Editor's note: We are posting the text of this article from the August 1977 issue of Scientific American for all our readers because the author has won the 2009 Nobel Prize in Physics.

Three months ago the Bell System began the, commercial evaluation of a light-wave communication system in which messages are coded into pulses of light transmitted through hairthin glass fibers. The new system carries voice. data and video signals over one and a half miles of underground cable interconnecting two switching offices of the Illinois Bell Telephone Company and a large commercial building in Chicago's business center. The light-guide cable. only half an inch in diameter. contains 24 fibers in two ribbons of 12 fibers each. The information capacity of each fiber is 44.7 megabits per second, meaning that the light source feeding into the fiber is turned on and off 44.7 million times per second. At this pulse rate a single fiber can carry 672 one-way voice signals; thus the 24 fibers have a capacity of 12 X 672, or 8,064. two-way conversations. To match this capacity with conventional pairs of copper wires would require a cable many times larger. Apart from such technological advantages. the light-guide system will save copper and greatly increase the potential capacity of existing underground duct systems.

There is nothing particularly new in using light for communication. After all, the American Indians sent up smoke signals and the English built bonfires to warn of the approach of the Spanish Armada. In the 1790's Claude Chappe built an optical telegraph system consisting of semaphore stations on hilltops throughout France. The system. which reputedly could transmit messages a distance of 200 kilometers in 15 minutes, remained in service until it was superseded by the electric telegraph. In 1880 Alexander Graham Bell invented the "photophone," with which he demonstrated that speech could be transmitted on a beam of light. In one system Bell focused a narrow beam of sunlight onto a thin mirror. When the sound waves of human speech caused the mirror to vibrate, the amount of light energy transmitted to a selenium detector varied correspondingly. The light reaching the detector caused the resistance of the selenium, and therefore the intensity of the current in a telephone receiver, to vary, setting up speech waves at the receiving end. And at least until World War II it was common for naval vessels to exchange messages with Morse-coded light signals.

What is new today are the techniques available for generating a light beam that can be modulated at extremely high rates and, equally important, for transmitting the resulting signals through a glass fiber several miles long with an acceptably low loss of energy. The modern interest in light-wave communications dates from the first demonstration of the laser in 1960. This device, which can emit a nearly monochromatic beam of intense visible, or infrared radiation, opened up a region of the electromagnetic spectrum whose frequencies were 10,000 times higher than the highest frequencies then in service for radio communication systems. Since potential informationcarrying capacity increases directly with frequency, communication engineers had expended great ingenuity over many decades developing systems of ever higher frequency. From the early days of radio they had pushed useful frequencies gradually upward by about five orders of magnitude, from about 100 kilohertz (100,000 cycles per second) to about 10 gigahertz (10 billion cycles per second). Now the laser provided an increase of four more orders of magnitude to 100 terahertz (100 trillion cycles per second). By utilizing only a small part of the full range of light frequencies generated by the laser a single light-wave system could in principle simultaneously carry the telephone conversations of every person living in North America.

The early lasers, however, were cumbersome and unreliable; the best of them failed after a few months of operation. Furthermore, it soon became obvious that the concept of projecting laser beams from point to point through the atmosphere, in analogy with microwave radio-relay systems, was unsatisfactory because the signal would be attenuated by fog, smog, rain and snow. Actually it is easier to transmit light pulses reliably from Arizona to the moon than it is to transmit them between downtown and uptown Manhattan.

There has been steady progress in making lasers compact, reliable and long-lasting and in circumventing the transmission of the light beams through the atmosphere. Moreover, although for some demanding applications lasers are still preferred, for others a simpler and cheaper device, the high-intensity light-emitting diode (LED), is adequate. The first promising alternative to transmitting light signals through the atmosphere consisted in sending optical signals through a light pipe: a carefully fabricated tube a centimeter or so in diameter provided with optical means (possibly local variations in the density of a gas in the pipe) for bending the rays wherever the pipe had to diverge from a straight line.

As an alternative to the light pipe, which presented many practical problems, communication engineers began studying the possibility of transmitting light through glass fibers. Bundles of glass or plastic fibers had been used for some time to carry light short distances, for example to light an instrument panel or to examine the interior of the stomach, but they were not nearly transparent enough for the purposes of lightwave communications. The materials commonly employed were less transparent than water. The glass fibers ultimately developed for communications are so transparent that if seawater were as clear as they are, one could easily see to the bottom of the deepest ocean.

