Fiber optic systems inevitably require splicing. Sometimes a splice may be necessary because the span is longer than any single available cable; other times the splice will be necessary because shorter runs are more convenient to install. Sometimes a splice will be necessary because no one told the backhoe operator that he might want to try ten feet to the left. Whatever the reason for splicing, all splices are judged by the same criteria:
Acceptable loss depends upon the function of the link. Low cost systems may tolerate a loss of 1 to 3 dB, while a LAN should aim for 0.3 to 0.75 dB. Long haul telecommunications systems generally want 0.2 dB or less, because the cumulative splice loss (the sum of all splice losses in the system) can make costly repeaters and amplifiers necessary.
One major source of splice loss is the differences between the two fibers. Although optical fiber is made to very precise specifications, there are still very slight differences between fibers. When the numerical aperture (NA) of the transmitting fiber is larger than that of the receiving fiber, loss occurs. When the core diameter of the transmitting fiber is larger, loss also occurs. If the cores are not perfectly circular, loss will occur unless they are perfectly lined up. Even the cladding (through which the light is not intended to travel) makes a difference: differences in the cladding can cause the cores to be misaligned. Fiber manufacturers have made much progress in reducing these problems. Fiber produced to a 5 micron tolerance has a maximum loss of 0.6 dB; newer manufacturing techniques allow tolerances of 1 (0.1 dB) to 2 (0.28 dB) microns.
The other source of loss is the splice itself. Lateral and angular displacement allow light to escape the core. End separation allows Fresnel reflection to disturb the path of the light. In air, this loss is 0.34 dB. Fresnel loss is combated by index-matching fluid that maintains the core's index of refraction across the gap. The air gap also allows the high order modes of multimode fiber to escape through the gap in the cladding. The final source of splice loss is the surfaces of the two ends. Any imperfections in the ends of the fiber will disrupt the light, so the end must be made as smooth as possible.
Loss from NA-mismatch: loss = 10log(NAr/NAt)2
Loss from diameter-mismatch: loss = 10log(dr/dt)2
Types of Splices
There are two main types of splice: mechanical and fusion. Mechanical splices are cheap and easy to do, but are lossy. Fusion splices have ridiculously low losses, but require expensive equipment.
Mechanical splices are simply braces that hold the ends of two fibers together. The technician inserts the two fiber ends (which should have been stripped, cleaned, and cleaved), aligns them, and crimps the splice. This type of splice often requires index-matching gel to eliminate air gaps, and has a loss of around 0.1 to 0.2 dB. Many mechanical splices are reusable, as long as there is still gel in the splice.
Fusion splices require the technician to place the (stripped, cleaned, and cleaved) fiber ends in the splice machine, align them using the microscopic display, and activate the device. The splice machine uses a high-voltage electric arc to weld the glass together, creating a splice so good that when done properly, even the microscope doesn't show that there was ever a cut. The loss on a fusion splice can be as low as 0.05 dB.
When cleaving the fiber for a splice, care must be taken to ensure that the cleave is not more than 2 degrees away from being perpendicular with the main axis of the fiber, and that the end is free of defects. The most common way to achieve this in a splice is the scribe and break method. The fiber is held (usually using a groove in a specialized cleaving device), and nicked with a hard, sharp blade (diamond, sapphire, tungsten carbide, etc.). The fiber is then pulled, breaking the glass cleanly. When done properly, the surface is smooth and mirror-like; when done improperly, the surface may be rough (hackled) and may have a protrusion (lip).
Node your homework
Thank to BlackPawn, spiregrain, PaynenDiaz, and rootbeer277 for corrections and questions.