What Makes a Speaker Sound Good?

Bill Waslo, Liberty Instruments, Inc. April 2005

OK, no, I'm not really going to be able to resolve that question here, once and for all,  in a web article.  After all, the premise, that a given speaker driver "sounds good",  is a subjective description.  I can't prove it, and I can't say for certain what it is about it that makes me like the sound of a driver.  Others (such as Floyd Toole, Earl Geddes, and others) have done scientific studies or proposed theories on that subject, applicable to more than one listener.  What I am doing here, instead, is much less general, merely reporting some results of a brief personal study to try to find out what technical characteristics seem related to what pleases my ears.

I have been impressed with the midrange and highs sound of the larger panel Monsoon (MM1000 and MM700) multimedia planar satellite drivers over the years.  The bass sound was ok, but not exceptional, -- but what really grabbed me was the feeling of  being "transported" when I'd heard decent program material through the satellite panels.  I had never gotten around to buying a pair of these when they were still available but I recently, on a whim and with luck, obtained a pair of used MM1000 panels at a good price via Ebay (without the woofer/amplifier module, unfortunately).  The larger Monsoons seem to have somewhat of a cult following, now, as the used (and sometimes even non-working!) final selling prices on Ebay for these units rival and sometimes exceed the prices they carried when they were new.  The situation reminds me of that of the Quad ESL63 hi-fi speakers: also no longer made, but to the ears of many, still in a class by themselves.

I made a quick series of measurements of a MM1000 driver, guided of course by what I suspected might be the source of the subjective sound characteristic.  Since the Monsoon drivers are no longer made (and since the company that made them is apparently no longer in existence*), it seemed that I was in no danger of offending anyone or of appearing to plug a commercial product, and could safely publish the resulting graphs here on the Liberty Instruments' web page.  But still, let me add this disclaimer:  there are other good multimedia speakers out there, some better in some ways than the Monsoons; and there is no guarantee, anyway, that you (or most listeners in general) would agree with my assessment of the sound of any loudspeaker.  But for me, none of the currently available desktop speakers (notably including those simulating "flat panel" appearance) seem to have the sound of the old Monsoons.

Frequency Response

Frequency Response is the loudspeaker measurement that first comes to mind.  So, how does the Monsoon panel on-axis frequency response shape look?

...not really very impressive, it seems.  It stays within about a 6dB spread from 1kHz to 12kHz, but it is far from being the smoothest looking curve I've ever seen. The phase response is relatively well behaved, but that would be expected from a single driver and no crossover being used.  And the response drops like a rock, really, above 12kHz.  But I think that anyone who has played with a lowpass filter on an audio system would probably agree that the ubiquitous desire for  "20Hz to 20kHz" response (a hangup that the industry seems to have) is mostly based on marketing fiction.  I didn't really expect that the highest (supposedly) audible octave plays much of a part in my attraction to the sound of  this driver.  Response smoothness might be expected to play more of a part, but the measurement suggests otherwise.  (Or maybe I subjectively LIKE the ripples?).

The system spec for the MM1000 (with powered amps and bass module) states that these panels are used crossed-over third order at 200Hz.  I'm guessing that the high frequency amplifiers housed in the bass module must also apply some sort of shelf EQ below 1kHz to bring up the lower frequencies.  There's no way to check that, for the moment, though.

Impulse Response and Waterfall

The impulse response of this driver is shown next.  The mic is about 18 inches, on axis perpendicular to the face of the panel, with the rear of the mic and panel are clear for about 3 feet or more. Nothing exceptional to see here (again, the single dominant peak is what you'd expect from a single-driver measurement).  Seems reasonably well damped, though.

A CSD waterfall plot made from the IR shows that some resonances are detectable (at 2.5kHz, 8kHz, 9kHz, right where the response ripples peak... no surprise).  Although many speaker engineers do, I've never been able to make a better/worse comparative assessment with any confidence whatsoever, of a loudspeaker based on a waterfall plot.  I can make a waterfall plot look anywhere from great to terrible by just mildly tweaking various display parameters.  I usually just tweak them to best show the frequency placement of the resonance ridges (which usually makes the graphs tend toward the terrible-looking).  The ridges do show where damping might help (if there should be a way to add it).  In general, though, I can't say this waterfall really suggests much about why I might like this driver.

