Key payload technologies for future satellite personal communications: A European perspective

zyxw zyxw zyxwv INTERNATIONAL JOURNAL OF SATELLITE COMMUNICATIONS, VOL. 13, 117-135 (1W5) KEY PAYLOAD TECHNOLOGIES FOR FUTURE SATELLITE PERSONAL COMMUNICATIONS: A EUROPEAN PERSPECTIVE* VENTURA-TRAVESET, I. STOJKOVIC, F. COROMINA, J. BENEDICTO AND F. PETZ European Space Agency (ESTEC), P.O. Box 299, 2200 AG Noordwijk, The Netherlands J. SUMMARY zyxwv This paper deals with satellite payload architectures adequate for the provision of universal mobile telecommunication services (UMTS) by medium altitude earth and geostationary orbit ( M E 0 and GEO) satellites. The payload concept introduced is based on a transparent transponder with onboard digital signal processing (DSP) operations and digital beamforming at the traffic carrier level (narrowband beamforming). The emphasis of the paper is on key design aspects and technology considerations of the payload processor unit (PPU), which includes multiplexing/demultiplexing and digital beamforming operations, and on the antenna and RF signal handling subsystems. For completeness, a review on the European digital technology is also presented. The development of such a payload architecture is in line with present and planned activities of the European Space Agency (ESA) for M E 0 and GEO orbit systems, which are also implicitly discussed in the paper. KEY WORDS: satellite personal communications; digital beamforming; digital signal processing; European technology; multiplexing and demultiplexing; payload; repeater architecture; spacecraft antenna 1. INTRODUCTION Land-mobile satellite communications are evolving towards providing compatibility with the services offered by terrestrial digital cellular systems and complementing them in low population density areas (e.g. rural) and developing countries, where terrestrial coverage cannot be provided economically. Satellite opportunities arise when considering wide roaming requirements, and in particular to provide world-wide user roaming capability, since only about 15 per cent of the Earth’s geographical area is expected to be covered by cellular terrestrial services by the year 2000 (see Figure 1). Key factors for improving the satellite penetration include the cost/size reduction of the user terminals, reasonable user tariffs and the interoperability with the terrestrial systems. These key issues have a direct impact on the performance and flexibility required by future satellite payloads for the provision of universal mobile telecommunication services (UMTS). During the last few years, several systems for the provision of personal and mobile communications have been proposed by private companies/consortia and international organizations. A rather general classification is given by the orbit altitudes of constellations: low earth orbits (LEO), medium earth orbits (MEO) geostationary orbits (GEO) and highly-elliptical orbits (HEO). Orbit selection is the most crucial system driver with a direct impact on the payload concept. With a view to reinforcing the European industry position in this revolutionary scenario, the European Space Agency (ESA) is actively pursuing system studies and technology developments for all the above-mentioned systems. Some of the results have recently been One of the major outcomes of these studies, concerning payload architecture and technology, is the similarity among the architectural solutions for GEO and M E 0 systems, upon which we shall concentrate and describe in more detail in this paper. Our discussion will be centred on the technological and key design aspects of the different subsystems of the satellite repeater. We will introduce an advanced transparent repeater concept with on-board digital signal processing operations that include narrowband digital beamforming (DBF) and digital multiplexing and demultiplexing operations. The development of such a payload concept is in line with currently running ESA studies on architecture definition and sub-unit developments for M E 0 and GEO orbits. The paper is organized as follows: Section 2 presents the general architecture concept and the identification of the different subsystems that will be further discussed in the subsequent sections. Section 3 describes various antenna options suitable for M E 0 and GEO payloads. In Section 4, the impact of the RF-front-end sub-units will be addressed together with the identification of the crucial areas for technology improvements. In section 5 , key design aspects of multiplexing and demultiplexing subsystems will be discussed, including mixed ana- zyxwvu zyxwvutsrqpon zyxwvut * A reduced version of this paper was presented at the AIAA 15th International Communicationssatellite Systems Conference, held at San Diego. U S A . on 28 February-3 March 1994. CCC 0737-2884/95/020117-19 0 1995 by John Wiley & Sons, Ltd. Received September 1994 Revised 21 December 1994 118 zy zyxwvutsrqpon zyxwvu zyxwv zyxwv zyxwvutsrqp J . VENTURA-TRAVESET ET AL. J V' zyxwvut Figure 1. Terrestrial cellular coverage by the year 2000 logue/digital and fully digital solutions. Section 6 deals with the digital beamforming operation with a major discussion on its system advantages with respect to more conventional analogue wideband beamformers, the algorithm possibilities and different technology aspects to be considered for a DBF implementation. Some general comments on European digital technology and on its availability for future personal communication satellite designs are included in Section 7. Finally, Section 8 summarizes the main points of the paper. 2. REPEATER ARCHITECTURE A number of different architectures have recently been studied and compared via characteristics such as payload mass, DC power consumption, flexibility in matching traffic to beams, power-bandwidth flexibility, frequency reuse potential, compactness of feeder link, etc. The generic architectures studied include (1) the well-established architecture with SAW filter channelization, crossbar MMIC switching and RF beamforming (as used in Inmarsat-111, EMS, LLM), (2) transparent architecture with alldigital or hybrid (SAW-CFTldigital) channelizing, routeing and beamforming and (3) architectures with full on-board regeneration. System parameters involved in personal satellite communications drive towards solutions with many beams in the coverage area, primarily because of the low antenna gain and limited transmission power at the user terminal. This is true for all orbital altitudes, but becomes more pronounced for high orbits. High numbers of beams, high numbers of antenna feed elements and the need for high granularity to retain feeder link compactness, all drive towards solutions based on digital signal processing: the transparent payload architecture with narrowband digital beamforming, routeing and channelization was found to be a well matched solution. If there are no bandwidth constraints on the feeder link spectrum, the requirements on the granularity of the demultiplexers can be relaxed and thus architectures based on wideband beamforming can be more efficient than the narrowband ones. This paper limits its scope to transparent architectures with narrowband digital beamforming, routeing and channelization. The main advantages of this approach