Abstract
We present a broad review of the field of Optical Wireless, from early experiments through to today's high performance systems. Emphasis is placed on understanding the benefits and limitations of optical wireless, all of which ultimately define the applications. A variety of systems are examined, each incorporating a different technological solution to suit the particular application.

1 Prologue
It's a new day and work beckons. While eating breakfast you catch up on news delivered to your laptop via an indoor infra-red wireless link. Several wireless systems exist throughout your home, ranging from short distance point-and-shoot systems to cellular systems that cover whole rooms. Your laptop automatically selects the link that offers the best performance at that time. At breakfast it selects a short distance infra-red link, the base station for which is built into the table. As you roam through your home taking your laptop with you, it remains linked to the network by seamlessly switching between different infra-red cells. All these cells are interconnected within your home and linked to the high speed external network via a line-of-sight infra-red system to a communal 'pole' in the street.

As you enter your car it creates an in-car infra-red link with your laptop and interrogates your diary for the day. Your car learns that your first journey is to your office in the city so it accesses traffic information via its GSM link and displays on the dashboard navigation system, or speaks, a recommended route for you to follow. On your way to work you pull into a gas station. As you do so your car enters an infra-red cell and is instantly linked to the high speed network. Your car uploads its performance data to an agency which will contact you later if your car needs attention. At the same time your laptop uses the in-car infra-red system to establish a connection between itself and the external cell. This done, it takes the opportunity to contact your office to update your dairy and to up/download files from your desktop machine that it knows you will need. By the time you reach your office your laptop has exchanged Gbytes of data via a variety of wireless infra-red links and you are ready to face the challenges of the day.

During the day you make several personal calls on clients, all of whom have infra-red cells at their location, so you are on-line the instant you arrive. The buildings that you travel between are all located in the down town business area and are interconnected via high capacity line-of-sight infra-red systems that overlay the conventional cabled network. This enables you to up/download vast amounts of data very quickly, thereby maximising your effectiveness while with your clients.

And so on throughout the day.

Although this vision lies in the future, much of the implicit optical wireless technology is in fact available today, and in this paper we call upon our own work and that of others to present a broad overview of the field. We begin by describing the underlying principles of optical wireless, paying particular attention to the intrinsic benefits and limitations. We then examine a broad selection of systems, some experimental, others now available commercially, that are tailored to a variety of operational requirements. Performances and deployment issues are compared and contrasted, and we offer informed opinions on optimum solutions for a variety of applications.

2 Design fundamentals
A decade ago the optical fibre systems of the day were running at Gbit/s rates over single span distances exceeding 100km. A promising foundation for the development of optical wireless was therefore in place. However, the high performance of optical fibre systems is due in large part to the properties of the fibre itself. Remove the fibre, as in a wireless system, and the stable, low loss, guided propagation path is no longer available. Conveying light between the terminal stations in a controlled, reliable manner then becomes a challenge. In addition to this, optical fibre receivers, if applied to a free space environment, incur a very high coupling loss which reduces the overall power budget of the system. For these reasons and others the transition from optical fibre to optical wireless is not straight forward and new design solutions are needed.

2.1 Transmitter :
The optical fibre systems industry has spawned semiconductor lasers with broad bandwidths and high launch powers, features which are equally attractive to optical wireless systems, particularly for outdoor applications. However, for indoor applications lasers pose a potential safety hazard because they are a point source emitter (see below on eye safety). LED's, on the other hand, are large area emitters so can be operated safely, even at relatively high powers. They are therefore the preferred emitter for most indoor applications. To compensate for the lower powers that LED's generally emit, arrays of them can be used as the optical source. The penalty however is bandwidth. Whereas laser speeds extend into the Gbit/s regime, LED's are typically limited to 10Mbit/s, perhaps extending to 50Mbit/s for some specialist devices.

2.2 Eye safety :
Optical wireless systems, like their radio counterparts, can pose a hazard if operated incorrectly therefore a laser safety standard [1] has been established in which optical sources are classified in accordance with their total emitted power. The principal classifications are summarised in Table 1 for a point source emitter such as a semiconductor laser.

