Optical switching system

The present invention relates to packet switched digital telecommunications systems, and more particularly to an optical packet switching system.

Among techniques presently under test for implementing 5 broad-band telecommunications networks, networks using asynchronous time division techniques (ATD) are of particular interest, since at least from a theoretical point of view, they can integrate the widest range of services at different speeds in a single system. Since the basis of such a system is the switching of packets with destination labels, it is also known as Fast Packet Switching or Label Addressed Switching.

Promising networks allowing the use of label addressed switching techniques are described in a paper entitled ”ATD Switching Networks from Proceedings of GSLB-Seminar on Broadband Switching - Albufeira, Portugal, 19-20 January 1987, pages 225-234. Such networks are based on small switching nodes (typically 2X2), organized so that a pakcet is self-routing through the network; each stage

0 must examine just one bit of the label, deciding on the basis of its value to which of two outputs the packet is to be forwarded.

Since such a network is typically blocking, each network node needs a buffer memory to hold any packet whose proper

-- 2 output is already seized by the other input. The network is consequently not time-transparent: its efficiency is higher if the entering traffic is randomly distributed and the ratio between presence and absence of signals is low. Using current fabrication technologies (C-MOS) and 8 channels in parallel, the global throughput of a 128X128 network can reach a few Gbit/s.

Throughput can be increased by using faster technologies, for example by an order of magnitude using ECL (emitter coupled logic). An alternative is optical technology, but from the switching point of view, the latter technology offers the system builder rather limited performance: switching matrices combine a small number of inputs and outputs (8X8 is already a considerable achievement), with rather large size (a few cm), high input/output. attenuation (several dB) , and high crosstalk (a few tens of dB) . The most promising devices for optical switching currently commercially available are based on directional couplers or X-junctions, usually obtained by diffusing titanium optical guides into a substrate of lithium niobate. Devices with X junctions are described in a paper entitled Survey of Optical Switching, published at pages 143-151 of the previously-cited Proceedings, and directional couplers are described in a paper entitled High Speed ptical Time-division and Space-division Switching in the Proceedings of IOOC-ECOC 85, Venice, 1-4 October 1985, pages 81-88. Devices of this type permit limited-capacity matrices to be implemented (e.g. 12X12 with directional couplers, 16X16 with X-junctions). The X-junction seems to be better adapted to matrix organization, since it has no bending losses, but it needs higher driving voltages.

The low performance of these elements is compensated by a very large bandwidth (some tens of GHz) and a substantially infinitesimal switching time (some tens of picoseconds). These characteristics render the use of optical elements of interest in small size switching networks in which the bandwidth can be used to increase the ratio between presence and absence times of the signal.

So far neither optical logic devices for implementing self-routing functions in a network nor optical memory elements of practical usefulness are available. Hence it does not appear currently practicable to use optical switching elements in the same manner as in electrical technologies.

Thus, a switching network based on an optical asynchronous technique requires a system organization which carries out logic functions at the network periphery and does not require memory elements within the network.

The optical switching system described herein addresses these problems, so as to allow high speed packet switching by means of a switching network consisting of an optical device. Undesirable cross talk effects are suppressed and the network can be easily implemented.

Accordingly the invention provides an optical switching system, in which input channels carrying data packets are switched on a one to one basis to output channels, wherein a time compressor receives packets from each input channel, and a transducer converts the time compressed packets into optical signals on receipt of a timing signal, an optical switching network receives the signals and switches them to desired output channel, and in each output channel a time expander reconverts the switched optical signals into electrical signals and expanded them to their original duration, a central processor being

- 4 provided to control the switching of input channels to output channels and the timing signal controlling the application of the optical signals to the switching network so as to avoid conflicts in the network.

Further features of the invention will become apparent from the following description of an exemplary embodiment thereof, with reference to the accompanying drawings, wherein:

Fig. 1 is a block diagram of a switching system;

Fig. 2 is a set of timing diagrams showing digital signals present at various points of the system;

Fig. 3 schematically represents a number of possible states of switching elements?

Fig. 4 is block diagram of the switching network;

Fig. 5 is a view of a substrate supporting the optical switching network ; and

Fig. 6 is an electrical block diagram of a driver circuit for the switching network.

Referring to Fig. 1, references II, 12,...Ip...In and 01, 02,...Oq... On denote input and outputs respectively of a switching system, comprising time compressors TClm TC2,...TCp...TCn and time expanders TE1, TE2,...TEq...TEn, as well as an optical switching network RC. Each compressor and expander has a connection with a driving circuit LC, slaved through connection C to a centralized processor unit EL. The processor obtains information necessary for system control through a connection S from appropriate signalling circuits.