can be summarized as follows: a very high number of overlapped beams and use of near-peak antenna gain leads to best RF power efficiency and satellite G/T; fine demultiplexing granularity and digital switching lead to high flexibility to match traffic needs; improved frequency reuse through use of peak antenna gain and interference cancellation techniques; phase and delay adjustments of the feed chains to calibrate or correct the antenna system can be implemented in the digital BFN; one hop mobile-to-mobile connections. Although the payload processor in this case is adapted to operate with narrowband access schemes (e.g. FDMA or narrowband TDMA), the inherent transparency enables the introduction of services based on CDMA. The concept of this transparent digital payload is illustrated in Figure 2. In addition to the antenna subsystems, the payload is basically composed of three more blocks: the feeder RF front-end, the payload processor unit (PPU)-with the forward (FWD) and return (RTN) processors-and the RF front-end at the mobile side. Mobile-to-mobile single hop connection can be achieved by routeing KEY PAYLOAD TECHNOLOGIES zyxwvu z 119 ......................................... zyxwvut ................... IF : LNA D/c I * : : ...................... C/S-band - s-band Tx zyxwvutsrqp MOMk-bMObo. 1-band RX ................... t : : : ,.................: f : HPA . * Ufc , ................... : ....................... (redundancy not shown) RTN Proceuor Ne zyxw ........... ................... Figure 2. Generalized architecture for a satellite payload in GEO/MEO orbits providing personal communication services traffic from the return to the forward processors. In the following sections, a discussion on key design and implementation aspects of these units is included. 3. SPACECRAFT ANTENNA DESIGNS Different spacecraft antenna designs have to be applied to GEO and M E 0 constellations, due mainly to the large difference in their respective scanning angles. Whereas from a geostationary orbit the scanning angle is confined to approximately +9", from the M E 0 orbits this angle could be more than +20" to 230". An optimized solution for GEO has been found to be a slightly defocused reflector antenna. One of the design drivers is that the number of antenna elements should be minimized, as it has a direct effect on the power requirement of the PPU (at the same time the number of beams is not important, as it does not significantly affect the mass or the power consumption of the payload). On the other hand, it is convenient to use as many elements per beam so as to be able to spread the power over the numerous amplifiers (SSPAs) and thus retain power/beam flexibility with small Butler( -like) matrices, or avoid such matrices altogether. The best compromise solution has been found to be a defocused reflector antenna (FAFR), which combines the relatively small number of elements with a simultaneous spreading of power over a number of feed elements. For M E 0 constellations, planar array solutions are well adapted to the coverage requirements. The large scan requirements for a M E 0 payload antenna make the use of reflector antennas very difficult. Minimizing the number of radiating elements, therefore, has to be compromised and solutions are sought with direct radiating arrays. Internal ESA/ESTEC studies resulted in baseline designs for GEO' and M E 0 2 payloads based on transparent digital architectures. The results, with regards to antenna design, will be briefly presented here. zyxwv 3.1. GEO antenna design The spacecraft multibeam antennas are required to provide reconfigurable coverage of land masses from several positions on the geostationary orbit and to accommodate changes in traffic to beams, with maximum DC to R F efficiency. Over 33 dB gain is required in both the forward and return links, with 20 dB sidelobe isolation for frequency re-use. It is further assumed that the same beam footprints are used for the up- and down-links. Direct radiating arrays. Active arrays can provide the required flexibility. The use of separate transmit and receive antennas is conceptually simpler than the re-use of the same aperture, but implies complex deployment. For the same aperture, either interleaved or co-located (dual frequency) elements are possible. A configuration with separate antennas, 8 m x 2.7 m at L-band for receive and 5.1 m X 1.7 m at S-band for transmit, each with 192 subarrays of electromagnetically coupled annular slots, has been evaluated. The beams are elliptical and, even with optimum subarraying, sidelobe control requires a power inef- zyxwvu zyxwvu zyxwvu 120 J. VENTURA-TRAVESET ET AL. ficient excitation taper or use of different amplifiers. Beamforming is complex since all elements are involved for each beam. zyxwv 1 2 8 Feeds Reflector antennas. Multifeed reflector antennas are the other alternative. In the receive mode, where amplitude control at feed level has no power efficiency impact, focusing reflector antennas using beam synthesis5 lead to the smallest feed and reflector sizes. Each beam is formed by optimal weighting of pre-amplified signals from only some of the feeds. For transmit, amplifiers must operate close to nominal power for optimum DC to RF efficiency. Active focusing reflector antennas, with overlapping feed clusters, and one power amplifier at each feed, require complex power switching to cope with changes in beam loading. Image antennas6 where a feed array is magnified by one or two reflectors, suffer from reflector oversizing and require inefficient feed illumination tapering for sidelobe control. One preferred option is semi-active multimatrix antennas,’~~ as used for the INMARSAT I11 series, which provide the required performance with optimum power efficiency, together with minimum reflector and feed sizes. The same feeds are shared between several beams and are powered from identical amplifiers via Butler-like matrices, which direct the power towards the selected outputs depending on their input phase law. A design with 35X by 49X (4.2 m x 5.8 m at Sband) offset reflector (FID = 0.5) and a 128 element feed array placed on the satellite wall (Figure 3) fed via 16 8 x 8 hybrid matrices (Figure Sixteen 8 x 8 Hybrids 128 Amplifiers I Low Level Beamforming , I Beam Inputs Figure 4. Multimatrix principle 4) has been analysed for global coverage. As only land mass and 10” elevation coverage is required, the number of feeds and matrices is reduced accordingly, but not shown. Since a very large number of channels is transmitted into approximately 85 beams for land coverage, optimized complex excitations (with a limited dynamic range to simplify beamforming) can be used, as each amplifier contributes to many beams and, therefore, its power is averaged. With a 10 dB range, central beams use 3 to 7 feeds and outer ones up to 16. The cross-over levels between beams vary from -3 dB (centre) to -1.3 dB (edge). Computed contours of typical beams over the Earth’s surface with these excitations are shown in Figure 5 for the antennas of Figure 3. A scaled version of this antenna (6.5 rn x 9 m), operating in the beam synthesis mode, is proposed for the receive function. With digital beamforming it is envisaged to generate a large number of repositionable beams crossing over around 1 