Distortion

I didn't bother to measure harmonic distortion of the panel.  First, it is difficult to do effectively  in my very small office without reflection problems.  Second, I have little if any confidence that harmonic distortion at some arbitrary continuous tone level has much to do with the sound characteristic of a normally functioning loudspeaker operating in its intended range.  And what's the typical tone level of music or most program material, anyway? -- it is all over the map, a huge variation in ranges seldom even vaguely resembling a continuous tone, so while perhaps a 3D plot of harmonic distortion vs. level vs. frequency might relate to sound quality, that was more than I was up for measuring.  My subjective assessment of this panel seemed to apply about the same at low or moderately high level (the panels don't handle very high levels well), so simple distortion wasn't expected to be an explanation for the sound quality.

I did make a quick and simple "spectral contamination" type measurement, below, at what seemed a good moderate (subjective to my ears) drive level.  For comparison, I made an equivalent plot of a homebrew 2-way cone-based speaker system (that I also like the sound of, though for a different characteristic than for the Monsoon panel).

One obvious difference is that the 2-way cone speaker, having a woofer,  has a lot more output at 100Hz than does the monsoon panel.  It also looks smoother through the midrange. But there is a strange shelf of hashy/noisy output  (40dB down or so) in the cone speaker, that is not seen in the Monsoon panel (in which the noise floor falls in a continuous way suggestive of room noise).  The upper end of the hash shelf is, perhaps coincidentally, near the crossover point of the 2-way system.  The Monsoon panel, though, does show  higher close-in IM products around the higher tones, perhaps from being driven below its intended range (no crossover is used, remember). 

The hash/noise floor may be related to the sound of the speaker.  Todd Beachamps has said that the Monsoon drivers seem to do better on these kinds of spectral contamination tests than many other multimedia systems he has tested. 

A further test that is warranted (when conditions around here are more quiet) would be a "distortion isolation" test.  This kind of test can separate distortion and noise from "linear" output using actual program material, but does require reasonably quiet conditions.  

(addendum, April 24, modified April 25): Distortion Isolation 
 I had a quiet opportunity this morning to make some time domain distortion isolation runs on the Monsoon and on the 2-way panel.  The test was done with the first 68 seconds of the song "Jacksons, Monk and Rowe" by Elvis Costello and the Brodsky Quartet.  The WAV file was first converted to mono and highpass filtered (3rd order, Butterworth at 200Hz) to avoid driving the Monsoon below its range.  In hindsight, it would have been better to put an additional 1kHz highpass on the cone 2-way to better match the response shapes, but the speaker was operated with its normal range.

Distortion isolation is a very touchy process.  I disabled as many noise sources (HVAC, etc) as possible.  As I had to remain inside the room during the test (I have no isolated studio room available), I tried to keep very still during each entire process to keep room reflections constant.  Also, as distortion isolation is very sensitive to even slight crosstalk effects, I made an adaptor to route the "tip" signal of the soundcard output (the stimulus signal) around the AudPod.  The AudPod uses analog switches which can allow trace amounts of high frequency signal to bleed into the microphone input.  The "ring" signal from the soundcard output, however, MUST be routed to the AudPod, as that (different) timing signal is required to maintain precise time matching for the synchronous stimuli used in distortion isolation processes.  Also, there should be no probes connected to the AudPod during these tests, as that is another possible source of crosstalk.

I converted the resulting files to WAV files, which can be downloaded below.  In each, the Left Channel is the "linear part", as predicted from the impulse response and the program wave file.  The Right Channel is the "distortion and noise" part, that is, the difference between the actual signal as played through the speaker and the predicted "linear part". 

