RaneNote

Dynamics Processors -- Technology

Chapter 4 -- Specialized Compressors

This chapter explores very useful variations on the basic compressor technology. Adding a parametric equalizer section in the side-chain creates a frequency sensitive compressor; using a crossover allows split-band compression; putting a tracking filter into the main signal path and side-chain gives you a dynamic EQ; comparing broadband and bandpass energies produces a relative threshold dynamic EQ, which makes a terrific de-esser; while other clever additions solve the problems of automatic gain control and peak limiting. Here are the details:

Frequency Sensitive Compression

Frequency sensitive compression is broadband compression as described above with the addition of side-chain equalization to make the detector more or less sensitive to certain frequencies. The basic topology is shown in Figure 6. Side-chain equalization may take the form of a parametric filter (with variable boost, cut and bandwidth), high-cut filter, low-cut filter or all three. In some cases, multiple parametric filters or a multiband graphic are used in the side-chain. If the amplitude of a frequency in the side-chain is reduced, the broadband compressor is less sensitive to it. If the amplitude of a frequency is boosted in the side-chain, the broadband compressor is more sensitive to it.

Figure 6. Frequency sensitive compressor block diagram.

Split-Band Compression

Split-band Compression divides the incoming signal into two or more frequency bands as shown in Figure 7. Each band has its own side-chain detector and gain reduction is applied equally to all frequencies in the passband. After dynamics processing, the individual bands are re-combined into one signal. The handling of in-band signals is the same as for the general-case compressor shown above.

Figure 7. Split-band compressor block diagram.

This configuration is easily done with Rane's Drag Net software and DSP processors like the RPM 2m as shown in Figure 8. This quickly expands into three, four or more frequency bands as required. Moreover, adding side-chain EQ and filters is just a drag and drop away.

Figure 8. Drag Net split-band processing.

Dynamic EQ

Dynamic EQ differs from the forms of compression listed above in that it dynamically controls the boost/cut of a parametric filter rather than broadband frequency gain. The basic dynamic EQ uses a bandpass filter in the side-chain with variable center frequency and bandwidth. The side-chain detector is sensitive only to the passband frequencies. A parametric filter with matching bandwidth and center frequency is placed in the main signal path and the boost/cut of the filter is controlled the same way a broadband compressor boosts or cuts broadband gain. The basic topology is shown in Figure 9.

See the Application section in Part II for the many creative uses of Dynamic EQ.

Figure 9. Dynamic EQ block diagram.

Relative Threshold Dynamic EQ

Relative Threshold Dynamic EQ is a special form of dynamic EQ where the rms level of the bandpass signal in the side-chain is compared to the rms level of the broadband signal. The difference between the bandpass and broadband levels is compared to the threshold rather than the absolute rms value of the bandpass signal. The advantage of this type of dynamic EQ is that the relative amplitude of a band of frequencies, as compared to the broadband level, is maintained regardless of broadband amplitude. The typical topology is shown in Figure 10.

Figure 10. Relative threshold dynamic EQ block diagram.

De-essers

De-essing limits or controls the sibilant content of speech. Sibilance produces a hissing sound. English sibilant speech sounds are (s), (sh), (z), or (zh). De-essing is often confused as a type of dynamics processor. It's actually a specific application that is accomplished using many different types of dynamics processors. And contrary to popular belief, successful de-essing is not as simple as placing a bandpass or treble-boost filter in the side-chain and calling it done. Frequency Sensitive Compression, Split-Band Compression, Dynamic EQ and Relative Threshold Dynamic EQ are all used for de-essing.

True de-essing involves comparing the relative difference between the troublesome sibilants and the overall broadband signal, then setting a threshold based on this difference, therefore it is our experience that Relative Threshold Dynamic EQ (as described above) is the best dynamics processor for this task as it is able to maintain proper sibilant to non-sibilant balance regardless of level.

A good de-esser looks at the average level of the broadband signal (20 Hz to 20 kHz) and compares it to the average level of a bandpass filter in the side-chain. The threshold setting defines the relative threshold, or difference, between broadband and bandpass levels, which result in compression of sibilants. Because de-essing depends on the ratio of sibilant to broadband signal levels, it is not affected by the absolute signal level, allowing the de-esser to maintain the correct ratio of broadband to sibilant material regardless of signal level, as shown in Figure 11.

