Audio's Crescendo Test (Audio magazine, may 1976)

by Richard C. Heyser

Sound is dynamic. And sound is what speaker reproduction is all about. One of the challenges which dynamic sounds place on a speaker is the presence of brief peak-energy components considerably above the average level. These bursts of sound can place enormous demands on any speaker. With the advent of super power amplifiers, it is not unusual to find cases where a listener can adjust the level high enough that several hundred watts of power are used on peaks. Yet the speaker doesn't go up in smoke because the average power may be low enough to prevent damage.

This condition is not solely the province of the sound freak or the person wishing to break a lease. Natural sound has very high peak energy, and as recording and reproducing methods get better the peaks begin to inevitably rise. A natural sound you may not consider loud, such as a typewriter or piano chord, may in fact have a very high peak sound pressure level.

We would like to test a speaker for its behavior under such dynamic conditions. Audio can, and does, observe the acoustic waveform for high power tone bursts. However, this doesn't get to the heart of at least one aspect of the subjective effect of peaks. The question we pose is, given a normal reproduced sound at moderate level, what is the affect on that sound of a suddenly applied signal much greater in level and incoherent with that sound? If you could imagine concentrating on an oboe in an orchestra, this is like asking whether the sound of the oboe is changed when the horns let go triple forte. If the oboe is modulated by these peaks, we know the sound will be unnatural. The test we have developed for probing this condition we call the crescendo test.

Sound Measurement

We use a musically based sine wave, such as 440 Hz, as our "inner voice" and use a very narrow bandwidth filter to measure the sound energy of that musical note. A wide band incoherent noise source is then added at an average electrical level 20 dB higher than that of the tone, and the level of the tone, with this added crescendo, is compared to its prior value.

We say the speaker has reached a subjective crescendo limit when the inner voice is reduced 1 dB by the added crescendo. The instantaneous peaks of the complete signal are measured and used as the peak crescendo handling capability of the speaker.

Now, let's go back and take a look at why and how. If an incoherent signal cross-modulates a coherent signal, sidebands will be produced in a frequency about that coherent signal.

The energy in these sidebands will itself be incoherent whether the process was pure amplitude modulation, angle modulation, or some combination thereof. By using a very narrow filter with a total noise bandwidth of 1 Hz centered on the frequency of the tone, we are measuring the energy of the "carrier" which is being noise modulated. When the carrier drops 1 dB, we know that 21 percent of the carrier power has gone out of band due to modulation. We don't know where it's gone but we do know it shouldn't have left.

Our oboe is now spread out thin in the frequency spectrum. If you listen monaurally, this is similar to what is known in broadcast parlance as "ducking" and it may or may not be unpleasant. If you listen in stereo, however, it can be quite a different effect. A phantom center-image oboe will have its energy shared in both channels. If the horns let go only in the left channel, the oboe component of that channel will be modified in a manner different than the component of the right channel. The result can be anything from the oboe sliding to stage right, to a ten-foot-wide oboe, to a "song and dance" oboe which glides around the stage keeping time with anything loud enough to shift it.

Of course, all the other instruments of the orchestra are playing a similar game of musical chairs.

Image Distortion

This seldom considered aspect of distortion in reproduction can lead to a blurring of the stereo or quadraphonic image. You're not quite sure what is wrong but the reproduction seems to lack sparkle as you raise the level. Actually, polar pattern characteristics, transient response, and non minimum phase-response properties interact with this crescendo handling to provide a very complex behavior.

There is not enough space in this brief correspondence, or 10 such articles, to go into those interactions which have been identified in speaker reproduction.

The tweeter and woofer may differ in crescendo handling. We therefore use tones which separately test different drivers. If a tweeter gives up before a woofer, this can lead to differential spreading of fundamentals and overtone structure in music.

Back to the "how." A Gaussian white noise source is used as the incoherent signal and its average level for a 20-kHz band is made 20 dB higher than the average test tone power. This means that peaks rise to 30 dB above the average tone level. A white noise signal is used because it has constant spectral energy density and the Gaussian distribution of amplitude is a good all-around model of many sound processes.

A Wiener matched filter is used since it will optimize the mean square difference between the coherent tone and incoherent noise. One second "chunks" of time are processed and average acoustic power divided by average electrical drive power is plotted as a function of average electrical drive power. Three superimposed plots are made. The tone without noise is run at reference level and a second run made at 1 dB below reference level. Then the tone at reference level is plotted when noise is added. This tells us three things. First, if the acoustic transfer gain is constant (and quite often it is not). Second, if adding noise at any level causes cross-modulation, and third, how the tone changes with increase in total level.

Acoustic Patterns

If the acoustic gain decreases with drive level whether noise is present or not, the speaker will lack punch on the crescendos themselves. The acoustic gain is similarly monitored during the harmonic and inter modulation test, so there is not much chance of a speaker with this property getting past measurement without detection. If the gain increases with level, this gives, not surprisingly, a punchy sound, particularly if it occurs in a midrange and not in the woofer, since a vocal or musical crescendo will "bark" right out and the sound will tend to go bright. Some people like this sound, others do not.

If the crescendo test shows a drop of signal when noise is added regardless of level, it generally signifies a curvature of transfer characteristic at low level. This may give a "class B amplifier" type of sound at low levels and usually correlates with a low level harmonic or intermodulation distortion that doesn't go to zero with decreasing level.

After all of this analysis, the result may be the laconic phrase, "speaker X passed the crescendo test with flying colors." The one-liners we hang on this test seldom does complete justice to the amount of work expended to see how well the speaker performs in reproducing dynamic sound or to the fact that the system under test is behaving properly. We think it's worth it. After all, accurate sound is what this is all about.

--------------

(Audio magazine, 1976)

Also see:

Build A Mike For Binaural Listening/Gene A. Nelson

Record Cleaners Revisited/B.V. Pisha

= = = =

Top of Page    Home