Light-emitting Diode (LED)
Light emitters are a key element in any fiber optic system. This component converts the electrical signal into a corresponding light signal that can be injected into the fiber. The light emitter is an important element because it is often the most costly element in the system, and its characteristics often strongly influence the final performance limits of a given link.	Figure 1 - LEDs Convert an Electrical Signal to Light
LEDs are complex semiconductors that convert an electrical current into light. The conversion process is fairly efficient in that it generates little heat compared to incandescent lights. LEDs are of interest for fiber optics because of five inherent characteristics: 1. They are small. 2. They possess high radiance (i.e., They emit lots of light in a small area). 3. The emitting area is small, comparable to the dimensions of optical fibers. 4. They have a very long life, offering high reliability. 5. They can be modulated (turned off and on) at high speeds. Table 1 offers a quick comparison of some of the characteristics for lasers and LEDs. These characteristics are discussed in greater detail throughout this article and the article on laser diodes.
Table 1 ¨C Comparison of LEDs and Lasers Characteristics LEDs Lasers Output Power Linearly proportional to drive current Proportional to current above the threshold Current Drive Current: 50 to 100 mA Peak Threshold Current: 5 to 40 mA Coupled Power Moderate High Speed Slower Faster Output Pattern Higher Lower Bandwidth Moderate High Wavelengths Available 0.66 to 1.65 µm 0.78 to 1.65 µm Spectral Width Wider (40-190 nm FWHM) Narrower (0.00001 nm to 10 nm FWHM) Fiber Type Multimode Only SM, MM Ease of Use Easier Harder Lifetime Longer Long Cost Low ($5-$300) High ($100-$10,000)
Light-emitting diodes use GaAlAs (gallium aluminum arsenide) for short-wavelength devices. Long-wavelength devices generally incorporate InGaAsP (indium gallium arsenide phosphide).
Light Emitter Performance Characteristics
Several key characteristics of LEDs determine their usefulness in a given application. These are: Peak Wavelength: This is the wavelength at which the source emits the most power. It should be matched to the wavelengths that are transmitted with the least attenuation through optical fiber. The most common peak wavelength are 780, 850, and 1310 nm. Spectral Width: Ideally, all the light emitted from an LED would be at the peak wavelength, but in practice the light is emitted in a range of wavelengths centered at the peak wavelength. This range is called the spectral width of the source. Emission Pattern: The pattern of emitted light affects the amount of light that can be coupled into the optical fiber. The size of the emitting region should be similar to the diameter of the fiber core. Power: The best results are usually achieved by coupling as much of a source¡¯s power into the fiber as possible. The key requirement is that the output power of the source be strong enough to provide sufficient power to the detector at the receiving end, considering fiber attenuation, coupling losses and other system constraints. In general, LEDs are less powerful than lasers. Speed: A source should turn on and off fast enough to meet the bandwidth limits of the system. The speed is given according to a source¡¯s Rise or fall time, the time required to go from 10% to 90% of peak power. LEDs have slower rise and fall times than lasers. Linearity is another important characteristic for some applications. Linearity represents the degree to which the optical output is directly proportional to the electrical current input. Most light sources give little or no attention to linearity, making them usable only for digital applications. Analog applications require close attention to linearity. Nonlinearity in LEDs causes harmonic distortion in the analog signal that is transmitted over an analog fiber optic link. LEDs are generally more reliable than lasers, but both sources will degrade over time. This degradation can be caused by heat generated by the source and uneven current densities. In addition, LEDs are easier to use than lasers. LEDs are found in a wide variety of consumer electronics products. LEDs are used as visible indicators in most electronics equipment, and laser diodes are most widely used in compact disk (CD) players. The LEDs used in fiber optics differ from the more common indicator LEDs in two ways: 1. The wavelength is generally in the near infrared (because the optical loss of fiber is lowest at these wavelengths). 2. The LED emitting area is generally much smaller in order to allow the highest possible modulation bandwidth and improve the coupling efficiency with small core optical fibers. LEDs and laser diodes are very similar devices. In fact, when operating below their threshold current, all laser diodes act as LEDs.
