Geddes et al on Measuring Loudspeakers

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hi All --

Earl's thread on waveguides has recently veered into touching on measuring, and I figured it's worthy of a thread of its own. My hope is that not only with Dr. Geddes give us his thoughts, but also for Lynn Olson, Duke, SY, ScottG, GM and others who know what they're doing to drop off a pearl of wisdom or two on the topic.

As for me, I'm laboring through my first complicated speaker project. I've built some widerange-on-an-OBs before, and now I'm working on a 2x12" on OB + horn/compression driver. I made a preliminary post here, but that was a while ago and it's changed since then. In addition to the Jensen A12s, I picked up a couple of the Tone Tubby 12s. And, as well as the Altec 802s I now have one each of the BMS 4550, B&C DE250, Beyma CP380M, and Radian 475PB in the mix.

I have a copy of D'Appolito's "Testing Loudspeakers" and I'm working my way through that. And, I've picked up a Behringer ECM8000, a Rolls MP13 preamp, a licensed copy of ARTA, and built a LM3886 amp (because I didn't have a single SS amp in the house) and a bunch of jigs.

I've made some measurements, but how to know if they're worth anything? Also, while ARTA I'm sure is a boon to those who know what they're doing, its flexibility is a bit daunting for those without the background. What kind of signal generator to use? What are the relative advantages/disadvantages of periodic noise, swept-sine, MLS? Sequence length? What sample rate? How to figure out the delay in order to get usable phase data? Anyways, you get the point.

In addition to a theoretical disucssion on the merits of different measuring techniques, I would think it would also be helpful if those that can speak on some of the practical hurdles in measuring chime in. For instance, I live in suburban Philadelphia. The only time of day quiet enough for me to contemplate measuring drivers outside would result in the police being summoned. I'm stuck doing measurements in my 13x18 living room -- any helpful suggestions?

Hopefully this is enough of a jumping point for a discussion that will prove a boon to both experienced and n00bs alike.

Regards,
John
 
Hi John

You beat me to it. :)

I'm in the same boat - I've found a speaker system that I like enough to want to get that last 5% from, but I'm going to need a much better understanding of meaurement tools to do it. I'm confident that I'm getting good FR data, but beyond that, I'm unsure. This is partly due to my misconceptions, lack of knowledge, basic stupidity, and other personal traits that make me the lovable guy that I am. :D

On Earls' suggestion, I went back and took a fresh look at how I've been using the tools I have, and for the life of me I can't repeat the sound card issues that I was sure existed. There is still a warm-up period needed for the preamp, but from a cold start, I was able to get stable readings within 30 seconds everytime. I think I got things in my head from my initial learning curve that stuck, in spite of them not being totally correct, and probably stem from changing too many things at once and assigning the result to the wrong one. Anyway, I'm ready to move on. The card is good. :)

" What are the relative advantages/disadvantages of periodic noise, swept-sine, MLS? Sequence length? What sample rate? How to figure out the delay in order to get usable phase data? Anyways, you get the point."

Perfect. I would like to see some discussion on these things as well. My grasp of them is limited to the pictures - I can understand the graphs and the waterfalls, but I need to know more about the settings and parameters used to create the output (preferebly without needing a Phd to get a practical understanding of them).
 
There is much more made of the signal IMO than is necessary. Anything that gets us to an impulse response is fine.

Once you have the impulse response then interpretation becomse the issue. If you can't get an impulse response that is believable, repeatable and clean, then thats the first goal.

In theory any and all signals MUST yield the same impulse response. It used to be that time and processing were major limitations and some signals excel at speed, etc. But today, those things are not such an issue, and for us they are not so important. On an assembly line where you have to take a measurement ever few seconds time can be a factor. But none of us is in that much of a hurry.

Don't worry about the signal type, use whatever gets you an impulse response.

Below is a fairly typical impulse response - note reflection starting after the initial response. Note that the waveforem MUST be above zero as much as it is below zero. The woofer is raised by 10 compared to the waveguide. You need to be looking at nice clean impulses like this before you can begin to do anything. You need to be able to find the direct sound and the reflections. If you can't tell where one starts and the other ends then you have too small of a space.

I get back later when I have more time.
 

Attachments

  • impulse.pdf
    34 KB · Views: 422
AJ said:

...I can understand the graphs and the waterfalls...

You're a step ahead of me, AJ. CSDs still pretty much confound me.

gedlee said:
Below is a fairly typical impulse response - note reflection starting after the initial response... You need to be looking at nice clean impulses like this before you can begin to do anything. You need to be able to find the direct sound and the reflections. If you can't tell where one starts and the other ends then you have too small of a space.

Thanks Earl. I'll take another look at my results over the weekend. I don't remember seeing such a strong first reflection -- I did have a 2-foot deep pile of pillows and such to dampen the floor reflection, at LO's suggestion.

