1 A METHOD OF ROUTING MESSAGES BETWEEN ACCESS POINTS The present invention relates a method of routing messages between access points. It is more particularly used in the telephony art to construct and improve 5 switching centers. An improved method can be used in the mobile telephony art in particular. The object of the invention is to reduce the cost of telephone exchanges. In the telephony art, it is usual to divide switching systems into systems referred to as circuit 10 switched systems and systems referred to as connection switched systems. Circuit-switched systems set up a physical circuit in the form of a pair of copper wires, an optical fiber link, with all the necessary plant, or some other form of permanent link between two access 15 points to be connected. Users on either side of the two access points to be connected have unrestricted use of the permanent link. They can therefore exchange information over the line made available to them at bit rates that are as high as they need, provided that this 20 is compatible with the technology of the link. In connection-switched systems, on the other hand, a network is organized between multiple access points and, at any given time, the link between one user and another via the network uses one channel on one connection of the 25 network. A channel can therefore be defined as a reserved time slot of a frame and/or an allocated frequency band within an overall band. It can also be defined as a temporarily assigned coding law in the case of a code division multiple access (CDMA) connection. 30 This being so, the essential difference between circuit-switched systems and connection-switched systems is that messages sent in connection-switched systems include the address of the user to whom the information sent is addressed. In circuit-switched systems it is not 35 necessary to introduce an address into the message to be transmitted once the connection has been set up, since the message can be transmitted between the connected 2 points only from one point to the other. In contrast, in a connection-switched system, routers are installed at each node of the network. Each time a router receives a fraction of a message to be 5 transmitted it extracts the address part and transmits the message onward (in the general direction of the destination user) by assigning a connection which connects it to another router so that it can transmit a packet of information contained in the message. The 10 message with its packets of information then progresses across the network and is finally transmitted over the destination user's line. The latter system has the advantage over the former system that it requires a much simpler infrastructure. 15 On the other hand, for each packet of information to be transmitted from one user to another, it is at all times necessary to extract the destination address, to determine a path by which the information packet must be transmitted onward, and to insert the information packet 20 into a channel of a connection corresponding to that path. Determining the path involves designating time slots, bands, coding keys and, usually, destination routers in which the remainder of the routing processing will be effected. The routing of messages between the 25 connected users is constantly changing and constantly being reorganized because of the conditions of use of the network and on terminating a call between two users and setting up a call between two new users. The invention relates to this reorganization. 30 Providing the routers entails providing the access points, which are also referred to as nodes, with switching and transmission electronic circuits capable of setting up on demand sufficient numbers of transmission connections and channels for the calls to be routed. The 35 problem with providing the above electronic circuits is to optimize the hardware resources employed, i.e. to provide the minimum number of electronic circuits needed 3 for a nominal number of calls to be routed. The most serious problem to be solved is then optimizing the use of the connections. Simplifying, in the situation referred to throughout 5 the remainder of the description, it may be said that a connection in a router is made up of a particular number of time slots of a frame, for example L = 32 time slots. In other words, this kind of connection can transmit L = 32 different messages simultaneously. The connection 10 therefore includes 32 channels. The circuits of this kind of router include add time division multiplex circuits and drop time division multiplex circuits. During a session, a connection of the above kind is assigned to the call between an access point A and an 15 access point B. The connection requires the presence of an add multiplexer that can receive L = 32 messages. The add time division multiplexer can be implemented on an electronic circuit card. For statistical reasons, if there is only one message still to be exchanged 20 between a router A and a router B at any given time, the connection must nevertheless be maintained. The electronic circuit card concerned then routes only one time slot (one channel). In practice the other time slots are filled with filler signals. In the case of 25 digital transmission, the filler signals are 1 or 0 bits or connection consistency signals. The connection already set up cannot be used if L - 1 = 31 other users wish to set up calls between themselves and other users who are no longer accessible via the access point B but 30 only via other access points C or D. In this case other electronics (other electronic circuit cards) must set up other connections. The document US-A-4 718 058 discloses a channel switching system in which short messages are concatenated 35 to occupy an entire channel on a connection (a virtual link between two access points). This approach, which in the final analysis is similar to setting up channels on a 4 connection between two access points, tends to solve the problem of non-optimum use of connections. However, it has the drawbacks already referred