaker+room?

>I bought a very nice microphone and built a box with foam on the
>inside. My idea is to place the loudspeaker and the mic in the box and
>measure the power level at different frequencies by playing a sine
>tone at those frequencies.

OK, I assume you're interested in the response of just the loudspeaker.


>Or is it better to have the microphone and loudspeaker in an open
>space where the mic is 10 inches away from the loudspeaker?

It is all about minimizing the influence of reflections so if your box does
a very good job of suppressing reflections then maybe you can go with
that.

>I'm sure neither of the 2 methods above are valid for making such
>measurements (if so please explain why), but how should I do then?

This works for me so maybe it will also work for you. The couple of times I
had to measure a loudspeaker I start by putting the speaker on a table
right on the edge of the table so that reflections from the surface of the
table can be neglected. Then I do my best to clear the area in front of the
speaker in a radius of say 1.5 to 2 meters. Then I play MLS sequences and
record them. For each recording I move the microphone a little. Then I
average the impulse responses obtained from the recorded MLS sequences to
get ONE averaged impulse response. When you FFT this impulse response and
plot the magnitude spectrum you will most likely see that it is not smooth.
You will probably see the effect of reflections in your impulse response,
they have a comb filter like effect on your spectrum. So you want to
truncate your impulse response to get rid of the reflections. You can
probably find the place to truncate your impulse response by just looking
at a plot of the impulse response or you can find it by finding the first
notch in the magnitude spectrum. After you truncate your impulse response
you will see that your speaker response becomes much smoother. 

Cheers

Just a quick note that Floyd Toole wrote about this extensively, for example

http://www.aes.org/tmpFiles/elib/20121010/5276.pdf

Or

http://www.harmanaudio.com/all_about_audio/loudspeakers_rooms.pdf

I recall that many people use a technique where the low frequency response is done using very close mic spacing or possibly in an anechoic chamber,  and then this is stitched together with the higher frequency response, which might be obtained from the truncated impulse response. I forget the details on this but it does seem to be common practice.  

There is a lot of debate about whether or not you should apply EQ to compensate for the room + speaker or just the speaker. Most authors recommend just applying EQ to compensate for the speaker, as all the details and "grass" in the frequency response curve are a result of reflections, and reflections are not naturally modeled by a cascade of 2nd-order equalization filters.

On a side note, years ago I had a conversation with Floyd about the way that Consumers Reports measures loudspeakers, and he felt very strongly that they didn't know what they were doing, and that listeners did not prefer their top-rated models.  


Bob


Bob

On 10/9/2012 11:18 PM, Vladimir Vassilevsky wrote:

   ...

 > It would be difficult to get meaningful measurements without anechoic 
room.

   ...

I have gotten results that agree well with an anechoic chamber by 
setting up my (battery-operated) equipment in a wheat field where the 
nearest structures or trees were about a quarter mile away. The wheat 
stalks minimize reflections from the ground, and there are no other 
reflectors. Such fields are rare nowadays here in the East.

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