Regards,
John
 
What I would do also, besides playing around with gating, is to see the effects of various windowing schemes. Don't worry about "good" data yet, just get a feel for what spectra look like when you do crazy things like start the FT somewhere after the impulse start or include vs exclude the first refection. If you're creative, you can even put jumps and discontinuities in the impulse response. You just want to get to the point where if you see something funny, you'll know what caused it.

You've got an incredibly valuable resource in the d'A book.
 
The key to taking good measurements in smallish rooms is to set up the system as far from reflections as possible. here is what I do.

First I clear the room, which isn't so tough.

Then I put the speaker on a stand that gets it at about the rooms mid-point. I have found that its just not possible to get rid of a reflection, so I don't bother with damping etc. I put the line from the mic to speaker diagonally across the room. This gives the greatest distance between the source and the first reflection. The floor or ceiling tend to be the limiting factors.

It really is important to get a decent time from first arrival to first reflection as this limits the LF capability. Making the speaker to microphones distance shorter helps this but at the sake of getting too close to the source. I would never get closer than about three feet and prefer more like ten. I use a stand that rotates to get polars.

I go into the nearfield for the low frequency and then mate the two measurements together. This yields a very reliable and useful set of polar responses from which to develop a crossover or test the final design. I can't over emphasize how important it is to do more than just the axial response. No single response could ever tell you how a loudspeakers sounds.
 
nullspace said:

Thanks Earl. I'll take another look at my results over the weekend. I don't remember seeing such a strong first reflection -- I did have a 2-foot deep pile of pillows and such to dampen the floor reflection, at LO's suggestion.

Regards,
John

Thats another reason why I just leave the reflections, it makes them obvious. There is nothing worse than finding that the crummy data you have was because you didn't get rid of all the refections, but didn't realize it. Even a small reflection - barely visible in the impulse response, can have major effects. Thats because the impulse is linear and FR's are log. This is why some people square and take the log of the impulse (Energy-time) because it highlights the reflections and things that make differences in the FR in dB.
 
nullspace said:
Great stuff Earl; thanks very much.

A 'lazy susan' is on my project list. Maybe next weekend...

Regards,
John


Listen - its trivially easy. Just drill a hole in a board and place it onto the top of the stand with a pin in the middle. I use a 1/2 dowel - brass in my case. Now grease the layer between the two boards and mark off the angular increments to a nail put in the side. I have used this "polar test stand" fo several years and it works so well that I don't see any reason to change. Although in Thailand we had a remote actuated one which was fun, but certainly not ncessary to do the job. Hey, its all about getting the job done, not spending money.

My test stand is a few 2X4s and a MDF top - net cost about $10.

And oh yes definately buy the $49 Beringer mic and I also use their preamp. I've had $1000+ mic and preamp and they got exactly the same response. Absolute mic accuracy is not nearly as important as doing good comparative measurements.
 
Another very easy stand for polar response can be made out of 2 melamine rounds, a 24" stacked on an 18". You can bolt them together in the center and the upper slides very easily on the lower. You measure and mark gradations in the melamine and you have a very quick rotating platform.

You can get these melamine rounds (just a particleboard circle covered in melamine) at home depot etc.

You have to be careful that the axis of rotation is on the baffle center, or the baffle will "pull away" from the microphone as you go further off axis, giving exaggerated curves. Some of my earlier curves posted on my site suffer from this.

I agree with the cheap mic. The ECM is fine. For a little more, you can by a calibrated one from KimG (~150) or you can sent KimG your ECM and he'll do a cal curve for ~$40. Not bad to check/fine tune.

Soundcards can be tricky. Most work fine but some have issues. Just check out the SE yahoo board:bigeyes:

I also agree with not using fiberglass, pillows etc. These absorb differently at different frequencies, and make the marker position a bit challenging. More trouble than it's worth.

Now, If you can measure outside, say, on the back of a hillside, well...:D (Still, if you can find a church, a friends warehouse, etc, you can make some very detailed measurements.

One meter measurements are fine for a small 2 way, but not at all acceptable for larger speakers. 2m is often just ok, and 2.5-3 is better. (but hard to do!)

Everyone should get D'A's book. It's easy to read and covers the essentials. Everyone should also play around with known drivers on larger baffles, try different lengths off the ground, include and exclude reflections, look at nearfield and farfield.

I see a lot of cruddy 1m very short windowed measurements posted on the diy boards-it's often hard to know what to make of them...
 
I use melamine for the sliding surfaces too - it won't absorb the grease, and slides easily. But with heavy speakers the grease is almost essential or they don't turn very easy. I don't see the need for them to be round though. Maybe easier to mark the angles I suppose.