Before considering the kinds of communication system one can build from lasers, light-emitting diodes and glass fibers let us examine how information from a source, such as a telephone, a television camera or a computer, is made suitable for transmission by light. In conventional "analogue" transmission systems the wave pattern of the original signal is used to modulate the amplitude of the energy entering the transmission line, in this case the amplitude of the light beam emerging from a light source and entering a glass fiber. At the far end of the fiber the light enters a photodetector that converts the varying intensity of the light into a corresponding electrical signal. The signal is amplified as needed to reproduce the incoming electrical waveform for presentation to the ear, the eye or an inanimate device such as a computer.

Even in the best fibers some of the light is lost by absorption and scattering, so that the strength of the light signal decreases geometrically as the signal travels from the source to the detector. For example, if the strength of the light signal falls to half its original value after the signal has traveled a kilometer, the strength will fall to a fourth of the original value at the end of the second kilometer, and so on. Thus for long-distance transmission the light source should be as powerful as possible and the detector as sensitive as possible, other things being equal.

At present this requirement is best met by high-intensity lasers and ultrasensitive photodetectors of the "avalanche" type, that is, detectors in which each incoming photon triggers an avalanche of electrons. It should be observed, however, that the maximum transmission range is much more dependent on losses in the fiber than on the power of the source or the sensitivity of the detector. For example, decreasing the loss by a factor of two will exactly double the range, whereas increasing the power of the source by a factor of two will normally increase the range by only about 10 percent (to be precise, by that length of fiber which will increase the loss by a factor of two).

The chief drawback of the analogue transmission system I have been describing is that if the amplitude-modulated signal is distorted in any way during its passage through the fiber, and a certain amount of distortion is unavoidable, the distortion will be superimposed on the signal that is extracted and amplified at the receiver. One of the most effective means for transmitting an essentially distortion-free signal is to encode the signal into digital form before transmitting it. This is done by sampling the amplitude, or height, of the continuous signal wave electronically at regular intervals. If the wave is to be represented accurately, it must be sampled at twice the rate of its highest frequency component. Hence a voice signal with a maximum frequency of 4,000 hertz will be accurately represented if it is sampled 8,000 times per second. The individual sample measurements are coded into binary form, represented by a series of 1's and 0's. The binary numbers are now transmitted according to a prearranged code. For example, 1 can be transmitted as a pulse of light and 0 by the absence of a pulse. At the receiver the pulses are detected and used to reconstruct the original wave.

The most important advantage of digital transmission comes in dealing with weak signals. Every detector has an inherent internal noise that corrupts the signal entering the detector to a greater or lesser degree. Thus communication engineers commonly talk about signal-to-noise ratios. The ratios are measured on a logarithmic scale to the base 10 and in units of the decibel. A decibel is defined as 10 times the logarithm of the ratio of two power levels. For example, a signal-to-noise ratio of 20 decibels signifies that the signal level is 100 times higher than the noise level. Since digital pulses are either present or absent, they can be detected with a low probability of error even in the presence of significant noise. For example, with a signal-to-noise ratio of 21 decibels only one pulse in a billion will be lost in the background noise. For analogue signals, on the other hand, any noise tends to distort the message; hence if the signal is to be satisfactorily reproduced, the signal-tonoise ratio must be much higher than 21 decibels. Typically a signal-to-noise ratio of 60 decibels is needed, that is, a signal a million times greater than the noise.

The digital transmission system's greater tolerance of noise means that digital signals can be transmitted farther than analogue signals before amplification is needed. Another great advantage of digital transmission lies in the ease with which digital pulses can be detected and regenerated. Since minor distortions in the shape of the pulse are of little consequence, the weakened pulses can be detected and regenerated without imposing stringent requirements on the amplifiers.

It has become increasingly common in telephony to send voice signals over cable or microwave transmission systems by means of digital pulses. The voice signal is sampled 8,000 times per second, with eight binary digits specifying

the "height" of each sample. Since eight binary digits are capable of specifying 28, or 256, levels of amplitude, they provide an accurate specification of the wave pattern. This means that in order to reproduce the original voice wave, which has a frequency bandwidth of 4,000 hertz, the digital system must be able to transmit 64,000 pulses per second. The large bandwidth of lightwave systems makes it attractive to be somewhat generous in the use of bandwidth in return for a vastly improved signal-to-noise performance, which pays off handsomely in range, or the distance the signal can travel before it must be regenerated.

Thus we see that in a practical lightwave communication system the range depends on the power of the source, the attenuation per unit length of fiber, the noise level of the detector and the kind of modulation or coding that is employed. The capacity of the system in bandwidth, pulses per second or any other measure of information capacity depends on the speed with which the source can be turned on and off, the response speed of the detector and also on the pulse-spreading characteristics of the fiber.

