SatSR is a mobile edge computing (MEC)-based super-resolution (SR)-enhanced adaptive video streaming system for mobile networks with satellite backhauls. We open sourced the chunk-level simulator and DRL-based SR Scale Factor adaptation Algorithm.

• Install prerequisites (tested with Ubuntu 18.04, Tensorflow v1.1.0, TFLearn v0.3.1 and Selenium v2.39.0)

• To train a new model, put satellite and RAN link training throughout traces in cooked_traces/sat and cooked_traces/user, and corresponding validing throughout traces in cooked_test_traces/sat and cooked_test_traces/user. The trace format for simulation is [time_stamp (sec), throughput (Mbit/sec)]. We have put the data sample into the corresponding folder.

• Make sure actual video chunk files are stored in video_server/video[1-6], then run

  python get_video_sizes
  

• Run this code to start training the SR scale factor adaptation algorithm.

python train_main.py

The reward signal and meta-setting of videos can be modified in train_main.py and train_env.py.

The training process can be monitored in sim/results/log_test (validation) and sim/results/log_central (training). Tensorboard (https://www.tensorflow.org/get_started/summaries_and_tensorboard) is also used to visualize the training process, which can be invoked by running

python -m tensorflow.tensorboard --logdir=./results/

where the plot can be viewed at localhost:6006 from a browser.

Trained model will be saved in results/.