Convolutional processing has been around in an academic form for a long time, but it was the rapid advances in DSP technology about five years ago that enabled the first practical products using this technology to reach the pro-audio market. I'm thinking here of devices like Sony's DRE S777 and Yamaha's SREV1 — both sampling reverb units — and the mighty Sintefex FX8000 'Replicator' and its siblings, the FX2000 equaliser and CX2000 compressor. All of these products employ very sophisticated digital signal processing to 'convolve' the input audio signal with a 'sample' derived from some desired process. If you're not familiar with the principles of DSP convolution, it's worth checking out the Sony preview and Sintefex review, which contain a lot of background material (see SOS June 1999 and September 2002 respectively, or see www.soundonsound.com/sos/jun99/articles/sonydres.htm and www.soundonsound.com/sos/sep02/articles/sintefex2000.asp).
In essence, the characteristics of an audio signal processor (or indeed an acoustic environment, in the case of the convolutional reverb units) are captured by measuring their unique impulse response. An impulse — a click to you and me — theoretically contains all signal frequencies at the same time, and if an impulse is fed into an audio signal processor, the output signal will be a modified impulse. The impulse level may change, there may be 'echoes' or phase-shifts, and the impulse may be stretched in time. The nature of these variations is unique to each process and so defines every aspect of the signal processing in a precise way — it is the sonic equivalent of a fingerprint.
Since a single digital sample is very similar in its form to a single impulse, this convolution technique is well suited to digital signal processing. In essence, all that is required is that each digital sample input to the convolutional processor has to be modified to replicate the same impulse pattern obtained from the test described above, which means that a single input sample may have to generate a vast train of output samples of different levels — all of which have to be added to the train of samples generated by the preceding and following samples. Consequently, convolution is an extremely DSP-hungry technology, requiring a huge amount of high-resolution digital signal processing. The Sintefex FX8000, for example, employs 44 SHARC DSP chips.
However, provided this is all done with sufficient precision, when the new input signal is convolved with the impulse response of the required audio process, the resulting output will provide a perfect replica of the required sound, just as if the input signal had been processed through the original device. This approach enables a degree of fidelity and accuracy that can not be achieved in any other way, which is why the technique is becoming so popular. It also explains why those companies who make use of these techniques tend to speak of digital 'replication' rather than 'modelling'. The latter uses conventional digital signal processing to provide a sound similar to the intended original unit by 'modelling' its characteristics. 'Replication', on the other hand, uses the more sophisticated convolutional signal processing to provide a precise and totally accurate recreation of the intended unit's characteristics.
More Than Convolution
A convolutional processor requires a huge amount of data to define the characteristics of the product it is intended to replicate. If you consider an EQ, different impulse responses have to be taken for every possible variation of control setting — every frequency value, every gain increment, and every Q value. Devices with non-linear characteristics — those with transformers, valves, or level-dependent distortion effects, for example — also have to be analysed at a variety of different input signal levels to produce an accurate replication.
A compressor is even more of a challenge to analyse, because of the dynamic nature of the device — its performance changes intentionally with both input signal level and the signal envelope. So, in addition to taking impulse responses for every possible control parameter setting, each one has to be repeated for a wide range of input levels, and the dynamic response has to be analysed too.
However, complex though all this is, the problem actually gets a lot harder when trying to replicate a mic preamp. Not only are there all the sonic characteristics associated with the circuit topology, active devices, input transformers and so on to analyse, measure and replicate, but there is also the interaction between the microphone's output circuitry and the preamp's input impedance and circuit design. This interaction affects the sonic characteristics of a preamp in a significant way, as anyone who has played with a switchable-impedance preamp can attest, and varies with different microphones because of their own impedance characteristics. So, the only way to accurately replicate a particular preamp is to match its input impedance characteristics exactly — the precise resistance, capacitance and inductance — as well as the electronic or transformer-coupled input topology, and all the other characteristics associated with its amplification circuitry. Naturally, this adds another suite of analytical tests when collecting data on a preamp for replication.
