Researchers from the Laboratory of Photonics Technologies and Machine Learning for Sensor Systems at the Novosibirsk State University Physics Department, in collaboration with researchers from the Skolkovo Institute of Science and Technology, have created a self-tuning fiber laser source for ultrashort pulses. This work was conducted on the basis of the Russian Science Foundation grant “Nonlinear photonics for optical communications and laser applications” in the “Optical Communications and Laser Technologies” category. A publication on the results will appear in the Nanophotonics scientific journal in the near future.
Evgeniy Kuprikov, Junior Researcher at the Laboratory of Photonics and Machine Learning for Sensor Systems, explained,
Mode-locked fiber lasers are nonlinear optical systems that generate ultrashort pulses with a high repetition rate. Depending on the strategy for setting the resonator parameters, different pulse modes can exist for the same parameters. The setup for these systems is technology that is difficult to automate, which is why many of these laser sources can only exist in a laboratory. However, humans can learn to tune mode-locked lasers. The setup process is somewhat reminiscent of a game in which instrument handles act as a joystick and to win you need to get the desired generation mode. Our plan is to replace a person in this process with an intelligent agent that, through trial and error, can independently find a winning configuration strategy and use it in the future. To do this, we use a deep reinforcement learning approach. This is a kind of interdisciplinary research that combines two different areas - laser systems and artificial intelligence algorithms.
In the literature these systems are called “smart lasers.” They can be configured without human intervention but they require a preliminary setup, which is called training.
Researchers at the Skolkovo Institute of Science and Technology’s Nanomaterials Laboratory have developed and assembled a new fiber laser with a saturable absorber based on an ion cell made of carbon nanotubes. It turned out that by applying voltage to a nanotube cell it is possible to control the characteristics of the saturable absorber. This provides an additional degree of freedom controlling the laser. It was demonstrated that in such laser modes, harmonic mode locking can be obtained in which the pulse repetition rate in the mode increases by a multiple. However, to obtain high-order harmonic mode-locking modes, human intervention is required to ensure laser tuning. At this stage, specialists from the NSU Physics Dept. Laboratory joined the research remotely and began developing an algorithm for tuning the fiber laser. The installation itself was located at the Skolkovo Institute of Science and Technology in Moscow.
Kuprikov continued,
Our Laboratory proposed using the Soft Actor-Critic reinforcement learning algorithm to solve the problem. After automation and preliminary preparation, the training process took one day. During this time, the agent learned to independently launch a laser without human intervention and find modes with a high order of harmonic synchronization. At the same time, the laser adjustment occurs by itself much faster than with human participation. In addition, the agent was able to learn to obtain harmonic mode synchronization modes of the 11th order, while a person managed to achieve only the 9th order.
Pulsed smart lasers can be used in metalworking, communications systems, and high-tech medicine, as well as in engineering and scientific research. Laboratory staff are focused on creating laser systems that would work flawlessly, be easy to use, and do not require human adjustment.