dB. zyxwvuts zyxwvut zyxwvu I Figure 3. Reflector antennas on spacecraft Figure 5. Selected optimized 35 dBi directivity contours zyxwvu zyxwvutsrqp 121 KEY PAYLOAD TECHNOLOGIES 3.2. M E 0 antenna design The mobile on-board antenna is to work at Lband and at S-band. Dual frequency operation is achieved with either two separate arrays (for receive and transmit) or a single array which operates at the two frequencies. The use of two antennas allows a single design, using common existing technology, to be scaled to ensure beam congruence. However, one array will need to be deployed. Dual frequency arrays can be achieved with either dual frequency radiating elements or interleaving of transmit and receive arrays within the same aperture. Separate TxIRx arrays. The design consists of array elements on an equiangular triangular lattice. The array element spacing is chosen to be 1.4h in order to avoid grating lobes falling on the Earth. This allows for beam broadening due to an aperture amplitude taper introduced to reduce sidelobe levels. Figure 6 shows the grating lobe of a beam scanned to near to the edge of the coverage region. At this stage, a 'continuous' Taylor distribution taper has been applied-the elements in each hexagonal ring of the array have the same level. A 61-element array has been considered and coverage is achieved with 37 beams. This assumes an element aperture efficiency of 90 per cent and 1.5 dB for other losses. Figure 7 shows the 37 beam coverage. The beam contours are isoflux contours relative to a gain of 24.5 dBi for users with a 30" elevation angle and represented in the Figure by the inner of the two concentric circles. The outer represents the Earth's rim. The antenna sizes and masses are 2.4 m/18 kg (L-band) and 1.5 m/7 kg (S-band). These figures include a backing structure, but not the deployment mechanism. Digital beamforming allows a larger number of beams to be used resulting in an increase of the edge-of-coverage (EoC) gain by up to 1.5 dB. For I . /----- zyxwvu Figure 6. Grating lobe for outer beam Figure 7. MAGGS-14 coverage with 37 beams the above 61-element array, the maximum EoC gain that can be obtained is 26 dBi. In this case full coverage requires generating at least 91 beams and the beams near the edge of the coverage area will cross close to their peak (a minimum gain of 25.5 dBi requires 79 beams, whereas 61 beams would fully cover with a minimum EoC gain of 25 dBi). The array elements in the above designs are assumed to be subarrays of printed elements (patches). The elements lie on a triangular lattice and the patches can also lie on a triangular lattice but of smaller spacing. An example of subarraying could be triplets of patches forming a triangle (Figure 8). In this case the patch spacing is 0:808h, corresponding to an element spacing of 1.4X. Other subarraying schemes are currently under consideration. zyxwvu Figure 8. Triplet subarrays zyxwvu zyxwv 122 J. VENTURA-TRAVESET ET AL. zyxwvu zyxwv Arrays on square lattices have not been considered in detail for the separate array case. Element spacing, to avoid grating lobes on the Earth, is about 1.25X, and these elements would then be subarrays of four patches, spaced 0.625X and forming a square. Combined TxlRx array. The use of dual frequency elements will result in far from ideal geometry for the receive band due to the element spacing, determined by grating lobe considerations required for the transmit band. Another solution employs interleaved Tx and Rx arrays occupying the same physical aperture. In one example, shown in Figure 9, both lattices are square and the Rx element spacing is 1.2% The corresponding Tx spacing is l.1X. Here an element is a subarray of four of the patches shown whose spacings are half of the element spacing. The square lattice geometry has been chosen as this determines the ratio of the element spacings at 1-414A, which is not too dissimilar to the ratio of the frequencies. The performance of the interleaved array will be degraded from that of the separate arrays, due to non-ideal patch design and lattice geometries necessary to accommodate the arrays in the same aperture. The antenna dimension/mass is expected to be about 2.5 m/20 kg assuming a low dielectric substrate as used in the separate array design. The use of high dielectric constants, in order to reduce patch physical sue to facilitate interleaving, will increase the mass. However, this mass increase can be limited by using a high dielectric constant only in the central region populated by L- and S-band elements. 3.3. Conclusion For ME0 constellations separate Tx/Rx arrays currently provide the simplest solution to dual frequency operation, albeit with the need to deploy one of the arrays. Of the two dual frequency array w e o w Figure 9. Interleaved microstrip antenna options studied, the interleaved array is preferred at this time. However, this is a new area and the array performance is difficult to predict due to unknowns regarding the radiating elements design and performance in the dual element environment with digital processing. For GEO constellations the optimal solution is a focused array-fed reflector antenna, presenting a compromise between the lowest number of feed elements and the spreading of power among them. Key technologies, required to be developed to space standard, are large deployable reflector antennas and planar phased array technology for either dual frequency or interleaved antennas. 4. TX AND RX SUBSYSTEMS To comprehensively present today’s available RF technology is outside the scope of this paper. However, we aim to indicate where the potential improvements might be achieved and emphasize the need for further developments. As we have noted in Section 2, MEO/GEO payload architectures for personal communications are driven by the large number of beams to be generated. As a consequence, these architectures are characterized by high demultiplexing granularity and antennas with many radiating elements, even when the radiating structure is very near the focal plane. The number of radiating elements to be considered ranges from at least 50-60 (for M E 0 payloads) to well over 100 (for the GEO payloads) in each direction. This means that all the RF elements in the antenna chains are replicated many times for both the forward and the return link repeaters. The mass and power consumption of these elements, within a transparent digital payload architecture, is quite significant, typically forming up to one quarter of total payload mass and at least three quarters of the overall payload power consumption. It follows that any savings on mass and improvements in power efficiency, that can be achieved in these blocks, will have a very pronounced effect overall. Main contributing elements in the transmit and receive chains on the mobile side are SSPAs, upconverters and output filters at L- or S-band (for transmit), and input filters, LNAs and down-converters at the return link side at L-band (for receive). Because of the high incidence of these blocks, huge savings can be achieved in both mass and power by paying appropriate attention to their design. The possible approach for further development would encompass some or all of the following: 1. Integrating the three transmit blocks (up-converter/SSPA/filter) as a single miniaturized package, thus reducing the mass. 2. Improving the SSPA efficiency, particularly under multicarrier or pulsed conditions of operation. 