Two kinds of light source are currently in service. The first is a refined version of the LED's that are found in the visual-display units of pocket calculators. For light-wave communications what is needed is a source that is not only much more intense than that of visual displays but also comparable in size to the optical fiber, which is only a few hundredths of a millimeter in diameter. In LED's designed for light-wave communications a small hole, or well, is etched into the face of the LED in order to bring the fiber as close as possible to the active region of the semiconductor junction, where the light originates. The fibers have their lowest loss in the infrared region of the spectrum, so that one selects a semiconductor material that emits infrared radiation. LED's made from gallium arsenide, which emit at a wavelength of about .8 micrometer, are satisfactory, but they would be even better if their emission wavelength were somewhat longer. Semiconductor materials that promise a better wavelength match for present optical fibers are under active investigation.

The other type of light source is the semiconductor diode laser, which has a more complex structure than the lightemitting diode. No bigger than a grain of salt, a diode laser consists of several layers of semiconductor material, each one of a different composition. The sandwich structure helps to establish the conditions necessary for laser action; it provides a region that confines the charge carriers giving off light when they recombine and also helps to guide the light in a preferred direction.

It was once difficult to lay down successive layers of material in such a laser without spoiling the crystal structure of the layers. The early devices were therefore notorious for their rapid decay in light-emitting efficiency; some failed completely within a few hours. New techniques were gradually developed so that the composite structures can now be fabricated without introducing imperfections in the crystalline layers. Accelerated aging tests demonstrate that the most recently developed devices should last for several years at room temperature. Ultimately laser diodes should be fully as reliable as other solidstate devices.

The laser source has two main advantages. The first is its directionality. Because the stimulated emission from the laser emerges in a narrow beam it is possible to couple a large fraction of its radiation directly into the end of an optical fiber. The second advantage is its small spread in color, or wavelength, which is typical of a laser source. In traveling through a light guide rays of different wavelengths travel at slightly different velocities; hence the broadening of the pulses in a light guide varies directly with the width of the band of wavelengths transmitted. Laser sources are therefore able to transmit pulses at a higher rate over a given distance than LED's, which emit a broader band of wavelengths. The spectrum of a typical diode laser has a width of only 20 angstroms, compared with 35 0 angstroms for a light-emitting diode of the type suitable for light-guide communications. After traveling a kilometer through an optical fiber, the laser pulse will show a dispersion in time of 200 X 1012 second, equivalent to a dispersion in distance of four centimeters at the reduced velocity with which light travels through glass. For an LED source the dispersion is nearly 20 times greater. This dispersion due to a lack of spectral purity places an important limitation on the pulse rate and thus on the information capacity of light-wave communication systems. Another major limitation is a type of pulse-spreading due to modal dispersion, which arises because some of the rays entering an optical fiber travel slightly longer paths than other rays. As we shall see, modal dispersion can be greatly minimized but not entirely eliminated.

In order to obtain the extraordinary transparency needed for a light guide the optical fibers are designed so that the light never comes near the outside surface of the fiber, where dust, scratches or contact with other surfaces would cause serious losses. Each fiber actually consists of three layers. The outer layer is a coating, usually of plastic, that provides protection from scratches and abrasion, which could weaken the fiber and lead to breakage under stress. Within the protective coating the glass fiber itself has a core region with an index of refraction slightly higher than that of its surrounding cladding. Because of this higher refractive index, rays that enter the end of the fiber at a shallow angle to the central axis are reflected back into the core when they strike the interface between the core and the cladding_ Rays that enter the fiber at large angles to the axis simply escape without being reflected. One can see from geometric considerations that if a ray is reflected back into the core at its first encounter with the interface, it will continue to be confined indefinitely, provided there are no sharp bends in the fiber. Such bends can be avoided by carefully encasing several fibers in a fairly stiff cable sheath.

One can now visualize the source of modal dispersion: a light ray that enters parallel to the central axis of a fiber will normally travel a shorter distance than a ray that enters at an angle and caroms from side to side as it travels down the fiber channel. As a result a light pulse made up of a combination of rays becomes spread out over time.

