文章目錄
題目:
Reinforcement Learning for Decentralized Trajectory Design in Cellular UAV Networks With Sense-and-Send Protocol
Citation:
J. Hu, H. Zhang and L. Song, “Reinforcement Learning for Decentralized Trajectory Design in Cellular UAV Networks With Sense-and-Send Protocol,” in IEEE Internet of Things Journal, vol. 6, no. 4, pp. 6177-6189, Aug. 2019, doi: 10.1109/JIOT.2018.2876513.
文章的 ieee 鏈接:
https://ieeexplore.ieee.org/document/8494742
創新點:
In literature, most works focused on either the sensing or the transmission in the UAV networks, instead of considering UAV sensing and transmission jointly.
系統模型
其中每個 UAV 對應一個事先已知的 task , 不存在 UAV-user association 的問題
回傳的過程要求達到一個要求的信噪比,如果達到了,可以認爲再 1 個 time slot 就可以完成傳輸
sense and send cycle
the process is divided into cycles, which are indexed by k. In each cycle, each UAV senses its task and then sends the collected.
每個 cycle 的結構可直觀地反映在圖裏
beaconing :
- sensory data to the BS In the beaconing phase, each UAV sends its location to the BS in its beacon through the control channel.
- the BS then broadcasts to inform the UAVs of the general network settings as well as the locations of all the UAVs. : UAVs can obtain the locations of other UAVs in the beginning of each cycle.
Based on the acquired information, each UAV then decides its trajectory in the cycle and informs the BS by another beacon.
A. Sense-and-Send Cycle
beaconing結束後UAV勻速直線運動直至下一個cycle開始
transmission state :
B. Uplink Subchannel Allocation Mechanism
channel allocation state :
the BS allocates the C available uplink SCs to the UAVs with uplink requirements , in order to maximize the sum of successful transmission probabilities of the UAVs.
sense-and-send protocol analysis
Outer Markov Chain of UAV Sensing
the state transition takes place among different cycles . for each UAV,
it has two states in each cycle
Inner Markov Chain of UAV Transmission
符號提示
A. UAV Trajectory Design Problem
上面我們一直研究和分析的是每個cycle 內的具體細節,本節我們把cycle當成一個單元考慮
Single-Agent Q-Learning Algorithm
we first set up the model to describe the UAVs’ trajectories.
we consider the utility of each UAV to be the total number of successful valid sensory data transmissions for its task.
Multiagent Q-Learning Algorithm