ANTENNA SYSTEM This invention relates to an antenna system; particularly, but not exclusively, to a multiple beamformer for a satellite mobile communication system. Such systems are described (in general terms) in, for example, WO93/0 902 9; WO93/095 77; WO93/09578; WO93/09S13; WO93/09614; WO93/09624; EP-A-0510789; EP-A-035885; EP-A-421722. The proposed Inmarsat (TM) P21 system consists of a number of earth stations, which may be mobile stations, having antennas for communication with a constellation of communications satellites. Each satellite carries an antenna system designed to transmit and receive an array of multiple beams, each directed towards a portion of the surface of the Earth, the beams together covering the entire surface of the Earth. Each beam carries a number of frequency-multiplexed channels; for example, the bandwidth of each beam may be 5 MHz, enabling each to carry a large number of user channels (typically carrying voice telecommunications) . The satellite typically also carries an antenna for connection to a fixed earth station, communicating for example with a public telecommunications network. To synthesize a plurality of beams in the far field beam pattern, if the transmit and receive antennas are directly radiating antennas (i.e. without reflectors) consisting of a large array of radiating elements, a conventional beam forming network requires, in principle, a phase shifter for every radiating element, for each beam position, and a power divider for every beam. Thus, for one hundred elements and on hundred beams, 10,000 phase shifters and 100 power dividers are needed, and the number of components grows roughly exponentially for large numbers of beams and elements. This represents a considerable weight of RF components, and the power losses of the feeding system is also high. Both weight and electrical power consumption are at a premium in satellites. One alternative type of beam former for array antennas is the "Butler matrix", described in US 3255450 (Butler), which consists of a butterfly cascade arrangement of four-port power dividers with associated phase shifters, receiving N input RF signals and feeding a linear array of N spaced elements. The dividers each receive two input analogue RF signals, one of which is phase shifted, and output two RF signals with a 90 degree phase difference. The effect of the array of dividers and phase shifters is Chat the RF signal supplied to any one of the inputs is fed, in progressively incrementing phase shifts, to each of the elements of the array. Thus, the array acts as a phased array, generating a beam at an angle dependent upon the phase shift increment (which depends upon the number of radiating elements) and the element spacing. By selectively exciting each input in turn, an incrementally scanning beam can be generated which may be used in radar applications. Alternatively, the beamformer can be used to generate a grid of multiple fixed beams from a common aperture. A beam can be scanned in one of two orthogonal directions by providing several such linear matrices in aligned rows and columns, the outputs of the row matrices feeding the inputs of the column matrices, and the outputs of the column matrices feeding a two dimensional array of radiators. Butler matrices are virtually lossless, and this tends to be the reason for their use. JP-A-59-44105 discloses a two-dimensional beam-forming network comprising two orthogonal stacks of Butler matrices, for forming beams lying aligned along angles on a rectangular array. EP-A-0056205 discloses a large Butler matrix formed from two orthogonal stacks of Butler matrices. WO88/04837 discloses a steerable beam reflector antenna used on a communications satellite in which a Butler matrix is used for beam steering. EP-A-0468662 discloses an antenna (which may be a directly radiating antenna) in which a Butler matrix is used as a power splitter to distribute power between antenna array elements to form a single unidirectional composite beam, the progressive phase shifts provided by the matrix being cancelled by phase shifter elements. A feature of linear Butler matrix array antennas is that the crossover point between adjacent beams is over 3dB down, so that the power between the beams drops off to half the maximum beam level. For a square array, the power minima between 4 adjacent beams are 8dB down, which would, of itself, render a conventional square Butler matrix unsuitable for forming multiple satellite communication beams, since it is desirable to provide uniform coverage of the Earth surface. According to the present invention, there is provided an antenna system using a passive power splitter matrix (e.g. a Butler matrix) as a beam former for a hexagonal array antenna to create a hexagonal array of beam directions. The use of a hexagonal array gives a better coverage of the Earth suiLaus Liian wouia an equivalent square array, since the power between adjacent beams does not drop off so deeply. Preferably, the aperture around the array is smoothed, which, in the f ar field pattern., reduces the drop off in power between adjacent beams. Preferably, the matrix