To overcome pulse-spreading many of today's fibers have a core whose refractive index is graded, or shaped, to compensate for the different distances the rays travel. In such fibers the refractive index decreases with radial distance away from the center. In the regions of lower refractive index the light travels faster. It is possible to arrange the radial decrease in refractive index so that all rays arrive at their destination at more nearly the same time. In a fiber of uniform refractive index the pulse-spreading amounts to about 25 X 10-9 second per kilometer, which is equivalent to about 500 centimeters. Graded-index fibers now being tested in the field reduce this dispersion by a factor of 25, and in laboratory samples an improvement of 100 to one has been demonstrated. The first high-transparency fibers were made by the Corning Glass Works out of a material whose principal component was silicon dioxide. The first successful fiber with a graded refractive index was made by the Nippon Sheet Glass Co., Ltd. In a process developed by Bell Laboratories a graded-index fiber is made by heating and collapsing a three-foot tube of quartz glass that has previously been coated on the inside with dozens of precisely controlled layers of silicon dioxide doped with germanium. Each layer is only about a hundredth of a millimeter thick. The tube is collapsed into solid rod called a preform, which is.then drawn into a fiber a few kilometers long.

In the best fiber specimens the transmission losses can be as low as one decibel per kilometer, which is equivalent to 80 percent transmission of the input energy. Such low losses, however, cannot be achieved at the operating frequency of available light sources; a morerealistic average loss figure is four or five decibels per kilometer, or about 30 percent transmission of the input energy. Even at this value laser-light pulses can be transmitted a distance of 14 kilometers before amplification is needed. (At that distance only 10-7 of the input energy survives.) Undoubtedly as the sources and detectors are "tuned" to the region of the spectrum where the fibers show a minimum loss at wavelengths somewhat longer than one micrometer, and as the fibers themselves are improved, the distance between amplifiers can be extended substantially beyond 14 kilometers.

The hair-thin light guides are readily assembled into cables. After they are coated for protection against humidity, abrasion and losses due to bending, the fibers are assembled into fiat, color-coded ribbons, each ribbon containing 12 fibers. Up to a dozen ribbons are enclosed in a cable that cushions and protects the individual fibers against damage in field service. Considerable ingenuity was required to devise efficient splicing methods. A technique was finally developed that can align all the fiber ends in a cable to an accuracy of within two micrometers.

Light guides offer a number of advantages over transmission by metallic conductors. Since the light in a light-guide transmission system is tightly confined to the inner core of each fiber. signals cannot leak between adjacent fibers and give rise to "cross talk." Moreover. since light guides are not affected by electrical interference from other sources. lightwave systems should show advantages in carrying information in electrically noisy environments. such as between switching apparatus in telephone central offices.

Cables for light-wave communications offer distinct savings in materials compared with metallic cables of equivalent capacity. At present the optical fibers are much more expensive than copper wire. but this is to be expected when a technologically complex new product is first put into production. Just as there are two kinds of light source for light-wave communications. so there are two types of detector in service. Both are solid-state devices. One is a simple device of the junction type known as a PIN detector. rather similar to a solar cell. in which photons of light generate an electric current. (The letters P, I and N stand for the electronic properties of the semiconductors used in the junction of the detector.) The other device is the avalanche photodetector mentioned above. All signal detectors have a background noise that increases in proportion to their operating speed. For example. the background noise in a PIN detector increases from 10-11 watt when the device is operated at one megabit per second to 10-9 watt at 100 megabits per second. At the same operating speeds the background noise in an avalanche detector is lower by a factor of 10. It follows that the transmission distances for low-speed systems are greater than those for high-speed systems. In light-wave communications the signal detector is the first stage in a receiver mod ule that contains the circuitry needed to adapt the signals for transmission through the existing telecommunication network.

Let us now assemble these various L kinds of information about sources. detectors and fiber properties and see what kinds of communication capabilities are available. First let us calculate the range for a low bit-rate system. one capable of transmitting 106 bits per second. In order for detection to proceed with few errors the signal arriving at the detector must be 100 times larger than the detector's internal noise. If an avalanche photodetector is employed. the arriving signal must have a power level of at least 10-10 watt. For maximum range we. would choose a laser with a power output of 10-3 watt in preference to an LED. which is an order of magnitude less powerful. As we have seen. with digital coding the maximum allowed attenuation of the light passing through the fiber is a factor of 107, or 70 decibels. Since present production fibers have an attenuation of less than five decibels per kilometer. we can expect satisfactory transmission for a distance of 14 kilometers before amplification is needed. (If fibers with an attenuation of only one decibel per kilometer are perfected. the range could be stretched to 70 kilometers.) In practice it seems doubtful that fibers will ever be available with continuous lengths much greater than a few kilometers. Therefore the extra loss introduced at the junction of two fibers must be added to the loss figure. Present plug-type connectors introduce a loss of about .5 decibel. If one were to need six connectors for a 14-kilometer route. the additional loss would come to only three decibels. (The total loss could be held to 70 decibels by shortening the route by three-fifths of a kilometer.)