Let it be assumed that activity at a generic input Ip is as represented in Fig. 2, diagram Ip: namely data blocks with random times of arrival and variable duration. The probability of the duration exceeding any particular value

1235040 decreases exponentially. Time compressor TCp receives the data block and retransmits it in a shorter time, as indicated in diagram I’p=O'q of Fig. 2; the inverse operation is performed by the time expander TEq, whose input 0 ' q is connected to output I ' p of TCp through the optical switching network RC (Fig. 1) . The switching network, compressors and expanders are driven by the driving circuit LC, controlled by the centralized processor unit EL.

The timing of blocks re-expanded by expander TEq is shown in the diagram Oq of Fig. 2.

Assuming that the capacity of the network RC is infinite and the degree of time compression/expansion is unlimited, then by increasing the time compression of the blocks, the probability of transit through the network of two packets at the same time can be reduced indefinitely; in pactice, there is a lower limit to this reduction, determined by the physical limitations of the apparatus. If the time compressors also have temporary storage capability, then when the network is already occupied by a data block, others can be delayed for the necessary time for the preceding block to clear the network; this entails no problem, since system time transparency is neither ensured, nor required. Under these conditions the network does not require storage capability in its switching elements.

A network of this type can be made up of elements capable of assuming two states, indicated by symbols x and =” in Fig. 3. In the ”x state two signal flows pass directly through the element without interfering with each other from inputs a and b to outputs a’ and b' respectively; in the ·»=»» state the outputs are exchanged so that inputs a and b are connected to outputs b¹ and a’ respectively.

A network with n inputs and n outputs (nxn) is shown in Fig. 4. It has two columns of n switching elements each. References Xll, X12,...Xlp...,Xln indicate elements of a first column and X21, X22,...X2q...,X2n indicate elements of a second column. n inputs I'l, I'2,...I'p...I'n are connected to inputs a of n elements of the first column and n outputs O'l,O'2,...O'q...,O'n are connected to outputs a’ of n elements of the second column. Outputs a' of the elements of the first column are connected to the corresponding inputs a of elements of the second column, whilst in each column each output b' is connected to input b of the adjacent element. It should be noted that the elements of the second column are inverted with respect to those of the first column.

The theoretical network capacity is equal to the capacity of a single connection when used without idle periods: 10 Gbit/s can be achieved with optical components. Routing is performed one packet at a time according to very simple rules, by examining input and output indices contained in the label of a packet present at an input. If a packet is present at input I'p with output O'q, routing strategy is reduced to three cases, as follows:

<td> a)</td><td> p&lt;q:</td><td> points Xlp and Xlq are activated;</td>

<td>^b&gt;</td><td> p&gt;q:</td><td> points X2p and X2q are activated;</td>

<td> c)</td><td> p=q:</td><td> no point is activated.</td>

After establishing the internal circuit, high-speed packet transmission through the network is then enabled, following which another packet is examined. Even though no simultaneous connections are possible, Fig. 4 shows for illustration two possible cases of connections between inputs and outputs with different ordinals, namely, with

1295C40 input ordinal inferior (I¹1-0’2) and superior (I'n-O'q) to the output ordinal respectively.

Such a switching network can be implemented in optical technology using switching elements consisting of x junctions, which can be placed side by side on the same substrate, as in Fig. 5. A number n of optical parallel guides is formed in a substrate of suitable material, for example by diffusing titanium into lithium niobate (LiNbOg) . Two further guides are formed in the same way at an angle β and an angle 180-β to the parallel guides and metallic electrodes are deposited at each crosspoint for applying electrical fields controlling the junctions.

If the angle β is small enough, L=n*d//3, where L is the minimum length of the parallel optical guides, d their specing and n their number, i.e. the number of input/output lines. With L-100 mm, d=50/xm, and /3=0.0174 rad (1°), n is equal to 34. The distance between two at which driving electrodes are If a 100-input/output network is desired having the same physical size of substrate, a crossing angle of 0.05 rad (2.86°), with the electrodes by 1 mm, would be necessary. These values are not compatible with the characteristics of LiNb0₃, but are delivered within reach of anticipated technological improvements attainable in fabrication materials.

adjacent crosspoints, deposited, is 2.8 mm.