Outdoor point-to-point systems generally use high power lasers that operate in the Class 3B band to achieve a good power budget. The safety standard recommends that these systems should be located where the beam cannot be interrupted or viewed inadvertently by a person. Roof top locations or high walls are usual for this type of system.

Indoor systems pose a particular challenge because the safety standard recommends that they must be Class 1 eye safe under all conditions. Table 1 shows that for systems employing laser sources, launch powers must not exceed 0.5mW at the short wavelengths where most low cost devices operate. Indoor systems that use lasers therefore find it difficult to achieve a good power budget. However, by using LED's instead of lasers a much higher launch power can be used and still remain Class 1 eye safe. This is because LED's are not 'point source' devices as are lasers, they are large area devices, so if viewed the image on the retina of the eye covers a large area, and hence the power is diffused. Indeed, arrays of LED's can produce substantial launch powers and yet be Class 1 eye safe. This and cost are the two primary reasons why indoor systems mostly use LED's as the emitter, either singly or in arrays.

Lasers operating inside the Class 3B band can in fact be rendered Class 1 eye safe by passing their beam through a hologram which is sealed within the overall laser enclosure. The hologram breaks up the wavefront in the optical beam which diffuses the image of the laser spot on the retina of eye. The same hologram also shapes the beam to create a definable footprint, e.g., square or rectangular. We have demonstrated that a 40mW Class 3B laser can be rendered Class 1 eye safe [2], and with certain refinements to the hologram the same outcome could be achieved with a 100mW laser. This technique, although proven in the laboratory, has yet to be taken up by the industry.

2.3 Atmospheric loss :
The power budget and overall performance of a point-to-point free space link is strongly determined by atmospheric loss along the propagation path, which itself comprises free space loss, clear air absorption, scattering, refraction and scintillation.

Free space loss defines the proportion of optical power arriving at the receiver that is usefully captured within the receiver's aperture (Fig 1). A typical figure for a point-to-point system that operates with a slightly diverging beam would be 20dB, whereas an indoor system using a wide angle beam could have a free space loss of 40dB or more.

Fig 1. Schematic representation of free space loss.

Clearly free space loss is experienced by all forms of optical wireless systems, from short distance indoor systems to long distance outdoor systems. However, the remaining sources of atmospheric loss described below are entirely the domain of the latter.

Clear air absorption (the equivalent of absorption loss in optical fibres) is a wavelength dependent process that gives rise to low loss transmission windows centred on 850nm, 1300nm and 1550nm [3]. This process is essentially the same as that present in optical fibres, and hence the low loss windows coincide with those in fibres. This in turn means that the same opto-electronic devices can be used, which from a commercial point of view is crucial.

Scattering and refraction due to water droplets refers to the attenuating effect of rain, fog, mist and snow on the power reaching the receiver. This attenuation is continually in a state of flux so is most meaningfully quantified in terms of the percentage of time that it exceeds a given value (usually taken to be the point at which the error rate in the recovered data is just acceptable). Over a period of a year we undertook detailed measurements of attenuation over a 100m path in a rural area. Fig 2 shows some of the results obtained.

Fig 2. Seasonal atmospheric attenuation over a 100m path for a 1 year period.

Not surprisingly the worst performance was achieved during the winter months due to the high prevalence of rain, fog, mist and snow. Corresponding charts for an urban area show similar trends but with rather less variability between seasons and higher average attenuations.

A further study used data provided by the UK Weather Centre, spanning a period of 8 years from Jan'75 to Dec'83 [3]. Fig 3 shows some of the results obtained for three major UK cities.

Fig 3. Atmospheric attenuation averaged over 9 years.

It is apparent that atmospheric attenuation is consistently low for all three cities for 99% of the time. Put another way, a free space link spanning say 1km in any of these cities would achieve an availability (i.e., up time) of 99%. Fig 3 suggests that this figure could be improved to 99.5% given a suitable power budget etc, however 99.9% could be difficult to achieve. Of course, by reducing the link length and/or increasing the power budget, the availability would improve correspondingly.

