The present invention relates to an optical network and in particular to a network suitable for carrying TDMA (time division multiple access) signals. A network embodying the present invention might be used, for example, in an access network within a broadband optical telecommunications system, or in a local area network (LAN) for interconnecting computer systems.

As discussed in our co-pending European application no.9604020.9 (BT case A25109), the increasing power of computer systems in terms of processor speeds and storage capacity has made it possible for conventional personal computers to handle multimedia applications which involve real-time video and computer graphics. The high bandwidth data associated with such applications place heavy demands on the network and the performance of conventional LANs has failed to keep pace. Optical networks using synchronous TDMA potentially offer a far higher bandwidth than conventional LANs and so might be used as a high speed LAN. However, in practice, current proposals for optical TDMA networks suffer a number of limitations.

A key design problem in the construction of optical TDMA networks is the need to distribute an optical clock signal to the different nodes on the network. This clock signal is required to enable a receiver in a node to read data from an appropriate slot, and to enable a transmitter to write to the appropriate slot. Conventionally, the optical clock signal has been transmitted together with the data The clock signal is distinguished from the accompanying data by, for example, a different polarisation state or by a different wavelength. Each node then requires a stage for separating the clock from the data. One example of this conventional approach to optical TDMA is disclosed in the paper by J G Zhang in Pure Appl. Opt. Ill (1 994) 609-614. This discloses a network which uses a star topology and in which a clock signal at the frame rate is transmitted at a different wavelength to the data. In this system, the need to demultiplex the incoming frames to recover the clock adds undesirably to the complexity of the node and places constraints on the network topology.

A further example of a prior art network is disclosed in the paper by Nowatzyk & Prucnal published in the proceedings of 22 Annual International Symposium on Computer Architecture, June 1 995. This proposes the use of a clock or "framing pulse" which is distinguished from the data pulses by a different polarisation state. The receiver section of each node then requires a device to separate the data from the framing pulse. However, using such polarisation multiplexing techniques, it is found that polarisation characteristics of all the optical elements of the network need to be tightly controlled if the network performance is not to be degraded by interaction between the clock and the data. European patent application number 873081 69 9 (EP-A-0268355) discloses an optical communications network in which an optical synchronisation signal is transmitted on one optical, modulated at the customer premises, and transmitted onto a second optical line which is locateα separately from the first optical line. However, while such an approach might have been possible for use at relatively lower bit rates, recent work has been directed to achieving much higher bit rates, in excess of 1 Gbit per second. At these higher bit rates schemes such as that disclosed in EP-A-0268355 fail as a result of timing variations between the two optical lines.

According to a first aspect of the present invention there is provided an optical network comprising:

(a) an optical bus including at least two co-located optical fibre waveguides;

(b) a plurality of transmitter nodes, each transmitter node including a transmitter stage connected to the optical bus and arranged to receive an optical clock signal from one of the fibre waveguides and arranged to write an optical data signal, in a time slot synchronised to the optical clock signal, onto another one of the fibre waveguides; and

(c) a least one receiver node, the receiver node including a receiver stage for receiving an optical data signal which was output onto the optical bus by the transmitter stage of a transmitter node. The present invention provides an optical network having a construction and topology markedly different from those of conventional networks Through

using a bus including two optically separate but physically co-located waveguides, the invention provides a local clock at each node which provides an accurate timing reference for reading or writing data, irrespective of the position on the bus of the node. The different waveguides may be different fibres located within a single cable, or may be different cores of a single fibre. The use of separate optical paths avoids the problems of crosstalk and susceptibility to noise associated with conventional networks in which clock and data are multiplexed on a single path using, for example different polarisation states. It is found that by co- locating the waveguides timing errors can be reduced to low levels, allowing operation at very high bit rates.

Commonly, a transmitter node and receiver node may both be present at a single location. For example, this would be the case where the transmitter and receiver nodes are part of a network interface for a computer which is connected to an optical LAN and which needs both to transmit and to receive data via the LAN. The invention is not however limited in this respect, and also encompasses implementations in which transmitter nodes and receiver nodes are at different locations.

Preferably the transmitter stage includes a modulator and is arranged to generate the optical data signal by modulating the clock signal received from the said one of the waveguides .

The preferred implementation of the present invention further simplifies the construction of the network nodes, by avoiding the necessity for a local optical source for data generation. Instead, the transmitter in a node takes an optical clock signal from one fibre in the bus, modulates it, and places the signal back on another fibre in the bus in the appropriate time slot.

