Advances, Systems and Applications
DQN-based joint channel power allocation algorithm |
---|
1 Initialize scene parameters and algorithm parameters |
2 Obtain channel assignment information, user distribution information, and new user service information |
3 for episode = 1:max_ episode |
4 Initialize state space st |
5 State Reconfiguration \({s}_{t}^{*}\) |
6 for t = 1,2,3……,T -1 |
7 Select action by \(\varepsilon -{\text{greedy}}\) algorithm |
8 Execute the action \({a}_{t}\), get the reward value\({r}_{t}\), and observe the next state \({s}_{t+1}\) |
9 Reconstruct \({s}_{t+1}\) as \({s}_{t+1}^{*}\) and put experience data \(\left({s}_{t}^{*}, {a}_{t},{r}_{t},{s}_{t+1}^{*}\right)\) into the replay experience pool |
10 Randomly selected sample data from the replay experience pool |
11 Calculate the error function |
12 Updating Q-network parameters using gradient descent \(\omega\) |
13 Update the target Q network parameters \({\omega }^{-}\) |
14 end |
15 end |
16 Get deep reinforcement learning network parameters |
17 Output the channel and power assigned to each new user |