Speaker Measurements 101 Page 3

The Acceptable Compromise: Quasi-Anechoic

Pros: Reasonably affordable, fairly accurate
Cons: Complicated, only fairly accurate

As I said before, few speaker companies-and no reviewers or hobbyists-have the resources to buy an anechoic chamber. (Although some manufacturers and audio publications are lucky enough to be able to borrow or rent one on occasion, especially if they're in Canada and in the vicinity of the National Research Council's facilities in Ottawa.) Quasi-anechoic technique seeks to simulate an anechoic chamber by electronically removing reflections from nearby objects. As we'll see, it's not perfect, but it's sufficiently accurate and affordable that it's now the preferred technique for reviewers and small speaker companies.

Here's how it works. Imagine a speaker placed on a stand 2 meters high, and a measurement mic placed on a stand 2 meters high and 1 meter from the front of the speaker. At sea level, it takes sound 2.9 milliseconds to travel from the speaker to the mic. The sound that reflects off the floor-the next closest object, assuming a high ceiling or outdoor setting-takes 12.5 ms to get from the speaker to the mic. So if you were to shut the mic off before that reflected sound hits it, you'd have a nice, clean measurement, right? This is called a gated measurement, and it's the core of the quasi-anechoic method.

The problem is, you have only 9.6 ms between the time the sound from the speaker hits the mic and the time the reflected sound hits the mic. To measure an audio signal, you generally need to capture at least one entire plus/minus cycle of it. And if you have only 9.6 ms seconds to work with, that means the lowest frequency at which you can grab a full cycle before the mic shuts off is 104 Hz.

But it gets worse. To ensure a good measurement, you have to shut off the microphone well before that floor bounce reaches it. So figure we're now up to 120 Hz. But the lowest frequency you can catch before the gate closes is also the resolution of your measurement for the first octave. The resolution doubles with each octave, so you're up to 1/2-octave resolution between 240 and 480 Hz, 1/4-octave resolution between 480 and 960 Hz, etc.

The result is what I call "the squiggles": wavy lines in your frequency response measurements below about 1 kHz that you know from looking at our anechoic measurement on the previoius page shouldn't be there. You can see them in the green trace on the graph at right, in the region below about 500 Hz. You can get rid of these by adjusting the timing of the gate, but you're reducing resolution.

The graph you see at right is what Allan and I measured in my backyard, where I do all of my speaker measurements, using the quasi-anechoic mode of HATS. And by the way, even though HATS produces a measurement all the way down to 20 Hz, the measurement at lower frequencies is useless, as you can see if you compare it with the anechoic measurement on the previous page.

How, then, do you accurately measure the response of a speaker at low frequencies? This part's not so hard. There are two pretty good techniques you can use:

1) Place the mic close to the woofer, which makes the woofer so loud relative to the room acoustics modes that the effects of room acoustics don't appear in the measurement. Repeat this with the other woofers, the port(s), and the passive radiator(s), then sum those measurements together to get the total bass response.

2) Run a ground plane measurement by placing the speaker on the ground and the microphone at a distance of 1 or 2 meters (depending on the size of the speaker), then using an EQ correction to remove the effects of the room acoustics (although if you do this in a large open space you don't have to worry about the EQ correction). You can also apply smoothing to the ground plane measurement (I use 1/3-octave or 1/6-octave, depending) to minimize the effects of room acoustics.

I use both of these techniques, the choice depending on the size and configuration of the speaker. Close-miking produces a smoother-looking response curve, but it also tends to produce a bump in the bass response, typically around 100 Hz, that's not actually there.

In both cases, the bass response is then spliced to the quasi-anechoic response at a frequency between about 150 and 300 Hz. There's still one problem, though: The level of the bass response curve will be higher because of the closer mic position, or because the ground-plane technique boosts output by +6 dB. There are formulas you can apply to get the "proper" scaling, but I and most of the technicians I know just eyeball it, lowering the bass curve until a big part of it overlaps with the quasi-anechoic curve, then making the splice.

All of the quasi-anechoic measurement packages I've used (LinearX LMS, Audiomatica Clio 10 FW, PureBits Sample Champion, and Listen Inc. SoundCheck) employ some combination of averaging and filtering (i.e., filtering out frequencies other than the one being tested at that moment) to reduce or eliminate the effects of environmental noise. This usually works very well. Once my neighbor's gardener started up a leaf blower while I was running sweeps with LMS in my backyard. To see how well LMS could filter out the noise, I ran a speaker measurement while he was working and another after he stopped. There was no significant difference between the two curves.

When you're using quasi-anechoic technique, it's easy to put the speaker on a turntable, just as they do in anechoic chambers, to measure the response at various off-axis angles. These can then be averaged and compared with on-axis response, as I've done in many of my measurements for S&V.