Ask the Doctors: Behind the Precision Maximizer and the Precision De-Esser
by Dave Berners

This month, we will cover the working of the two new plug-ins for release 4.8: the Precision Maximizer and the Precision De-Esser.

UAD Precision Maximizer

Precision Maximizer
The basic operation of a maximizer was covered in the May, 2007 edition of "Ask the Doctors." The maximizer is a saturation effect in which a memoryless, nonlinear mapping is applied to the input signal. In many ways, the saturator does the same job as a brick-wall limiter, by decreasing the peak-to-RMS ratio of the signal while preventing the output signal from exceeding a pre-determined level. Both types of effects are nonlinear, and as such, produce distortion components. In the case of discrete-time systems, the distortion components can be aliased, producing various secondary sources of distortion within the audio band.

For the limiter, the primary distortion-producing mechanism is gain modulation. Distortion is minimized by choosing a (possibly program-dependent) release mechanism that limits the bandwidth of the gain signal. Slower release times will reduce the amount of distortion present at the output, but will also reduce the device’s effectiveness at reducing peak-to-RMS ratios. Limiters are most effective for signals that have a relatively high peak-to-RMS ratio to begin with. For signals with lower initial peak-to-RMS ratios (signals without large transients), it becomes difficult to build a limiter that can further reduce peak-to-RMS ratios without adding significant amounts of distortion, spread over broad parts of the spectrum.

By proper choice of saturation curves, the distortion can be mostly confined to low harmonics of the incoming signal.

By contrast, maximizers can be effective for decreasing peak-to-RMS ratios even if the initial peak-to-RMS ratios are relatively low. Because the maximizer is implemented using a memoryless nonlinearity, there is no attack or release, and the form and amount of distortion is determined instead by the shape of the saturation curve. Overall distortion levels are typically higher than those produced by limiters or other gain-reduction devices, but, by proper choice of saturation curves, the distortion can be mostly confined to low harmonics of the incoming signal. There are several advantages to centering the harmonic distortion around lower harmonics:

• Distortion is not as easily perceived at lower harmonics.
• Distortion at lower harmonics appears to be less objectionable, according to perceptual tests.
• Distortion confined to lower harmonics is less likely to be aliased.

The Precision Maximizer uses a saturation curve, which is chosen in an attempt to minimize the possibility of aliasing and to distribute most of the distortion energy into the lower harmonics. For inputs at or below 0 dB at the input to the saturator, distortion energy will be absolutely confined to a small number of harmonics of the original signal energy. For input signals that exceed 0 dB, the rate at which distortion components extend to the upper harmonics as a function of harmonic number is bounded. The "Shape" control allows for a range in the amount of peak-to-RMS reduction while keeping the maximum allowed output level constant. Oversampling is used within the algorithm to minimize possible aliasing.

The Precision Maximizer can be run in multi-band mode, where the signal is separated into three bands before the nonlinear mapping is applied. Bandsplitting is accomplished via a perfect-magnitude-reconstruction filter bank. The motivation for splitting the signal into multiple bands is that intermodulation distortion is reduced by processing the bands separately. Because bandsplitting makes it difficult to predict the maximum possible level of an output sample, a separate single-band saturator is provided which can guarantee no overshoot when running in split-band mode ("Limit" control).

UAD Precision De-Esser

Precision De-Esser
De-Essing is usually accomplished by frequency-selective compression or by multiband compression. The Precision De-Esser can be used in either frequency-selective mode or multiband ("Split") mode. In frequency-selective mode, filtering is performed at the input to the sidechain, so that the compressor is more sensitive to the selected frequencies (highpass or bandpass). In multiband mode, the signal is fed to a perfect-magnitude-reconstruction filter bank, with one band being fed through the compressor.

Ess detection is traditionally thought of as a binary decision. That is, an ess is either present or it isn’t. We do not usually conceive of a continuum of states between ess and non-ess, although offending esses can be produced at a continuum of levels. When  considered within the scope of a single ess, de-essing is more a ducking process than it is a reduction of dynamic range. For the Precision De-Esser, a compression scheme has been designed specifically for de-essing; the attack and release dynamics of the compressor have been chosen to be suitable for ducking. Two choices for attack and release times are available by user selection.

Ess detection is traditionally thought of as a binary decision. That is, an ess is either present or it isn’t.

The filter bank for the Precision De-Esser has been chosen on the basis of flexibility. In two-band mode, a lower-order filter is used. This is intended for de-essing vocal tracks that have not yet been mixed with program material. This gentle filter bank is sufficient for processing unmixed vocals, and the low order reduces phase distortion and resonances within individual bands. A three-band mode is implemented for tracks that have already been mixed. Here, a higher order (and sharper) filter bank is used, allowing for more control over the extent of frequencies affected by gain reduction. This filter bank is a three-band implementation of the minimum-phase filter bank used in the Precision Multiband Compressor.