Scintillation is the result of solar energy heating small pockets of air to slightly different temperatures, thereby creating regions of varying refractive index along the propagation path. This causes the optical signal to scatter preferentially at very shallow angles in the direction of propagation, whereupon multiple signals, all phase shifted relative to each other, arrive simultaneously at the receiver. This in turn causes the amplitude of the received signal to fluctuate rapidly by as much as 30dB if conditions are unfavourable. The power spectral density of these fluctuations typically spans 0.01-200Hz and hence can give rise to long bursts of data errors. Scintillation also distorts the wavefront of the received optical signal, causing the focused image at the photodiode to dance around the surface of the device. This requires a larger than normal photodiode to be used in order to ensure that the signal is never lost. Scintillation is generally not significant over distances less than 500m but increases rapidly with distance.

2.4 Receiver :
Optical fibre receivers mostly employ a transimpedance design because this affords a good compromise between bandwidth and noise, both of which are influenced by the capacitance of the photodiode. Because optical wireless receivers must use photodiodes with a significantly larger active area, designs which are tolerant of very high input capacitances are required. A transimpedance design could still be used, but in order to achieve a useful bandwidth in the presence of a high input capacitance the feedback impedance must be very low, which in turn increases noise and reduces sensitivity. An improved approach that is common in optical wireless receivers combines transimpedance with bootstapping, the latter of which reduces the effective capacitance of the photodiode. This allows a relatively high feedback impedance to be used which reduces noise and increases sensitivity.

Fig 4. Receiver sensitivity (at 155Mbit/s) in relation to photodiode type and area.

We have developed a variety of bootstrap receivers using PIN and avalanche photodiodes (APD) of different dimensions. The sensitivities of some of these receivers are summarised in Fig 4, all for a 155Mbit/s data rate. As expected the sensitivity improves (i.e., reduces in numerical value) as the photodiode area reduces because of the correspondingly lower capacitance. However, small area photodiodes incur a greater coupling loss due to the small aperture they present to the incoming beam, so a careful trade off between these factors is necessary to optimise the final performance.

Fig 4 also shows that a receiver with an APD gives a 10dB sensitivity advantage over a corresponding PIN receiver, which is consistent with optical fibre receivers. APD receivers however are more costly and require high operating voltages, hence are predominantly used in specialist systems where performance is key. Systems in which economy is a priority, such as most indoor applications, favour PIN receivers.

2.5 Interference from ambient light :
Because of the wide field of view (i.e., aperture) of a free space optical receiver, stray light in addition to the wanted optical beam can reach the photodiode. This is particularly the case with some indoor systems (see later on diffuse systems) in which the receiver can 'see' nearly everything above the plane of the photodiode. Ambient light raises the level of photonic noise in the receiver and hence can impair performance. Fortunately the properties of ambient light sources lend themselves to straight forward remedies.

First, by placing a narrowband infra-red filter over the photodiode the level of ambient light relative to the wanted beam will be significantly reduced. Consequently, photonic noise in the receiver will predominantly originate from the wanted signal, which is the optimum condition. Infra-red filters may be fabricated in glass or plastic, depending on the optical quality required and the application.

Secondly, the power spectral density of ambient induced noise extends from DC to typically a few 10's of kHz, exceptionally a few 100's of kHz, depending on the type of source. By conveying the data over the optical beam on a high frequency subcarrier, or more commonly by applying a line code that contains no low frequency components, interference is entirely avoided.

The strong DC content of ambient light also means that the dynamic range of the receiver could be impaired. By using the above infra-red filter and a receiver design which cancels or blocks any DC from the photodiode, this form of impairment is avoided.

3 Long distance systems (i.e., 100m to 5km).
Earlier in this paper we outlined the strong influence that atmospheric loss has on spans greater than say 500m. These systems must therefore have a good power budget to ensure error free transmission, particularly so if high speed operation is required (e.g., 155Mbit/s or higher). This can be achieved by addressing each of the three system elements: transmitter, propagation path, receiver.