to. The expected saving in hardware, which could be by a factor of the 5 order of 4, for example, if four 8 kbit/s messages were concatenated to form a 32 kbit/s global message (which would use one channel of the connection), is in fact illusory. The hardware saving obtained in practice is no better than 1.3: 1, rather than 4:1. The statistical 10 nature of channel occupation and the diversity of the demands for channel occupation mean that the connections are underused, which leads to considerable use of fillers. The document WO-97-27720 describes adding an 15 auxiliary switching circuit to constitute messages with a higher bit rate, for example 32 kbit/s, from messages with a lower bit rate, for example 16 kbit/s or 8 kbit/s. In theory, as in the preceding situation, the system should work well. In practice the hardware saving, in 20 terms of the implementation of the multiplexers, is no better than 30%. It can be greater, but a greater saving is achieved only at the cost of a considerable increase in the size of the auxiliary circuit. This leads to a false assessment of the reduction in the cost of the 25 telephone exchange. To solve this problem, and in particular the problem of underuse of the connections set up between two access points, the invention monitors the setting up of connections. If an abnormally large number of 30 connections for an overall number of channels between two access points is noted the connections are reorganized to create full connections, each with a nominal number of channels, if possible. This frees up connections. In practice it frees up electronic circuits (the electronic 35 circuit cards which set up the connections). Those electronic circuits can therefore be used to set up other connections, in particular connections with other access 5 points. It will be shown that this approach increases the saving from 1.3:1 to 3:1 or 4:1. The average saving is 3.5:1. The example discussed is that of constructing 5 32 kbit/s channels from 8 kbit/s channels. The invention approaches the theoretical limits. For simplicity, it can be said that, in order to free up a connection, and therefore the corresponding electronic circuits, reorganization in accordance with the invention entails 10 moving the message from an assigned time slot of one connection to a time slot of another connection, for example the time slot with the same number. The channel freed up can therefore be used to create calls with other access points. In the mobile telephony art, if 15 synchronization problems arise when the routing must be changed it is in practice preferable to perform a handover (change of base station) for a mobile telephone whose call must be routed on another connection. It will be shown that this approach does not lose any information 20 that should be transmitted, does not send the same information packets twice, and requires no additional auxiliary circuit. The invention therefore provides a method of routing messages between access points wherein: 25 - a frame is constructed and messages from diverse sources are grouped in the frame by time division multiplexing, - connections having a nominal capacity expressed as a number of messages are set up between a first access 30 point and a second access point, and - frames are sent from the first access point to the second access point using the connection set up between the access points, characterized in that: 35 - the number of messages sent between the first and second access points is measured and the number of connections set up between the first and second access 6 points for sending the messages is measured, and - the connections set up between the first and second access points are reorganized if the number of messages sent by the connection set up between the first 5 and second access points is less than a nominal capacity of all of the channels set up minus 1. The invention will be better understood after reading the following description and examining the accompanying drawings. The drawings are provided by way 10 of illustration only and are not limiting on the invention. In the figures: - Figure 1 is a diagrammatic representation of a routing circuit (router) provided at an access point to receive messages and to transmit the messages to other 15 access points by a method in accordance with the invention; - Figure 2 shows a frame which enables communication between access points; and - Figure 3 shows the hardware resources required at 20 the access points and connections that the hardware resources can set up between different routers. Figure 1 shows a system for implementing a message routing method in accordance with the invention. It essentially shows an access point A. In practice the 25 access point is at a geographical location in a territory. It is a telephone exchange in a small town or a district of a large town, for example, or a relay station installed at a routing node of a network. The access point A is equipped with circuits for setting up 30 connections with other access points, for example the access point B, C or D. The circuits provided for this purpose in the routing circuits at the access point A include N time division multiplexers, for example. To simplify the 35 example the value of N will be taken as 10. The multiplexer circuits include N combiner circuits 1 connected at their input to N time-delay circuits 2. The 7 time-delay circuits 2 have L inputs (L = 32 for example) and L outputs connected to L inputs of each of the N combiner circuits 1. The time-delay circuits 2 each include L programmable delay lines 3 for constructing 5 frames of L messages. The N combiner circuits 1 each have an output 4 connected to an input of a switching circuit 5 for connecting the output 4 to a line 6 to 8, for example a physical line. The lines 6 to 8 route the messages from the access point A to the access point B or 10 C or D, respectively. The circuit 5 makes a choice between these lines and therefore chooses a destination. The system shown symbolically in Figure 1 can take various