Atmospheric Propagation Characteristics of Highest Importance to Commercial Free Space Optics

Transcription

1 Atmospheric Propagation Characteristics of Highest Importance to Commercial Free Space Optics Eric Korevaar, Isaac I. Kim and Bruce McArthur MRV Communications Roselle St. San Diego, CA ABSTRACT There is a certain amount of disconnect between the perception and reality of Free Space Optics (FSO), both in the marketplace and in the technical community. In the marketplace, the requirement for FSO technology has not grown to even a fraction of the levels predicted a few years ago. In the technical community, proposed solutions for the limitations of FSO continue to miss the mark. The main commercial limitation for FSO is that light does not propagate very far in dense fog, which occurs a non-negligible amount of the time. There is no known solution for this problem (other than using microwave or other modality backup systems), and therefore FSO equipment has to be priced very competitively to sell in a marketplace dominated by copper wire, fiber optic cabling and increasingly lower cost and higher bandwidth wireless microwave equipment. Expensive technologies such as adaptive optics, which could potentially increase equipment range in clear weather, do not justify the added cost when expected bad weather conditions are taken into account. In this paper we present a simple equation to fit average data for probability of exceeding different atmospheric attenuation values. This average attenuation equation is then used to compare the expected availability performance as a function of link distance for representative FSO systems of different cost. Keywords: free space optics, FSO, laser communication, optical wireless, infrared, atmospheric attenuation, visibility, availability, link distance, link range 1. INTRODUCTION The concept of optical wireless communications has been used for millennia, examples including smoke signals and semaphore flags. In 1880, Alexander Graham Bell patented the photophone, which modulated light reflected from the sun with a voice signal and transmitted that across free space to a solid state detector (figure 1.) Thus was born the first Free Space Optics (FSO) link. Given that Bell described the photophone as the greatest invention I have ever made; greater than the telephone, we can see that the hype for FSO is also nothing new. Alas, the weather availability of the photophone was less than 50%, as it required direct sunlight for operation. Figure 1. Alexander Graham Bell s Photophone, It s the greatest invention I have ever made; greater than the telephone.

2 During 2000 and 2001, a number of analysts put out reports trying to quantify the size of the worldwide Free Space Optics market, and predicting its future growth. Typical of these reports was one put out by the Strategis Group in 2001 entitled Free-Space Optics: Global Trends, Positioning, and Forecasts. Typically the projections made in 2001 were lower and had slower growth curves than those made in Still, the global market size predicted by this report was $118M in 2000, $250M in 2001, $425M in 2002, $690M in 2003, $1,074M in 2004, $1,536M in 2005 and $2,031 M in Before using these or similar numbers to raise capital for a new FSO company, it is worth pointing out that these projections have no basis in reality, and even the supposedly historic numbers are wildly off base. It is most likely (in our opinion based on our experience) that the actual current annual worldwide market for FSO equipment is between $15M and $30M, with most systems installed by enterprise users. While significant growth is possible if the technology is adopted by telecommunications carriers or internet service providers, to this date that growth has not materialized. FSO equipment, when installed properly, is highly reliable from an equipment failure standpoint. As the leading worldwide provider of FSO equipment with over 5,000 links sold, MRV has been able to compile statistics on system reliability, and standard products such as the TereScope 1000X have a mean time between failure of greater than 10 years. Figure 2 shows pictures of two typical installations. On the left, two TereScope 1000 products are installed for short range links in New York City, demonstrating that multiple FSO terminals can work in close proximity without interference. On the right, a TereScope 3000 link is installed over a 2500m range in Costa Rica. Nearby microwave equipment is not a problem. Typical installations are outdoors. By mounting to solid parts of the building, and by appropriate choice of transmitter divergence, active tracking systems are not necessary. In spite of equipment robustness and reliability, in most locations with link ranges beyond 200 meters, there will be weather conditions which will cause temporary link outages. The FSO community uses the term availability to describe the percentage of time that a customer could expect a link to operate in a particular location. As will be seen in the rest of the paper, FSO equipment is most appropriate in those applications where link availabilities of around 99% are adequate, and is not well suited to % availability requirements (such as those of the North American Telecommunications carriers) except at short ranges of less than 200 meters. Figure 2. Typical Free Space Optics Installations in New York City, range 120m (left) and Costa Rica, range 2500m (right). Usually outdoors, FSO terminals may be mounted in close proximity to each other or other equipment such as RF antennas. 2. BACKGROUND WEATHER AVAILABILITY INFORMATION To predict the weather dependent availability of an FSO link, it is necessary to know something about the weather. In particular, we would like to know the statistics of the total integrated attenuation to be expected along the line of sight path between the two transceivers in a link. Based on the known values of the amount of attenuation which can be