Cone DI of JMR.WAV   :the distortion isolation file made with the cone 2-way loudspeaker
Monsoon DI of JMR.WAV : the distortion isolation file made with the Monsoon MM1000

The best way to listen to these is to open them into a PRAXIS plot, and use the "File, Listen" menu to play them back using headphones.  You can choose the (R) channel to hear the isolated distortion part in both ears, the (L) channel to hear just the linear part in both ears, or (s)tereo to hear the linear part in the left ear and the distortion part in the right ear.  The distortion of the cone speaker has a distinctly garbled, crunchy kind of effect, while that of the Monsoon sounds more like the original program source.  But the comparison isn't completely fair, as the Monsoon speaker is not playing any lows at all, so I'm hesitant to make too much of this result.  In both files, the sound of the computer fan and some outside traffic (a highway is about a mile away) can be heard in the isolated (right) channel.

An objective evaluation of the two distortion isolation files was less definitive than I hd hoped.  I made a 2 mega-sample FFT of both results (shown below, smoothed), and, using a custom Script, calculated the integrated signal levels ratios of the "linear" to the distortion/noise parts in the frequency range 200Hz to 10kHz.  The results for the two speakers, done this way, were within about 2 dB.  This may mean only that the result is somewhat dominated by noise in both cases, which I suspect to be true based on the "isolated" signal result being a relatively smooth curve from low frequencies to higher frequencies.  If the lower integration limit is increased in frequency, the Monsoon result begins to look better than the cone speaker (by about 5dB if 500Hz is used), implying that the weak low frequency output of the Monsoon is affecting the calculated result.  This could perhaps be significant to the sound of the speaker.

Interestingly, the "linear" spectrum shown for the Monsoon was smoother than for the cone speaker. One possible explanation for that is the different directional pattern of the monsoon driver.  These spectra include all room effects, which would be excited differently via different radiation patterns.

Green curves are the spectra of the "linear" parts, yellow are the spectra of the "distortion and noise" parts.

Polar Pattern

The radiation pattern is what I suspected would be the main player in the Monsoon's sound. 

The Monsoon satellite is a dipole driver, which should have a figure-of-eight polar pattern, and better direct the sound to my ears and away from the surfaces and objects to the sides..  "Should have" --  but the panel sits slanted back at an angle so the measured effect (when the mic is placed at listening level,  above the panel but on-axis relative to its plane) doesn't  show a figure-of-eight shape at higher frequencies.  That is because when the inclined driver is rotated 180 degrees, the back of the dipole then points down below the mic. 

Similar measurements were made of both the MM1000 panel and the two-way speaker, for comparison.  It would take a large number of pictures to show the effects at all frequencies, but to summarize the general results:

• The front lobe width is generally more narrow for the cone speaker at midrange and high frequencies than for the dipole MM1000, getting wider than the dipole speaker near and below 1kHz.

• The relative width and shape of the front lobe varies much more with frequency with the cone system than with the Monsoon panel.

All graphs:  5dB per division. 0 degrees is on-axis relative to the front surface plane of the loudspeaker

I suspect that the perceived sound quality of this speaker may be more related to the consistency (over frequency) of its radiation pattern, than to just its basic directivity (which seems to show a wider forward pattern than does the cone system at higher frequencies).  This would be in line with some of the results published (by Toole, I believe). 

The response off-axis is more consistent than for the cone speaker, and this should result in the inevitable room reflections having a similar character to the directly radiated sound.  Maybe similar sounding reflections are better masked than are reflections which differ significantly (?) -- such as might occur from a very much varied off-axis response as in by the cone 2-way speaker). 

The polar response may help explain why all the non-dipole "flat panel" speakers on the market seem to me to lack the engaging effect of the Monsoon driver. 

Conclusions (or maybe "suggested explanations")

The most significant characteristic of the Monsoon driver identified in these tests is its relatively consistent frontside radiation pattern, due to the dipole design.  There is also possibly an improvement over some cone type drivers in low-level nonlinear intermodulation hash produced by the driver.  The sound of the distortion produced by the cone speaker definitely sounds (to me) more objectionable than does the Monsoon, but the effect seems to be more in the lower mid-frequencies (which are mostly lacking from the Monsoon).

Of course, another possibility is that maybe there are non-sound-related issues affecting the perceived quality of the speaker.  It is not impossible that the dipole shape and appearance of the Monsoon panel could affect my (and others') impression of the sound, a matter perhaps better approached through double-blind listening tests.

The www.monsoon.com website has now only a message "This domain is not for sale.  Do not call, do not write.."!!