I'm looking for some in‑depth education on the subject of parallel compression, with respect to its application in the mastering process. Can you help?
Via SOS web site
SOS Technical Editor Hugh Robjohns replies: Parallel compression is a 'bottom up' arrangement that lifts the quieter elements in the dynamic range in a relatively gentle and benign way, without crushing top‑end dynamics or introducing a dulling effect, which is a side‑effect of many compressors.
In essence, the signal to be processed is split, one path feeding the output directly while the other feeds a compressor. The output of the compressor is mixed into the main output along with the direct signal, and this is why it's called parallel compression. If analogue gear is used for parallel compression, there are usually no timing or phasing problems, but in DAW‑based setups there can be, if the plug‑in delay compensation isn't spot‑on.
The compressor is normally set up with a relatively modest ratio of 2:1 and the threshold adjusted so that the compressor is providing perhaps 20dB of gain reduction on the loudest peaks. You can then fine‑tune the threshold, ratio and output level of the compressor against the direct signal to get the desired effect.
The way it works is that when the signal is quiet, the output is comprised of both the direct and compressor‑path signals. The compressor won't be doing anything for a quiet signal, so the direct and compressor outputs are going to be roughly the same level. Mixed together, the actual output will therefore be about 6dB louder than the source. For high‑level signals, the direct path will be loud, but the compressor will be applying 20dB or so of gain reduction, such that the contribution from its output is relatively small. As a result, the output will be only slightly louder than the original signal.
So quiet signals are made louder, while loud signals aren't: bottom‑up compression. The big advantage is that the louder signals don't sound congested and squashed as they would with a conventional compressor setup. Of course, it's vital that the compressor can handle 20dB of gain reduction (or more) without sounding nasty. This shouldn't be a problem with software, but can be with analogue hardware.
One other risk, because of the summing of the two parallel paths, is phasing. The solution is to place a short delay in the direct signal path and adjust to remove the phasing. If the parallel compression is being done in a DAW, the delay will need to be a handful of samples. If you're using external hardware, it could be a couple of milliseconds.
It's often easiest to calibrate matching delays using sine tones and a (temporary) polarity inversion in the direct channel. Inject a tone (of any frequency) with opposite polarities in the two paths, and the combination should cancel out completely if the delays in each path are identical. If they aren't identical, only a partial cancellation will result. However, when you're trying to set up a delay in the direct path to match the processing delay in the compression path, and using a pure tone, there's a danger that you could end up delaying the signal too much and still get a perfect cancellation, because the delay could end up introducing 360 degrees of phase shift instead of zero degrees. The way to make sure that doesn't happen is to start with a very low frequency (which has a very long wavelength, so would need a huge delay to get a 360‑degree shift), adjust the compensating delay for maximum cancellation, and then increase the frequency in stages, fine-tuning the cancellation as you go. That way you can't accidentally end up 360 degrees out.
Start with a sine wave at a low frequency and adjust the delay to obtain the maximum null (silence). Then increase the frequency as you focus in on the correct delay time. The higher the frequency, the more accurate the matching delay needs to be to maintain the quietest or deepest null. Once you get up to about 15kHz with a very deep null, switch off the tone, restore the polarity inversion and enjoy working with your time‑aligned parallel paths. This is an old technique that was used to align tape‑head azimuths, where the same potential error problems existed.