The consumer market is beginning to exert its influence on the pro-audio business, and it's not all good news.
Mixing studios are becoming a strange technological admixture of the old and the new. Alongside computer-controlled mixing desks and virtual faders sliding up and down on screen by themselves, replacing the engineers that would once have slid around on chairs to make a fade at the right time, are pieces of vintage hardware.
Studios will often have one or more Pultec equalisation (EQ) units to fine-tune the balance of mixes or an Urei 1176 compressor to massage a guitar solo or vocal into shape.
Bigger studios will have the originals; smaller ones, official recreations or unofficial clones. The Pultec EQ-1P, for instance, is very limited in what it can do – it only operates on very low and very high frequencies – but it remains a favourite among audio engineers despite its 60-year-old circuit design. The odd – but prized – behaviour of the Pultec has encouraged software companies to try to match it in the digital domain with a combination of circuit and behavioural modelling. Unlike most digital EQs where a boost and cut at the same point will cancel each other out, the Pultec's response does not – it is this slight unpredictability that makes it a studio favourite. Its internal amplifier, used to compensate for losses from its passive EQ circuitry, also imparts its own character to the sound.
Where even a large studio might have only one or two hardware Pultecs or an 1176, emulating them in software makes it possible to deploy many of them at once, just as long as your computer has enough MIPS (million instruction per second). What's more, there seems to be an endless thirst for more MIPS. Even the lossy inaccuracy of tape has become a prized sound – largely for the way that it compresses loud sounds. 'Looking at the systems from Universal Audio, they are doing A/B comparisons of an original Studer tape recorder to an emulated Studer, and it's dead on,' says Denis Labrecque, marketing programmes manager for pro-audio at Analog Devices.
Labrecque points to another audio tool, the reverb unit, as something that has multiplied inside the digital mixing environment. In the 1960s, the reverb unit was a large, unfurnished room, or a plate of steel attached to microphones. It was turned into a hardware box by companies such as Lexicon in the 1980s, and shared among audio tracks. Now, mixing engineers will routinely set up multiple parallel reverbs using software equivalents, each with a subtly different treatment to try to place each sound within a distinct space.
'They might perform reverb processing on 128 channels now where before they were doing it on a lot fewer,' says Lebrecque. The result is a seemingly endless thirst for MIPS. Labrecque explains: 'There are two sides to this. One is the more you give people the more they want to do. The other is the level of sophistication of that these modelling processes demand, which demands even more processing power.'
A decade ago, the only realistic option for this kind of processing – often called 'in the box' by mixing engineers – was the digital signal processor (DSP). Products such as the Analog Devices Sharc and the Motorola DSP56K family were mainstays of plug-in cards made by Digidesign for its Pro Tools environment, as well as Creamware and Universal Audio – the successor company to 1176-maker Urei.
To get enough horsepower, these cards were armed with multiple processors linked through a bus or network for transferring streaming-audio samples. Competition has intensified from mainstream processors such as Intel's Pentium family, and a new generation of graphics processors from nVidia and AMD subsidiary ATI, that have enough resolution in their floating-point units to handle audio samples.
Because people tend to upgrade their PC and Mac hardware more frequently than studio equipment, the available horsepower from these new single-chip multiprocessors has convinced mixing engineers to use these platforms more often. Avid, which bought Digidesign several years ago, responded by separating the processing from its audio interfaces so that the plug-ins for the professional version of Pro Tools could be moved entirely to the PC-based digital audio workstation (DAW).
With a much larger market to go after and so refresh their chip-level offerings more frequently, Intel, AMD and Nvidia look to be in a strong position. 'It's a question that comes up about every two years or so and has for the past 10 to 15 years. There is a continual leapfrogging of audio processing technology,' says Labrecque.
Ian Dennis, technical director at pro-audio hardware supplier Prism Sound, says: 'Processing power is increasing dramatically on all types of platform: per buck, per watt, per square millimetre.'
Because users do not necessarily have to buy dedicated processing cards, the market has opened up for audio-processing software. 'The decamping of audio production from the exclusive recording facility to the desktop PC means that more music is being made than ever, and so there are more customers out there if we adapt our offerings to the changing practices,' Dennis believes.
There is, however, a downside, Dennis adds: 'On the technical side, we undoubtedly find it harder to guarantee the same high-quality, high-reliability experience for DAW users – because so much of the system is now beyond our control.'
One of the casualties is latency. Typically, the equipment used in studios – whether professional or at home – works with a few hundred samples at most. Even that is undesirable, but longer latencies become too distracting for musicians and engineers to work with. Mastering engineers can work with longer latencies, but this is where all of the music is pre-recorded. Even then it can be distracting to make a change to a setting and have the results only appear a second later.
Handling long latencies
Latency becomes a serious issue with some architectures, such as GPUs. These parts are optimised for situations that involve lots of data being processed in parallel. A typical audio function, such as a filter within an EQ, involves long streams of serial data. Because the execution units in graphics processing units (GPUs) are not designed to read data held by other units, it is difficult to split the filter into parallel elements.
