Clear glass can still lose nearly all light over a long path
A small piece of glass can look clear and still lose nearly all light over a long path. Kao and Hockham’s 1966 work studied two things: the modes a waveguide supports, and whether material loss could be cut enough for useful communication. The key question was which losses were fundamental and which came from impurities that could be removed.
In the lecture passage we read, Kao describes a target of about 20 decibels per kilometre near 0.6 micrometres, and the role of iron contamination. The paper argued that glass could be improved to reach that target. It did not report that the 1966 apparatus had already achieved a modern low-loss network.
Sources: [1]
Decibel loss adds up with distance
When the power loss per unit length is constant in decibels, transmitted power falls off in powers of ten. A loss of 20 decibels over one kilometre leaves one percent of the launched power, before any connector or coupling loss. Two kilometres at the same rate leave one ten-thousandth.
These are arithmetic examples, not measurements of any particular fibre. Whether a link works also depends on how well light is launched and how sensitive the detector is. If the input coupling differs between two measurements, coupling loss can be mistaken for propagation loss.
Sources: [1]
Guiding light and keeping pulses sharp are separate problems
A core with a higher refractive index than the cladding around it can support confined modes. A ray that stays inside the core shows the geometric limit of guiding. It does not measure attenuation or guarantee a single mode; the waveguide dimensions and the wavelength decide which modes propagate.
Even with low loss, different modes or frequencies can arrive at different times and spread out a data pulse. Kao and Hockham considered mode properties and dispersion as well as impurities. Checking only whether a ray escapes leaves the communication problem unsolved.
Sources: [1]
A graded core lets steep rays catch up
Guiding the light still leaves a timing problem. In a core with one uniform index, a ray that zigzags steeply covers more glass than one running straight down the axis, so the parts of a short pulse carried by different modes arrive at different times. For a core of index about 1.5 that sits 2 % above the cladding, the spread is about 100 nanoseconds per kilometre.
In 1973 Gloge and Marcatili worked out what happens when the index instead falls gradually from the axis to the edge. Steep rays then spend much of their path in the outer glass, where the index is lower and light is faster, and nearly catch up with the axial ray. For a parabolic profile their theory cuts the spread from 2 % of the travel time to 0.02 %, about 1 nanosecond per kilometre in the same fibre. A profile slightly different from a parabola brings it to 0.25 nanoseconds per kilometre, and they warned that an error of one part in ten thousand in the index more than doubles that best value, to 0.53.
The gradient-index Lab shows the same effect with rays in a flat slab, whose index step is much larger than a fibre’s so that the bending is easy to see. In its parabolic slab the optical paths across the end face differ by about 5 % of their mean, against 18 % in a uniform core of the same peak index. A slab whose index follows a hyperbolic secant brings every ray in the plane to the end face together.
Sources: [3]
Later materials work turned the idea into working systems
The 2009 prize recognized achievements that made light transmission in fibres usable for communication. Later progress in glass purification, fabrication, light sources and detectors turned the idea into working systems. Their performance should not be read back into the original paper.
Our source is Kao’s own retrospective passage, and the full lecture was not accessible. We have not digitized an original 1966 attenuation curve or any later measurement of production fibre.
Try both fibre models
Use the guiding model to see which paths or modes stay confined, and the pulse model to see how a signal spreads in time. Neither replaces a calibrated loss measurement. Together they show why transparency alone was not enough.
Sources: [1]