physical forms. The physical lines 6 to 8 can be channels of connections with an even higher bit rate. 15 Corresponding plant is provided at the access points B, C and D. Note that with N multiplexers the number of access points B, C or D must be limited to N, at least in practice, because if this were not the case it would be possible for an access point to be the destination of a 20 call and for it to be impossible to assign a connection to the call because all the connections would already have been assigned to other destinations. A processor 9 delivers instructions 0 to the circuits 1, 2 and 5 of the access point A and to this end is connected by a bus 10 25 to those circuits, to a program memory 11, to a data memory 12 and to a clock 13. The program memory 11 contains a program 14 which includes a multiplexing subroutine 15. This is known in the art. Figure 2 shows a frame. In this example, 125 ps 30 frames each include 32 time slots. Timing the circuits 1, 2 and 5 synchronized to the clock signal H at 8 000 Hz therefore enables L = 32 messages from diverse sources to be routed. The messages originate from mobile telephones 16, 35 for example. The diagram shows an uplink from a mobile telephone 16 to a base station 17 itself connected to the access point (network node) A at a downstream point in 8 the network. For example, in the case of a speech message, the mobile telephone 16 includes for the uplink a microphone 18 connected to an analogue-to-digital sampler-converter 19. The sampler-converter 19 is 5 conventionally clocked by an 8 kHz signal but there is no a priori parallel between this timing of sampling of the speech signal and the timing imposed on the routing circuit of the access point A. The sampler-converter 19 produces one 13-bit word per sample. This is known in 10 the art. The 13-bit words are compressed into 8-bit words in A Law logarithmic compressors 20. The 8-bit words are then fed to the input of a sender-receiver 21 connected to a send antenna. The 8-bit words are therefore sent successively via the antenna to a base 15 station 17 which the mobile telephone 16 is logged onto. It is therefore feasible for L mobile telephones like the mobile telephone 16 to be connected to a time-delay circuit 2 connected to the combiner circuit 1. In this case it is possible to construct an entire frame to be 20 routed between the circuit 1 and an access point, for example the access point B. In practice there are switching circuits between the circuits 1 and 2. Figure 3 shows the possibility of creating connections 6, 7 or 8 terminating at the respective 25 access points B, C or D for each of the N multiplexers at the access point A shown in Figure 1. Clearly, once a 1 - 2 multiplexer has been assigned to a given link, for example the link A - B, the network is in a situation equivalent to a circuit-switched network throughout the 30 duration of frame 22. It can therefore be assumed that at most N multiplexers provide N links between the access point A and the access point B, at least in theory. In practice, each of the N multiplexers at A can set up a link with the access point B. 35 An arbitrarily simplified numerical explanation can now be given of the problem solved by the invention as encountered in reality. If N x L (10 x 32 = 320) calls 9 are to be set up between A and B, the N multiplexers at node A are dedicated to a link between A and B. If N x (L - 1) calls then terminate but one call continues, occupying a channel of each connection between A and B, 5 then the N multiplexers all continue to be occupied (although each routes only one message!). Thus all the resources of the access point A are in use. A subsequent call cannot be set up if the corresponding message must be routed to access point C. On the other hand, if the 10 message must be routed to access point B, then the message can naturally be incorporated into one of the connections already set up between A and B. However, since it is statistically possible to request access point C, this amounts to it being necessary to create 15 additional multiplexers. The total number of multiplexers (and of electronic circuit cards) required at access point A is then greater than N. The additional multiplexers are indispensable even though the access point A already has a nominal capacity of N x L calls. 20 It can be shown that in practice the number of multiplexers is of the order of N x L x 3, although ideally it should be only of the order of N. The invention monitors the connections set up between the first access point A and the second access 25 point B. For all the connections monitored in this way, the number of messages (the number of channels actually employed on each of the connections) is measured. To this end, as shown here by way of example only, it is possible to assign each of the connections that can be 30 set up a database record in the memory 12 which includes a connection name in a field 23 and an indication 24 that a corresponding channel on that connection is occupied in L additional fields each corresponding to one database record. 35 The number of databases in the memory 12 is the same as the number of links between different access points. In the present instance there are three, for example, one 10 for the link between A and B, one for the link between A and C and one for the link between A and D. In practice monitoring is not complicated because the processor 9 assigns the various time slots of the connections to the 5 messages to be transmitted using the multiplexing program 14. The memory 12 and the databases that it contains are therefore already available in the prior art system, at least implicitly. A monitoring program 25 according to the invention 10 can therefore determine the number of time slots occupied and the number of connections set up between the N multiplexers for a link between two access points, for example the access points A and B. This is preferably done for each of the links AB, AC or AD. The program 15 then determines whether the number of calls actually