The solution Focusrite have developed mimics the actual input-impedance characteristics of each replicated preamp directly at the analogue input of the Liquid Channel. Using an incredibly elaborate arrangement of extremely high-quality signal relays, various inductors, capacitors, resistors and a special transformer can be switched into the input circuitry. This is what led to the idea of a 'liquid' channel — the fluid way in which the input stage can be altered. Using this unique technology, the microphone will 'see' the precise input impedance of the genuine preamp, and thus the tonality created by the input stage's loading of the microphone will be replicated accurately as well.
One problem encountered when trying to replicate vintage equipment is that no two units sound the same. These inherent variances are caused by the ageing of components, or where components have been substituted by others of slightly different characteristics (eg. different valves). Such variances generally result in a different level of harmonic distortion, and so Focusrite have incorporated a facility for the user to dial in a controlled amount of second-harmonic distortion to control the perceived 'warmth' of the replication.
In theory, there is no limit to the kinds of signal processing that can be replicated using convolution as the heart of the processing — although the legal aspects of this kind of technique are yet to be formally addressed. The existing convolutional reverb units demonstrate the startling accuracy that can be achieved in replicating the acoustic signatures of real environments, and the Sintefex products have provided stunningly accurate and very cost-effective replicas of a wide variety of classic EQs and compressors.
However, the one other item of revered studio outboard that frequently carries the 'classic' title is the mic preamp. Whether modern or vintage, valve or solid-state, electronically or transformer balanced, different mic preamps sound very different, and engineers and producers often go to extraordinary lengths to find the best-sounding preamp for each project.
So, wouldn't it be nice if the convolution technique could be applied to replicating the best classic mic preamps? Well, that is exactly what Focusrite have done with their new 'Liquid Channel', which was officially launched at the AES Convention in New York in October, and mentioned in last month's SOS News pages. The production units should be in the shops by the New Year, but shortly before the AES launch, I was given a privileged preview of a production prototype, and the opportunity to learn more about this revolutionary product directly from the designers.
The Liquid Look
The Liquid Channel's front-panel styling is pleasing to the eyes, with its attractive rotary encoders and subtly illuminated buttons.
The preamp section starts with a vertical bar-graph meter showing input levels over a 20dB range, with a separate Clip LED. A button cycles the input selection between analogue mic or line level and the digital input, while a fourth LED illuminates when the selected preamp replication is using the input transformer.
A rotary encoder determines the input gain, with a ring of LEDs to show the current setting. The mic input gain spans +6 to +80dB (even if the original preamps being replicated didn't), and line inputs span ±10dB. Phantom power, polarity inversion and a high-pass filter are enabled by three more buttons at the bottom of the panel section.
The next part of the front panel is concerned with clocking. A button cycles through the available internal clock rates (from 44.1 to 192kHz), while another selects internal or external word-clocks, or the digital input as the clock reference. A button at the top of the section activates the 'session saver' function — essentially an automatic gain control which prevents digital clipping of the output of the device, regardless of compressor or EQ settings.
The central portion of the unit is dominated by a large, two-line, backlit LCD window and six associated rotary encoders, each with a ring of LEDs to give an idea of the current settings. The first encoder determines the amount of second-order distortion introduced to provide a controllable degree of 'warmth' (as explained in the other box elsewhere in this article). The next five controls handle the compressor parameters: Threshold, Ratio, Attack, Release and Gain makeup. The numeric values of each of these six parameters are presented clearly above each encoder knob. The top line of the display shows the name of the preamp replication, the preamp gain, the preset program name and number, and the compressor replication.
The preamp and compressor replications are selected by first pressing the associated button at each end of the window to indicate which element to change. The data-encoder wheel in the bottom right-hand corner of the unit is then rotated to scroll through the available options. Pressing the data encoder enters the selection, and there is a brief mute while the new convolution data is loaded and the input circuitry is reconfigured.
A second vertical bar-graph meter to the right of the LCD shows the amount of gain reduction, along with two further buttons to activate the stereo link function with a second unit, and to switch the compressor into circuit. The gain-reduction meter is a rather odd because of its unusual scaling — more than half the display spans 3dB, and a quarter covers just the first dB. Although extremely precise, the result is a strange visual illusion where the meter appears to lag behind the audio peaks.
