Real Time FPGA-Based CNN Training and Recognition of Signals

Training a machine learning model requires many resources and a fast, efficient processing system. This work proposes an FPGA based machine learning model for fast and efficient signal recognition, allowing for a mobile application of a training model. This is composed of several processes and steps. First is receiving the signal and running it against the current model. Second is applying several filtering methods, as well as noise, to change how the signal is represented. Finally, training the current model based on the data generated from the original signal, updating the list of recognized items. This is run on an FPGA using python on a Linux environment, utilizing a CPU based training algorithm.