The UAD 4K Channel Strip and Buss Compressor

April 22, 2009 11:11:11 AM PDT

The UAD 4K is an emulation of the SSL 4000 channel strip, which comprises high and low cut filters, four-band EQ, compression, and expansion/gating. The EQ includes high and low bands which can be switched between shelving and peaking filters, and two mid peaking filters with variable Q.

The buss compressor is modeled after the SSL 4000 buss compressor. It features feedback topology, VCA-controlled gain, a wide range of attack and release times, and an “Auto” release mode with program dependence. Compression ratios are also selectable.


The SSL EQ is particularly interesting in that the peaking/shelving filters are purposefully cross-linked in pairs: there is significant interaction between the high and high-mid bands, and between the low and low-mid bands. Figure 1 shows an example of cross-linking between the high and high-mid bands. The high band is boosted as much as possible at its minimum frequency, resulting in a gain of about 19 dB (blue trace). On a separate trace (black), the high-mid band is adjusted to the same frequency and approximate Q as the high band, and boosted as far as possible, resulting in a gain of about 16 dB. The third trace (green) shows the result of engaging the high and high-mid bands, both at the same time. As can be seen, there is interaction between the bands. Rather than showing a boost of 35 dB, which would be produced in the absence of interaction, the resulting curve only achieves a maximum boost of about 26 dB. A similar interaction occurs if the two bands are both used to cut in the same frequency range. A dotted trace (red) shows the response that would occur in the absence of interaction between the bands.

Figure 1: SSL 4000 EQ high and high-mid band interaction
Figure 1: SSL 4000 EQ high and high-mid band interaction

Passive EQs typically display interaction between bands, because passive filter sections often have low enough input impedances and high enough output impedances to affect each other. In the case of the SSL, the band interaction is obviously intentional, because the EQ is built around amplifier stages that have low output impedance and high input impedance. There is precedent for this type of behavior; several active vintage Neve EQs, including the 1073, have multiple EQ bands built around a single amplifier. Usually, bands sharing amplification stages are chosen to work in different frequency ranges.

In the case of the 1073, the low and high shelving filters share an amplification stage, and interact with each other. For the SSL, coupled sections operate in the same frequency ranges. One potential benefit of this design is that if two nearby bands are boosted simultaneously, separation is maintained between the areas of boost; without inter-band coupling, boosted spectral regions would be more likely to overlap.

All four bands of the SSL are designed to produce a convenient, behavior-relating filter Q to the amount of gain prescribed. In addition, the middle two bands have a continuously adjustable Q control.

"The SSL EQ is particularly interesting in that the peaking/shelving filters
are purposefully cross-linked in pairs…"

Cut Filters

The 4000 console includes high and low cut filters. The high cut filter is a two-pole response which approximates a second-order Butterworth filter, with cutoff frequency ranging from 3 kHz to 21 kHz. Figure 2 shows plots for the SSL 4000 high cut filter, with plots from the UAD 4K plug-in superimposed.

Figure 2: High cut filters
Figure 2: High cut filters

The low cut filter has a third-order response similar to a third-order Butterworth filter, with cutoff frequency ranging from 15 Hz to 400 Hz. Figure 3 shows plots for the SSL 4000 and UAD 4K low cut filters.

Figure 3: Low cut filters
Figure 3: Low cut filters


The dynamics section of the channel strip uses a common RMS detector for the compression and expansion sections. Feedforward topologies are used for both compression and expansion, so the RMS detection is performed on the incoming signal. In order to provide the capability for very fast attack times, the integration time for the RMS detector is set much lower than it would be for a typical application. This results in considerable ripple appearing on the RMS-detected signal. The RMS output is generated by the Analog Devices AD536 RMS-to-DC converter. The AD536 produces a log-encoded estimate of the RMS signal level, according to the externally selected smoothing time constant.

By definition, the RMS estimate of a signal is formed by averaging the square of the instantaneous signal level. However, squaring a signal increases its dynamic range, which can lead to difficulties with noise. The AD536 uses an ingenious technique to minimize the dynamic range of the signals produced: The input to the averaging circuit is the squared input signal divided (instantaneously) by the averaged output. Multiplication and division are performed, together, in log-space, to minimize dynamic range of these signals. Thus, the square of the input signal appears only implicitly.