3. Reducing the power consumption of the upconverter block. zyxwv zyxwvut z zyxwvutsrqp zyxwvutsrqponm 123 KEY PAYLOAD TECHNOLOGIES 4. Integrating the three receive blocks (filter/ LNA/down-converter) into a common package, or even a certain number of these into a common package. 5 . Sharing the power supply circuitry among a number of units. The European Space Agency (ESA) is exerting a significant effort aimed at developing equipment which satisfies the requirements of personal communications payloads. As a step in this direction, an L-band transmitheceive module based on European technology has been developed under an ESA contract.9 Table I shows the overall system parameters associated with this development. The transmit module has gain, phase and gain-slope externally controlled, as well as a power stage with a DEBS bias ASIC for improved linearity and efficiency. The receive module, on the other hand, also has gain-phase and gain-slope control, as well as a receive bandpass filter, and the input stage is realized in MIC for best noise performance, using European HEMT transistors. As a result of research within the ESPRIT programme, significant progress has been achieved in the solid state arena, where several GaAs manufacturers are now very active in Europe, and MMICs can be delivered at competitive price and in quality levels. Likewise, the availability of low noise HEMTs and medium power MESFETs is good. Making use of these technologies, miniaturized low noise amplifiers (LNAs), operating at L and K , bands, have recently been developed under ESA contract. lo They incorporate discrete HEMT front ends and MMIC gain blocks, simultaneously achieving very low noise performance and a high degree of miniaturization. New solid state technologies, particularly those targeted at increased efficiency and power capability of SSPAs, have become available. Good examples are heterojunction bipolar transistors (HBT) and pseudomorphic multi-junction HEMTs (PMHEMT), both of which ESA is presently evaluating for their applicability to space. The availability of integrated technologies together with the need for miniaturized equipment is modifying the space microwave equipment scene at a very rapid pace. It is now easy to envisage payloads that are much more complex at a system level, owing to the fact that it is possible to build full subsystems based on one, or at most a few integrated circuits. Finally, improvements in filter design and manufacture are leading towards less tuning of subsystems, at least for wide-band applications. In the near future new technologies will also be applied to reduce their mass and size. 5 . MULTIPLEXING AND DEMULTIPLEXING TECHNOLOGY As indicated in the generalized architecture diagram of the narrowband-oriented satellite payload (see Figure 2), demultiplexing and multiplexing operations will respectively precede and follow the digital beamformers in the forward and return processors. The function of the demultiplexer, in both forward and return links, is to extract each individual channel from the input FDM multiplexed stream for its subsequent processing which obviously includes the frequency mapping and beamforming operations. Alternatively, the multiplexer will generate the complementary operation, i.e. the FDM multiplexed stream generation, after the beamforming and mapping processing have been performed. Though conceptually acting as single blocks, multiplexing (MUX) and demultiplexing (DEMUX) operations are usually performed in several stages by cascading several mux and demux sub-units, respectively. Depending on the system specifications and the mission technology constraints, solutions for the mux/demux routeing operations range from fully analogue ones (by means of SAW/CFT devices) through hybrid analogue/digital architectures to fully-digital ones. In the following two subsections digital and analogue multiplexing/demultiplexing operations will be reviewed at the individual block level. zyxwvu 5.1. Digital demultiplexing Introduction. The operations of multiplexing and demultiplexing basically require the same considerations when dealing with a digital implementation solution. In fact, both units are based on common algorithms, and single ASIC architectures can be conceived for the dual mode (mux/demux) operation." Based on this, we will consider only the demultiplexing operation, the same conclusions being applicable to the multiplexer counterpart. Table 1. Basic parameters of the L-band transmitheceive module9 Transmit subsystem: Receive subsystem: Gain Output power Secondary efficiency Mass Gain Noise figure target DC power Mass 40 dB 12 w >30 per cent C0.15 kg 40 dB 1.5 dB 1.1 w zyxw <0*11 kg 124 zyxwvutsrq zyxwvut zyx zyxwvuts J . VENTURA-TRAVESET ET AL. The use of digital solutions for a demultiplexing operation is unavoidable for such situations where a fine channel granularity (around 5100 kHz) is required. In these situations, SAW/CFT demultiplexing solutions are not suitable, owing to the size limitations of the quartz-crystal forming the SAW chirp delay line (see Section 5.2). Digital solutions, on the other hand, seriously compete with hybrid analoguejdigital solutions for such situations in which the total band to be processed is of the order of a few MHz (around 510 MHz). For larger processing bandwidth an analogue pre-demux processing is generally more convenient. In our discussion, we shall concentrate on the last demultiplexing units (or stages) from the several demultiplexing stages that conform to the channelization function of a payload, since they usually are the most complex devices and since their principles are also applicable to their previous demultiplexing stages. Algorithms. In a first general classification, two possible algorithm approaches must be considered: 1. per-channel processing approach 2. block processing approach. In the per-channel approach each channel is processed independently of all the others. In a block processing approach, though, all channels are processed in common, sharing particular elements such as filtering or a FFT operation. Block-processing solutions are very much superior (in terms of complexity) to the per-channel ones for such situations requiring the demultiplexing of uniform slots (all slots with equal bandwidth) or where the individual channel-slot bandwidths are related by an integer factor (e.g. related by a factor of two). We are assuming here, irrespective of the access scheme used, that the channels to be processed by our satellite are grouped in uniform slots of equal bandwidth, and that each slot is individually beamformed in the narrowband beamforming operation. With these two assumptions, then, block processing solutions are the only ones to be considered. * Having selected the block-processing approach, two families of algorithms will be considered: ture (see Figure 10) consists of a polyphase bank of filters, followed by an FFT block with a number of points equal to the number of channels to be demultiplexed (or double if a complex representation is used) and a second filter stage (channeldefinition filter) separate for each