It is possible to dedicate each sample within a block to a thread and then download the entire block into the GPU's processor cores. There is a catch, though: 'You need to send all the samples at the same time,' says audio engineering researcher Tiziano Leidi, based at EPFL in Lausanne, Switzerland.
You can only do that if you have buffered them already, which implies a long latency between sampling the audio input and actually processing them. Leidi says the range of sample latencies for GPU-based audio processors can be as high as 32,000. Even at high sample rates, those latencies are significant. 'Of course, in ultra-demanding applications where large channel count or processing complexity dictates dedicated hardware rather than a PC, we can deploy huge processing power at a fraction of yesterday's cost,' Dennis adds.
Because of the latency issue, some companies, such as Universal Audio, are sticking with DSPs rather than being tempted over by the promise of more MIPS. Consumer audio is now beginning to see the same tension between dedicated and general-purpose processing. Wolfson Microelectronics systems architect Ian Smith says: 'More hi-fi products are becoming a PC in a box.'
The effect is that consumer hi-fi resembles more a professional DAW, running software plug-ins that apply functions such as noise cancellation, room correction and artificial reverb. But some of these functions demand low- or fixed-latency response, which can be much easier to deploy using dedicated processing, says Smith. 'For functions such as noise cancellation, latency is critical.'
Labrecque says: 'Compared to 15-20 years ago, it is very different. Your great system then might have an eight-band EQ. Now it's synthesising 5.1 material from a stereo source. There is a tremendous amount of processing going on in the consumer market.' The home-user is not expecting to have modelled Pultec adding some bass and treble to a mix, but Labrecque adds: 'EQ management, noise abatement, these kinds of things are appearing and not just in homes. I sat in a car that seemed to have better acoustics than a studio the other day.'
Pro-audio versus home users
Other aspects of pro-audio design are trickling into the consumer space. 'We're finding the people heading up consumer audio divisions now are primarily ex-pro audio, and they carry the mindset from there,' continues Labrecque. 'Markets that used to have low specs are now beginning to get pro-like specs,' agrees Bob Adams, a fellow at Analog Devices, and specialist in analogue-to-digital converter design.
Nick Roche, director of global applications at Wolfson, says: 'We now have smartphones that can drive full HDMI interfaces. It makes sense to use the phone as a high-end audio source. We can't quite say our smartphones are as good as mixing desks. But we are starting to see a bleedthrough. We do see more of the technology migrating into the high-volume products.' Applications such as Apple's Garageband have helped to popularise home recording and hobbyist audio products have improved dramatically since the days of recording to multitrack cassette tape; but this has encouraged chipmakers to focus on the lower sampling rates, such as 44.1kHz and 48kHz rather than the 96kHz and even 192kHz rates that professional recording studios use.
Home-cinema owners expect their DVD and Blu-ray players to be able to work with audio sample rates way in excess of the CD audio standard of 44.1kHz. These A/D and D/A converters are not designed in quite the same way as the devices aimed at pro-audio users. Compromises in design have led to some scepticism as to the value of going above 96kHz among studio owners, an environment where higher sample rates are the norm to minimise the impact of processes that can generate unwanted aliasing, a phenomenon caused by high-frequency signals being captured by the converter and reflected down into hearing range. Aliasing is an ugly sound that designers take great care to reduce as far as possible. Analogue-modelled compressors, filters, and distortion units can be big culprits.
Filtering and going digital
The main difference between consumer and pro-audio converters lies in the filtering performed once the signal is in the digital domain. Practically all audio-rate converters used a sigma-delta architecture – a technique Philips called 'bitstream' in the early days of CD – that uses a very low-resolution conversion stage in combination with sampling at many times the nominal sample rate. In an A/D converter, a digital 'decimation' filter processes the sampled waveform to build a much higher resolution waveform out of the stream of 1bit samples at the target sample rate. The filter in a D/A reverses the process. A key decision in designing decimation and interpolation filters lies in their complexity and, with that, how much space on the chip to allocate to them.
If a D/A converter is designed primarily for operation at 48kHz in consumer equipment, the filter itself cannot simply be run faster to cope with a higher sampling rate. So, the net effect of running the D/A converter at 192kHz is that the interpolation filter will run through fewer stages per sample, which leads to a slight reduction in overall quality at the higher sample rates.
Converters designed for the pro-audio environment have more complex decimation and interpolation filters that offer better performance at higher sample rates. Because of the filter trade-off, controversy reigns in pro-audio as to whether the move to 192kHz from 96kHz – which is now commonplace in studios – is worth it. Dan Lavry, head of Lavry Engineering, authored a white paper several years ago arguing against its use on the basis that 192kHz was stretching the performance of A/D and D/A converters, and that the overall noise floor increases.
'When you move from 48kHz to 192kHz, you increase the pass-band by a factor of four,' says Roche. 'So the noise, overall, increases.' However, says Adams, 'it is a stretch to say you can hear above 20kHz and even if your hearing was that sensitive, an even bigger stretch to hear the effect at 50kHz', so the noise that can be picked out by the ear and brain is far lower. *