Just guessing, but I think you could implement 16 variable delay lines, with delayed outputs simply summed to produce the on-beam signal. An X-Y joystick could be read (say every 0.1 sec), the results subjected to a bit of trigonometry to produce the individual delay values (0..100%) for each delay line. A centered joystick should produce 50% of max delay on all delay lines.
Max delay would need to be roughly equal to the maximum time difference for a sound travelling across the array from the side (diagonally, worst case). The leading edge mike would need to be delayed by max, the trailing edge mike delay = 0. The other mikes would have intermediate delays. If you were happy with a fixed beam, and mechanical rotation of the mike array, just connect all the electret mikes in series to get the highest possible signal for on-axis sounds. Almost free gain. No DSP. You may need load-balancing resistors across each mike. No need for a load terminating Z if the resistors are well chosen. Some interesting quirks arise with the system. Directionality drops off (main lobe widens) for low frequencies. If you want bass response, the array needs to be wide. This is a function of wavelength. However, as the spacing between the mikes becomes large, you get sidelobes on the polar sensitivity plot creeping in from the high frequency end. Possibly some non-linear spacing rule might reduce the mike count, allowing for a sparse array. Of course if you go the DSP route for smart steerability, the computation would be horrendous for anything but a linear array. Any thoughts? Jim Adamthwaite
