What about focusing each LED source separately?
However, the diameter of each LED optic would need to at least equal the diameter of the single optic used for a point source light. The result would be an enormous searchlight with a very broad beam.
I can't find any solid information on the quantitative light output of the AN/TVS-3, but xenon arc lamps typically produce 30 lumens per watt. Therefore, the 20KW AN/TVS-3 should produce around 600,000 lumens.
Readily available high-power LEDs are capable of 1700 lumens, so equaling the AN/TVS-3 would take--as a very rough ballpark that does not take into account the different light distribution between LEDs and discharge lamps--350 LEDs and optics. The AN/TVS-3 uses 30" optics, so those optics would also need to be at least 30" in diameter.
350 30" optics is a lot more like a wall of light than a searchlight. Aiming that many separate optics would also be a nightmare.
https://en.wikipedia.org/wiki/Luminous_efficacy
What I'm not sure of is if the lumins/watt rating is given for the light output or the power input.
The input power rating is 450 Amps * 200 Volts * 1.732 = 155880 electrical watts. That's nearly 7.75 times as much electrical power as light output. That makes sense since the peak efficiency is about 7.3 for a xenon arc lamp.
I probably am missing something about optics though.
Bigger optics = less beam divergence = more useful range.
This is a function of the diffraction limit[0] and is a property of physics.
If you want to match the beam divergence of a 30" optical system, you have to use 30" optics; there are no shortcuts.
[0] https://www.edmundoptics.com/knowledge-center/application-no...
I can, however, imagine that a multi source system would require in-phase light sources to kind of mimic a phased-array receiver on the send side.