For ever more efficient optical fibres

EPFL researchers have taught a type of machine learning algorithm to reconstruct images that became blurred while being transmitted through an optical fiber. The work could increase the amount of information transmitted through telecommunications networks, improve endoscopic imaging used in medical diagnosis and enhance the capacity and quality of optical fibers.

“We use modern deep neural network architectures to retrieve the input images from the scrambled output of the fiber,” said Demetri Psaltis, the head of EPFL’s Optics Laboratory, who led the research in collaboration with colleague Christophe Moser from the Laboratory of Applied Photonics Devices. “We demonstrate that this is possible even for fibers 1 kilometer long,” he added, calling the work an important milestone. Their research has now been published in the journal Optica.

Deciphering the blur

Optical fibers have long been used to transmit information with light. Multimode fibers have much greater information-carrying capacity than single-mode fibers. Their many channels – known as spatial modes because they have different spatial shapes – can transmit different streams of information simultaneously.

While these fibers are well suited for carrying light-based signals, they have not successfully been used to transmit images over long distances before. This is because images travel through all the channels, and what emerges at the other end is a pattern of speckles that the human eye cannot decode.

To tackle this problem, Dr. Psaltis and his team turned to a deep neural network, a type of machine-learning algorithm. Deep neural networks can give computers the ability to identify objects in photographs. They have also helped improve Google’s speech recognition systems, for example. The design of these algorithms is inspired by the way neurons transmit information in the human brain. Input is processed through several “hidden layers” of artificial neurons, each of which performs a small calculation and passes the result on to neurons in the next layer.

Our brains develop mental models for objects by being exposed to many different examples, so that, for example, when encountering a new type of tree we are able to recognize it as a tree instead of a telephone pole or a bush. Similarly, when a deep neural network is exposed to a large enough set of training data, the machine learns to identify the input by recognizing the associated patterns of output.

Caption: A speckle pattern from an image transmitted through a multimode fiber passes through the hidden layers of a deep neural network and is reproduced as the number 3. / Demetri Psaltis, EPFL

A simpler method

Navid Borhani, a scientist who participated in the research, says this machine learning method is much simpler than other approaches to reconstructing images that have been transmitted through optical fibers, which require making a holographic measurement of the output. The deep neural network was able to cope with distortions caused by environmental disturbances as the signal passed through the fiber. Random fluctuations in temperature along the length of the fiber together with movements caused by air currents can add noise to the image – and the noise worsens the farther the signal has to travel.

“The remarkable ability of deep neural networks to retrieve information transmitted through multimode fibers is expected to benefit medical procedures like endoscopy and communications applications,” according to Dr. Psaltis. Telecommunication signals often have to travel through many kilometers of fiber and can suffer distortions, which this method could correct. Doctors could use ultrathin fiber probes to collect images of the tracts and arteries inside the human body without needing complex holographic recorders or worrying about movement.

“Slight movements because of breathing or circulation can distort the images transmitted through a multimode fiber,” adds Dr. Psaltis. “The deep neural networks are a promising solution for dealing with that noise.”

Dr. Psaltis and his team plan to try the technique with biological tissue samples. They hope to conduct a series of studies using different categories of images, to figure out all the possibilities and limits of their technique.