Active Speaker Terminology. (Part 5)
Active crossover technology involves lots of new and exotic terms, so to complete our brief look at active speaker systems (HSR Dec/Jan), here's a succinct clarification of some of the more common terminology.
All crossover networks are built from filters. In a low pass filter, signals above a certain frequency (say 300Hz in a bass crossover) are attenuated. Conversely, high pass filters attenuate signals below a given frequency, say 4kHz for a tweeter. If three or more drivers are used, then the midrange driver(s) will be fed via a bandpass filter, which can be thought of as cascaded low and high pass filters, working at different frequencies of course.
The bass and treble sections of the crossover may also be configured as bandpass filters. For instance, a high pass filter is often incorporated in the bass to remove potentially cone-ripping signals below 30Hz, say, which can arise when a microphone is dropped. Similarly, at the top end, radio frequencies (RF) and ultrasonic garbage is often tempered with a low pass filter, set to attenuate (or 'roll off') signals above 20kHz.
Because the effect of the roll-off at the two frequency extremes is potentially audible (unlike the complementary roll-offs at the crossover points, which should come together to create a flat frequency response), the 'rate of attenuation' or filter slope will often only be -6dB or -12dB/octave. A more rapid roll-off can have untoward effects on the quality of the sound, especially at the top end.
Top end roll-off is, incidentally, invaluable for protecting tweeters, particularly when tape machines are accidentally rewound or fast forwarded with the monitoring system connected. At the same time, the crossover may add to the inherent roll-off above 20kHz provided in the power amp and/or outboard equipment, resulting in a significant drop within the audible range; say -2dB at 17kHz, so beware.
The ultimate rate of attenuation of any filter or crossover is given by its slope, which is also known as its order. The rate of roll-off is normally constant with frequency and generally comes, oddly enough, in multiples of 6dB/octave. This becomes more meaningful when we notice that -6dB represents a two-fold change in signal voltage level (a halving), whilst an octave represents a two-fold change in frequency (either doubling or halving). A 6dB/octave slope is thus shorthand for saying 'the relationship between signal level and frequency is inversely proportional', whilst the higher order slopes (-24dB/octave etc.) involve square or cubic relationships. Table 1 helps clarify this, and cast light on the developing rate at which the steeper slopes remove unwanted, out-of-band energy.
Although the ultimate slope (ie. after the first octave) is regular, filters differ in their behaviour in the region where they begin to take effect. The behaviour here is determined by damping, the amount of positive feedback applied and the 'Q' of the filter networks.
What's important though is that exotic names are given to the various response curves around the roll-off point in order of descending damping factor eg. Bessel, Butterworth, 3dB Chebyschev. These 'initial curves', named after the passive filter originators, are known as filter functions and can apply to any order of filter. For example, 4th order Chebyshev, represented as C4, equals -24dB/octave with an underdamped roll-off.
Broadly speaking, for crossovers and most things audio, the Butterworth slope, with just the right amount of damping, is optimum: other filter functions aren't capable of a flat amplitude response when summed.
Filters also have interesting names, such as 'state variable', 'equal-component Sallen and Key' and 'elliptical', to name but a few, but they're really only of interest to designers and are best ignored when discussed seductively in manufacturer's brochures. The previously discussed parameters are far more crucial to the way a crossover performs.
On terminology again, remember that turnover frequency and roll-off point (or frequency) have similar meanings, and are often quoted at -1, -3 or -10dB points. However, they are not necessarily the same as the crossover frequency (or point) which may occur at -3dB, -4dB, or -4.82dB or other seemingly arbitrary levels, depending on a host of factors. In fact, the designer can choose the attenuation in decibels (dB) which defines the crossover point, within broad limits. So, essentially, you can't accurately discover the crossover frequency setting of a unit by simply looking for a -3dB (or whatever) roll-off. It must also be noted that the theories discussed apply equally to both passive filters and crossovers.
A perfect loudspeaker should have a perfectly flat amplitude response ie. the acoustic output should be the same at any audible frequency, although a slight dip of 1dB in the 2-4kHz region is subjectively preferable to many people. Real speakers, especially high-power, high-efficiency types, don't as a rule achieve this requirement, although a response fluctuating by 2dB between the extremes of the frequency range is typical of the best speakers. Whilst still imperfect, this sort of response is acceptable, as long as changes in response remain gentle, occuring gradually over half an octave or more: remember that the ear tolerates changes of less than -6dB/octave with little trouble.
In less expensive speakers, response curves deviate to a greater extent - Figure 2 gives examples. Speaker 1 has a slightly dull, bass-heavy response whilst speaker 2 has a more exaggerated bass lift giving an initially impressive, hard bass 'thump'. However, below 100Hz the bass falls off rapidly and sooner or later, you'll notice that something is missing; hence the phrase 'one note bass'. This curve is typical of micro-speakers and bookshelf models where bass lift is used to give the impression of an extended bottom end.
The third speaker in Figure 2 will have a brilliant and slightly 'thin' sound, but this larger-than-life response tends to become fatiguing after a while. Speaker 4 has a presence peak, like some cardioid microphones, and as usual its effect is dependent upon the sharpness of that peak. A 2dB peak in this region, for example, will create forward imaging and make details in the sound acceptably more prominent. Too much of a peak (more than 3dB) is bad news, as the sound will take on a shrieky quality that's very uncomfortable to listen to for prolonged periods, especially at high volume.
Lastly, speaker 5 features a complementary curve - a recessed midrange. Again, a small dip over a broad region is fine; your ears will soon adjust to it, but too large a dip sucks out the life from vocals, making the sound dull and the stereo imaging uneven.
The inherent frequency response of a speaker is normally very different to the measured response in the room in which it's used. The perceived response, the sum of reverberant and direct sound that you hear, is also different, though related to both response curves. Our hearing mechanism makes some very astute computations and comparisons between the direct and reverberant sounds; so the inherent sound of a speaker is relevant, even though measuring instruments in the room may suggest otherwise.
With room-induced anomolies in mind, the intrinsic response of a speaker is best determined in a place where only the direct sound from the drive units is measured. This is done either outdoors with the speaker and measuring microphone mounted at the tops of poles (typically 10 feet from ground level) or in a heavily damped room with virtually no reverberation - known as the anechoic chamber.
To maintain these conditions for low frequencies, a very large (and thus expensive) chamber is called for, so the less elaborate outdoor method can offer more accurate free-field measurements below 100Hz. The frequency response curves generally published by manufacturers are made in anechoic chambers of a finite size (it's no fun measuring speakers outdoors in January!), so you should take into account the inevitable inaccuracies below 100Hz.
A rough response measurement can be achieved by applying a pink noise signal to the speaker and the results over several minutes are averaged and displayed on a spectrum analyser. More accurate measurements are made by sweeping a very pure sine-wave from 20Hz up to 20kHz. Simultaneously, the response is plotted on a chart recorder and you end up with the familiar piece of graph paper plus a squiggly black line.
This method reveals the very sharp peaks and dips which account for audible differences between speakers with otherwise similar response curves. At the same time it's important not to place too much emphasis on the smaller bumps and wiggles - they're insignificant when set against the superb error-correcting abilities of our hearing system.
Feature by Ben Duncan
Previous article in this issue:
Next article in this issue: