What if money doesn’t really come into it? What if a ‘virtuous’ system could be built from ordinary components connected together in a sensible way?
One such virtuous system, I would suggest, might comprise the following:
- Digital source material (CD, FLAC, DVD etc.).
- DSP-based crossover filtering and driver correction in both the frequency and time domains
- Multichannel DAC (one DAC per driver)
- Solid state amplification (one amp per driver)
- Three-way box speakers that do not use bass reflex
On the face of it, this would reduce, or provide mechanisms to address, all of the problems listed in The Signal Path of Shame.
Such systems are still quite rare. Those that exist often elicit superlatives from those that hear them, but compared to the visual appeal of valves and vinyl, a black box of DSP holds little appeal for audiophiles.
Taking the various elements one by one:
Unlike various ‘legacy’ analogue sources, the digital audio system stores the signal in the same way that we would store documents or images in a computer: as numbers. Unlike those systems, however, it reconstitutes the stored data using an extremely accurate curve-fitting system which includes an analogue element at the lowest level – to create a continuous output waveform.
Many myths abound, and even technically literate people fail to understand that the system is fundamentally perfect, even down to the lowest levels. Dither is an integral part of the system; there isn’t a horrible distorted waveform lurking down in the lower few bits of resolution – merely the signal with added ‘hiss’, like tape hiss but far lower in relative level. The only questions are: how much hiss can we tolerate, and what are the highest frequencies we wish to reproduce? No one has yet shown that CD doesn’t actually meet all requirements of human hearing. Higher resolutions and sample rates are available, but amplifiers and speakers are generally not capable of reproducing frequencies much higher than CD anyway – and can distort in audible ways if asked to.
DSP for Crossovers and Driver Correction
This is all part of the simple concept of wishing to reproduce the signal unchanged in the sound pressure variations that emanate from the speakers. This basically requires a diaphragm or “piston” to accelerate in proportion to the amplitude of the recorded signal.
In practice, we find we need to combine the outputs of several pistons of different sizes because a single driver cannot reproduce extremes of low and high frequencies simultaneously without problems. If the drive units are close enough together and producing suitable signals, their outputs will sum acoustically to create the effect of a single piston.
The conventional solution of using passive crossover filters is fundamentally imprecise, even if they are made of NASA-grade materials. With DSP, several problems are ‘mopped up’ in one go, and the basic system can be seen as:
- Define a target response.
- Split the signal into frequency bands suitable for the individual drivers
- Measure the output of the system.
- Program the signal processing to make the necessary compensations to meet the target response.
Even better, if possible, is to substitute:
3. Model the system.
At its simplest, DSP can give extremely accurate versions of conventional crossover filters. These do not guarantee to give you the acoustic response you desire, however, because the individual drivers impose their own frequency and phase responses.
The next step up is to produce filters that cannot be built from conventional analogue means. These can, for example, be linear phase, meaning that they do not distort the transient response of the system, and may effectively need to see into the future. This is achieved by adding an overall delay or latency into the system.
Beyond that comes driver correction, which is usually restricted to corrections of the phase, frequency and time responses of the drivers so that their acoustic outputs better match a simple target response. A driver can be rendered genuinely ‘flat’ and phase neutral in its response – at a single point in space. DSP cannot influence the dispersion characteristics of a single driver in combination with its baffle – a diaphragm of finite size will always ‘beam’ as frequency rises – but DSP can influence the overall dispersion of the sum of multiple drivers through the slopes of the crossovers between drivers and at what frequencies they occur.
A very interesting next step may be nonlinear driver correction, where we get even closer to a completely neutral system, but this is not commonly done – yet. However, in some speakers, bass drivers are already corrected for distortion using feedback of cone motion.
For an active DSP system we need to reproduce multiple channels simultaneously: one per driver. A multichannel DAC is just like a conventional stereo DAC but with more outputs. Samples are passed to it in “frames”, with samples for all drivers reproduced simultaneously and locked together to the same sample clock. Such devices are commonplace and cheap, usually running from USB or some other digital interface.
A distinction is made between isochronous and asynchronous interfaces. Only asynchronous DACs can be said to be truly free of jitter caused by multiple clocks in the system and cable effects – but this is a theoretical fact rather than something audible (unless there is something very seriously wrong in the implementation). However, it makes sense to use asynchronous DACs in domestic hi-fi systems – because we can.
Solid State Amplification
Solid state amplifiers have been available for several decades but, amazingly, are not universal – some people still use valves.
Various topologies of solid state amps exist, and most have the possibility of low distortion and low output impedance. For an active system we need one amp per driver. A relatively recent development is Class D amplifiers that are very compact and efficient, making active systems more practical and attractive.
Three-Way Box Speakers
This is just one of several possible options but, for all its simplicity, has many advantages.
The chief areas of interest are:
- power handling
- acoustic interaction with room and listener
An active system with more than two ‘ways’ already gives us advantages for the first three items. The last item is the one where speaker design deviates from a simple engineering problem. Measurements may not actually represent what people think they do.