Communications over bursty optical channels

Our modern society depends entirely on global networks for high-capacity data communications, in which almost all data is carried by optical fibers.

Traditionally, the fiber links, including both hardware and software, are carefully optimized to transmit bits as reliably as possible.

The standard performance metrics of bit reliability are bit error rate (BER) and more recently generalized mutual information.

Furthermore, the fiber link, which has a long memory due to the physical properties of lightwave propagation in combination with signal processing algorithms in transmitters and receivers, is usually made virtually memoryless by interleaving.We claim that these standard metrics are partly obsolete and that interleaving is a waste of resources.

For the end users in communication networks, the quality of service (QoS) is governed not by the BER but by the packet loss rate (PLR). This is because standard network protocols include mechanisms for error detection and retransmission on a packet level.

If bit errors are not independently distributed in time but occur in bursts, then there is a big difference between systems optimized for BER and PLR.

By introducing a network awareness in the physical layer design, the project aims to (1) mathematically model how the fiber channel memory creates error bursts, (2) optimize coding and shaping schemes based on the derived models, and (3) evaluate the QoS for network users. The project will lead to energy savings and higher QoS.