Hi there!
Filter clipping is an interesting topic, and one with many layers to consider, most of which have already been addressed in a way or another in this thread.
First off, let me stress the fact that IIR is no different than FIR in this regard: implementing the same minimum-phase correction in IIR and FIR will produce the same output levels and potential overs. In fact the same goes for analog filters, albeit of course without the risk of clipping (but still, saturation can happen).
The first obvious thing to consider is of course the correction itself (ie filter response): ideally you don't want any EQ point to exceed the 0dB mark, as any 0dBFS signal at these frequencies would get clipped. In rePhase this can be checked directly on the magnitude curve with any loaded measurement bypassed, or with the
max response level value as
@Geert noted above. This can be considered good practice to handle this adjustment within the FIR (ie adjust the fader in the
General tab), but can also be done on the input, or anywhere around the convolution as long as the signal is still in floating point format.
Now, even in this case you can still get clipping when the filter is applied (and once again the same would occur with IIR), because any non linear-phase correction will change the timing at which the different frequency components of the signal are summed. An all-pass filter is a good potential example, but any EQ or filter will also be subject to that phenomenon. This cannot be triggered with a sweep (or only marginally so) as this is essentially a single frequency signal. You can indeed imagine complex test signals that will produce higher output levels, but such worst-case signals would be correction-dependent. Real-world signals will most probably produce less overs than you might fear in practice.
On top of all that, and as mentioned above, you also have to consider that keeping your output signal clear of any clipping is not a guaranty that it will not clip in the ASRC or when applying the reconstruction filter in the DAC. This can be checked by resampling your signal to some high sample rate (eg 384kHz or 768kHz) and check for clipping: you basically turn inter samples into real samples.
All in all you take the risk of clipping your signal when applying filters, but the good news is that when using digital volume control (again either on the input, or anywhere around your filtering porcess where the signal is still in floating point) you don't really have to address it directly with dedicated headroom. You can simply integrate it in your "global" headroom.
If you set your levels so that a 0dBFS signal at the output of your processor produces the maximum level your amp/speakers/ears/neighbors can handle, and if in this situation you still get an idle noise level low enough that it is not audible at your listening positions, then you are good to go. This calls for devices with a high enough dynamic range, and good level matching between them to maintain it throughout. Of course you also need a well-behaved system, ie no on/off noises, no bug susceptible of producing 0dBFS signals, no cabling issues, etc.
When approaching that max level you will increase the probability of hitting some corner cases and clip the signal during the filtering process, but in the same time your output stage, amp, speakers and ears will also see their distortion increase. This can all be factored-in and handled together, and the filtering process will most probably not be the worst offender there. In this scenario you don't really need a DAC with headroom for inter sample overs, and you can even use a DAC with rising distortion levels towards 0dBFS.