3 tolerated by a given system at a given range (its link margin in the absence of atmospheric attenuation), it would then be possible to calculate the expected availability of the system. Unfortunately, the exact data which is needed is not generally collected, and therefore a number of assumptions are generally necessary. The best data available on a worldwide basis is historical visibility data which has been collected at airports. Such data yields statistics that give the percentage of time that visibility exceeded a particular distance. The smallest distance over which this visibility data is usually presented is 400 meters. The data tabulations do not give information on whether visibility was impaired due to haze, fog, rain or snow, nor what the lighting conditions were like. The assumptions we make about this, and the manner in which the data was collected (by people trying to discern targets at different ranges or by automated instrumentation) will affect the amount of attenuation we calculate for a given visibility. In addition, attenuation conditions in a city between buildings may be different from those experienced at the airport; the attenuation may not be constant along the whole path of the link; and attenuation may be different at the top of a building than at the bottom. Nevertheless, FSO vendors typically use the airport visibility data to predict performance in a given location. As an example, figure 3 shows the expected availability for a TereScope 3000 system (manufactured by MRV) operating at 155 Mb/s. As can be seen, the availability is very high out to a range beyond 3 km in a city like Las Vegas, Nevada, where there is very little fog. More typical of the major population centers in North America and Europe is the expected availability in San Diego. Although San Diego does not have much rain, there is a fairly high incidence of heavy coastal fog. (As might be expected, weather conditions such as fog which affect FSO link availability occur more in some months than others). Finally, St. John s in Newfoundland represents a location with particularly poor average visibility. Availability (%) TereScope Availability Link range (km) Las Vegas, NV San Diego, CA St. Johns, NF Figure 3. Link Availability due to weather is heavily location dependent. Las Vegas has a very high percentage availability out to 3 km for typical long range equipment, while St. Johns Newfoundland has poor and San Diego has average availability. One way to approach the dependence of visibility and availability on location is to create a geographical contour map showing expected availability at a given range, or expected range at a given availability. For RF and microwave systems, the main weather limitations are due to heavy rain, and there are standard models which divide the world into different rain regions. Using these, maps of range for a given availability are fairly straightforward to generate for microwave systems. Although no such comprehensive maps exist for FSO systems, a first attempt was presented by Isaac Kim at SPIE s Optical Wireless Communications IV conference in 2001, and is shown in figure 4. For this map over most of the United States, contours represent percentage of time airport visibility was measured to be over 400 meters. While this approach is useful for pinpointing more optimum areas for FSO deployment, it is difficult to use for trading off system parameters in the design of an FSO system to be sold worldwide.