The parralax problem of rotation is one good reason for greater distances of measurement as it is minimized by distance. I have not found this to be a huge problem, although it is noticable.

There are so many questions with ground plane that I have never tried it. Many use it though. In theory the plane has to be perfectly reflective and this would almost never be the case. If its not perfectly reflective then its very hard to say what effect it has. Too many questions for my taste.

Pits are common and very useful if you can do this, but digging a pit in my living room was not seen as acceptable.
 
Measurement Signal Types, revisited

The following article was linked in one of the posts of the thread I posted here earlier, but it was a few posts down and some may have missed it. It's a very good article that outlines the strengths and weaknesses of each signal type and measurement system, including MLS, noise, dual-channel FFT, impulse, TDS, swept and stepped sines.

As for ground plane measurements, I find them to be perfectly suited for subwoofer testing using swept sines. You can't get decent bass measurements indoors, because you'll measure the room, not the speaker. Find a large outdoor space free of reflections or noise. No anechoic chamber is as echo free at bass frequencies as a large outdoor space.
 
gedlee said:
The key to taking good measurements in smallish rooms is to set up the system as far from reflections as possible. here is what I do.

First I clear the room, which isn't so tough.

Then I put the speaker on a stand that gets it at about the rooms mid-point. I have found that its just not possible to get rid of a reflection, so I don't bother with damping etc. I put the line from the mic to speaker diagonally across the room. This gives the greatest distance between the source and the first reflection. The floor or ceiling tend to be the limiting factors.

It really is important to get a decent time from first arrival to first reflection as this limits the LF capability. Making the speaker to microphones distance shorter helps this but at the sake of getting too close to the source. I would never get closer than about three feet and prefer more like ten. I use a stand that rotates to get polars.

I go into the nearfield for the low frequency and then mate the two measurements together. This yields a very reliable and useful set of polar responses from which to develop a crossover or test the final design. I can't over emphasize how important it is to do more than just the axial response. No single response could ever tell you how a loudspeakers sounds.


You are not kidding. Even in a warehouse, where I have substantial room I prefer to take measurements outside in the parking lot with the loudspeaker lifted 8' in the air to get the first reflections out of the way.
 
This is a fantastic thread. When I started measuring loudspeakers it helped me not only build better speakers, but taught me how to hear better. Once I could correlate the measurements with what I was hearing, it opened up a new world of appreciation for speakers.

Can't wait to see where this goes, and thanks to everyone for contributing.
 
For off axis measurements I just use a piece of MDF marked with angles as shown. Speaker cannot move or possibly vibrate like it can on a pivot. However, you might get some small reflective artifatcs from the board itself and it can be prone to tipping the speaker.

I am lucky in that I live in a quiet street with a 4 metre (13 foot) balcony. I reckon I can get a 17+ msec gate for a 3 way design pending. I will place the mic on a boom extending out from the balcony - pref. 2+ metres to get point summation of all 3 drivers. For off-axis measurements, the mic will stay put and I will tilt the speaker (placing horizontal) using shims to effect the angles.

David
 

Attachments

  • picture.jpg
    picture.jpg
    45.6 KB · Views: 1,116
nullspace / Patrick,

If you want to design a speaker with a +/- 1dB response through most of the range, I personally believe microphone calibration is essential in one form or another. My ECM8000 begins to rise by 1dB+ from 3KHz up.

You might find this thread useful for how to perform a mic calibration against a reference tweeter using SpeakerWorkshop (or any measurement software that allows a frequency division to give you a 0dB relative offset file to use as a calibration file). In my case due to where I live, i couldn't find a local person who could calibrate against a known source.

http://www.diyaudio.com/forums/showthread.php?s=&threadid=119164&highlight=

Cheers,
David.
 
Patrick Bateman said:
Once I could correlate the measurements with what I was hearing, it opened up a new world of appreciation for speakers.

Until you can do this, you are just shooting in the dark and hoping to hit something.

Dave Bullet said:
If you want to design a speaker with a +/- 1dB response through most of the range, I personally believe microphone calibration is essential in one form or another.

While this might be true in principle, in reality its not that important. Its the relative measurements that matter most. In other words - getting the on axis and off axis responses to hold together (1 dB in this context is impossible) is far more important than 1 dB in absolute response. So yes, 1 dB error is an issue for "absolute" measurements, but it is more than likely irrelavent when designing a loudspeaker to sound good. 1 dB is barely perceptable in absolute terms.

+- 1 dB along a single axis is a pointless criteria if no other axis is considered. And once the other axes are considered, +- 1 dB is way beyond the capability of any loudspeaker that I have ever seen.

One needs to keep the goal in context and make measurements that are meaningful to that goal. Higher accuracy does not yield a better design if the measurements are the wrong ones.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.