The resulting RMS estimation process can be modeled incrementally using a form of Bernoulli’s equation. Through a change of variables, the exact solution to this equation can be derived with relatively low computational cost.

For the compression circuit, the log-encoded RMS estimate is fed into a peak-type detector. The detector is a second-order filter which provides program dependence. The program-dependent release is always relatively fast for transient recovery, with the slow component of the release adjustable on the front panel from one tenth of a second to four seconds. Two attack times are available by means of a switch. Figure 5 shows a plot of the channel compressor’s response when set to the highest possible ratio, with a slow attack and the slowest possible release. The response for the SSL hardware channel is shown in blue, and the UAD 4K response is shown in red. The input signal used is shown in Figure 4.

Figure 4: Input signal used for compressor response plots.
Figure 4: Input signal used for compressor response plots.
Figure 5: Hardware and software channel compressor response.
Figure 5: Hardware and software channel compressor response.

The peak-detected signal is passed through a nonlinear circuit which produces a "soft knee" for the static compression curve. Gain is applied at this point in the circuit to produce compression ratios ranging from 1:1 to infinity:1.

For the expander, the RMS-detected signal is fed to a peak detector, whose state is used to calculate the expander gain. With attack and release times set to their smallest values, RMS ripple can be transmitted through the gain calculation to produce a significant amount of modulation distortion. Attack and release times are adjustable from the front panel. The expansion ratio approximates 1:2.

For the gate, a Schmitt trigger is inserted between the RMS estimate and the peak detection circuit. The Schmitt trigger provides some degree of anti-chattering for the gate, and produces fast transitions between "open" and "closed" for the gate. However, for low frequencies or highly dynamic signals, excessive RMS ripple can result in gate chatter.

SSL recently released a modular channel strip based on the 4000 series consoles. The dynamics module (SSL XR418 E Series Dynamics Module) has a slightly modified gating circuit which includes a fast peak detector between the RMS estimate and the Schmitt trigger. This peak detector allows for undiminished response time on attack, but greatly reduces the ripple on the RMS estimate, which eliminates most gate chatter. The combination of peak detector and Schmitt trigger can be viewed as a built-in hold off time for the gate, in addition to providing a fixed amount of gate hysteresis. As a result, the XR418 gate is less prone to chatter than the original 4000 series gate. However, for certain signals, the original gate circuit is more responsive.

UA has provided models related to both the original 4000 series console gate and the XR418 gate, switchable by means of the expansion Select control. Gate G1 models the original 4000 series gate, while G2 models the modernized version.

For linking between multiple channels, gain-reduction control voltages are combined between sidechains before being applied to channel VCAs. The combination of signals is done in such a way that the channel with the highest degree of gain reduction dominates. This means that, for compression, the "hotter" channel will determine the amount of gain reduction for all channels. For expansion/gate behavior, if any linked channel has no signal present, no linked gates will open. If signal is present on all linked channels, all gates will open.

UAD 4K Buss Compressor

This buss compressor is modeled after the buss compressor section of the SSL 4000 console. This compressor uses a VCA to effect gain control. Peak detection is used, with various first-order filters used for an array of attack and release times, and one second-order filter for "auto" (program dependent) release. The compressor uses a feedback architecture, with an auxiliary VCA being used to generate a “virtual output,” from which level estimates are derived. Two additional VCAs provide gain reduction and makeup gain for the audio channels. These VCAs are ganged to the auxiliary VCA.

The compressor has a soft knee which becomes more narrow at higher ratios, which is typical behavior for a feedback compressor. Figure 6 shows the static compression curves for the SSL buss compressor, along with the curves for the UAD 4K buss compressor.

Figure 6: Static compression curves for buss compressor
Figure 6: Static compression curves for buss compressor

The “Auto” release setting displays a significant amount of program dependence, which is plotted in Figure 7 for the SSL hardware and the UAD 4K plug-in.

Figure 7: Program dependence for buss compressor
Figure 7: Program dependence for buss compressor.

The SSL 4000 has many features that give it a distinctive flavor in use. Coupling between EQ bands, nonstandard peaking sections, convenient Q control, significant program dependence on compression, and very fast gating response all contribute to a special character that is easily recognized. By carefully studying these features, we have been able to reproduce that same character in plug-in form.

— Dr. Dave Berners