4 Figure 4. First attempt at FSO availability map (from I. Kim et al., in Optical Wireless Communications IV, SPIE Vol (2001) p. 84.) Contours represent percentage of time visibility is greater than 400m. Before moving on to the development of an average attenuation vs. availability model, it is important to address the issue of wavelength dependence of attenuation for a given visibility or weather condition. As has been discussed previously (for instance Korevaar et. al., Debunking the recurring myth of a magic wavelength for free-space optics, in Optical Wireless Communications V, Vol (2002), p.155), the scattering of optical radiation in heavy fog, clouds, rain and snow is not very dependent on wavelength from the visible out to 10 microns and beyond. This is because those weather conditions contain a large percentage of scattering particles with radii larger than the wavelength. Although for higher visibility conditions, such as clear air or haze, the scattering loss falls off quickly with higher wavelength, all commercial FSO systems must be designed for the lower visibility conditions where the loss is not wavelength dependent (other than for easily avoidable molecular absorption lines). Figure 5 shows an example of the attenuation to be expected as a function of wavelength for a typical west coast (U.S.) fog. 3. AVERAGE ATTENUATION EQUATION In order to trade off different design parameters in the development of FSO equipment, it is useful to have an analytical expression for the probability of encountering different atmospheric attenuation conditions. Although in reality the visibility statistics are very location dependent, it is important to have an idea of what attenuation will be encountered for an average installation before the deployment location is known. For instance, most FSO vendors specify a recommended maximum range for their products. However, the basis on which that maximum range is calculated varies from vendor to vendor, so comparison can be difficult. In this section we present a model which can be used to calculate the probability of encountering a given attenuation in an average sense. To our knowledge, this is the first time such a model has been published, and we hope that such a model can be used for general comparison of different FSO systems. The model is presented using a log/log scale in figure 6 so that the severe limitations which would be placed on FSO equipment by a % (Five 9 s) availability requirement can be readily seen.

5 Figure 5. MODTRAN calculation of the total atmospheric attenuation as a function of wavelength for a typical spring west coast fog. Other than for molecular absorptions, attenuation is similar for all wavelengths contemplated for Free Space Optics. (Figure courtesy of Eric Woodbridge, AirFiber Inc.) There are three sets of data used in figure 6, and they are generated as follows. The first set of data is shown as diamonds, and comes from standard airport visibility data at airports near the 10 largest cities (by population) in the United States. These airports are Newark (for New York), Los Angeles, Chicago, Houston, Philadelphia, San Diego, Phoenix, San Antonio, Dallas and Detroit, and have a mix of different weather conditions. Percentage of time that visibilities are exceeded are tabulated against visibility ranges of 400 meters up to 16 km. The percentage numbers for the ten cities were then averaged for each of the visibility ranges. These numbers were then converted to outage fractions. For instance, at a visibility range of 0.5 mile or 800 meters, that visibility was exceeded 99.2% of the time on average at the ten airports, corresponding to an outage fraction of It was then assumed that the attenuation corresponding to propagation over one visibility range was 13 db. (Thus, if the visibility range is 800 meters, the attenuation is 16.2 db/km). In general, the attenuation corresponding to one visibility range seems to vary between 8.5 and 17 db. (The best article we have seen on this is by R.M. Pierce et al., Optical attenuation in fog and clouds, in Optical Wireless Communications IV, Eric J. Korevaar, Editor, Proceedings of SPIE Vol (2001) p. 58.) We have chosen to use 13 db as the best compromise, rather than the more standard Kruse model definition of 17 db corresponding to 2% contrast. We have also ignored any wavelength dependence, since this will only be apparent for very high visibilities which are not the limiting weather condition for FSO systems. The data points were then generated by plotting the outage fraction as a function of the attenuation (calculated as 13 db divided by visibility range).

