T
Twiki
Text
Spiking-Neural-Network Deep Reinforcement Learning for Energy-Efficient ISAC in V2X Networks
A roadside unit models a dynamic V2X network as a Markov decision process and, using only its current sensing information rather than pilot signals or CSI acquisition, jointly optimizes integrated-sensing-and-communication beamforming and power allocation with a deep reinforcement learning agent whose policy network is a spiking neural network, whose event-driven sparse spikes cut energy consumption while sustaining communication rate and sensing accuracy.