A high power Class 3B semiconductor laser can be used as the source, emitting perhaps 100mW (+20dBm). Alternatively, a lower power device can be coupled to an erbium doped fibre amplifier (EDFA) to achieve the same end result. We utilised the latter approach in a trial 155Mbit/s link that spanned 4km between Imperial College and University College in London [4]. The use of an EDFA dictated a laser wavelength of 1550nm, which enabled the system to be Class 1 eye safe while still affording a useful launch power of 10mW (Table 1). The overall system design is depicted in Fig 5.

Fig 5. Experimental 4km point-to-point system.

A good power budget requires that the overall propagation loss is minimised. Little can be done to alleviate the various atmospheric losses, however steps can be taken to minimise free space loss (Fig 1). In our experimental system we achieved this by using an astronomical telescope at each end (Schmidt-Cassegrain, 20cm aperture). The resulting beam diameter at the receiver was typically 2m, which equates to a free space loss of 20dB. It was possible to reduce the beam diameter to only 0.5m (8dB free space loss) but then maintaining beam alignment became more difficult.

Beam alignment is a critical factor in long distance point-to-point systems because it will drift with temperature changes and fatigue in the anchorage assemblies. Some means of automatically maintaining alignment by mechanically steering the beams is desirable, particularly in commercial installations where maintenance intervention must be kept to a minimum or even avoided altogether. A long distance system developed and marketed by Canon [5] embraces this requirement to the fullest extent. After crude manual alignment, each end of this system locks on to the other and maintains alignment even in the event of the transmitter/receiver units being jolted or moved off-axis by several degrees. Of course, such robustness comes at a premium.

Receiver sensitivity is the final factor in achieving a good power budget. Since the best sensitivity is delivered by an APD fronted receiver (Fig 4), these are generally used in long distance systems. The relatively high cost of these receivers is offset by the commercial advantage afforded by the greater reach. In any case, receiver cost tends to be a small factor in the cost equation for long distance systems.

Fig 6. Capacities and distances for various point-to-point systems.

The date rate -v- distance of our 4km system and others we have developed are plotted in Fig 6, together with commercial systems from Canon [5], AirOptics [6] and CableFree [7]. It is unlikely that long distance point-to-point systems will afford reliable spans much beyond 4-5km because of the extreme vagaries of atmospheric loss. Systems such as the Canon and our own are indeed capable of longer distances (e.g., 8km has been reported) but only when the weather, season and time of day combine to produce ideal conditions. This might equate to an availability of only 1-10% averaged over a year, whereas users require at least 99%. Indeed, some applications in telecommunications require 99.99% or even 99.999% (referred to as 5 nines). As Figs 4 and 5 show, obtaining that extra percentage point requires a significant improvement in the power budget parameters, which in turn will be reflected in a higher cost. We therefore believe that the largest market for long distance point-to-point systems will involve sub-1km spans, with the >1km market largely confined to specialist applications. Of course in countries such as the USA which enjoy a predominantly dry climate and contain large open expanses, the balance and size of these markets will be different.

4 Short distance systems (i.e., less than 100m).
All of the design and performance factors described above for long distance systems apply equally to short distance systems but with one obvious exception - atmospheric loss has a much reduced effect, and none at all for indoor systems. The power budget is therefore determined almost entirely by the transmitter launch power, free space loss, and receiver sensitivity.

4.1 Point-to-point systems :
Outdoor short distance systems, used perhaps to deliver high capacity links between neighbouring buildings, enjoy the luxury, as do long distance systems, of utilising a high power Class 3B emitter since these systems will be located away from people. A good power budget can therefore be achieved even with relatively simple components. This, combined with the low atmospheric loss, means that a higher free space loss can be tolerated in return for the convenience of a larger beam diameter at the receiver. Setting up and aligning these systems is then significantly easier than their long distance cousins, and furthermore, automatic alignment and tracking is not required. Consequently, overall complexity and cost is significantly reduced. Fig 6 includes the results of some of our experiments in his area. The 1Gbit/s result over 40m [8] is particularly interesting as it shows the order of capacity that can be achieved with today's technology.