channel. Conceptually, the demultiplexer structure could be performed with a single filter stage preceding the FFT block, so avoiding the channel-definition stage. However, for our mobile satellite concept, a single-filter-stage approach would result in impractically long filters. A reported variation'* on this general two-stage polyphase-FFT algorithm consists of replacing the second bank of filters (the filter-definition stage) with a single front-end network based on an interpolated Hilbert transform FIR filtersm A similar solution, recently proposed in an ESA contract," consists of a first-stage tight filter separation between even and odd channels in the FDM multiplex (via a pair of imaged, lowpass and highpass, halfband filters) such that the filter requirements of the second stage polyphase-FFT (consisting now of two blocks, one for the processing of even and one for the processing of odd channels) are very relaxed. In fact, by working at twice the required Nyquist sampling rate and having rejected alternate channels of the input frequency multiplex, the transition width between slots is correspondingly increased. The filter structures of the FFT-oriented demultiplexers are usually FIRS since they can easily be designed to have linear phase and to operate at the reduced decimated output rate. Furthermore, FIR architectures are more convenient when in addition reconfigurability is required for the polyphase architecture. A reconfigurable architecture can deal with such situations requiring some sort of flexibility in the system specifications (e.g. different access schemes with associated different slot granularities, contiguous or non-contiguous filtering, etc.) but in general is not simple to implement. For tree-oriented algorithms, on the other hand, the demultiplexing operation is based on a successive division of the input signal into smaller frequency bands at each stage of the tree using two complementary halfband filters and a four-point DFT. zyxwvutsrq z zyxwv 1. FFT-oriented algorithms 2. tree-oriented algorithms. FFT-oriented algorithms are basically the DSP implementation of the SAW-CFT processors described in Section 5.2. The generalized architec- * Note that other payload solutions, e.g. with wideband beamforming, may require flexible demultiplexers to allocate several slots to a given beam, and thus other solutions, including fixed analogue filters, may be more suited. PolyphaseFilters Channel DefinitionFilters 0 Figure 10. Illustration of the polyphase FFT demultiplexing approach zy zyxwvutsrqp zyxwvutsrqp zy 125 KEY PAYLOAD TECHNOLOGIES The common processing cell is a polyphase-m structure for two channels, which is repeated as many times as needed (Figure 11 illustrates for the case of N = 18 channel^'^). Imaged half-band filters have the property that passband and stopband have the same width and peak-to-peak associated ripple values. This results in a symmetric unit sample response where every even coefficient (except the central one) is zero. Having designed the lowpass halfband filter, the complementary halfband highpass filter is very simple to obtain by multiplying by (-1)" the filter coefficients of the lowpass version. This means that the two imaged halfband filters are contiguous, i.e. their transition bands add to provide an unit response. The four-point DFT, on the other hand, involves only additions and trivial multiplications by j , resulting in an efficient hardware implementation. An important feature of the tree algorithm approach is the possibility that it offers for the demultiplexing of an FDM multiplex input consisting of non-uniform channel slots, whenever their respective bandwidths are related by powers of two. This is accomplished by 'cutting' the tree at such branches and depth where the associated slots need no further filtering, even though other branches need further splitting. Comparison of the algorithms. There is no simple answer as to the most suitable solution for the on-board multiplexing/demultiplexing implementation. In particular, having selected a block-oriented solution, the selection of the most adequate algorithm from the two above-mentioned families depends very much on the system requirements for the demultiplexer. As a general guideline the following considerations can be made: 1. In general, implementations based on the FFTbased algorithms become more efficient than the tree-algorithm solutions as the number of channels to demultiplex increases. This is basically due to the efficiency of the FFT. When the number of channels is small, on the other hand, tree-based implementations are reported to be ~ u p e r i o r . 'The ~ ~ ~crossing ~ point is not clear and depends very much on the particular hardware approach selected for each case. 2. The antialiasing filter shape factor determines the portion of wasted slots in the input stream. It has been noted that tree-oriented solutions are best for high shape factors since branches O1 zy O2 O3 O4 O5 6' O7 zyxw 8' Figure 1 1 . Illustration of the tree-based dernultiplexing approach (Nc= 8) 126 zyxwvu zyxwvu zyxwvu J. VENTURA-TRAVESET ET AL. zyxwvutsrq zyxwvutsrq zyxwvu of the tree which have non-useful outputs can be predetermined and conveniently cut on the implementation. This is not possible in an FFT implementation where all input channels (useful or not) will be demultiplexed with subsequent extra-complexity. 3. The superior flexibility of the tree-algorithm, in the sense of handling slots of different bandwidth (BW), has already been commented upon. Some solutions based on the FFT approach that allow some degree of flexibility in the slot BWs have been reported,16 but they are not as hardware efficient as the uniform solutions. 4. With respect to redundancy considerations, the tree algorithm has been reported to be superior. l4 A polyphase-FFT implementation will fail completely if the first filter stage or the processor fails, so these two units should be duplicated if redundancy is considered. The tree algorithm, on the other hand, will completely fail only if the first tree-cell fails. Failure on a deeper cell stage of the tree will provoke the loss of such slots still associated with such a cell. So, an efficient redundancy can be achieved by replacing only the cells associated with the first tree levels. 5 . Finally, regarding the hardware design some arguments in favour of the tree algorithm are based on its greater regularity (repetition of common cells) which is favourable to an ASICoriented implementation. l5 As can be seen from the previous paragraphs, the issue on the selection of the best demultiplexing algorithm is a complex one. In general, we can conclude as follows: 1. Block-processing algorithms are more adequate than per-channel-processing for the muxIdemux operation in mobile Personal Communication payloads based on narrowband digital beamforming processing uniform slots. 