Figure 2a shows the behavior of an LED, and Figure 2b shows the behavior of a laser diode. The plots show the relative amount of light output versus electrical drive current. The LED outputs light that is approximately linear with the drive current. Nearly all LEDs exhibit a ¡°droop¡± in the curve as shown in Figure 2b. This nonlinearity in the LED limits its usefulness in analog applications. Figure 2 - Emitter Characteristics,  (a)  LED            (b) Laser
The droop can be caused by a number of factors in the LED semiconductor physics but is often largely due to self-heating of the LED chip. All LEDs drop in efficiency as their operating temperature increases. Thus, as the LED is driven to higher currents, the LED chip gets hotter causing a drop in conversion efficiency and the droop apparent in Figure 2a. LEDs are typically operated at currents to about 100 mA peak. Only specialized devices operate at higher current levels.
LED Types
There are two basic types of LED structures: edge emitters and surface emitters. Figure 3 - LED Structures
Edge emitters are more complex and expensive devices, but offer high output power levels and high speed performance. The output power is high because the emitting spot is very small, typically 30-50 µm, allowing good coupling efficiency to similarly sized optical fibers. Edge emitters also have relatively narrow emission spectra. The full-width, half-maximum (FWHM) is typically about 7% of the central wavelength. Another variant of the edge emitter is the superradiant LED. These devices are a cross between a conventional LED and a laser. They usually have a very high power density and possess some internal optical gain like a laser, but the optical output is still incoherent, unlike a laser. Superradiant LEDs have very narrow emission spectra, typically 1-2% of the central wavelength and offer power levels rivaling a laser diode. These devices are popular for fiber optic gyroscope applications. The second type of LED is the surface emitter. Surface emitters have a comparatively simple structure, are relatively inexpensive, offer low-to-moderate output power levels, and are capable of low-to-moderate operating speeds. The total LED chip optical output power is as high or higher than the edge-emitting LED, but the emitting area is large, causing poor coupling efficiency to the optical fiber. Adding to the coupling efficiency deficit is the fact that surface-emitting LEDs are almost perfect Lambertian emitters. This means that they emit light in all directions. Thus very little of the total light goes in the required direction for injection into an optical fiber.
LED Drive Circuits
LED optical output is approximately proportional to drive current. Other factors, such as temperature, also affect the optical output. Figure 4 shows in greater detail the typical behavior of an LED. Two curves are shown. The top curve represents a 0.1% duty cycle with the peak current as shown on the horizontal axis. The bottom curve shows the output with 100% duty cycle. Note the light versus current curve droops below the linear curve. Figure 4 - Optical Output vs. Current in a LED
LEDs are usually driven with either a digital signal or an analog signal. Analog LED Driver Circuits Figure 5 shows three configurations for analog LED drive circuits. Figure 5 - Analog LED Drive Circuits For more information on VCSELs see the article Laser Diodes.
Circuit 5a illustrates the simplest of the three configurations. It uses a transistor, Q1, and a limited amount of resistors to convert an analog input voltage into a proportional current flowing through the LED, D1. Also referred to as a transconductance amplifier, this configuration converts a voltage into a current. In LEDs, the light output equates proportionally to the drive current, not the drive voltage. While the drive current varies, this circuit illustrates the voltage dropping across that LED and remaining constant. LEDs exhibit a peak drive current at about 100 mA, and the voltage drop is typically 1.5 Volts. Circuit 5a works as follows: the small resistor, R1, prevents oscillations in Q1. The input voltage, VIN, appears on the base of Q1. VR2 is the voltage at the emitter of Q1, and it equals the base voltage minus 0.6 Volts. Since these base and emitter voltages only differ by a DC offset voltage, the AC portion of the base equals that of the emitter. The emitter voltage VR2 causes a current equal to VR2/R2 to flow through R2. Due to the nature of transistors, the Q1 collector current approximately equals the Q1 emitter current. (To be precise, the collector current equals b/(b+1) times the emitter current. The transistor current gain, b, is usually 10 to 100.)  Collectively, we find that the LED current, and thus the output light, relates to the input voltage VIN as follows: A drawback of the simple circuit is that the base capacitance varies with the base voltage, which introduces nonlinearities that limit the circuits linearity.  