Indoor point-to-point systems do not differ from the outdoor variety in their operating principles, however in practice their designs are invariably very different. Firstly, they must be Class 1 eye safe, which generally means that the optical source is an LED. This in turn limits the capacity to typically a few Mbit/s. On the other hand, indoor systems need none of the weather proofing that outdoor systems require and only operate over short distances, hence they can be produced very cost effectively. A good example of such a system is manufactured by JVC [9]. Here a capacity of 10Mbit/s can be conveyed over a span of around 20m (Fig 6). Such systems could be used, for example, to extend a 10Mbit/s LAN port to a different part of an office where no convenient port exists, or to link two separate offices via a link corridor.

4.2 Telepoint systems :
Indoor applications, being fundamentally confined to short distance spans, are uniquely appropriate to optical wireless systems that utilise wide diverging beams rather than the narrow beams of point-to-point systems. Such systems are sometimes referred to as optical telepoint systems (Fig 7) and afford a number of attractive features. Most importantly, each 'cell' created by an optical telepoint base station can be shared by as many users as can be sensibly accommodated within the cell.

Fig 7. Optical telepoint.

We have experimented with a number of telepoint systems, creating cell diameters of 10m down to 0.5m. The larger cells might be deployed in open offices and public areas such as foyers, libraries, waiting rooms, hospital wards, etc. The smaller cells (e.g., 1m or less) are particularly suited to applications in which a cell is dedicated to a single user, for example a desktop or a telephone booth.

A good example of a commercial telepoint system, again manufactured by JVC [9], generates a cell of about 10m diameter, which is sufficient for up to perhaps 6 simultaneous users. Each cell delivers a capacity of 10Mbit/s (Fig 6) shared between the users via conventional LAN protocols. Line of sight paths are required between the base station and all the user terminals. As we will shortly discover, line of sight propagation is in fact vital if capacities more than a few Mbit/s are to be reliably achieved.

Fig 8. Diffuse optical wireless system.

4.3 Diffuse Systems :
The converse to indoor systems that employ line of sight paths are those that employ diffuse paths (Fig 8). In this group of systems the beams, as well as radiating over a wide angle, are also allowed to reflect off surfaces and objects in the vicinity such as walls, ceiling, floor, furniture, etc. In addition, the field of view of the optical receivers in the base and user stations is widened to such an extent that line of sight as well as reflected light can be detected with the same ease. Such an arrangement goes some way towards guaranteeing that a signal path is always present, regardless of obstacles or people impeding the line-of-sight path. This arrangement also allows roaming to some degree. The penalty however of diffuse transmission is a much reduced capacity compared with line-of-sight systems. This is entirely a consequence of the multiple signal paths reaching the receiver which cause classic pulse spreading and inter-symbol interference. Research has shown [10] that the theoretical capacity of a diffuse system is a function of many factors such as room size and geometry, the fabric and distribution of furnishings, the placement and orientation of the base and user stations, etc. Generalising, this work predicts an upper bound of about 25Mbit/s in a room 10 metres on a side, although higher rates have been demonstrated under particular conditions.

Interference from ambient light is a particular issue for diffuse systems because of the extremely wide field of view of the receivers. However, as outlined earlier, the use of infra-red filters and robust signal formats minimises any performance penalty that may arise.

A good example of a diffuse system is manufactured by Spectrix [11]. A single base station can comfortably irradiate a room 10 metres on a side and deliver 4Mbit/s, which is shared by the users within the cell. This system, like the JVC system described above, utilises LED emitters to be Class 1 eye safe and to keep cost low.

5 Tracking architecture for indoor systems.
In many indoor applications it is desirable to deliver capacities far beyond the 4-10Mbit/s described above. However, as explained earlier, to do so requires lasers to be used as the emitter because of their higher speed, and to remain Class 1 eye safe they must be operated at low power levels (Table 1). Given these constraints it becomes necessary to reduce the diameter of the down-beam such that it illuminates only a single user station. Then, by designing the base station to produce several narrow down-beams simultaneously, with each affording a line of sight link to one of several user stations, a high capacity can be delivered to multiple users within the same cell. The beams would be reused as users enter and leave the cell, the maximum number of users at any time being determined by the number of simultaneous down-beams that the base station can produce. In practice this number would be chosen to match the maximum number of users that can be sensibly accommodated within the cell. The beams could also be made steerable thus allowing a degree of roaming within the cell.