2. Two basic algorithms remain for the demultiplexing implementation: the FFT and the tree-oriented algorithms. The performance of those two algorithms in terms of operation and complexity is very similar. For a final selection, some system specifications (described above) need to be considered together with the particular experience of the designedmanufacturer company. Implementation issues. The most crucial system specification in the design of a digital demultiplexer is the equivalent individual slot-filter characteristic. Pass, transition and rejection bands are determined by the slot separation in the input stream. The amount of transition band, and thus the bandwidth efficiency per slot, is clearly an important system specification which will directly affect the demul- tiplexer complexity. To finally determine the number of taps of the demux filter we must specify the stopband rejection together with the maximum allowed passband ripple. Similar procedures are described in References 11 and 13 for the derivation of these parameters. To determine the stopband rejection it is important to know the number of interfering slots that will alias into the wanted channel as a consequence of the demultiplexer output decimation. The effective number of interferers basically depends on the number of slots processed by the demultiplexing unit together with other important system specifications, such as the space filtering rejection of the preceding beamformer (so reducing the effect of the interfering slots), the voice activation factor and the traffic statistics (determining the number of slots that are simultaneously occupied at a given time). These interfering channels together with the stopband rejection of the demux filter will determine the resulting CI I level at the wanted channel. Given the CII value we can determine the overall CIN degradation at the demultiplexer output associated with such stopband rejection, by approximating the effect of interferers as internally generated white Gaussian noise and adding it to the system AWGN (e.g. for a 0.5 dB degradation, internally generated aliasing ‘noise’ should be approximately 10 dB below the AWGN). Alternatively, having fixed a level of degradation for the demultiplexer block, we can determine the requirements of the stopband rejection of our filter. This specification, together with the passband ripple will determine the filter length. Note that the above procedure is based on several approximations. To determine the actual filter coefficients, simulations are required. Furthermore, for a given filter length, the filter coefficients can be optimized with the criterion to minimize the overall CIN degradation at the output of the demultiplexer. In this way, passband ripple and stopband attenuation can be traded off against each other so as to optimize the overall CIN degradation. In the case of a reconfigurable demultiplexer architecture, the previous analysis will of course be performed for the most demanding filter mask. It is important to note that given an overall figure for the allowable PPU degradation, an analysis must be performed to efficiently allocate the various contributing degradations to each one of the subblocks of the digital PPU (degradation budget). For instance, it is convenient to overdimension (and thus to allocate very little degradation) such blocks of the PPU that are less repetitive (e.g. the blocks at the feeder section) thus relaxing the specifications of other PPU blocks that are used many times (e.g. the blocks at the mobile section). This approach can lead to an efficient optimization of the payload power and complexity for a given value of the overall PPU degradation. Other important specifications concern the number of quantization bits required at the demux z zyxw KEY PAYLOAD TECHNOLOGIES inputs/outputs, and the precision levels and truncationhounding philosophy at the internal subprocessing units of the demux block (e.g. filter taps representation, internal metrics, etc.). Again, in a first approach these specifications can be obtained through analysis. The procedure is based on determining such quantization levels that meet a given degradation specification. For instance, one can determine the associated quantization and clipping noise due to the A/D converter operation. It is well known that these noise terms will depend on the number of bits that are used to represent a sample (which must include the sign bit) and the ratio between the r.m.s. signal level and the A/D converter range. As a first approximation, the noise terms can be considered to be independent and their effect to be uniformly spread across the whole processing band. In this way, given a value of the CIN ratio at a given point of the PPU, one can determine the degradation by adding the quantization and clipping noise as additional and equivalent contributing thermal noise terms. Simulation with real scenarios is required to obtain actual degradation values, and thus to finally fix the quantization parameters. In Reference 13 a complete procedure is described relating the quantization noise in the demultiplexer (and so determining the number of bits required) to the level of degradation allowed on the demux, the number of antenna array elements, the frequency-reuse factor, aliasing products and the CIN values of the link budgets. Other implementation issues are related to the input/output interfaces. For a digital demultiplexer a complex representation of the input signal seems to be more suitable than real representation with respect to the internal demultiplexing filter operations and following processing blocks (DBF).l1 The demultiplexing output will be more convenient when represented as a TDM stream with time multiplex complex samples associated with each demultiplexed slot. In this way the output connections as well as the input connections to the following processing blocks (digital switch, DBF) are minimized. We conclude this section by mentioning some ESA developments in this field. ESA has already developed a 16-channel demultiplexer ASIC for onboard application^'^ based on the tree algorithms. For completeness, the design of the dual 16-channel multiplexer is currently under development under another ESA contract.” Furthermore, ESA is now pursuing” the ASIC implementation of a dual multiplexing/demultiplexing architecture based on the polyphase-FFT algorithm, to be integrated in a demonstration payload processor unit (PPU) based on narrowband beamforming. Finally, a similar payload concept has been studied for ESA in Reference 18 and the associated demultiplexer technology and ASIC implementation is under development in Reference 19. 127 5.2. SAW-CFT demultiplexing General considerations. The hybrid analogue/ digital demultiplexer described here relies on an analogue transmultiplexing process based on the chirp Fourier transform, where the input signal is transformed from the frequency to the time domain and vice versa. It is basically an analogue technique, which can be used to perform relatively narrowband frequency demultiplexing (and multiplexing) to a large number of slots in a single hardware unit. The coarse demultiplexing achieved in this way is further enhanced by digital means in terms of narrower slots and very small guardbands. Also known as compressive spectrum analysis, the CFT is a particular case of the chirp-2 transform, originally developed for improving the processing resolution of Doppler radar systems. Much of the early work in this field was performed in the late 1950s (at Stanford University), but now it is a well established method making use of advancements in surface acoustic wave (SAW) technology and is used for a wide variety of applications ranging from early warning radar detectors to FDM/TDM transmultiplexer~.