Figure 11. De-esser unaffected by level.

This means that the de-essing performance is consistent and predictable, regardless of how loud or quiet the singer/talker is. Taming sibilance when the talker is quiet is just as important as when the talker is at a fevered pitch.

Figure 12 shows what happens using a primitive de-esser with a side-chain EQ. Sibilance during loud passages is attenuated, but there is no gain reduction during quiet passages, even though there may still be a significant amount of "sss" in the person's voice. For a given threshold, this often results in an overly aggressive effect during the loud choruses, and a completely ineffective result during the hissy, whispered verses.

Figure 12. Primitive de-esser with a simple side-chain. Varying input levels adversely affects de-essing.

Automatic Gain Control (AGC), or

Automatic Level Control (ALC)

Automatic Gain Control (AGC), also known as Automatic Level Control (ALC), is a specialized form of compression. It is a circuit or algorithm that varies gain as a function of the input signal amplitude. Commonly found in pro audio applications where you want to automatically adjust the gain of different sound sources in order to maintain a constant loudness level at the output. One of the most common applications is for speech. Another example, on better DJ mixers the gain adjusts automatically when the DJ switches sources between records, CD, or MP3 files. Not only do signal levels differ greatly between different source technologies but also between any two examples of the same technology, e.g., between CDs, or between MP3 files, etc.

AGC is more similar to older compressor designs which compressed a signal about a threshold value (see Appendix). In these designs, gain was reduced for signals above the threshold and increased for signals below the threshold. One of the problems encountered with this type of compressor is the possibility of very high gain at low signal levels.

A typical modern AGC implementation is shown in Figure 13a. Note that the traditional threshold control is now labeled target. The target is the desired, nominal level. As with early compressors, gain is reduced for signals above the target and increased for signals below the target as shown in Figure 13b. Note the gain is unity (input = output) at the target. The threshold is defined as the level below which the AGC circuit will not increase the gain. Note the gain is unity (input = output) below the threshold. The implementation shown in Figure 13a indirectly determines the threshold for given maximum gain and ratio settings. Other implementations exist, but the basics are the same. The hold parameter determines how long the above threshold gain is held in the absence of a signal. For speech, this feature allows the correct gain to be held during pauses. Figure 13c shows the control screen for Rane's Drag Net AGC module.

Figure 13a. AGC block diagram

Figure 13b. AGC response graph.

Figure 13c. Drag Net AGC control block.

Peak Limiter

As expected, the basic topology of the limiter is similar to that of the compressor. Unlike the compressor, the limiter must ensure that a signal never exceeds the set threshold. This requires the use of a peak responding detector and a fixed ratio of infinity:1. Figure 14a illustrates the basic topology. The response of a limiter with a threshold of -20 dB is shown in Figure 14b.

Figure 14a. Peak limiter block diagram.

Figure 14b. Peak limiter response graph.

Figure 14c. Drag Net Limiter control block.

While the basic operation of a limiter is straightforward, coaxing one to sound good is challenging. Abrupt limiting causes significant alteration of the sound and determining the best release rate for a particular signal is problematic. Digital signal processing enables two additions to the basic topology which go a long way toward resolving these issues. First, adding delay in the main signal path allows the side-chain to "see what's coming", and start to respond prior to the threshold actually being reached. The result is a softer leading edge resulting in a more natural sound. Second, looking at recent history gives the system knowledge of where the signal has been and where it is likely to go. With this knowledge, the best release rate is set dynamically. Figure 15 illustrates the topology.

Figure 15. Look-ahead peak limiter block diagram.

Primarily used for preventing equipment, media, and transmitter overloads, a peak limiter is to a compressor as a noise gate is to an expander (more on this later).

The most useful dynamics processor designs incorporate a separate peak limiter function independent of the compressor. A separate peak limiter frees the compressor from the task of clamping wild excursions. The peak limiter plays level-police while the compressor persuades more gently.

Peak Limiter Uses

Prevent clipping and distortion in power amplifiers.
Protection of loudspeakers from damage resulting from destructive transients (like a dropped microphone).
Preventing overs (digital clipping) during recording.
Preventing overmodulation of the transmitted signal in broadcast.