Optical links have many advantages
BlogJanuary 11, 2018
Rodger Hosking
Mercury Systems
Advances in optical interface technology boost performance levels to help meet increasing data rates and signal bandwidths. New specifications define how to deploy these optical links within open industry standards, affording improved interoperability and supporting future upgrades. Offering many advantages over traditional copper connections, optical links will boost data rates, improve signal integrity and security, and greatly extend distance between system components.
Optical links offer many benefits
One major shortcoming of copper cable is signal loss, which becomes a serious limitation for higher frequency signals and longer cable lengths. Across a span of 100 meters, optical cables can sustain data rates as much as 100 times higher than copper cable.
Because copper cables radiate electromagnetic energy, eavesdropping on network cables is a major security concern, not only for military and government customers, but also for corporations, banks, and financial institutions. Advanced signal sniffers in vehicles and briefcases are hard to detect and restrict. Optical cables are extremely difficult to "tap" without damaging the cable, resulting in immediate detection.
Signals flowing in copper cables are also susceptible to contamination from nearby sources of electromagnetic radiation, such as antennas, generators, and motors. This is critical for military and commercial aircraft and ships, as well as manned or unmanned vehicles, which are often packed with dozens of different electronic payloads. Optical cables are completely immune to EMI [electromagnetic interference] and even lightning discharges.
Physically, optical cables are much smaller and lighter than copper cables, especially important for weight-sensitive applications such as weapons, unmanned vehicles, and aircraft. Optical cables will operate just as well when submerged in seawater, and are completely immune to electrical shorting—especially important where explosive vapors may be present. To ease installation through conduits and passages, optical cables have smaller diameters and can withstand up to ten times more pulling tension than copper cables.
Driven by huge commercial markets for data servers, storage networks, telecom systems, and home and office internet and entertainment systems, optical interfaces are replacing older copper connections for good reasons: cost and performance.
As the use of optical cables becomes more widespread, the cost per length can be much lower than copper cables that depend on commodity metal pricing. As is often the case, industrial, military and government embedded systems are now taking advantage of the many benefits of this rapidly advancing commercial technology.
Optical cables
An optical cable is a waveguide for propagating light through an optical fibre. It consists of a central core clad with a dielectric material having a higher index of refraction than the core to ensure total internal reflection. Optical cables use either multi-mode or single-mode transmission.
Multi-mode cables accept light rays entering the core within a certain angle of the axis. They travel down the cable by repeatedly reflecting off the dielectric boundary between the core and the cladding. The core diameters are typically 50 or 62.5 microns, and the wavelength of light is typically 850 nm.
Single-mode cables propagate light as an electromagnetic wave operating in a single transverse mode straight down the fibre using typical wavelengths of 1310 and 1550 nm. The core diameter must be no greater than ten times the light wavelength, typically 8 to 10 microns. Although single-mode cables can carry signals over lengths 10 to 100 times longer than multimode, the transceivers are more expensive.
Hundreds of different types of optical cable connectors exist, each addressing specific applications and environments. The challenge is connecting the ends of two optical cables to retain the maximum fidelity of the light interface, in spite of human factors, tolerances, contamination, and environments. Special tools and kits for cleaning the ends of each optical fibre are essential for reliable operation.
Optical transceivers
Coupling electrical signals to light signals for transmission through optical cables requires optical transceivers. Most systems require full duplex operation for each optical link to support flow control and error correction. A pair of optical fibers, often bonded together in the same cable, supports transmit and receive data flowing in opposite directions.
Although several analog light modulation schemes (including AM and FM) have been used in the past, now almost all transceivers use digital modulation. Optical emitters simply translate the digital logic levels into on/off modulation of the laser light beam, while the detectors convert the modulated light back into digital signals. This physical layer interface for transporting ones and zeroes can support any protocol.
The latest transceivers use laser emitters to support data rates to 100 Gbits/sec and higher, and each generation steadily reduces the power, size and cost of devices. Different technologies are required for emitters and detectors, but both are often combined in a single product to provide full-duplex operation.
Optical transceivers thus provide a physical layer interface between optical cables and the vast array of electrical multi-gigabit serial ports found on processors, FPGAs, and network adapters. As a result, optical transceivers are transparent to the protocols they support, making them appropriate for a virtually any highspeed serial digital link.
Electrical signals of the optical transceivers connect to the end point device, which must then handle clock encoding and recovery, synchronization, and line balance at the physical layer. Data link layer circuitry establishes framing so that data words can be sent and received across the channel.
Choosing the right optical protocol
Protocols define the rules and features supported by each type of system link, ranging from simple transmission of raw data to sophisticated multi-processor support for distributed networks, intelligent routing, and robust error correction. Of course, heavier protocols invariably mean less efficient data transfers and increased latency. Generally, it is best to use the simplest protocol that satisfies the given system requirement.
As an example of a lightweight protocol, Aurora for Xilinx FPGAs features onboard link-layer engines and high-speed serial transceivers. Aurora is intended primarily for point-to-point connectivity for sending data between two FPGAs. It includes 8b/10b or 64b/66b channel coding to balance the transmission channel, and supports single- or full-duplex operation. Aurora handles virtually any word length and enables multiple serial lanes to be bonded into a single logical channel, aggregating single lane bit rates for higher data throughput. Data rates for each serial lane can be 12.5 Gbits/sec or higher. Extremely simple and with minimal overhead, Aurora is very efficient in linking data streams between multiple FPGAs within a module, or between modules across a backplane.
Stepping up in complexity is the Serial FPDP protocol defined under VITA 17.1. It addresses several important needs of embedded systems, including flow control to avoid data overruns, and copy mode to allow one node to receive data and also forward it on to another node. The copy/loop mode supports a ring of multiple nodes eventually completing a closed loop. The nominal data rate on each lane is 2.5 Gb/s, but advances in device technology now support rates almost twice that speed.
Infiniband defines a flexible, lowlatency, point-to-point interconnect fabric for data storage and servers with current rates of 14 Gbits/sec, moving up to 50 Gbits/sec in the next few years. Channel speeds can be boosted by forming logical channels by bonding 4 or 12 lanes.
The venerable Ethernet protocol still dominates computer networks, with 10 GbE now commonly supported by a vast range of computers, switches, and adapters. Even though Ethernet suffers from high overhead, making it somewhat cumbersome for high-data-rate, low latency applications, its ubiquitous presence virtually assures compatibility.
