The simplest audio declicker is an attenuator which, at the precise moment that a click is detected, mutes both channels, thereby reducing the impact of the click. The minimum duration of the mute (actually a high speed fade-out and fade-in) is typically 2.5ms, and in any case must be greater than the scratch length. Therefore even a small number of mutes seriously affects the percieved sound quality.
Also, since the method only seeks to make the clicks less obtrusive, it does not in any way restore the underlying signal. In addition, it is only applicable when the energy contained within a click is very much greater than the energy within the signal. (Primitive though this concept is, it has also been utilised in the digital domain. When an error correction system is unable to cope with high density or long data errors many digital systems will mute.)
The most sophisticated analogue click-removal algorithm is used in a device known as a switcher. Using two sources of nearly identical signals (the opposite groove walls of a monaural record replayed using a stereophonic cartridge) this monitors for the cleaner signal (i.e. the lower energy) and switches the output source between groove walls as appropriate.
This method removes large clicks but, like all non-digital solutions, is unable to distinguish small ticks from genuine signal components. Also, the switcher assumes a perfect monaural source. If the groove walls differ significantly, or suffer degradation simultaneously, then the assumption (and therefore the restoration) fails.
Digital technology has made it possible to implement ideas that could not be realised using analogue electronics.
The first of these is Sample & Hold (S&H) which, in many ways, is the same algorithm as used in a perfect muting system. However, instead of creating a signal plateau at zero amplitude, this method assumes that a plateau at the level of the most recent valid signal will be closer to the true signal. S&H removes the largest manifestations of clicks and scratches, but the resulting waveform contains unpleasant distortion and many audible 'bumps' and 'pops'.
While these low amplitude thumps may be preferable to the high amplitude clicks of the untreated data, the signal will show signs of severe break-up if the click density is high. Many listeners complain that these artefacts and side effects (as, for example, implemented in the error correction of domestic CD players) are more unpleasant than the clicks that they replace.
Figure 1: Sample and Hold.
The next stage is Linear Interpolation. Whereas Sample and Hold takes account of one good sample adjacent to the scratch, linear interpolation takes account of two adjacent good samples and the elapsed time between them.
In this algorithm the corrupted data is replaced by a straight line joining the last good sample and the next available good sample. It is impractical to implement this method in the analogue domain, but relatively straightforward in the digital domain.
The audible result of this method is less offensive than Sample and Hold, but suffers from low frequency artefacts and a reduction in audio bandwidth over the interpolated region.
Figure 2: Straight line interpolation following sophisticated scratch detection algorithm.
CEDAR's interpolator is based on the technique of signal modelling. By analysing the signal over a region of several milliseconds it can build a model of the underlying resonant signal (such as music or speech) and then use the information to replace each click with an interpolation that fits this resonant model.
The CEDAR declicker embodies the latest developments of this algorithm (along with a sophisticated click detector) to remove up to 2500 clicks per channel per second in real time. The performance of this system is so good that, in almost all cases, it is not possible to hear that the signal was damaged prior to restoration.
Figure 3: A high order interpolation following a sophisticated click detection algorithm.