exchanged in this way is less than a nominal capacity of the set of connections set up minus one. If all the multiplexers are capable of M calls (which is evidently not obligatory), the program determines whether the 20 number of calls for M connections is at least L x (M - 1). If so, at least the equivalent of one connection can be saved by reorganizing the channels on the connections. In this case the invention uses the program 25 to perform the reorganization. It is 25 preferable to determine which connection is the least occupied. Let M denote that connection. In principle, reorganization entails transferring a channel using the time slot 26 of the connection of the multiplexer M to a time slot 27 of another connection, for example the 30 connection of the multiplexer 1, and later than (or earlier than) the time slot 24. The time slot 27 preferably has the same position on the connection of the multiplexer 1 as on the connection M. In this way the calls routed by a connection (and 35 therefore by a given multiplexer) can be progressively transferred to another connection. The connection that is to be freed up is preferably the connection carrying 11 the lowest number of messages. The receiving connections are preferably the connections of the remaining M - 1 connections that are most occupied. It is nevertheless possible for there to be no 5 equivalent free slot in other available connections to which the time slot of a message in a frame can be transferred. In this case a problem arises at the time of transfer. Without going into detail, it is known in the art that this type of situation routinely leads 10 either to dispensing the same information twice or to not sending the content of a message corresponding to a time slot. In the case of transmitting speech messages, the presence of such transmission errors can be tolerated. In this case, only the content of a speech message 15 corresponding to a period of 125 pis would be degraded in this example. Although hardly satisfactory, such degraded operation can be tolerated. In this case, rather than transferring the contents of the time slot 26 into the time slot 27, it can be transferred into a time 20 slot 28 on another connection but at another time relative to the frame start synchronization. However, the above solution is not acceptable if binary data to be transmitted is not speech data but pure digital data. As the nature of the data to be 25 transmitted is not known at the access point A, the invention prefers a different approach. The aim is still to set up transmission connections 1 to M - 1 that are as compact as possible, in order to minimize filler data and to save the most resources. If this problem is to be 30 solved, major management of the information packets transmitted is then required at the access point A. The time-delay circuits 3 of the 1 - 2 multiplexers can delay a signal by at most the duration of a frame. If transfer would lead to loss of the content of the time slot 26 35 (because it would have to be placed in an earlier time slot 28 of another connection), it would be necessary to store the content of the time slot 26 for longer than the 12 duration of a frame at the time of transfer. Solving the problem in this way makes the multiplexing circuits 1 very complex, which loses the benefit of the simplicity that their structure naturally 5 provides. On the other hand, the problem does not arise if the content of the time slot 26 must be transferred to a later time slot 28 on another connection. In the context of mobile telephony, the invention solves the above problem very simply without having to 10 add additional plant and without having to add delay lines with a very long time-delay to the time-delay circuits 3. It has been realized that the problem of lost time slots has already been solved in the context of mobile 15 telephony, in particular GSM mobile telephony, in which it occurs at the time of handover (which correspond to transferring a call from a link between a mobile telephone 16 and a base station 17 to a link between the mobile telephone 16 and another base station 29). The 20 base stations 17 and 29 already contain all of the necessary plant for the user not even to realize that the connection has been switched from one base station to another. This handover method is particularly rigorous for transmission of data. 25 When it is necessary to move a message from a time slot 26 on a channel to be freed up to another time slot, the call between the telephone 16 and the base station 17 is handed over. As seen from the node A, the channel 26 therefore appears as a channel that ceases to exist in 30 favor of another channel 28 that is created. The new channel 28 is of course assigned to one of the existing connections between the access point A and the access point B. This is known in the art. Of the M connections that are candidates to receive the channel, the invention 35 eliminates connection M and retains only M - 1 connections. The invention looks for the least occupied connection (here the connection M), and then prohibits 13 the re-use of that connection if the M - 1 other connections set up are not all fully occupied. If necessary, the prohibition can be removed before all the M - 1 other connections are used. 5 In this case, the control circuits of the access point A send a signal S constituting a handover instruction via a HO (handover) program 30 to the base stations 17 and/or 29 or possibly to a base station controller. To be more precise, the signal S is seen as 10 a handover request. There is no need to satisfy the request immediately. It can be satisfied one or more frames later. What is important is that during a long time period, in particular that of conversations exchanged with mobile telephones, the data specific to 15 the calls is routed on a defragmented connection and not on very diverse and underused connections. If necessary, another connection can be occupied during the same frame 22 for the same call but for another channel before the connection M is released.