Uncovering the truth behind high-pass filtering

From day one of music production school, you’re instructed to apply filters to non-bass elements in a mix, in order to remove clashing low-end frequencies and therefore reclaim mix headroom. Yet when applying a high-pass filter, a signal’s peak level can actually increase! Why exactly is this? And does this mean it’s bad to apply individual filters throughout a mix? Andrew Simper, coding expert and filter guru, fills us in.

“This is a good point, and is definitely something to keep in mind. One drawback of using high-pass filters is that, although frequencies are cleaned up, you end up having a more ‘spiky’ signal, which eats up some of your dynamic range – but this is where nonlinear filters can help, as they keep these peaks down.

“It’s probably easiest to consider a square wave as the worst-case example. If you flick back to our opening diagram (on p27), you can clearly see that the high-pass peak level is around twice that of the input signal! A lot of the time, a high-pass filter is used to clean up extraneous low-frequency noise from a recording – such as unwanted bumping the body of the guitar, rumble from wind, or some other content below that of the notes being played. In this case, there isn’t anything further to do, because the spikes won’t happen.

“However, if you’re cutting into the fundamental frequency of the instrument with the filter, and this is required for the mix, then this will start boosting the peak level of the sudden changes. So, back to the worst case situation of the square wave: usually this isn’t going to chew up much dynamic range because the perceived loudness of a square wave will mean the volume of the track will be so low that it won’t cause any issues. Also. the placement of the individual spikes will not align with any other transients, so these won’t add together to push us over 0dBFS.

“Yet if you do need to keep these sorts of peaks down, the only option is to do some non-linear processing to tame them, either from a fast-acting compressor/limiter or a nonlinear low-pass filter. This will add distortion of some kind. In our example of the square wave, this isn’t an issue, since the entire square wave is one big distorted shape anyway, so adding more distortion won’t significantly alter the existing harmonics; it will just subtly change the amplitude of a few of them – though if we’re doing digital processing, we do still need to worry about aliasing.

“If your sound isn’t as bright, you won’t have these spikes jumping out, but you may still have large transients on the attack portion of the instrument. The good news here is that large transients already add a big splash of frequency to the mix, so distorting them a bit won’t make too much difference, as long as there isn’t too much harmonic content riding on top of the transient – which there won’t be for most monophonic sources.”

So surely we can minimise the impact of filtering individual signals by instead grouping them together and filtering the collective signals in one go? Perhaps not, as Andrew explains: “If there are multiple signals added together, then we could possibly be adding different sets of harmonics together, or having harmonic content riding on top of a large transient. If we high-pass this summed signal, and then use non-linearities to tame the peaks, we will end up with lots of inharmonic distortion. This will contribute noise to the mix, even with all-analogue processing, so it should be kept to a minimum.”

Andrew’s solution? “I recommend doing any high-pass filtering as early as possible, then using a fast compressor or limiter to tame the peaks of those signals that you need to be loud. With this workflow, you have a tidy signal that will fit into the mix with minimal processing required down the line – perhaps some gentle compression or EQ depending on the context. If you have two overlapping signals, and you don’t want to lose the low frequencies from either of them, then the best thing to use is sidechain compression, so only one of them is being heard at a time. This preserves the dynamic range and keeps things from getting muddy, and is always something to keep in mind if high-pass filtering isn’t working.”

If your sound isn’t as bright, you won’t have these spikes jumping out