Fig 9. Concept behind the tracking system.

We developed an experimental system, operating at 155Mbit/s, to investigate the practicality of this concept [12]. The emitter in each transmitter was a laser array, and similarly the receivers used a PIN photodiode array (Fig 9). Device availability dictated that only a single base station and user station could be constructed, and the laser and photodiode arrays were one dimensional rather than the ideal two. Upon arrival of the user in the cell, pilot LED's on the base and user stations enable the down and up-beams to locate and lock on to the corresponding receivers. Thereafter alignment is maintained by monitoring the power detected by adjacent photodiodes in the array then re-aligning the beam as required by activating the appropriate laser in the array. At this point a single laser in each array would be active, the particular choice depending on the required launch angle of each beam. This process would be repeated for other users entering the cell, resulting in additional lasers in the base station array being activated. As a user begins to move within the cell the detected beams would migrate from one PIN to the adjacent one in the array. This data, when fed into a tracking algorithm, would cause different lasers to be activated so that the beams remain directed at the corresponding receivers. This process would continue until the user was again stationary or had left the cell.

Of course the converse approach to a tracking architecture is to flood the floor area with stationary overlapping beams, with each cell wide enough to accommodate only a single high capacity user. Roaming is then achieved by handing over from cell to cell, akin to cellular mobile telephony. We are investigating this technique for use in indoor locations such as finance centres and trading rooms where there are many mobile users, each requiring a very high capacity link.

These experiments have proven that very high capacities can be delivered to multiple users in an indoor, Class 1 eye safe environment. Commercial systems with these capabilities have yet to appear for reasons that include the immaturity and high cost of some of the devices, particularly the arrays for the tracking architecture.

6 The IrDA standard for very short distance point-to-point systems (e.g., 1m or less).
An exciting development in recent years has been the arrival of very short distance 'point and shoot' optical wireless systems for laptop computers, PDA's, palmtop's, printers, calculators and mobile phones. These wireless systems are commonly referred to as IrDA systems in recognition of the IrDA standard that they embody. The Infra-red Data Association (IrDA) was established in 1993 with the aim of producing an open standard for wireless data communication using mature, commercially available infra-red components. The resulting protocol provides a simple, low power, low cost, reliable means of wireless infra-red communications for a broad range of computing, communications, and consumer devices. The IrDA now has some 160 member companies around the world, and IrDA systems are already resident on over 97% of new portable computers (this includes laptops, PDA's, palmtops, etc).

The IrDA standard presently specifies the following parameters :-

• data rates from 128kbit/s up to 4Mbit/s (raising this upper limit to several 10's of Mbit/s is being considered),
• maximum operating range of 1m (increasing this to 10m is being considered),
• 30 degree cone from the transmitter that equates to a 50cm beam diameter at the receiver positioned 1m away,
• very wide receiver field of view.

All of these features ensure that a useful data rate is available over a wireless link that requires only the crudest of alignment.

7 Comparison with radio systems.
It is worth comparing the performance of the various short distance (primarily indoor) optical wireless systems we have discussed with their radio cousins because recent advances in the latter have narrowed the gap to the point that, in some instances, the optimum choice is not obvious. As we have seen earlier in this paper, commercial indoor optical wireless systems can deliver 4Mbit/s (diffuse) and 10Mbit/s (line of sight) into a room of some 10m on a side. The technology exists to improve on these rates and cell size but commercial systems that exploit this have yet to appear.

Today's radio (or wireless) LAN systems can deliver 2Mbit/s to multiple users over distances of typically 50-100m depending on terrain, with the promise of 8Mbit/s in the near future. Standards are being produced for next generation systems to deliver 10's of Mbit/s.

Radio systems will almost always afford a greater reach and wider coverage than optical wireless systems because a higher transmitter power can be used and the receivers can take full advantage of sensitive heterodyning techniques. On the other hand, radio will always be a narrower bandwidth medium than optical, although this is not apparent in today's commercial systems because optical wireless manufacturers have yet to fully exploit the available bandwidth.