The next element of the front panel concerns the simple three-band equaliser. There are six rotary encoders, again with LED rings, arranged in pairs for each band. The high shelf at the top features a frequency control spanning 200Hz to 20kHz with ±18dB of boost or cut range. The mid-range section covers 100Hz to 10kHz with the same gain range, while the low-frequency shelf ranges between 10Hz and 1kHz. The mid-range section can also be switched to a high-Q setting for more surgical duties, and the entire EQ section is switched into circuit using an illuminated button. Other buttons allow the EQ to operate before the compressor if required, or for it to act in the compressor side-chain (with a Listen facility to aid in tuning). If accurate EQ values are required, another button calls the current parameter values to the LCD window.
The last control section on the right-hand side provides all the housekeeping facilities. Three buttons at the top provide a global Bypass mode, with Compare and Revert facilities so that you can try out different parameter settings in a non-destructive way. The lower part of the panel carries the preset Save and Recall buttons, and a preset-naming facility. There are also Clear and Setup buttons to configure the unit, and the Data knob already mentioned, with its over-press function to act as the Enter button.
I found the operation of the unit very intuitive with clear and informative metering. The LCD display also provides all the critical information in a very understandable way.
The Liquid Channel is essentially a channel strip in a 2U rackmounting box that precisely replicates (rather than models) a wide range of classic mic preamps and compressors, in combination with a new Focusrite digital EQ and some top-grade digital converters thrown in for good measure. It uses digital convolution with some very sophisticated analogue techniques (see the box below) and is the product of almost three years of joint R&D effort between Sintefex — a company known for its experience in convolutional processing, of course — and Focusrite, who bring their knowledge of class-leading mic preamp technology to the party.
The Liquid Channel features stunning chrome rack ears and a distinctive new fascia style which will be continued on future Liquid series products (several are apparently planned). This is a single-channel processor equipped with analogue mic and line input connections (but no DI input), plus an AES-EBU digital input, and both analogue and digital outputs. It is also equipped with word-clock in and out, and a USB port for both remote control and data transfers to and from a Mac or PC. A pair of phono connectors is also provided to couple two units together for stereo working.
The signal path comprises a mic preamp — the most sophisticated Focusrite has ever built — followed by an A-D stage. The subsequent digital processing provides a convolutional preamp and compressor, plus a newly designed digital EQ stage. The processing order of the compressor and EQ can be reversed, or the equaliser can be allocated to the side-chain of the compressor.
The rapid advances in DSP technology are illustrated by the fact that the first Liquid Channel prototypes produced two years ago employed four SHARC DSP chips with a maximum audio sample rate of 96kHz. The production unit uses a single, high-powered SHARC chip for all its processing, and supports audio sample rates up to 192kHz!
With its digital I/O facilities, the Liquid Channel can be used entirely in the digital domain — but not only for the replications of analogue compressors. Another converter enables the digital input to be routed through the analogue preamp as a line-level signal, to benefit from a chosen preamp's character, if required.
By the time the product is in production, it will contain the convolutional data for 40 different mic preamps and 40 compressors, and even when fully loaded, the user will be able to change the library of preamps and compressors using the USB download facility to obtain new data from the Focusrite web site. As mentioned in last month's News item, Focusrite are being coy about exactly which vintage models will be available in replicated form, but it's not hard to draw up a shortlist of models — both discontinued and current, esoteric and familar — that could fill the wishlists of most engineers and producers. Names like the Neve 1073, Pultec MBI, API 3124 and Focusrite's own ISA110, plus compressors such as the Urei 1176 black face, Teletronix LA2A, Fairchild 660, and Amek 9098 spring to mind — well, to my mind, anyway!
The unit also provides 99 user preset memories, which can also be stored externally via a USB transfer to a computer. These enable every parameter — including mic gain — to be stored or recalled in an instant. The unit can also be operated remotely via the USB interface but there is, surprisingly, no MIDI facility at all.
Despite my short time with the Liquid Channel prototype (and despite not being in ideal conditions for a listening test), I was able to compare the Liquid Channel running its Focusrite ISA110 replication with a real ISA110 sitting next to it. The preamp replication seemed remarkably accurate; I was unable to tell them apart.
The Liquid Channel promises to be a producer's dream, providing accurate renditions of a wide range of hard-to-find or unaffordable preamps and compressors, in an easy-to-use format and with repeatable settings. Assuming the production units live up to the promise of the model I played with, I can see the Liquid Channel appearing in racks all over the world in double-quick time. Watch this space!