6 Average 10 Largest U.S. Cities Fraction of Time Attenuation Exceeded Average San Francisco, San Diego, New York, Los Angeles, Chicago, Boston Average London, Glasgow, Manchester General Fit Curve Attenuation (db/km) Figure 6. Standard Average Outage Curve. The fraction of time that attenuation is exceeded is plotted as a function of attenuation. Data for lower attenuations is averaged for airports at the 10 largest U.S. cities. Intermediate attenuation data is averaged for 6 cities, and high attenuation data is for 3 cities in the U.K. A generalized curve fit can be used for comparisons of different systems. The second set of data shown as squares was calculated in a similar manner using higher resolution data from 6 representative airports where it was available (courtesy of Scott Bloom at AirFiber). The six cities for which outage fractions were averaged for this data were San Francisco, San Diego, New York, Los Angeles, Chicago and Boston. This data is plotted for visibility ranges between 100 meters and 1000 meters, again using an attenuation of 13 db divided by visibility range. Finally, there is hardly any data available for visibilities less than 100 meters. However, a linear fit on a log/log scale to the first two data sets would lead to attenuation values exceeding the maximum attenuations observed at outage fractions of and lower. Therefore, it is important to take an asymptotic attenuation value into account for the model. It seems reasonable, based on general FSO experience, to choose an asymptotic attenuation near 265 db/km, corresponding to some dense fogs and clouds. To achieve this asymptotic value, the third set of data is averaged in a different way than the other sets. This data (also courtesy of Scott Bloom at AirFiber) presented attenuations for London, Manchester and Glasgow in the U.K. as a function of outage percentage

7 down to 0.001%. At this level, the attenuation in London was calculated as 170 db/km, and in Manchester and Glasgow as 310 db/km. For these data points (triangles), the attenuations for the three cities were averaged for each of the different outage percentages, so that the attenuation corresponding to 0.001% (or ) was 263 db/km. (As can be seen from this procedure, the low percentage, low visibility numbers are subject to the highest error in this model average). Finally, after plotting the average attenuation data on a log/log plot, an analytical fit was made to the data. For simplicity, a linear relationship was used for the low attenuation data, and this was modified with a factor to produce an asymptote at 265 db/km. This model fit should be useful across the whole range of data, except for attenuations of less than about 1 db/km, where the outage fraction will be underestimated for visible and near infrared wavelengths because of Rayleigh scattering even in clear air. The fit formula is plotted as x s with a line through them. The average attenuation equation is given as follows, with A = attenuation in db/km: OUTAGE FRACTION = 0.22 * A * 100 -(A/265)^10 (1) The first part of the formula gives the proper linear relationship between outage fraction and attenuation under the assumptions discussed above for the airport data. The final product causes an asymptotic behavior limiting the attenuation to near 265 db/km for very small outage fractions. 4. LINK AVAILABILITY VS. RANGE Using the average attenuation equation, along with design parameters for specific FSO equipment, it is possible to calculate the expected link availability as a function of link range for that equipment, assuming the average visibility statistics embodied in equation 1. Before providing an example of this calculation for some sample FSO systems, it is necessary to discuss the effect of scintillation, which will reduce the achievable distance of many FSO links under high visibility conditions. By scintillation, we are referring to fluctuations in receive intensity due to constructive and destructive interference within the optical beam as it propagates through the atmosphere. Pockets of air of differing temperature, and thus different refractive index, cause slight bending of the rays of light on their way to the receiver. The intensity pattern at the receiver will consist of dark and light spots. If these spots are larger than or comparable to the size of the receiver, there can be large drops or spikes in the receive intensity. These fluctuations occur on timescales of a few to many milliseconds as the wind blows the interference patterns across the receive aperture. It is well known that using a larger receive aperture will lower the amount of intensity fluctuation, which increase as the link range increases. (Scintillation fade also tends to be worse at longer wavelengths because the size of the dark and light spots increases). Finally, as shown graphically in figure 7, (and as patented in figure 8), increasing the number of appropriately spaced transmitters decreases the amount of intensity fluctuation because the scintillation patterns from the different lasers are somewhat decorrelated. Quantitatively, scintillation fade is a very complicated, and not perfectly understood process. For an atmospheric propagation theorist, there are a number of scary looking models (with arguments raised to the 5/3 power), none of which appear to give completely satisfactory results. In addition, there is not much useful experimental data where all of the parameters necessary for the models have been collected. Finally, as for attenuation due to varying weather conditions, the margin needed for scintillation fade is strongly dependent on location (such as proximity of the link to the ground or a hot roof), time of day, and weather condition. What is needed for the FSO equipment designer is a general formula which predicts the necessary link margin to allocate to scintillation as a function of number of transmitters, receive aperture size, link range, and atmospheric attenuation. We present such a formula in equation 2, where our formula is derived from experimentally measured scintillation fluctuations under a number of range, receive aperture size, and transmitter number conditions at the worst time of day for typical installations. This equation is by no means the final word on this subject matter, but is given in a form that is useful for the product designer. Experiments and theory to give a more accurate formula with the same adjustable parameters under some kind of appropriate worst case average scintillation condition would make an excellent Ph.D. thesis project. Margin (db) = 2 + (12/ApNum) * (100mm/Diam) 2 * (Range/1000m) (2)