Preferably the transmitter stage is arranged to receive an optical clock signal from a first one of the optical fibre waveguides which in use carries a write clock, and the receiver stage is arranged to read an optical clock signal from a second one of the optical fibre waveguides which in use carries a read clock. The use of separate clocks carried on different fibres enables savings in the power budget over the extent of the network. An optical data signal produced by modulating the write clock may be carried on the same waveguide as the read clock, in which case the write and read clocks are arranged to have different

wavelengths. Alternatively, the data may be carried separately from the two clock signals on a third co-located optical fibre waveguide.

The receiver nodes may be coupled to the bus via a star network. As described in further detail below, this offers the advantage of greater fault tolerance and robustness.

According to a second aspect of the present invention, there is provided a method of operating an optical network comprising an optical bus comprising at least two co-located optical fibre waveguides, a plurality of transmitter nodes connected to the optical bus, and a receiver node, the method comprising: (a) at a transmitter node, receiving an optical clock signal from one of the fibre waveguides, and writing an optical data signal, in a time slot synchronised to the optical clock signal, onto another one of the fibre waveguides; and

(b) at a receiver node, receiving an optical data signal which was output onto the optical bus in step (a) . The present invention also encompasses a local area network (LAN) suitable for interconnecting a plurality of computer systems and using a network in accordance with the first aspect.

Systems embodying the present invention will now be described in further detail, by way of example only, with reference to the accompanying drawings in which:

Figure 1 is a schematic of a network embodying the present invention; Figure 2 shows a first example of a node for use in the network of Figure 1 ;

Figure 3 shows a second example of a node for use in the network of Figure 1 ;

Figure 4 shows a third example of a node for use in the network of Figure

1 ,

Figure 5 shows a first example of the construction of an optical bus for use in the network of Figure 1 ; Figure 6 shows a second example of an optical bus;

Figure 7 shows a third example of an optical bus; and Figures 8a and 8b show a variable delay stage for use in one of the receivers of Figures 2-4;

Figure 9 is a schematic of a network including receivers connected in a star topology,

Figure 10 shows a second example of a network with receivers connected in a star topology, Figure 1 1 shows a third example of a network with receivers connected in a star topology,

Figure 1 2 is a circuit diagram of a transmitter stage;

Figure 13 is a circuit diagram for a receiver stage;

Figure 14 is a circuit diagram of a second example of a receiver stage; Figure 1 5 is an example of a master clock source;

Figure 1 6 is a timing diagram showing channels carried on the optical bus, and

Figure 1 7 illustrates a dark pulse source

An optical network comprises a number of nodes N 1 , N2, N3 connected to an optical fibre bus 1 . In the present example, the network is a local area network (LAN) and a number of personal computer workstations PC1 , PC2, PC3 . .. are connected via the optical bus to each other and to a network server 2. Although, for clarity, only three nodes are shown, in practice the network may support many more nodes The network uses a structure termed a re-entrant bus topology. Each node has two connections to the bus, with the transmitter connected to an upstream portion of the bus, and the receiver connected to a downstream portion Each transmitter outputs a signal in a different time slot, so that a TDMA frame is assembled on the bus. Figure 1 6 shows one example of the TDMA frame In this example the output frame contains 1 6 channels at a separation of 25ps

Figure 2 shows schematically the connections between a node 200 and the bus 1 in a first example. The bus includes two optically separate but physically co-located optical fibre waveguides 21 , 22. The transmitter includes a modulator 23 which is coupled to the first waveguide 21 carrying the optical clock. The modulator is coupled via a 50:50 coupler. The modulator 23 then modulates the optical clock signal with data, and transmits the modulated data signal out onto the second optical fibre waveguide 22.

Figure 1 2 is a circuit diagram showing in detail the construction of the transmitter in this example. In the diagram, dotted lines show electrical paths, and solid lines show optical paths. A first optical coupler 1 21 taps the optical clock signal from the clock fibre 21 . A second coupler 1 22 sends part of the clock to a receiver 1 23 and part of the clock into the modulator 23. The receiver 1 23 converts the optical signal into the electrical domain. Its output is filtered by an electrical bandpass filter 1 24. The bandpass filter is centred at the clock frequency. The output of the bandpass filter 1 24 is an electrical sine-wave at the clock frequency. This is amplified by a first amplifier AMP1 . The amplified signal is input to a network interface card IFC. The network interface card links the node, for example, to a local workstation which receives/transmits data via the bus The network interface card IFC synchronises its data output to the electrical clock signal received from the amplifier AMP1 The data output from the network interface card IFC is amplified by a second amplifier AMP2 and applied to the modulator 23 so as to modulate the optical clock. The modulator is advantageously polarisation insensitive and may, for example, be an electro- absorption modulator. The optical data from the modulator 23 is coupled into the data bus via a third coupler 1 26. Optionally, there is an optical delay line before coupler 1 26 The delay line 1 25 allows selection of the time slot on the bus into which the data is inserted.