~’The application of this method to a mobile/personal communications satellite repeater is novel and the idea has been pursued by ESA through a number of development contracts. l1 The principle of the chirp Fourier transform is based on the processing of a linear frequency modulated signal by a dispersive (compressive) delay line used in its inverse mode to compress such a signal into a short pulse. The input signal is mixed with a linearly swept local oscillator (chirp) and downconverted to the IF band (the operating frequency band of the dispersive filter). Each signal frequency component present in the input band during the LO sweep moves across the IF band as a sweeping tone that appears at the IF as a linear frequency modulated signal, i.e. a chirp. The sweep rate of the local oscillator is matched to the delay slope of the SAW dispersive delay line, so that a time compressed pulse appears at the output of the processor for every frequency component of the input. The overall result is a domain conversion of frequency to time. The output signal, in practice, has a considerably higher bandwidth than the input signal, but is compressed in time. When the output of the CFT is sampled at an instant in time inside a CFT frame, the sampled voltage is identical to the output of an equivalent bandpass filter.20 The impulse response of this filter is identical to the amplitude weighting function of the dispersive delay line. When the CFT is combined with DSP, the main design objective for the weight function is to suppress aliasing noise. If the slope and repetition rate of the chirp generator are correctly chosen, there is no loss of information in the transformed signal, other than that due to noise and distortion, which can both be kept within predefined limits. The most significant component of a CFT pro- zyxwvu zyxwvu 128 zyxwvutsrqpo zyxwvu zyxwvu J. VENTURA-TRAVESET ET AL. cessor, from a functional, as well as a design aspect, is the SAW dispersive delay line (reflecting array compressor, RAC), whose basic structure is illustrated in Figure 12. Its primary characteristic, the linearly variable group delay with frequency, has to be maintained in a relatively large frequency band, within the specified environmental conditions (temperature range, mechanical environment etc. ) Research projects conducted by ESA/ESTEC11*26 indicate that an operating input bandwidth in excess of 30 MHz, i.e. the full mobile allocation in L-band, is achievable. Current state of CFT technology in Europe and ESA involvement. The technology has been successfully operationally demonstrated and there are a number of units currently used for experimentation. However, there are still problems which need to be overcome before submission for space qualification. The main problem encountered has been related to packaging and the change of parameters when exposed to mechanical shocks or random vibration. To this end, novel developments for the correction of the phase and amplitude responses are considered involving real time measurements. Fine tuning and accuracy of phase tracking between different CFTs, which can be practically realized, is particularly important for payloads where a number of feed elements are used to form a beam (as would most probably be the case for a global personal communication mission, whether M E 0 or GEO), because the positioning, gain and sidelobes of the beam will critically depend on the phase tracking of the CFTs. Results so far achieved are encouraging and indicate that the total phase error introduced by the CFT can be of the same order as other errors in the antenna element chain, and thus not be an outstanding problem. On the other hand, this problem might be reduced if the beamforming is digital (as we expect would be the case), because compensation could be introduced on a per-channel basis and recalibration of the system performed fairly regularly and automatically. The present research activity supported by ESA, aimed at improving the packaging is near completion. If the results are encouraging and activities on CFT technology continue, space qualification could be expected in 1996, according to ESA estimates. Finally, in a separate contract" a SAW CFT device is being developed with the aim of eventually demonstrating its operation in a full payload processor environment. zyxwvut zyxwvutsrq An example of a CFT for mobile application. The technical complexity of the CFT is summed up by the time-bandwidth product. In general, the wider the bandwidth of the SAW device (dispersive delay line), (a) the shorter the SAW device (b) the shorter the impulse response (c) reduced degree of selectivity, i.e. wider minimum frequency slots and guardbands. In an ongoing ESA project,*l it has been demonstrated that temperature stable quartz SAW substrates allow time-bandwidth products up to 3000 to be realized for SAW chirp lines with typical weight functions. Table I1 shows typical parameters of a CFT unit suitable for the processing of the full L-band mobile frequency allocation (34 MHz) and optimized to perform, in conjunction with a fine digital demultiplexer, the breakdown of the input spectrum to slots of 30 kHz.20 The fine digital demultiplexing needed for further demultiplexing of each CFT slot is based on the same principles as described in Section 5.1. The order of the demultiplexing will be low for the CFT case which will favour the use of the tree-oriented algorithm. To make an efficient overall processor, the DSP chips must be capable of handling a large number of CFT slots in time multiplex. Compared to the chips for the all-digital case, the chips in this case will need more memory to be capable of time multiplexing between a large number of CFT slots, but will be simpler due to the overall lower order of demultiplexing. Conclusion. Based on earlier ESA studies, the CFT-based (de)multiplexing principles and their implementation for mobile and personal communications payloads are well known and the hardware has been demonstrated. Our system analyses have shown that CFT multiplexing and demultiplexing, although capable of coping with a very wide range of input bandwidths and slot sizes, yields the best results, compared to competitive all-digital techniques, when the overall processed bandwidth is relatively large. The breakpoint depends on a number of system parameters and ranges from several megahertz to 15 MHz. Processing of bandwidths in excess of 15 MHz will almost certainly be more efficient, especially in terms of DC power requirement, by employing CFT techniques rather than using an all-digital approach. Demultiplexing to slots narrower than approximately 250 kHz, although perfectly achievable with CFT only, will be better performed if the CFT process is augmented by digital demultiplexing, because smaller SAW devices zyxw zyxwvutsr Figure 12. Principal structure of the SAW reflecting array compressor (RAC)'" zyxwvu zy 129 zyxwvutsrq zyxwvut KEY PAYLOAD TECHNOLOGIES Table 11. Typical parameters of a CFT unit suitable for a full L-band mobile processing zyxwvutsrq zyxwvutsrqpo Bpass = 34 MHz fc k = 200 MHz 0.85 Sampling T, 5 ps T, = 2 ps 5 processed bandwidth centre