8 Figure 7. Received Intensity as a function of time for 1, 2 and 3 transmitters. Fluctuations due to Scintillation are reduced by using multiple transmitters. Figure 8. MRV patent for using multiple transmitters to reduce scintillation.

9 In equation 2, ApNum is the number of separate transmit apertures, and Diam is the diameter of the receive telescope (or equivalent diameter of multiple receive lenses if used). For example, at 1 km range with 1 transmit aperture and a receiver diameter of 4 inches (100 mm), a margin of 14 db should be allocated for scintillation fade. Alternatively, at 2 km range with 3 transmit apertures and a receiver diameter of 8 inches (200 mm), a margin of only 4 db is necessary. The approximate scaling of needed margin (in db) increasing linearly in range, inversely with the area of the receiver, and inversely with the number of transmit apertures appears to be borne out experimentally. In addition, an asymptotic minimum margin of 2 db seems necessary (even though one would not expect to need this in the laboratory). A final modification can be made to the required scintillation fade margin to take account of the fact that scintillation tends to be reduced in low visibility conditions, because there is less differential heating of the ground. We do that in an ad hoc manner by multiplying by an extra factor of exp(-10/attenuation) once the scintillation margin has been converted to db/km by dividing by the link range. Here, Attenuation is the atmospheric attenuation due to the weather condition in db/km. Thus, for instance if the visibility is 1 km, and Attenuation is 13 db/km, our scintillation margin for the single transmitter link above would have been 2dB + 12dB/km prior to adjustment, and 2dB db/km after the adjustment for the lower visibility condition. (Basically, we are saying that if the visibility is poor, more of the link margin can be used to overcome atmospheric attenuation, and less is necessary for the scintillation). Now we are ready to present system parameters for four sample FSO systems in preparation for a performance comparison. Relevant parameters are shown in Table 1. It is assumed that all four systems are designed for a data rate of 100 Mb/s (or almost equivalently 155 Mb/s). The Medium and Long range systems in the table correspond fairly closely to systems available from a number of vendors. It is expected that during 2003 end users could expect to purchase and install such systems for around the $10,000 and $25,000 link prices shown, depending on desired features. For comparison, a very short range system is shown which might (for instance) consist of a low power Class 1 fiber transmitter expanding into a lens, and being collected by a 2 inch aperture with a PIN detector. For the short, medium and long systems, the divergence of the transmitters and field of view of the receivers is large enough to allow for normal mispointing of the system due to building sway if the equipment is mounted properly with good initial alignment. Thus, no active tracking is needed for these systems. At the other end of the cost spectrum, we present a Deluxe system which maximizes system performance which might be achieved on an FSO link with current technology. This system has a very narrow divergence of 100 microradians, and an active link alignment system to keep the mispointing angle less than 40 microradians. Four transmitters are used, each broadcasting 500 mw of power, well above the 10 mw typical of the medium and long range system. This power can be kept Class 1M eye safe by using a transmit wavelength of 1550 nm. Parameter Short Medium Long Deluxe Data Rate (Mb/s) Number of Transmitters Transmit Power/Aperture (mw) Wavelength (nm) Eye Safety Class 1 1M 1M 1M Receiver Diameter (m) Transmit Divergence (mrad, 1/e) Receiver Sensitivity* (nw) Allowed Mispointing (mrad) Installed Link Price Estimate (2003) $3K $10K $25K $200K *Power needed at front of telescope for Bit Error Rate, laboratory conditions Table 1. Operational parameters for four sample Free Space Optics Systems.

