An optical communication system transmitter consists of a digital-to-analog converter (DAC), a driver amplifier and a Mach–Zehnder-Modulator. The deployment of higher modulation formats (> 4QAM) or higher Baud rates (> 32 GBaud) diminishes the system performance due to linear and non-linear transmitter effects. These effects can be categorised in linear distortions due to DAC bandwidth limitation and transmitter I/Q skew as well as non-linear effects caused by gain saturation in the driver amplifier and the Mach–Zehnder modulator. Digital predistortion counteracts the degrading effects and enables Baud rates up to 56 GBaud and modulation formats like 64QAM and 128QAM with the commercially available components. The transmitter digital signal processor performs digital predistortion on the input signals using the inverse transmitter model before uploading the samples to the DAC.
Older digital predistortion methods only addressed linear effects. Recent publications also compensated for non-linear distortions. Berenguer et al models the Mach–Zehnder modulator as an independent Wiener system and the DAC and the driver amplifier are modelled by a truncated, time-invariant Volterra series.[15] Khanna et al used a memory polynomial to model the transmitter components jointly.[16] In both approaches the Volterra series or the memory polynomial coefficients are found using Indirect-learning architecture. Duthel et al records for each branch of the Mach-Zehnder modulator several signals at different polarity and phases. The signals are used to calculate the optical field. Cross-correlating in-phase and quadrature fields identifies the timing skew. The frequency response and the non-linear effects are determined by the indirect-learning architecture.[17]
