Status date:
The capabilities of autoencoders are well known and space researchers and enthusiasts have developed multiple systems that improve the performance of the state-of-the-art compressor for space applications, limiting the bandwidth usage and consequently reducing costs.
In the framework of this activity, IngeniArs improves the useful data-rate by optimising the data to be transmitted via Variation Autoencoder AI models.
Such models can reduce the amount of data to transfer, encoding the information into a set of points within the AI model latent space. These points are then restored into the original information (the compression is lossy).
To validate these models, different use-cases (telecommunication, video stream, image transmission) are tested by highlighting how these AI models can enhance both transmission (upstream) and reception (downstream) systems.
Finally, for the upstream segment, a feasibility study related to porting the developed models into GPU@SAT, a space-grade AI accelerator developed by IngeniArs, is conducted.