However, the linearized, low frequency circuit shown in Figure 5b eliminates most of the nonlinearities associated with Q1. In this case, U1 forms a feedback loop that drives the base of Q1 in such a way that assures that VR2 equals VIN. In this case, LED current, and thus the output light, relates to the input voltage VIN as follows: The circuit shown in Figure 5b still experiences some lesser nonlinearities associated with Q1, but these do not represent the limiting factor. The circuit is limited by the delay associated with the feedback signal in the servo loop formed by U1, allowing the circuit to only achieve a bandwidth of about 10-100 MHz. This limitation makes the circuit in Figure 5b work well in application transmitting DC coupled analog signals. Figure 5c shows the highest performance analog LED drive circuit. In this case, resistor, R1 supplies the DC current through D1. Sometimes, a constant current source or a network that includes temperature compensation replaces R11. A wideband RF amplifier, U1, serves two purposes. First it amplifies VIN to allow the use of a small input signal. Second, it isolates the LED from the input circuit, allowing precise impedance matching at the input, VIN, which reduces reflections. The output of U1 is usually 50 Ohms or 75 Ohms. A typical LED may have an input impedance ranging from 5 Ohms to 10 Ohms. An impedance matching network is inserted between the amplifier and D1. Furthermore, capacitor, C1, serves to block any DC level associated with the output of the matching network. This circuit will drive LEDs to their highest possible frequencies. Circuit 5c usually delivers the highest possible linearity. In this case, the LED, D1, usually limits performance. Digital LED Drive Circuits When the drive signal is digital, as illustrated in Figure 6, there is no concern about LED linearity. The LED is either on or off. There are special problems that need to be addressed when designing an LED driver. The key concern is driving the LED so that the maximum speed is achieved. Figures 6a, 6b, and 6c show three popular digital LED driver circuits. The first circuit, shown in Figure 6a, is a simple series driver circuit. The input voltage is applied to the base of transistor Q1 through resistor R1. The transistor will either be off or on. When transistor Q1 is off, no current will flow through the LED, and no light will be emitted. When transistor Q1 is on, the cathode (bottom) of the LED will be pulled low. Transistor Q1 will pull its collector down to about 0.25 Volts. The current is equal to the voltage across resistor R2 divided by the resistance of R2. The voltage across R2 is equal to the power supply voltage less the LED forward voltage drop and the saturation voltage of the drive transistor. The key advantage of the series driver shown in Figure 6a is its low average power supply current. If one defines the peak LED drive current as ILEDmax and assumes that the LED duty cycle is 50%, then the average power supply current is only ILEDmax/2. Further, the power dissipated is (ILEDmax/2)•VSUPPLY where VSUPPLY is the power supply voltage. The power dissipated by the individual components, the LED, transistor and resistor R1, is equal to the voltage drop across each component multiplied by (ILEDmax/2).  The key disadvantage of the circuit shown in Figure 6a is low speed. This type of driver circuit is rarely used at data rates above 30-50 Mb/s. In general, there are two ways to design an LED drive circuit for low power dissipation. The first is to use a high-efficiency LED and reduce ILEDmax to the lowest possible value. The second is to reduce the duty cycle of the LED to a low value. Usually larger gains can be made with the second method. Figure 6 - Digital LED Drive Circuits
The second LED driver circuit, shown in Figure 6b, offers much higher speed capability. It uses transistor Q1 to quickly discharge the LED to turn it off. This circuit will drive the LED several times faster than the series drive circuit shown in Figure 6a. The key advantage of the shunt drive circuit is that it gives much better drive symmetry. LED¡¯s are easy to turn on quickly, but are difficult to turn off because of the relatively long carrier lifetime. In the shunt driver circuit in Figure 6b, resistor R2 provides a positive current to turn on the LED. Typically, R2 would be in the 40 Ohm range. This makes the turn-on current about 100 mA peak. Transistor Q1 provides the turnoff current. When saturated, transistor Q1 will have an impedance of a few Ohms. This provides a much larger discharging current allowing the LED to turn off quickly. The key disadvantage of the shunt driver is the power dissipation. It is typically more than double that of the series driver. In fact, the circuit draws more current and power when the LED is off than when the LED is on! The exact power dissipation can be computed by first analyzing the off and on state currents and then combining the two values using information about the operating duty cycle. The last driver circuit, shown in Figure 6c, is a variation on the shunt driver shown in Figure 6b.Two additional resistors and two capacitors have been added to the basic circuit. The purpose of these additional components is to further improve the operating speed. Capacitor C1 serves to improve the turn-on and turnoff characteristics of transistor Q1 itself. One has to be careful that C1 is not made too large. If this occurs, the transistor base may be overdriven and damaged. The additional components, resistors