Suitable commercially available components for implementing this circuit are as follows: couplers 1 and 3 - 50:50 coupler P22S1 5A50; coupler 2 - 90: 10 coupler P22S1 5A1 0 both from Sifam Limited. Receiver: type PDC2201 , BT&D/Hewlett-Packard; bandpass filter - BA-2488-100 Daden-Anthony Associates (CA, USA) Amp 1 : type 60_20_1 , Nucletudes SA (Les Ulis, France) . Amp 2 - WBA000108D ERA Technologies (Leatherhead, UK). Optical delay line - type ODL- 600 Santec (Aichi, Japan 485). Alternatively, the optical delay line may be a word-forming network as described in the present applicants co-pending International application no. WO 94/21088. The receiver stage of the node is coupled to a downstream portion of the bus. Data is read from the bus using an optical AND gate 24 and a variable delay stage 25 to select the appropriate TDMA channel. A clock signal is coupled from the first optical fibre 21 into a first input of the AND gate and the data signal is

coupled via the variable delay into the second input of the AND gate. Alternatively and equivalently, a variable delay may be applied to the data signal The AND gate may be an SLA-NOLM or may be an integrated semiconductor device. Appropriate devices are described in the applicant's International Application PCT/GB 95/00425.

Figure 1 3 shows in detail the construction of the receiver stage in this example. A first coupler 1 31 is used to tap the optical clock signal from one of the optical fibre waveguides. The clock signal passes through an optical delay line 25 and optical amplifier 1 33. The delay line is used to align the optical clock with a required data channel. The optical amplifier 1 33 increases the power of the optical clock to the level required to drive the optical gate. A second optical coupler 1 36 taps data from the other optical fibre waveguide. The output of an optical AND gate 1 32 is converted to the electrical delay by a receiver 1 34, amplified in electrical amplifier 1 35 and fed to the network interface card IFC. The components for implementing the circuit are as for the transmitter stage discussed above. The optical AND gate may use four wave mixing in an SLA as described in the above-cited International application PCT/GB 95/00425. Alternatively the optical AND gate may be an integrated Mach-Zehnder device as described by Ludwig et al, Electronics Letters, Vol. 32, page 327, 1 996. Figure 14 shows a second alternative implementation of the receiver stage using, in this case, an electro-optic gate. The optical clock is tapped at a first coupler 141 and converted to the electrical domain by a receiver 142. The signal is filtered in a bandpass filter 143 and amplified in a first amplifier AMP 1 . The resulting signal is input to an electrical channel selector (ECS) as described below with reference to figures 8a and 8b. The output from the electrical channel selector 1 45 passes through a second amplifier AMP 2 and a comb generator 146 before being applied to an electro-absorption modulator (EAM) 147. The applied electrical signal gates the electro-absorption modulator to select one data channel from the incoming data channels. These data channels are tapped from the optical fibre by a second coupler 148. The output of the electro-absorption modulator is converted to the electrical domain by a receiver 149 and is amplified by a third amplifier AMP 3 before being input to the network interface card IFC.

Suitable components for the receivers, filter and couplers are as described above in relation to the transmitter. Other suitable components are as follows: Amp 1 - type 60_20_1 - Nucletudes SA, 91 944 (Les Ulis, France); Amp 2: Mini- circuits type ZHL-42 - Mini-Circuits Europe (Camberley, Surrey), Amp 3: type WBA000108D - ERA Technologies (Leatherhead); comb generator: HP33000 series - Hewlett-Packard (Bracknell, Berkshire) . An appropriate EAM is described in the paper by D. G. Moodie et al., Electron. Letts., 3 Aug 1 995, 31 , no.1 6.