8 Seamless hand over.
It is clear that the optical and radio systems described in this paper are complementary in many respects. Optical wireless (telepoint in particular) can deliver a high capacity to multiple users within a small cell, whereas radio can produce a much larger cell but will deliver a lower capacity. It is therefore easy to envisage scenarios in which both types of system would benefit from being used in combination. For example, a large open plan office would deploy optical telepoint systems to establish several discrete high capacity cells, each spanning a small cluster of perhaps 4-6 desks, and additionally a single radio system would be deployed to cover the whole office. Each desk might also be equipped with an IrDA base station.

Clearly it would be desirable for the workers in such an office to access the network via whichever wireless link delivers the highest capacity to their laptops (or some other portable device) at that time. Furthermore, it would be equally desirable to have sufficient intelligence within these laptops to automatically and seamlessly hand over from one wireless scheme to the other as the workers roam throughout the room, thus maintaining a link at all times. A typical sequence might then be: a worker at their desk accesses the network from their laptop via 10Mbit/s telepoint or 4Mbit/s IrDA; upon leaving the cell their laptop hands over to the office-wide radio cell which delivers a 40kbit/s link (from 2Mbit/s shared between 50 users); upon returning to a telepoint or IrDA cell the laptop hands over and the original data rate is restored; and so on. The intelligence within the laptop, as well as managing the hand overs, would also take account of the different capacities by tailoring the quality of service delivered to the user. For example, streamed video (vision and sound) at the desk via telepoint or IrDA might be reduced to sound only in the radio cell, then back to vision and sound upon return to a desk.

Of course, hand over need not be confined to indoor wireless technologies. GSM can also be included to great effect. A typical sequence might then be: a worker at their desk accesses the network from their laptop via 10Mbit/s telepoint or 4Mbit/s IrDA; while roaming the office their laptop hands over to the radio cell which delivers a 40kbit/s link; upon leaving the office or even the building their laptop hands over to the 9.6kbit/s channel provided by the GSM network; and so on.

We are investigating a number of hand over schemes that can support these scenarios. For example, we have demonstrated seamless hand over between IrDA and GSM, and IrDA and DECT. By building on this work hand over schemes that have the ubiquity envisaged above and in the prologue to this paper will become practicable.

Closing remarks
We have summarised the principal developments to date in the field of optical wireless, with particular emphasis on the capabilities and opportunities presented by today's systems. It is clear that the technology has reached a good degree of maturity and in the coming years we will witness the large scale deployment of these systems.

References

1. IEC825:1993, Safety of Laser Products. Equipment classification, requirements and users guide.
2. P Eardley, D Wisely, D Wood and P McKee, "Holograms for optical wireless LAN's", IEE Proc. on Optoelectronics, Special Issue on Free Space Optical Communications, Dec'95.
3. P Smyth, P Eardley, K Dalton, D Wisely, P McKee and D Wood, "Optical Wireless - A Prognosis", SPIE'95.
4. D Wisely, M McCullagh, P Eardley, P Smyth, D Luthra, E Miranda and R Cole, "4km terrestrial line of sight optical free space link operating at 155Mbit/s", Proc. "Free Space Laser Communication Technologies VI", SPIE Vol. 2123, 1994.
5. CanoBeam (www.usa.canon.com).
6. AirOptics (www.jolt.co.il).
7. CableFree (www.cablefree.co.uk).
8. P Eardley and D Wisely, "1Gbit/s Optical Free Space Link Operating over 40m - System and Applications", IEE Proc. on Optoelectronics, Special Issue on Free Space Optical Communications, Dec'95.
9. JVC VIPSLAN-10 (www.jvcinfo.com).
10. F R Gfeller and U Bapst, "Wireless in-house data communication via diffuse infra-red radiation", Proc. IEEE, Vol. 67, No. 11, Nov. 1979.
11. SpectrixLite (www.spectrixcorp.com)
12. D Wisely, "A 1Gbit/s Optical Wireless Tracked Architecture for ATM Delivery", IEE Colloquium, London, Feb'96.