Unmanned Aerial Vehicles Control Commands Compression and Aggregation in Bandwidth-Limited Channels

Unmanned aerial vehicle first-person view control requires a control command transmission from an external pilot station to unmanned aerial vehicles. In order to control an unmanned aerial vehicle with telemetry and manage transmission protocols with protocols, such as CRSF and MavLink, via packet networks, it is necessary to ensure that bandwidth exceeds 50 Kbit/s. Therefore, control of a large number of unmanned aerial vehicles in remote areas, which are out of last generations’ mobile networks coverage, is impossible without reducing the bandwidth requirement. The purpose of the present research is to discover a possible compression ratio of control commands with their aggregation, error-correcting coding, and subsequent encoding with a neural network. The errorcorrecting codes examined are Reed-Solomon codes. The custom autoencoder neural network architecture is developed. The result of the work is a new control commands compression method based on command aggregation, error-correcting coding, and encoding with a custom autoencoder model. The specific feature of the method is that in order to correct control command errors after encoding and decoding with the neural network model, the error-correcting codes are used. The method can be applied to control commands protocols for unmanned aerial vehicles. Neural codec of control commands, based on this method, and a neural network for aggregated control commands compression are described. The developed method ensures a 18,75% compression ratio for control commands of the CRSF protocol. Hence, it also reduces the bandwidth required for FPV-control. The results can be extended for other control protocols.