One example of a variable delay stage suitable for use in the receiver stage of Figure 14 is shown in Figures 8a and 8b. As described in our above-cited co- pending European application, this delay stage is designed to carry out channel selection in the electrical domain, while the reading of the TDMA channel remains in the optical domain. Figure 8a shows how the electrical channel selector (ECS) drives a local picosecond source via an impulse generator. This then provides one of the two inputs to the optical AND gate. Figure 8b shows in detail the construction of the electrical channel selector. In the selector the optical clock signal from the first waveguide is detected using a detector 82 which might be, for example, a PIN photodiode. After amplification, the signal is filtered to generate a clean electrical sine wave. The signal is then input to a delay stage 83 comprising a series of electrical AND gates LG arranged in a linear array. The array is implemented as a low-cost chip available commercially as NELNLB6202. The AND gates control access to a microstπp delay line. The delay line is accurately stepped in delays to the channel separation of the LAN. For a system operating at 40Gbιt/s, the channel delay T equal 25ps. The array of AND gates is controlled via an input from a demultiplexer 54. In this example, the demultiplexer is a NEL4705 device manufactured by NEL. The demultiplexer converts an incoming serial delay select word generated by the workstation connected to the node into an appropriate gating signal for the AND gate. In this way the workstation selects the appropriate delay to transmit data on its respective channel.

The construction and operation of the variable delay stage are described in further detail in our co-pending European application cited above, the contents of which are incorporated herein by reference. As an alternative to the ESC, a VCO

(voltage controlled oscillator) may be used in an analogous manner to select the channel.

Figure 5 shows a first example of the physical construction of the optical bus. This example uses a blown fibre cable construction. Four optical fibres 51 a- 51 d, each of which is a standard telecommunications fibre of 1 25 microns diameter, are located within the outer jacket 52 of the cable. Each fibre is surrounded by a 250μm polymer jacket 53 Aramid yarn fibres 54 in the interstices of the optical fibres provide reinforcement for the cable. Any two of the fibres, for example those referenced 51 a and b, may be used to provide the two fibres 21 ,22 of the network of Figure 2. As described below, for networks using other configurations, 3 or more fibres may be used. Blown fibre cable construction is described in further detail in the present applicant's European patent no. EP 108590.

Using an optical bus constructed as described above, it is found that the relative jitter between optical signals in different ones of the optical fibres is negligibly small, for example over a cable of length 1 km the jitter is less than 2ps. This allows a clock transmitted on one of the fibres to provide a timing reference for data signals transmitted on one or more of the other fibres, as discussed above in relation to Figure 2.

Figure 6 shows an alternative construction for the optical bus. In this example the bus is formed from a steel-cored cable, with 8 optical fibre waveguides 61 a-h surrounding the steel core 62. As in the first example the waveguides are formed of standard telecommunications fibre of 1 25μm diameter. Each is surrounded by a 250μm inner jacket and a 900μm outer jacket. The waveguides and the core are surrounded by a 2-ply sheath 63,64.

Figure 7 shows another alternative construction for the optical bus. This example uses a twin-core optical fibre. Two cores, each of 3.8 μm diameter are separated by 37μm. Each of the cores provides one of the two waveguides 21 ,22 of Figure 2. A suitable twin-core fibre is described in the paper by B. K. Nayar et al., "All-optical switching in a 200-m twin-core fibre non-linear Mach-Zehnder interferometer", OPTICS LETTERS, 1 6 , 6, 3.91 , pp 408-410. Although in the example of Figure 2, a single clock is used for both writing data at the node transmitters and reading data at the node receivers, other arrangements are possible. Figure 3 shows an implementation in which separate synchronised clocks are used for reading and writing. The two clocks are at

different wavelengths. In this example, the write clock output onto the first waveguide 31 is at 1 560nm and the read clock output onto the second waveguide 32 is at 1 540nm The two clock signals must have low relative jitter and may be derived from a single master source by super-continuum generation. In this technique, a short pulse from an optical source is amplified to a level where non¬ linear effects such as self-phase modulation result in the spreading of the spectrum of the pulse. For example, a pulse having an initial bandwidth of around 3nm may be spread over 100nm. Optical filters may then be used to carve out from the spectrum two or more pulses in different wavelength regions. The relative timing of these two pulses is fixed. This technique is described in further detail in Moπoka, Uchiyama, Saruwataπ, "Multiwave Picosecond Pulse Source with Low Jitter and High Optical Frequency Stability Based on 200nm Supercontinuum Filtering " Electronics Letters, June 22 1 995, vol. 31 , no. 1 3, page 1 064. Figure 1 5 shows schematically a master clock source using super continuum generation. The non-linear medium in this case is a length of optical fibre joining the optical pulse source via a splitter to 2 optical bandpass filters which select the respective frequencies.