is made redundant, and only some output ports are connected to radiating elements; the other output ports are terminated. This aspect of the invention causes the Butler matrix to no longer achieve its usual advantage of being lossless. However, we have found that the loss is tolerable, for an improvement in power drop off at the crossover. In a preferred embodiment, the amplification or loss in the path to the radiating elements differs across the aperture of the array, so as to provide a gentle taper in the power fed to the edges of the array. This aspect of the invention raises the cross¬over level between beams and reduces the side-lobe level of the far-field radiation pattern. Preferably, the matrix comprises two orthogonally connected stacks of power splitter matrices. In another aspect, the invention provides a beam forming network for an array antenna system which comprises first and second orthogonally connected stacks of power splitter matrices, there being fewer matrices in at least one stack than the order of the matrices in that stack. Thus, "oversized" matrices may be employed to form non-rectangular antenna arrays, but without requiring matrices to the number of twice the order of each matrix. In another aspect, the invention provides an antenna system in which several different power splitter matrices are provided, and corresponding output ports of each matrix are connected jointly to elements of an array antenna, so that a single array antenna can generate multiple grids of beams. By phasing the outputs of the matrices differently, the different grids can be steered to offset positions, so that one grid can be interpolated at minima between beams of another. In another aspect, the invention provides a communication transceiver station (for example, a satellite) having a digital processor for performing channelization (i.e. multiplexing and demultiplexing) connected via an analogue beam former comprising a passive power splitting network (e.g. a Butler matrix) . This enables the load on the processing device to be greatly reduced, without the substitution of a highly complex beam forming structure, and thus reduces the mass, power consumption and volume of the signal processing system, and hence makes it more suitable for use in a satellite. In another aspect, the invention provides an antenna system in which several different Butler matrix devices are connected in parallel to the same array antenna, each device being arranged to generate an array of beam directions, the arrays being mutually offset so as to produce a combined array of beam directions having a smaller angular spacing. Thus, a single antenna can be used to generate a large number of beams, with improved beam coverage and reduced dropoff between beams as compared to a beam array producible from a single Butler matrix device. Other aspects and embodiments of the invention are as described in the following description and claims. Preferred embodiments of the invention will now be illustrated, by way of example only, with reference to the accompanying drawings, in which: Figure 1 illustrates schematically a satellite mobile communications system; Figure 2 is a perspective view of a space vehicle according to one embodiment of the invention; Figure 3 is a block diagram showing schematically the electrical components of the communications system of the sDace vehicle of Figure 1; Figure 4 is a block diagram showing schematically a Butler matrix used in the embodiment of Figure 3; Figure 5 is a schematic perspective view showing the arrangement of Butler matrices of Figure 4 to provide a beam former according to the embodiment of Figure 3; Figure 6a is a diagram showing the radiation pattern of the receive antenna of the embodiment of Figures 2 to 5; and Figure 6b is a corresponding plot of the radiation pattern of a transmit antenna according to this embodiment; Figure 7a illustrates the physical arrangement of array elements in the receive antenna, and indicates the relative gain of the amplifiers connected to each element; and Figure 7b illustrates the physical arrangement of array elements in the transmit antenna, and indicates the relative gain of the amplifiers connected to each elements-Figure 8 is a diagram illustrating the connection of transmit and receive beam ports to respective beam formers shown in Figure 5; Figure 9a illustrates the respective connections between elements of the receive antenna and the beam former of Figure 5; and Figure 9b illustrates the respective connections between elements of the transmit antenna and the beam former of Figure 5; Figure 10 illustrates a beam former and antenna arrangement according to a further embodiment of the invention; and Figure 11 illustrates schematically the radiation pattern obtainable with the embodiment of Figure 10. Figure 1 illustrates a known proposed satellite mobile communications system. The system comprises at least one satellite (SI and S2) in orbit around the Earth (E) , on which are located a plurality of terminals; for example mobile communications terminals (Ml and M2) and fixed communications terminals (Fl) . The latter