18.2.3 Interpreting the Data: Exercises in Detection
Having single-axis data along with data representing progressive increases in spatial averaging is highly useful. For example, Figure 18.7 illustrates the identification of three distinctive forms of misbehavior in an early prototype of the example loudspeaker shown in Figure 18.6.
• Off-axis misbehavior leading to a change in directivity. This is revealed in the fact that around 2 kHz the on-axis curve is quite flat, but as the data embraces more off-axis measurements into the spatial average, a shallow dip develops. In Figure 18.6, it can be seen that the DI shows a small bump at 2 kHz.
• On-axis misbehavior: ripples in the frequency response caused by enclosure diffraction. Proof that they are not resonances is that they are much attenuated by the moderate spatial averaging incorporated in the listening window curve, and have all but vanished in the increasing spatial averaging of the lower curves. It is highly improbable that this would be audible in a room, but it has a threatening appearance in the on-axis curve. Figure 18.6 shows that the problem was eliminated in the final product. This is the type of circumstance that led to a redefinition of DI, as just discussed.
• A low-Q (i.e., well-damped) resonance at the upper limit of the tweeter. It is visible in the top three curves, but there is little evidence of it in the sound power, meaning that it has a forward directional bias. Figure 18.6 shows that for the final product, the tweeter was improved, extending the frequency range and, in the process, eliminating even this innocuous resonance.
For perspective, this now-discontinued loudspeaker was included in numerous double-blind listening tests over several years. It always was a front-runner, either winning or being in a statistical tie with the best competitors. So although there are imperfections, one may conclude that this family of curves describes a highly commendable standard of performance.
Figure 18.8a shows an active/equalized loudspeaker at an early stage in its development. The well-mannered directivity seen here is characteristic of good constant-directivity horns. If directional control is required, as in an acoustically "live" room that cannot be altered, horns are an excellent solution, especially since horizontal and vertical directivity can be independently manipulated. Belief that horns and waveguides are inherently colored is an idea that good modern designs have put to rest.
Cone/dome systems are best suited to wide-dispersion applications, and there the challenge is to maintain a relatively constant, or a least smoothly changing directivity, as a function of frequency. In achieving that objective, it is increasingly common to add shallow horns, often called waveguides to mid-and high-frequency cone/dome drivers to subtly manipulate directivity so that they better integrate into the entire system.
This design is well optimized for small rooms in that above the transition region - 200–300 Hz - the directivity is quite constant. The small undulations in the curves are very consistent from the on-axis curve, down to the sound power curve. Normally, bumps that are seen in several spatially averaged curves are evidence of resonances. Here there are a lot, more than might have been expected from the transducers and enclosure. The question is: where did they come from?
Looking closely, it is seen that the bump around 700 Hz gets larger as spatial averaging increases and the DI drops, indicating that most of the energy is radiated off axis. This behaves like radiation from an enclosure resonance, possibly the large rear panel. But the other frequency response peaks and dips look much the same in all of the curves, and they do not show up in the DI curve. This is unusual.
FIGURE 18.8 (a) A large professional audio monitor loudspeaker: two 15-in. (380 mm) woofers vertically flanking a constant-directivity horn. It was active, with a dedicated equalizer. These data were taken at an early stage in the development and illustrate the challenges in separating the factors responsible for visible features in the family of curves. This is discussed in the text. The production version was significantly smoother than this representation (it is now discontinued). Note the high sensitivity of this and the following loudspeaker, about 95 dB at 1 m for 2.83 v input. (b) A large high-end loudspeaker with a 15-in. (380 mm) woofer, a midhigh horn, and a high-frequency horn. The benefits of a dedicated high-frequency horn are apparent compared to (a). This totally passive design is admirably free of resonant colorations, with superbly controlled directivity.
These directionally independent bumps can be fixed by equalization. But it turned out that some of the undulations were in fact caused by a preliminary casual equalization that had been done earlier, outside the anechoic chamber. After EQ adjustments, the loudspeaker sounded as it (finally) looked: very good. If there was a problem, it was a tendency to play it very much louder than is commonplace with consumer loudspeakers. That is one of the seductive characteristics of loudspeakers that do not power compress or distort at high sound levels; they don't sound loud until they are dangerously loud.
With two large woofers separated by a horn, the measurement distance of 2 m is within the near field, and there is evidence of directivity at bass frequencies. This is yet another advantage of spatially averaged measurements: they can provide meaningful data in the near field.
Horn-loaded loudspeakers are very well suited to large home theater installations; they deliver high-level crescendos effortlessly, and their directional control minimizes the amount of sound converted into heat in absorbers - which translates into significantly higher overall efficiency. Figure 18.8b is an example of a loudspeaker that is admired equally by audiophile stereo traditionalists and home theater enthusiasts (with deep pockets!).
There is a lesson to carry away from the example in (a). Equalization changes the intrinsic performance of a loudspeaker, and this can be good (if it is needed to repair the frequency response) or bad (if it was not needed).
Transducers, within their normal operating ranges, behave as minimum-phase devices (the misbehavior of the large horn above 10 kHz is evidence that it is outside of the predictable operating range). Parametric equalization of resonances in transducers is an effective solution within this frequency range. But the measurements must be made without reflections. Measurements done in a reflective space are nonminimum-phase, and they cannot be trusted to reveal accurate evidence of resonances in transducers and therefore of the corrective equalization appropriate to remedy them. Therefore, notions that "room equalization" can address the problems of inferior loudspeakers are optimistic.
Equalization cannot alter directivity, and steady-state measurements in rooms cannot reliably identify resonances. Such equalization is most useful for subtle adjustments to already well-behaved, loudspeakers, and then most likely at the lower frequencies.
All of this assumes that the measured data have sufficient resolution to reveal what can be heard. The next example begins with a prototype of an inexpensive loudspeaker that exhibited an easily audible problem, but there was no evidence in the measurements to explain it. At this time, it had been common for engineers to do time-windowed FFT measurements in their listening rooms. The dotted curve in Figure 18.9 shows what these people were looking at.
FIGURE 18.9 Measurements of a loudspeaker with an audible problem at 280 Hz. One measurement used a 10 ms time-windowed FFT, as done in a normal room (dotted), and the other measurement was performed in an anechoic chamber using 2 Hz resolution.
The problem was revealed when a female vocalist sustained a certain note. The loudspeaker "howled." From the listening experience one could deduce that a very frequency-specific, high-Q resonance was the problem.
A quick pitch match using an oscillator revealed that the problem was around 280 Hz. The time gating used for the measurement was 10 ms, chosen to eliminate room reflections from the data. This window yields data with 100 Hz frequency resolution (resolution = 1/window duration), so it was clear that any high-Q events around 280 Hz would simply not be visible.
The loudspeaker was then measured in an anechoic chamber where high resolution at low frequencies is possible; the solid curve in Figure 18.9 shows clearly that there was indeed a resonance. Gated measurements are very useful, but their limitations need to be kept firmly in mind. Howard (2005) discusses some interesting measurement options for those without access to anechoic chambers.