As in the first example, at the node transmitter a portion is tapped from the write clock fibre. This is modulated with data and added to the fibre carrying the read clock. The modulator applies a variable delay, so that the data is written in the time slot appropriate to the particular node. At a node receiver, the second fibre carrying the read clock and the modulated data, is tapped. The read clock is separated from the data using a wavelength dependent coupler WDC. A variable delay is then applied to the clock, as in the first example. The delayed clock is ANDed with the data signal to read a given channel.

Figure 4 shows a third example. This is closely similar to the immediately preceding example, in that it uses separate write and read clocks. Three waveguides are used. The node transmitter taps the write clock from a first waveguide 41 , and writes the modulated data onto a second waveguide 42. A third waveguide 43 is tapped by the node receiver, and carries a read clock used by the receiver in reading data from the second waveguide 42.

Although in the example illustrated in the Figures, only 2 or 3 network nodes are shown, in practice the present invention may be used in a network

having many more nodes. An important parameter in determining the number of nodes, and correspondingly the number of TDMA channels, is the extinction ratio of the modulator used to read channels in the receiver. The relationship between the extinction ratio and the number of channels is as follows. extinction ratio (in dB) = 20 + 1 Olog (number of channels - 1 )

This implies, for example, that a network having 1 6 channels and connecting 1 6 nodes requires a 31 .8dB extinction ratio This is obtainable with, for example, an electro-absorption modulator of the type identified above.

Although in the examples so far described all receiver nodes are connected directly to the optical bus, alternative configurations are possible. In particular, it is advantageous to connect the receiver nodes in a star-configured network. A star configuration has the advantage of increased resilience, in that a break in a single branch still leaves the other branches operational. The star network may be coupled to the bus by one or more 1 :N splitters, where N is the number of receiver nodes Figure 9 shows a first example of such a network. Here the head-end master clock source H transmits a single clock signal CK on one of the fibre waveguides, as in the network of Figure 2. Two splitters S1 , S2 couple clock and data signals respectively to the different branches B1 to BN of a star network. The branches are terminated by receiver nodes RN.1 to RN N. Each branch may have the same physical construction as the bus 1 . Figure 10 shows a second example. This is analogous to the network of Figure 3, in that separate write and read clocks are transmitted onto the two waveguides of the bus. In this example only a single splitter is required, since the read clock and data are received by each receiver node over a common path. Figure 1 1 shows a final example based on the network of Figure 4 and using separate waveguides for the read clock, for the write clock and for the data. As shown in the Figure, this requires two splitters to couple the star network to the bus.

A star topology for the receivers may be used both when the receivers are separate from the transmitters, as in, for example, the network of Figure 9, and also when transmitter nodes and receiver nodes are combined in the same locations, as in the network of Figure 10. In this latter case, the star network and the bus may run through a single cable.

Although the discussion above has assumed that the optical pulses are bright pulses from a modulated pulse source, the invention may also be implemented using dark pulses, as described and claimed in our copending International Patent Application PCT/GB97/00520 A dark pulse is a temporal gap, or a region of reduced intensity radiation in an essentially continuous burst of optical radiation. In this case the optical source may comprise a distributed feedback (DFB) laser operating at 1 555nm followed by an electroabsorption modulator (EAM) which has a maximum extinction ratio of 20dB and a mean absorption characteristic of 2.5dB/V. A 10ghz sinewave drive is then synchronised with and passively added to a 10Gbιt/sdata sequence from a data source 144 via a power splitter 140 (used in reverse to combine the two signals). A suitable power splitter is the Wiltron K240B, available from Anπtsu Wiltron. Both the sinewave and data sequence signal levels have a 2.5V peak-to-peak amplitude. The resulting signal comprises a sinewave with an offset voltage determined by the data signal, with the relative amplitudes arranged such that the maximum value of the cycle for a data 0 is below the minimum level for a data 1 . It will be appreciated that this precise arrangement is not necessary, and it would be sufficient to ensure that the entire 10 GHz cycle remains in the low absorption region of the modulator for a data 1 . The electrical signal is applied to the EAM 1 20, and the DC bias is adjusted to ensure that the entire cycle for a data 1 gives low extinction, whilst the troughs of the cycle for a data 0 give a high extinction. Consequently, dark-pulses are formed for data O's, whilst a low extinction is maintained for data 1 's.

An EAM suitable for use in the system is the one described in, for example, "Generation of 6.3 ps optical pulses at a 10 GHz repetition rate using a packaged EAM and dispersion compensating fibre", Electronics Letters, Volume 30, pp 1 700-1701 , which is incorporated herein by reference.