10 Finally, a large 12 inch receiver is used to collect most of the light out to a long distance. An alternate technology such as adaptive optics might allow for a smaller receiver to achieve this same function. Link margins as a function of range are calculated for these four sample systems as follows. (Results of the calculation are shown in figure 9). We present the calculation as a procedure rather than as an equation, both to show how easy it is (for instance to implement in Excel), and also to point out some things to watch out for. We steadfastly avoid using the normal microwave link equations, because those are generally based on the assumption that the divergence is simply a function of the diffraction limited transmit antenna aperture (whereas for FSO links the divergence is increased to the desired level to maintain pointing, and the aperture size is chosen for cost and eye safety reasons). Furthermore, for microwave calculations the geometric range loss factor is assumed to be wavelength dependent, which is simply due to an unphysical grouping of loss terms left over from the era of slide rules. The procedure we use is as follows. First we calculate the size of the transmit beam at the range of the link. To first order, this size equals the divergence of the transmitter multiplied by the range. Next we calculate the fraction of the transmitted beam (in vacuum) intercepted by the receiver. For this fraction we usually take the diameter of the receiver divided by the diameter of the transmitted beam and square the result. This procedure is accurate if the transmit beam is a cylindrically symmetric Gaussian (good approximation), if we define the divergence as the full width at the 1/e intensity points, and if the receive diameter is small compared to the beam diameter. More accurate calculations of beam overlap are sometimes necessary, especially if there are parallax or field of view constraints from a small detector. For the analysis here, we simply use the simple procedure, except that we make sure that the beam intercept fraction is never larger than 1 (which could give incredible link margins at close range if we do not make the adjustment). Next we calculate an excess power ratio which is the Transmit power per aperture multiplied by the number of transmitters multiplied by the beam intercept fraction divided by the minimum power needed at the receiver. From this, a preliminary link margin in db is calculated by taking 10 times the log of the excess power ratio. From this margin we subtract some loss due to beam mispointing. For a Gaussian this mispointing loss would be 8.7 db multiplied by the Margin (db/km) Short, No Scint. Medium, No Scint. Long, No Scint. Deluxe, No Scint. Short, w/scint. Medium, w/scint Long, w/ Scint. Deluxe, w/scint Range (meters) Figure 9. Available link margin vs. range for 4 sample FSO systems. Margin is shown with and without taking account of scintillation fade. At longer distances and for clearer weather, required scintillation margin causes the curves to roll down.

11 maximum mispoint angle divided by the divergence. (For instance, for 1 mrad mispointing and 2.5 mrad divergence, we would subtract 3.5 db). Then we subtract off 2 db for the fixed part of the scintillation margin loss. We are almost certainly being over-conservative by including this much fixed loss for mispointing and scintillation, but our systems have consistently exceeded the availability expectations we have set for our customers. Some vendors are less conservative in these calculations. Now we convert the link margin in db to a link margin in db/km by dividing by the link range. Then (finally) we reduce the link margin further to take account of the range dependent part of the scintillation. In figure 9, the link margin calculations are shown both with and without taking into account the margin needed for scintillation to show that this effect is most important at longer ranges in clearer weather conditions. As the systems become more expensive, the amount of link margin available for weather attenuation increases. To get to the most important results of this whole process, namely to compare link availability versus range for the different systems, we convert the attenuation in db/km to the Outage Fraction by using equation 1 (the average attenuation equation). The results of this transformation are shown in figure 10. On the right hand side, the equivalent link availability percentages are given, ranging from 90% (one 9) to % (the famous five 9 s amazingly claimed by some telecommunications carriers, corresponding to less than 1 hour of outage in 10 years, Ha!). Outage Fraction Link Range (meters) $3,000 $10,000 $25,000 Short Range Medium Range Long Range Deluxe $200, % 99.99% 99.9% 99% % 10 m 100 m 1 km 10 km Link Range Figure 10. Availability versus Range for 4 sample FSO systems. Availability is based on outages to be expected from an average of visibility statistics from different locations, calculated using the standard curve developed in this paper. For Four 9 s availability and higher, the cost and complexity of the deluxe system has minimal added benefit on achievable range. Availability 5. DISCUSSION AND CONCLUSIONS Figure 10 contains a large amount of very useful information for the FSO system design engineer and other personnel involved in setting product performance specifications. Such logarithmically scaled data has not generally been presented to the sales force or the end user, because most enterprise customers are concerned with ranges of a few hundred meters to a few kilometers, and availabilities between 95% and 99.9%. Before discussing this figure further, we