may, for example, comprise Earth stations connected to telecommunications networks (e.g. the PSTN). The satellites are preferably in intermediate circular orbits (ICO), at a height, for example, of 10,000km above the Earth. Twelve satellites may be provided, for example, in angularly distributed sets of four in each of three mutually normal orbits, so that each terminal is within the field of vision of several satellites at any time. Each satellite generates a plurality of spatially separated (but overlapping) radiation (e.g. radio frequency) beams, forming an array of beams (Tl-TN) which covers a contiguous area of the surface of the Earth; typically, the whole surface, in a manner described in more detail below. Each satellite also has an array of radiation reception directions (Rl-RN) which intercept the surface of the Earth; typically, the reception directions coincide with the beams. Thus, the beams and reception directions constitute spatially separate communications channels between the satellites and the terminals. As the satellites are non-geostationary, and as the terminals may be mobile, a given terminal may pass through multiple such channels over time, and may also need to cease to communicate with one satellite and switch to another ("handover"). In known fashion, one terminal can communicate with another, or with a telecommunications network, by establishing contact with a satellite, which then establishes contact with another terminal and transponds messages between the two. Referring now to Figure 2, a satellite according to the present embodiment comprises a housing 5, carrying a pair of deployable solar cell panels 4a, 4b (shown folded); at least one steerable high gain spot beam antenna 3 providing a feeder link for communicating with one or more fixed Earth stations connected to telecommunications networks; a receive array antenna 1 receiving on the plurality of reception directions Rl-RN; and a transmit array antenna 2 for generating the plurality of beams Bl-BN. The antennas 1-3 are provided on che side of the satellite which is maintained facing the Earth. A brief explanation of the modulation employed will now be given. The feeder link antenna 3 operates in the Ka band, at a transmit frequency of 20 GHz and a receive frequency of 30 GHz, over a bandwidth of 50 MHz. The receive antenna operates at a frequency of 2 GHz and the transmit array antenna at a frequency of 2.2 GHz, each with a bandwidth of 30 MHz. Each beam (in this embodiment there are 121) is allocated a block of 5 MHz within this spectrum, the blocks allocated to neighbouring beams being different so as to reduce the interference between adjacent beams. Each block includes up to 25 frequency slots (FDM and/or TDM) of 200 KHz bandwidth. Each frequency slot is split into 48 user channels. The bandwidth available for each user is thus 4 KHz, which is adequate for speech. Each beam can communicate with 1200 simultaneous users. Referring to Figure 3, the electrical arrangement provided within the satellite comprises a forward link, for communicating from an Earth station to a terminal, and a return link, for communicating from the terminal to the Earth station. The forward link comprises the feeder link antenna 3, the signals from which are bandpass filtered by respective filters 6a-6d and amplified by respective low noise amplifiers 7a-7d. The amplified signals are combined and down-converted to an intermediate frequency (IF) by a combiner/IF downconverter circuit 8, This IF signal is separated into two 25 MHz bands by complementary filters 9a, 9b, each of which is digitised by respective analogue to digital convertors (ADCs) 10a, 10b. The digitised IF signals are each then frequency-demultiplexed into 125 slots by frequency demultiplexers 11a, lib, thus giving a total of 2 50 slots each of bandwidth 200 KHz. Under the control of a digital control circuit 13, a switch or routing network 12 routes each of the 250 frequency slots to one of the 25 input ports of one of 121 (or, in general, N where N is the number of beams) frequency multiplexers 14a-14N, which multiplex the slots into a 5 MHz baseband signal which will be transmitted on one of the beams Bl-BN. The multiplexer 14 (and hence beam B) selected by the control circuit determines the geographical location on the Earth to which the frequency slot is transmitted, and the multiplexer input port (and hence modulation frequency) determines the user terminal which will receive the frequency slot. Each of the 121 multiplexed digital baseband signals is then converted to an analogue signal by a respective digital to analogue converter (DAC) 15a-15N, the outputs of which are each up-converted to a beam frequency lying within a 30 MHz range in the 2.2 GHz band by an array 16 of IF/S band converters- As mentioned above, the frequencies chosen for adjacent beams are different. The modulated signals are then supplied to respective input ports of an analogue beam former 20, which generates a plurality M (e.g. 109} of energising signals for energising respective radiating elements 200a-200M of the transmit array antenna 2. The energising signals are