R3 and R4 and capacitor C2, provide overdrive when the LED is turned on and underdrive when the transistor is turned off. The overdrive and underdrive accelerates the LED transitions. Typically, the RC time constant of R3 and C2 is made approximately equal to the rise or fall time of the LED itself when driven with a square wave. Figure 7 - LED Response to Digital Modulation Figure 7 shows the response of an LED to a digital modulation signal. The electrical signal shown is the type generated by more sophisticated LED driver circuits such as that shown in Figure 6c. Starting at time zero, we first see the digital signal go to a logic level 1. The most remarkable part of this event is the strong overshoot seen on the electrical drive signal. This overshoot may be two times the steady state logic 1 drive current. This overshoot accelerates the turn-on time or rise time of the LED. Even so, we see that the optical output lags behind the electrical signal. Typical values for very high-performance LED¡¯s and driver circuits would be 0.7 ns rise time of the electrical signal and 1.5 ns optical rise time. Later, when the digital signal goes back to a logic 0, we see the same process repeated. The electrical signal has a strong undershoot component which acts to accelerate the turn-off of the LED. The undershoot serves to reverse bias the LED, sweeping out the carriers. Even so, the turn-off time of most LED¡¯s is always slower than the turn-on time. Typical values for turn-off times are 0.7 ns for the electrical signal and 2.5 ns for the optical signal. Note that while in a logic 0 state, the drive current does not quite go to zero. It is common to provide a small amount of pre-bias current, typically a few percent of the peak drive current, to keep the LED forward biased and improve dynamic response. All of these tricks together can increase the operating speed of the LED and driver circuit to about 270 Mb/s. There have been numerous laboratory tests and prototype circuits that have achieved rates to 500-1000 Mb/s, but none of these have ever made it into mass production. Typically these levels of performance require a great deal of custom tweaking on each part to achieve the high data rates.
Energy Gaps in LEDs
When turned on, the LED will have a forward voltage drop of about 1.1 to 1.5 Volts. Shorter wavelength diodes (e.g. 850 nm) have the largest voltage drops. As the wavelength increases, the voltage drop decreases. This phenomenon can be related to the bandgap energy Eg of the LED. Equation 1 defines the bandgap energy Eg:
Eg=hc/l = 1240eV-nm/l	Where: h = Plank's Constant = 4.13 x 10-15 eV•s c = speed of light = 2.998 x 108 m/s l = wavelength in nm
Using equation 1, we can predict the energy gap of an LED based on its emission wavelength.
Table 2 - Common Light Emitter Materials & Characteristics Material Formula Energy Gap Wavelength Gallium Phosphide GaP 2.24 eV 550 nm Aluminum Arsenide AIAs 2.09 eV 590 nm Gallium Arsenide GaAs 1.42 eV 870 nm Indium Phosphide InP 1.33 eV 930 nm Aluminum-Gallium Arsenide AIGaAs 1.42-1.61 eV 770-870 nm Indium-Gallium-Arsenide-Phosphide InGaAsP 0.74-1.13 eV 1100-1670 nm
Table 2 lists some common light emitter materials, the emission wavelength and corresponding energy gap. The first materials, GaP and AlAs, are used to make emitters in the visible portions of the spectrum. The next three materials, GaAs, InP, and AlGaAs, are used to make emitters in the  near infrared portion spectrum generally referred to as the ¡°first window¡± in optical fiber. The last material, InGaAsP is used to make emitters in the infrared portion spectrum referred to as the ¡°second and third windows¡± in optical fibers. The energy gap corresponds to the energy of the emitted photons and also is indicative of the voltage drop associated with a forward biased LED. Knowing the voltage drop of the LED and the saturation voltage of the transistor we can compute the LED current. Equation 2 below shows the general form of the calculation.
ILED= VPower-VLED-VSAT/R3	Where: VPOWER = DC power supply voltage. VLED = forward voltage drop of the LED. VSAT = drive transistor saturation voltage R3 = series LED current limiting resistor ILED = peak LED current
Another common use of LEDs is to simply use their large forward voltage drop in some part of a circuit. In this case, the fact that the LED emits light is incidental. For instance, if one needed a 2.3 Volt drop in a circuit, then one could use three 1N4148 diodes in series or a single green LED. Obviously, only inexpensive indicator LEDs are candidates for this application. One important consideration for this usage is that all light emitters will also function as detectors. If the LED is in a sensitive portion of the circuit, then the circuit may become sensitive to ambient light conditions. It may be necessary to shield the LED or coat it with an opaque paint. It is also useful to note that many ordinary glass diodes, such as the 1N4148, also function as light detectors. Keep this in mind when using diodes in circuits that have high gains. One possibility pursued in the past was using ultra-low cost germanium diodes as long wavelength detectors. They in fact work very well, but are somewhat inconsistent from part to part.