A practical issue that arises when an array is allowed to undersample the soundfield is level normalization. An extreme example is the amplitude ( ) of a point source when it touches a microphone/speaker point on the array, which is well out of bounds for any reasonable practical system.
Level regularization is another reason to keep sources a few wavelengths or more away from the microphone/speaker array.
When the array becomes undersampled, say because a wideband source is approaching an extreme stage angle, the array can be regarded as transitioning from soundfield reconstruction by summing interpolation kernels to simple panning between/among available speakers (i.e., what we're normally always doing). Note that the high-frequencies must go into ``panning mode'' first, while lower frequencies may remain adequately sampled. We can implement an adaptive spectral partition between sample-based reconstruction below and panning above.21
http://arxiv.org/abs/1911.07575
.