This is a project designed to test if using AI for dynamic compression algorithm selection is possible. It uses either a ZMQ emulated channel or a UHD USRP B210 to send video data over a channel. Channel metrics are then fed to a deep reinforcement learning model, which selects the compression type.
action state:
- bfp
- zstd
- lz7
- no compression
reward state:
- total packet throughput
- compression ratio
- overall latency
input state:
- channel quality
- antenna gain
- frequency
- distance between radios?
install deps:
sudo apt install -y libopencv-dev libuhd-dev uhd-host libzmq-devbuild:
mkdir build
cd build
cmake ..
make -j$(nproc)