Harbeth M40.1 loudspeaker Manufacturer's Response

Manufacturer's Comment

Editor: Many thanks to Art Dudley for an excellent review of the Harbeth M40.1. Perhaps I could comment on the differences between the frequency-response measurements we made days ago at the BBC's anechoic chamber in London, the in-room listener-at-home curves, and those made by Stereophile.

Making high-quality loudspeaker measurements that can be confidently repeated by other researchers is not easy, and is far more difficult now that big, expensive anechoic chambers with controlled, reflection-free environments are rare—and expensive and inconvenient to hire. The advent of low-cost, computerized, audio measuring systems in the 1990s (to replace paper charts) led to the idea that one could synthesize a quasi-anechoic response on the kitchen table by placing the microphone near to the low/mid-frequency drivers, capturing their nearfield frequency-response measurements and then, at one's leisure, manipulating them mathematically to mimic a true (farfield) anechoic response. A brilliant cost-saving idea! When a microphone is placed very close to a drive-unit—in its nearfield—the sound received by the microphone is very loud indeed, and although there are still strong reflections from the local nonanechoic test environment, the concept is that the direct sound will be sufficiently loud to swamp these unwanted echoes. This is only partially true, and one of numerous assumptions that must be made when measuring nearfield without a real chamber.

Another variable at the heart of the maths is, "What is the radiating area of each driver or vent?" A ruler held to a drive-unit such as the M40.1's 8" midrange unit does not tell you with accuracy its effective area. All these horrible little variables and fiddle factors have to be considered—and be consistent and correct—when using a maths model, and I take my hat off to anyone who has the confidence to totally trust those results. Personally, when I really need to know what's going on, I cheerfully reach for my checkbook and pay the $2000/day to hire a conventional anechoic chamber. The bigger the speaker, the more drive-units, the less reliable the model, and the more confidence one has in the real anechoic measurement—or at least one made out in the open on a warm, windless day with the speaker on a tall pole many meters above the ground: a very daunting health and safety prospect with a big, heavy speaker.

The M40.1 is a three-way, twin-vented loudspeaker whose single pair of rear-panel terminals means it is not possible to externally isolate the drive to the bass, mid, or tweeter to permit measurement of their solo contributions, a precursor to accuracy with the maths-based approach. To properly isolate the M40.1's drivers would necessitate opening the speaker, soldering or unsoldering crossover components to disconnect the electrical feed to all but the driver being measured, and, ideally, replacing the undriven drivers with dummy electrical loads so that the crossover "sees" the same termination and behaves as designed. It is not stated if this is how the woofer and midrange responses were recorded for Stereophile. [It is not, with speakers like the M40.1 that do not have bi- or tri-wirable terminals.—JA.]

Then there is the tricky problem of how to, without a chamber, reliably measure in acoustic isolation the port(s) alone, as they are merely passive tubes, and obviously operate only if the woofer is being driven. Big problem!

Then there is the math itself: There is more than one paper, and in the case of the M40.1, different methods give a 6–10dB difference in the bass-level prediction! That's more than double the amount of bass output predicted, due to just one simple equation! Which is correct? What is the real frequency response? Easy answer: Trust the anechoic measurements above all others!

I've looked at this issue from the other angle in user forums and private correspondence with speaker-simulator designers. My question was, "Why is your low-frequency modeled response discrepant with my measured response? Surely, your simulator is in error." In fact, the issue transpires as a problem that in the real world there is significant bleed-through of sound from the nearfield vent(s) measurements into the nearfield measurement of the woofer, and vice versa, exacerbated by the omnidirectional microphone that is standard for loudspeaker measurement, the woofer and vents being so close together. With hindsight, it should have been no surprise to me—but it was—that cross-pollution of the measurements at low/middle frequencies would corrupt the maths-based quasi-anechoic method, which then usually overestimates the bass/mid level and/or corrupts the bass/midrange response shape. So in reality, at, say, 70Hz, there is some output from port 1, some from port 2, and some from the woofer, but these cannot be truly independently measured by placing a microphone even close to them. The result is that, when added together in the maths, there is double, triple, or quadruple counting of source contributions; ie, the bass shows a tip-up.

In contrast, the "traditional" holistic results from the (BBC) anechoic chamber, with the microphone set up 1m away, permits the sonic contributions of the vents and woofer to mix freely in the air on their way to the microphone. No pretense that they can be independently measured. There is no computer maths here, no fiddle factors, no rules of thumb. The only issues you have to think about there are how and where to clamp the microphone—and what to do if you become locked in. What the chamber mike records is just what you would hear standing there, or in your listening room at home.—Alan Shaw, Harbeth Audio, Ltd.

Harbeth Audio Ltd.
US distributor: Fidelis A/V
14 East Broadway
Derry, NH 03038
(603) 437-4769