With the selection of a source. a detector and a fiber. what will the information- handling capacity of light-guide systems be? Since it is desirable to transmit at the highest possible bit rate. one must consider a number of factors. As we have seen. the noise level of the detector increases with the bit rate. Thus if the signal power was just adequate for transmission at 106 pulses per second. it would have to be raised by a factor of 100 for transmission at 108 bits per second. In addition. as the pulses get shorter and closer together their spreading as they travel through the fiber becomes an important limiting factor.

In order to simplify the calculation let us decide somewhat arbitrarily that the pulse-spreading will not be more than half the interval between successive pulses. In a graded-index fiber the pulsespreading due to modal dispersion (the difference in path lengths) amounts to about 10-9 second per kilometer. which means that if one attempts to transmit

109 pulses per second. the spreading is equivalent to the entire interval between pulse peaks. Therefore to maintain a separation of half an interval the signaling rate cannot exceed .5 X 109, or 5 X 108, pulses per second. This is the limit if we have a laser source. which is so nearly monochromatic that pulsespreading due to wavelength dispersion can be ignored.

If we choose an LED light source. however. wavelength dispersion becomes the limiting factor in the signaling rate. For an LED the wavelength dispersion amounts to 3.5 X 10-9 second per kilometer, a figure 3.5 times larger than that for modal dispersion. In order to hold the spreading below half the interval between successive pulses the signaling rate with an LED source must therefore be slightly less than a third of the rate permissible with a laser, or 1.4 X 108 pulses per second. Naturally as the desired transmission distance is increased the signaling rate must be proportionately reduced. For example, for a nominal transmission distance of 10 kilometers the rate for the laser source would have to be reduced tenfold to 5 X 107 pulses per second, or approximately the rate (4.47 X 107) actually selected for the Chicago installation. These simple calculations illustrate what can be achieved with today's technology and also give a feeling for the kinds of design choice that can be made among range, capacity and device complexity. Undoubtedly there will be significant improvements in the future.

There are many promising areas of application for the new light-guide technology. For example. television signals could easily be carried over a single fiber, thereby opening up new possibilities for both entertainment and business purposes. Buildings could be "wired" with almost invisible fibers to provide internal communication services. The parts of computers could be interconnected with fibers. It is in telephony. however. that one can expect some of the first important applications.

Today much of the copper cabling that interconnects metropolitan telephone- switching centers goes through underground ducts where space is at a premium. Adding new duct space is both costly and inconvenient. Lightwave communication systems with their high capacity and small size could make better use of the existing underground ducts and help to postpone the need for new ones. Moreover. since adjacent switching centers in many cities are less than seven kilometers apart, light-wave systems might not require any amplifiers in manholes to boost signals along a typical route.

Before the completion of the Chicago installation Bell Laboratories and the Western Electric Company tested a prototype light-wave system under simulated field conditions last year in Atlanta. Two light-guide cables 640 meters long, each containing 144 fibers, were pulled through standard underground ducts and subjected to tests simulating a typical urban telecommunication environment. The installation work did not break any of the fibers, and the pulling operation, which required the negotiation of sharp bends, did not degrade the performance of the light guides. As in the present Chicago system, each pair of fibers carried the equivalent of 672 two-way voice channels. The light sources were gallium aluminum arsenide lasers operating at a rate of 44.7 million bits per second. At the receiving end the light pulses were converted into electrical signals by avalanche photodetectors. As part of the Atlanta experiment the ends of some individual fibers were joined to create a continuous communication path about 70 kilometers long. With the help of 11 regenerators, or amplifiers, virtually error-free transmission was achieved in one direction over the full path for a sustained period. The Chicago installation closely follows the Atlanta experimental system except that LED's are being used in addition to lasers as light sources.

Apart from some references to the anticipated red uction of fiber losses in the future, everything I have described here is based on current technology. It would be contrary to all previous experience to believe we shall not witness further dramatic developments. For example, a number of industrial and university investigators are conducting experiments with integrated optics, which include techniques for processing light signals within thin films, the optical equivalent of integrated microelectronic circuits. Such optical circuits may someday eliminate the need for converting light pulses to and from electrical signals in amplifiers along transmission paths. In addition both theoretical and experimental work is proceeding on the possibility of switching light pulses directly, obviating the need for first converting the light signals into their electrical equivalent. The hope is to develop optical switches to replace the present electromechanical and electronic devices, thereby making it possible to connect telephone calls in greater numbers and at higher speeds than ever before.