search

baturaysaglam / RIS-MISO-Deep-Reinforcement-Learning

Joint Transmit Beamforming and Phase Shifts Design with Deep Reinforcement Learning
MIT License
130 stars 37 forks source link

I have the doubt that i can't re current the results similar to the papers. #5

Closed Glen9010 closed 1 year ago

Glen9010 commented 1 year ago

I have downloaded this projects and run it. I found that in the first eps , the results can reach 9 or higher. But in the later eps, the result was stuck in about 1 to 4, this question puzzled me for a long time.

Glen9010 commented 1 year ago

if the agent is deployed in the real world and the channels is fast time-varying stochastic process， does it still need train for a long time maybe one or more episode ? Or this agent is designed for the static channels？

baturaysaglam commented 1 year ago

check these issues: issue 1 issue 3. I gave the corresponding answer previously.

baturaysaglam commented 1 year ago

to answer the question regarding channel representation, I suggest you read the paper. I'm not the author of the paper. but all I can say is if RL is used for learning, then the environment (channel) is assumed to be represented by a Markov Decision Process, which is a discrete-time stochastic control process.