In this paper we introduce an auto-encoder neural network that can be used to model a node-to-node radio frequency communication channel. With some simple assumptions the encoding layer of this neural network can be interpreted as a phase modulated signal which can be transmitted across a communication channel. When the auto-encoder is trained to minimize bit error rates within the transmitted messages, it effectively learns an optimal modulation scheme for that particular communication channel. To implement these signals in a physical link we developed GNURadio flowcharts along with custom signal processing blocks that transmit and receive the auto-encoder generated signals between a pair of Universal Serial Radio Peripherals. Finally, we show the methodology used for incorporating the GNURadio flowcharts with the auto-encoder training program so that the autoencoder can be trained on the physical communication channel itself.