Computation offloading method and device for integrated sensing and communication
A computation offloading method includes: establishing an associated model of a terminal for computation offloading; training the associated model by taking a to-be-computed task of the terminal, an uplink communication channel gain, a sensing pulse response and an angle difference between a communication beam and a sensing beam as input, to obtain an offloading parameter of the terminal for the to-be-computed task, wherein the to-be-computed task comprises to-be-computed communication data and to-be-computed sensing data, and the offloading parameter comprises a decision for offloading a computing task and a decision for offloading radio frequency transmission power; and offloading the to-be-computed task to an edge side according to the offloading parameter.
1 . A computation offloading method for integrated sensing and communication, comprising:
establishing an associated model of a terminal for computation offloading;
training the associated model by taking a to-be-computed task of the terminal, an uplink communication channel gain, a sensing pulse response and an angle difference between a communication beam and a sensing beam as input, to obtain an offloading parameter of the terminal for the to-be-computed task, wherein the to-be-computed task comprises to-be-computed communication data and to-be-computed sensing data, and the offloading parameter comprises a decision for offloading a computing task and a decision for offloading radio frequency transmission power; and
offloading the to-be-computed task to an edge side according to the offloading parameter,
wherein the associated model comprises:
an inter beam interference model, a sensing model, an uplink communication model of the terminal, a task model of the to-be-computed task, a computing model of the terminal, and a computing model of the edge side, and
wherein:
the associated model also includes a combined optimization model;
the combined optimization model is an optimization model for processing a time delay of the to-be-computed task and a sensing performance of the terminal; and
the combined optimization model includes: a constraint on a total delay of operating the to-be-computed task, a constraint on the sensing performance of the terminal, a constraint on the radio frequency transmission power of the communication beam of the terminal, a constraint on the radio frequency transmission power of the sensing beam of the terminal, a constraint on a relationship between the sensing beam and the communication beam, and a constraint on the offloading parameter.
2 . The computation offloading method for integrated sensing and communication according to claim 1 , wherein the inter beam interference model is a sector antenna model for characterizing a beam interference gain between the communication beam and the sensing beam.
3 . The computation offloading method for integrated sensing and communication according to claim 1 , wherein the sensing model of the terminal comprises:
a model for an orthogonal frequency division multiplexing sensing signal transmitted by the terminal and a model for an echo signal received by the edge side, and an output of the sensing model is conditional mutual information between a target pulse response and the received signal.
4 . The computation offloading method for integrated sensing and communication according to claim 1 , wherein the uplink communication model of the terminal is used to characterize an uplink transmission rate and the uplink communication channel gain between the terminal and the edge side.
5 . The computation offloading method for integrated sensing and communication according to claim 1 , wherein the task model is used to characterize a number of to-be-computed tasks within a preset time period, a size of each to-be-computed task, a number of CPU cycles required to execute one bit of the to-be-computed task, and a time delay threshold of each to-be-computed task.
6 . The computation offloading method for integrated sensing and communication according to claim 1 , wherein the computing model of the terminal is used to characterize a time delay of the terminal for processing the to-be-computed task and the sensing performance of the terminal.
7 . The computation offloading method for integrated sensing and communication according to claim 1 , wherein the computing model of the edge side is used to characterize a time delay of the edge side for processing the to-be-computed task and the sensing performance of the terminal;
wherein the time delay of the edge side for processing the to-be-computed task includes: a first time delay for transmitting the to-be-computed task from the terminal to the edge side, a second time delay for the edge side processing the to-be-computed task, and a third time delay for the edge side transmitting the processed to-be-computed task to the terminal.
8 . The computation offloading method for integrated sensing and communication according to claim 1 , wherein training the associated model to obtain an offloading parameter of the terminal for the to-be-computed task comprises:
taking the to-be-computed task of the terminal, the uplink communication channel gain, the sensing pulse response and the angle difference between the communication beam and the sensing beam as a state space, taking the decision for offloading the computing task and the decision for offloading the radio frequency transmission power as an action space, and establishing a reward function to obtain the offloading parameter;
taking the to-be-computed task of the terminal, the uplink communication channel gain, the sensing pulse response, the angle difference between the communication beam and the sensing beam and the reward function as input, through a reinforcement learning module based on Multi-DQN, to obtain the decision for computation offloading in a task offloading strategy; and
taking the to-be-computed task of the terminal, the uplink communication channel gain, the sensing pulse response, the angle difference between the communication beam and the sensing beam and the reward function as input, through a reinforcement learning module based on TD3, to obtain the decision for radio frequency transmission power in the task offloading strategy.