Bias Control for Mach-Zehnder Modulators: A Gradient-Based Optimization Approach

In this paper, we explore the use of gradient based
optimization algorithms for automated bias control in Mach-
Zehnder Modulators (MZM). We present and demonstrate, experimentally,
five gradient descent algorithms; Stochastic Gradient
Descent (SGD), Stochastic Gradient Descent with Momentum
(SGD+M), Adagrad, RMSProp, and Adam, applied to the bias
control problem in MZMs. We present a method of creating an
error signal from the measured output of an MZM with a low
frequency pilot tone, and provide a detailed explanation of how
each algorithm is used to both identify the set bias condition
and track the bias condition in the presence of disturbances.
Our implementation is capable of identifying and holding the
null condition and the quadrature condition. We evaluate the
bias point identification for each algorithm by measuring and
analysing the step response for each method. We test the bias
tracking of each algorithm using three forms of disturbance;
RF power disturbances, temperature disturbance, and long-term
bias drift. All tests were conducted at 20GHz. To the best of
our knowledge, this is the first investigation into the application
on gradient based learning approaches for MZM bias control.
This work has great importance on future bias control design
and implementations for telecommunications, the space sector,
Microwave Photonics (MWP), and defence.