A system for training a linearization compensation model includes a tone generator for providing at least two different RF tones, receiver path components for processing the RF tones, an analog-to-digital converter for converting the processed RF tones into digital signals, and a processor for using the digital signals to generate the linearization error compensation model. The resulting compensation model is particularly useful in a linearization system which includes a receiver for measuring a signal, an electro-optical modulator configured for converting the measured signal to an optical signal, an optical-electrical detector configured for converting the optical signal to an analog electrical signal, an analog-to-digital converter for converting the analog electrical signal into a digital signal with the processor being used for removing linearization errors from the digital signal.