amplified by a bank of M RF power amplifiers I7a-17M, and bandpass filtered by a bank of filters 18a-lSM, prior to being supplied to the respective radiating elements 200a-200M. The components of the return link are, in general, the reverse of those in the forward link. A plurality P of receiving elements, (for example 151) 118a-118P receive incoming radio signals in the 2 GHz band from terminals on the Earth. The signal from each element is filtered and amplified by respective filters 118a-118P and low noise amplifiers 117a-117P, and fed to the input ports of an analogue beam former 120. The signals at each of the N output ports of the beam former 120 are down-converted to a 5 MHz baseband signal by an array 116 of S/IF convertors, and digitised by a respective ADC 115a-ll5N. The digitised baseband signals are each frequency demultiplexed into 25 frequency slots of 200 KHz bandwidth by demultiplexers 114a-114N, and the frequency slots are routed, under control of the control circuit 13, through a switch 112 to a predetermined input (corresponding to a particular frequency) of one of a pair of frequency multiplexers llla,lllb generating 25 MHz output signals which are converted to analogue signals by a pair of DACs 110a,110b. The analogue signals are then combined into a 50 MHz signal by a pair of modulators 119a,119b and a filter 119c, and the combined signal is up-converted into 20 GHz signals by an IF/Ka converter and RF divider network 118. Each RF signal is amplified by an RF power amplifier (e.g. a travelling wave device} 117a-117d; filtered by a bandpass filter 116a-ll6d; and supplied to a feed link antenna 3 for transmission to a respective Earth station. Thus, the system shown in Figure 3 will be seen to consist of a feeder link communication subsystem comprising the elements 3, 6-9 and 16-19; a channel separation and combination subsystem comprising the elements 11-14 and 111-114; and a mobile link communication subsystem comprising the elements 16-13, 116-118, and antennas 1 and 2. Since, in this embodiment, the channel separation and combination (i.e. processing) subsystem is digital, ADCs and DACs 10, 15, 110, 115 are provided. The use of a digital channel processing subsystem, which is desirable, is facilitated in the present invention because, since the beam forming is performed by relatively wide bandwidth analogue beam formers 20, 120, it need not be performed by the digital hardware, which therefore does not need to operate over the full 30 MHz mobile link spectrum and can accordingly be realised using readily available components. Further, the power consumption of digital signal processing is roughly proportional to the signal bandwidth. Thus, since in the present invention the digital processor operates over 5MHz bandwidth signals at each beam port, and 30MHz bandwidth signals for the array elements are handled by the analog beamformer, the power consumption required by the digital processor is drastically reduced relative to an all-digital system. Referring to Figures 4 and 5, the analogue beam formers 20,120 will now be described in greater detail. Each analogue beam former 20,120 consists of two orthogonally connected stacks 30a,30b of Butler matrices 50, as shown in Figure 5. The stacks each consist of identical Butler matrix devices 50, each, as shown in Figure 4, comprising a 16 input, 16 output (i.e. 16x16) device comprising a substrate 51, a first side X carrying a plurality of first ports, a second side Y carrying a plurality of second ports, a plurality of identical 4 port couplers or hybrids 52, and a plurality of phase shifters 53, all interconnected by striplines. The term "side" in the document is used in a topological sense, without reference or limitation to the actual geometrical arrangement of the matrix; the physical positions of the ports and components are irrelevant (provided that the phase of the signals is not unduly distorted). Both beam formers 20, 120 use matrix devices as shown in Figure 4, but the number of such matrices making up the stacks illustrated in Figure 5 differs between the beam formers 20, 120. The structure of Butler matrices is well known in the art; it suffices to say thac each hybrid (which may have the structure disclosed in EP-A-005 62 05, incorporated herein by reference in its entirety) divides tne power input at two input ports thereof equally between two output ports thereof, which are separated in phase by 9 0 degrees. Where there are K first and second ports, and K is a power of 2, the matrix consists of (log2K) rows each of K/2 hybrids, and Ulog2K)-l) rows of K/4 phase shifters disposed between the hybrids, each hybrid being connected to two in the following row, the spacing between the two to which a hybrid is connected increasing as a power of two in each successive row. The phase shifts imposed by the phase shifters 53 are in increments of (180/N) degrees, and the phase shifters of "the first row apply shifts which are odd multiples of (1S0/N); those of the next row apply shifts which are odd multiples of (18 0/2N) : those of the next row apply shifts which are odd multiples of (180/4N); and so