Assuming that A(l) is the linear guide attenuation, A(x) the attenuation of the switch element in the x state and A(=) that in the = state, then if n is the number of inputs/outputs and L the guide length, the maximum attenuation A through the network is equal to:

A = 2-A(=) + (n-l)7-(x) + L-A(l)

For the case previously mentioned (L=100 mm, n=30), if A(l)=0.02 dB/mm, A(x)=0.15 dB and A(=)=0.5 dB, we obtain:

A- 2-0.5 + 29-0.15 + 100-0.02 = 7.35 dB

Crosstalk at an output with the same index as the input is due to the loss in a single X-junction in the ·'= state, whilst at the other outputs it depends on the loss through a single junction in the x state. A crosstalk of -25 dB, easily discriminated from the signal, is technologically attainable in a single junction.

The block diagram of a driving circuit suitable for the switching above network is shown in Fig. 6. It comprises the main blocks forming the time compressors and driving circuit LC of Fig. 1.

It provides for assignment of common resources and prevents the simultaneous access of more than one input signal. This can be done in a more or less sophisticated manner, for example by adopting collision-detection multiple-access techniques or cyclic-priority systems; however, for speed and simplicity of construction a simple scanning method is preferred.

Data packets consisting of a variable number of octets arrive in sequence at the generic serial input Ip of Fig.

6. Under the control of the central processor, each octet is written into the register of a FIFO memory MEp (FirstIn First-Out), with the addition of an extra bit signifying the presence of the data in one cell of each register. These cells are represented by shading in Figure 6. Simultaneously with the writing of the last octet of the data packet, the signal on line PR of a connection C is activated by the central processor.

A parallel-to-serial converter PSp, placed at an output of memory, transforms the parallel octets received from MEp into a serial signal with a high bit-rate, which is applied by line 1 to an electro-optical transducer EOp. An optical fibre FOp is connected to the output of EOp and to the corresponding input of the optical switching network already described.

A periodic scanning signal is sent by the processor through connection C and line SC to an input SET of a SETRESET flip-flop SRp through a four-input AND gate ASp. This signal has a period dependent on the desired transfer speed through the switching network and is phase-shifted with respect to the corresponding scanning signals of the other n-1 channels by a time equal to the period divided by n. A line BU, common to all the channels, carries a signal meaning switching network busy when in active state. If this signal is inactive at the end of writing of a packet as signalled by an active signal on line P, and in the proper time phase as indicated by an active scanning signal on line SC, the input SET of the flip-flop is enabled and its output passes to the active state. Through gate Pp, the line BU becomes active to indicate that the switching network is busy, so that the inputs SET of the other n-1 flip-flops and of the same SRp are inhibited, while RESET inputs are enabled.

The signal on line ABp connected to output SRp now enables the transfer of the octets from memory MEp, their parallel-to-serial conversion in register PSp and their electro-optical conversion by EOp, which outputs a corresponding high bit-rate optical packet. The signal on line ABp also enables the outputs of a register Rip, in which the central processor has written the destination address of the channel and two bits whose active states have the meaning of p&gt;q and p&lt;q respectively, q being the

- 10 order number of the output Oq.

A decoder DE, common to all the channels, reads the destination address supplied by register Rip on connection 2 and places an active signal on an appropriate one of the n output lines. In this case the active line will be the q-th, so that the combinational logic circuit, consisting of gates ORq, ADq and ACq, will be activated to provide an appropriate output while the combinational logic circuit consisting of gates ORp, ADp and ACp, will be activated by the signal on wire ABp to select an appropriate input. According to whether the condition at register Rip output is p&lt;q, p&gt;q or p=q on wires 3p and 4p, either gate ADp, or gate ACp or neither is activated, and hence lines Xlp or X2p activate the respective elements of the optical switching network. The same occurs when output gates ADq and ACq are selectively activated by signals on wires 3q and 4q, thus lines Xlq or X2q activate the output switch elements. When all the octets and corresponding presence bits have left memory MEp, the signal on line BPp assumes the inactive state, the input RESET of flip-flop SRp is activated through gate ARp, and the signal state on line BU indicates that the switching network is available.

With reference to the block diagram of Fig. 1, the compressors TCp can be taken as consisting of blocks MEp, PSp, EOp and of optical fibre FOp of Fig. 6, while all the other blocks of Fig. 6 can be taken as housed in the driving circuit LC of Fig. 1. The expanders TEq are easily implemented by circuits performing operations inverse to those performed in the compressors.

At the output of the switching network there is an optical fibre, connected to an opto-electrical transducer, followed by a serial-to-parallel converter and by a FIFO memory, which will regenerate a packet flow with the same

X295C40

- Il characteristics as that at the input. The flip-flop SRp (see Fig. 6) still provides an enabling signal on line ABp.

It is clear that the embodiment described has been given 5 only by way of non-limiting example. Variations and modifications are possible within the scope of the appended claims.