12 would like to point out that we are very cognizant of a wide range of technologies which can be used to enhance system performance at added cost. For instance, figure 11 shows hardware we built for a satellite to ground laser communications demonstration experiment at a range of 2000 km, a data rate of 1 Gb/s, and involving a satellite moving at 7 km/sec. In addition, we provided hardware for the first demonstrations of high speed ship-to-shore laser communications, 2.5 Gb/s data rate terrestrial links using FSO coupled to Erbium Doped Fiber Amplifiers, and 10 and 40 Gb/s multi-kilometer links using Wavelength Division Multiplexing. We have a number of patents relating to tracking technology which was incorporated in these systems. Nevertheless, over the years that we have manufactured FSO equipment we have seen the end user price point continue to drop, while data rates have increased. Thus, no fancy technology can be included in the hardware unless it is absolutely needed and the added cost is modest. Figure 11. MRV Satellite Lasercom Program. Hardware for a 2000 km satellite-to-ground lasercom demonstration experiment flew on BMDO s Space 1 Technology Aperture Research Vehicle 2. From figure 10 we can see that a fairly large performance improvement is possible by using a 10mW Class 1M system with a 4 inch lens and a high sensitivity Avalanche Photodiode (Medium Range) as compared to a 1mW Class 1 system with a 2 inch lens and a PIN detector (Short Range). However, a lot of product development will occur to provide cost effective systems intermediate in performance to these. Next we compare the Medium Range system to the Long Range system, which has four transmitters and an 8 inch receiver. If high availability (above 99.99%) is desired in a typical North American or European location, there is not a large increase in achievable range, and thus the lower cost medium range system would normally be chosen. If a range of more than 1 km is desired, however, the long range system becomes the only logical choice. In this case, the typical availability will be between 97% and 99.6%, depending on the range. This is generally adequate for enterprise applications, and can be enhanced to all weather performance for critical transmissions with a lower data rate RF backup. Further enhancements past the performance of the long range system will lead to availability curves between that of the long range system and the deluxe system, at ever increasing cost. For some applications these enhancements may be necessary as custom modifications, but they do not seem to make sense for standard commercial equipment. For instance, it is fairly straightforward and inexpensive to lower the transmitter divergence and increase the transmit power on a standard system to gain quite a bit of margin. The risk of lower transmit divergence is that re-alignment of the system may sometimes be necessary. At higher power, the system may no longer be eye safe. More expensive technologies will only be justified by a few niche applications where very long distance links at under 99% availability are needed, and the expense is not a limiting factor. For carrier applications which require high availability, the only way to achieve that in general in the population centers of North America is to use a network of short links. For instance, the $10,000 system achieves a link range of about 160m at % availability, while the $200,000 system achieves a range of about 280m. At this availability level, it will be much more cost effective to use two of the short range systems in series than to use something like the deluxe system. To summarize, we have developed a standard model average attenuation equation which can be used to compare the weather availability performance of different FSO systems in a realistic, consistent manner. Using this model, we have compared four different sample FSO links. We have shown that the added performance to be gained by incorporating expensive tracking or adaptive optics technologies will not generally justify the cost. We believe that simple parameterized models such as those presented here are most useful for the system designer, and we encourage the atmospheric propagation community to improve on the accuracy of these models while maintaining their simplicity. Time (sec)