on. Although in Figure 4, K/2 ((log 2K)-1) phase shifters 53 are indicated, only half of these apply non-zero phase shifts and the others may therefore be owitted in practice. The effect of this structure is that the signal at each first port X is divided into N signals of equal amplitude each appearing at one of the second ports Y, and each having a progressive phase displacement. For example, a signal applied at a first of the first ports, with an amplitude A, appears as a first signal at the first of the second ports with an amplitude A/N; a second signal at the second of the second ports of the same amplitude but shifted in phase by (360/N=A); a third signal of equal amplitude at the third port, shifted in phase by delta from the second;.... and a sixteenth signal at the sixteenth of the second ports shifted in phase by delta from the fifteenth, and in phase with the first. A signal applied at the second of the first ports X will likewise be evenly split between the second ports Y, but the phase shift increment is 2* A; the phase shift increment for the third input port is 3*A and so on. It will be apparent that if the second ports are connected to an evenly spaced array of radiating elements, the result is a phased array antenna, the beam offset angle depending on the array spacing, the signal frequency, and the phase increment, which itself depends upon the number of first ports and the identity of the first port at which the signal is provided. If multiple first ports are simultaneously excited, a grid of beams at incrementally shifted alignment angles are created, each one corresponding uniquely to the signal at one of the first ports. Each matrix device of Figure 4 is reciprocal, and hence the above description may be reversed; in other words, a plurality of signals applied at the second ports Y in a certain incremental phase relationship will result in the production of a signal at a single one, or multiple ones, of the first ports X so that a combination of the matrix device with an array antenna can provide a receive array antenna. If a stack of N Butler matrices of the structure of Figure 4 were positioned in parallel, in the manner shown in Figure 5, and each aligned row of output ports of all the matrices of the stack were connected to the input ports of a single Butler matrix of a second stack of N Butler matrices (in other words, the two stacks were orthogonally interconnected, where "orthogonal" in this document refers to the connection topology rather than any physical or geometrical arrangements of the devices), the two dimensional array of N: output ports of the second stack could be connected to respective radiating elements of a square two dimensional array of radiating elements, to provide a two dimensional grid of beams. In the present embodiment, however, a hexagonal grid of beams is generated. Referring to Figure 6, Figure 6a illustrates the beam pattern of the receive antenna 1, and Figure 6b illustrates that of the transmit antenna. It is apparent that the two comprise the same number of beams (121), and that the beams of Figure 6b are congruent with the receive directions of Figure 6a, but the latter are wider (and hence exhibit greater overlap, and reduced dropoff of strength between beams). To provide these beam patterns, the antennas 1, 2 each comprise a hexagonal array of radiating elements 100a-100P, 200a-200M, for example microstrip dipoles (not shown) printed on a common substrate. Referring to Figure 7, Figure 7a illustrates the arrangement of the radiating elements 20 0 of the transmit array antenna 2 and Figure 7b illustrates the arrangement of the radiating elements 100 of the receive array antenna 1. The receive antenna 1 consists of P=151 elements 100, arranged as shown, spaced at spacings of s, where s/L=1.04, and L is the antenna wavelength (2 GHz). The total antenna diameter is 2.2m The transmit antenna 2 consists of M=109 elements 200, arranged as shown, spaced at spacings of s, where s/L=1.04, and L in this case is 2.2 GHz. The total antenna diameter is 1.7m. The 109 amplifiers 18a-18M of the transmit antenna in this embodiment do not all apply the same amplification. The gain applied by each amplifier, relative to that connected to the central element of the antenna, is indicated in the circles denoting each element position in Figure 7b, from which it may be seen that the gain is tapered at the edges of Che array, by providing three outermost rings of elements 200 the gains for which are -3, - 6 and - 9 dB down relative to all the inner elements. Exactly the same is provided for the receive antenna of Figure 7a, by providing that the amplifiers 117a-117P for the three outermost hexagonal rings of elements 100 have progressively decreasing gains (again, -3, -6, and -9 dB) relative to those for the inner elements. The interconnection of the analogue beam formers 20, 120 and the antennas 1,2 will now be described with reference to Figures 5, 8 and 9. Referring to Figure 5, the first ports X of the stack 30a of matrix devices 50 form a two dimensional array of ports, each of which will be referred to by coordinates (i,j), where 0