Analogue to digital (A-D) and digital to analogue (D-A) conversion is at the heart of any computer audio interface, and quality conversion is something that Focusrite has been doing for almost as long as it has been creating microphone preamps. The first Focusrite digital converter, released in 1996, was part of the acclaimed Blue range of mastering processors: the Blue 245 A-D. It was designed by Dr Dave Malham, then at the University of York and an acknowledged expert in digital audio. Here he is talking about what took him into digital conversion system design and the work that went into the Blue 245. He also discusses the design criteria that go to make up a high quality digital audio conversion system, and the factors that affect the system's performance.
Foundations of digital audio
The original work that made digital audio possible was done over 60 years ago by researchers like Harry Nyquist and Claude Shannon, after whom the central digital sampling theorem is named. Early digital audio systems showed promise, but it was still a new technology – and one with some significant problems. The discovery of dither, the importance of stable, clean clock signals and the influence of jitter all improved digital audio quality bit by bit. Manufacturers often came up with unique solutions to the problems, such as special flavours of dither and special filters to get around quality issues.
Digital audio today
Today, many of the factors affecting digital audio quality are well-known and understood, although there are still some areas of contention. But by and large, the factors that were major issues in the early days are no longer a problem. We have no trouble deriving a clean clock signal from a noisy one, for example - though there is still controversy about whether internal or external clocking sounds best. We can remove jitter more or less completely. We know how to design good filters, and we know how to dither a signal successfully. So what makes one conversion system better than another?
Keys to converter design
It’s fair to suggest that in most cases there is no "special sauce" any longer that will make one conversion system better than another. But there are some key design considerations that will affect the quality of the end result.
It’s not a numbers game
All too often we tend to think that the better the numbers, the better the product. This, however, is simply not the case. A good conversion system is a matter of balancing competing factors, like the noise floor, dynamic range, frequency response and distortion. At Focusrite, the aim is to balance those factors for the best possible sound, and that doesn’t always mean the lowest numbers. It’s not worth squeezing a dB or so more out of a noise floor that’s already way below the noise level of any reasonable input signal if the result is that the distortion goes up, for example, and when you’ll hear one but not the other.
One other area where there’s a numbers game is a tendency in some quarters to give figures that are simply taken from the chipset’s data sheet. Needless to say, this doesn’t give you a true impression of the performance of the unit as a whole. When you see a number on a Focusrite specification, it’s what you’ll get when you are using the product on a real session.
One of the fundamentals of good converter design is not cutting corners. That means careful component selection for example, and it means good board layout and construction. Often you’ll find the converter chips in the centre of a board, with their digital support chips on one side and the analogue circuitry on the other. Keeping them physically as far apart as possible minimises fast rise-time digital spikes getting into the analogue side and degrading the noise floor. Further performance improvements can be obtained by using multi-layer boards with ground planes between layers. In many ways, the analogue circuit design and implementation is more critical than the digital and requires particular care and expertise for best results.
About the chips
Strange to say, if you’re dealing with a reputable manufacturer, knowing what converter chips they used is fairly immaterial. You should be able to expect that they will have chosen the best part for the job – and that will include knowing that it won’t become obsolete during the expected life of the product as well as performance. And of course price will come into it too. In the old days, when some chips gave markedly better performance than others, some manufacturers erased the identification on their converter chips so nobody else would know what they using. Today… there’s really no point.
The 'word length' or 'bit depth' of a digital system indicates how many different levels of amplitude the system can capture, and thus the maximum theoretical dynamic range. A 16-bit digital value, as used in systems like Compact Disc, can represent 65536 (2 to the power 16) different numbers, for a maximum dynamic range of around 96dB. Most converters offer 24-bit operation (16,777,216 different values), corresponding to a dynamic range of over 144dB – many times quieter than we can hear, and beyond the best performance that today's components can offer. The bit depth is commonly held to indicate the “resolution” of the system. This isn’t completely true: the better a digital signal is dithered, the greater the resolution. We can hear around 20-21 bits, so working at 24-bit makes sense. 32-bit… not so much.
It’s common to find conversion systems that offer a wide range of sample rates, from the “standard” 44.1kHz (CD) and 48kHz (DVD) up to 176.4 and 192kHz (as found on the majority of Focusrite interfaces) and sometimes beyond. While you may need a computer with a little more power, and you will need more storage space if you're running at higher sample rates, it makes sense to use them in a production environment, even though there may not be a lot going on in audio terms at the high frequencies these can capture. Here's why.
To begin with, higher sample rates mean lower latency – the time it takes a signal to pass through the interface. Any digital audio process introduces latency, but if the latency is low enough, you can, for example, monitor with your favourite plugins while you record. Low latency is a specific design goal in Focusrite interfaces, but you can make it even lower by working at a higher sample rate.
Another reason for considering higher sample rates is that the filters used in digital audio systems are 'further away' in frequency terms from the actual audio. This means that they can be smoother and more gentle, and less likely to introduce problems. As a result, you can expect improvements in sound quality.
96 kHz: a good balance
Overall, recording your masters at higher sample rates makes sense, and 96 (or 88.2) kHz sampling is a good balance. Latency is noticeably reduced; performance is optimised; the converters aren't operating at the edge of their range yet the filters are operating at well beyond the audio band, leading to improved sound quality. Above 96kHz sampling you will tend to experience less evident improvements while at the same time needing significant additional computer power and storage – but higher rates are there if you need them.
At the same time, it's worth noting that the streaming and download markets are becoming increasingly interested in providing “Hi-res audio” (which means 24-bit, 96kHz or better) to their customers, so it’s worth having 96kHz masters available.