frequency of SAW chirp line guard factor for CFT output frames complex duration of dispersive response CFT slot bandwidth CFT frame repetition rate =lo0 MHz 240 KHz Ripple in passband needs to be compensated by simple DSP filter Suppression of aliasing 40 dB Processing delay <30 ps (for full CFT/ICFT chain including cover filters) Order of digital demux 8 Maximum phase drift 54.2” for k 3 0 K temperature Power consumption 1.35 W (including all amplification and digital demux to 30 KHz slots) Total mass 275 g BSAW can be used and much tighter guardbands can be achieved. In most cases when FDMA or NB-TDMA access schemes are involved, the best overall solution, with respect to power requirements, will be a hybrid CFT/digital demultiplexer, if the processed bandwidth exceeds 10 to 15 MHz. The analogue CFT process should be used to (de)multiplex to slots of a size suitable for efficient digital demultiplexing and post processing to be implemented. In this way we can maximize overall performance: the power efficiency of broadband analogue processing and the accurate and tuning-free digital processing coupled with minimized guardbands achieved by digital filtering. 6. DIGITAL BEAMFORMING General description The basic idea behind digital beamforming is the same as that used in R F or IF beamforming: A signal impinging on the array antenna will arrive at different time instants to the different radiating elements, depending on its direction of arrival. Taking the signal present at one of the elements as a reference, and if the signal can be considered narrowband, the different times of arrival at each element are equivalent to a difference of phase.* The function of the beamformer is to first phaseshift each radiating element signal by the exact amount corresponding to the expected direction of * This is the assumption made for narrowband beamforming which is considered throughout this paper. arrival of the wanted signal and then sum all the signals. In this way, only the signals arriving from the expected direction of arrival will be added in phase, whereas any other signal will be subject to less enhancement (or even cancellation), depending on their direction of arrival. The thermal noise generated at each radiating element branch, being uncorrelated between different radiating elements, will be added incoherently, resulting in an ideal improvement of 10 log(N,) in signal to noise ratio, N , being the number of radiating elements. In order to control the sidelobes of the radiation pattern, a certain amount of amplitude control can be introduced into the beamformer branches. The same ideas apply in both receive and transmit directions. Digital beamforming performs this function in the digital domain, after sampling the complex envelope of the radiating element signals at the proper sampling rate determined by the Nyquist criterion. The basic block diagram of a receive digital beamformer is shown in Figure 13. The signals are first amplified, down-converted and sampled (either at baseband or at a suitable IF frequency) and then multipled by complex coefficients and added. The modulus and argument of the complex coefficients are equivalent to the amplitude and phase shift tapering on R F beamformers. Advantages of digital narrowband beamforming applied to mobile communications payloads Power consumption and mass of the digital beamforming function depends linearly on the total band- 130 zyxwvu zyxwv zyxwvutsrqp zyxwvuts J. VENTURA-TRAVESET ET AL. d Multi element antenna i 1 N elements 0 = Beam angle Figure 13. Basic block diagram of a digital beamformer width processed and on the number of radiating elements and does not depend on the number of beams. This is the major difference compared with classical RF beamforming; where the number of beams to be implemented is very high, RF beamforming would not lead to practical implementations. Then, if the total processed bandwidth is frequency demultiplexed to a number of frequency channels, different beamformers can be implemented in each of the frequency channels. In the limiting case, a single channel per beam could be implemented without a major cost increase in beamforming. From the beamforming point of view, the advantages of the implementation of a single channel per beam are twofold: First, assuming a fixed antenna size, to achieve a certain value of e.i.r.p., less RF power per channel is necessary on board the satellite, with the associated reduction in total power consumption (alternatively, a reduction of antenna size can be implemented keeping the RF power constant). Similar advantages are obtained from the G/T point of view in the receive case. Secondly, the frequency reuse of the available spectrum can be increased. This is seen in Figure 14. Two beams serving users on the same frequency channel can be spatially closer than with the classical beamforming approach, the reason being that the worst case of cochannel interference in a classical analogue fixed beam coverage concept is given by the signals arriving at the edge of coverage. When a single agile beam per channel is considered (as in digital beamforming), then the concept of edge of coverage has no meaning. Using digital beamforming to achieve single channel per beam coverage, the frequency reuse factor can be increased by a factor of approximately 2.3. Obviously, the advantages associated with the use of a single channel per beam can also be obtained if each beam serves not one but a group of channels, provided that all of these channels correspond to users very closely spaced on the ground and therefore arriving from close to the peak of the beam. In this way the total power consumption and mass on board can be significantly reduced, due to the reduction in the channelization requirements. Other advantages of digital beamforming are: 1. The calibration and compensation of phase and amplitude errors originating in the analogue part of the antenna and beamformer (LNAs, up/down-converters, SSPAs, diplexers, antenna deformations, etc. ) becomes much easier. 2. Active interference suppression techniques can be easily implemented on board to reduce the cochannel interference due to signals using the same frequency arriving from different beams on different directions. These techniques can additionally increase the frequency reuse capability of the payload. An alternative to active interference suppression would be to perform a beam synthesis so as to obtain optimum beam patterns that maximize the signal to interference ratio. This beam synthesis could be performed on the ground and based on the known direction of arrival of the active signals in the system. 3. Digital beamforming-related techniques allow the possibility of finding the direction of arrival of the different signals at the antenna aperture. This function permits the correct pointing of the beams to the mobile users and thus the optimization of the antenna gain. In order to perform this function, two main techniques have been identified (it is assumed that a user of the system will first transmit an unmodulated tone that will be used by the user location processor on board): zyx zyxwvuts zyxwvut 131 KEY PAYLOAD TECHNOLOGIES DBFN AGILE BEAM ANALOG FIXED BEAM Distributed ws Distributed INT ws INT 1 1 20 dB zyxwvuts I I k Min. Separation 1.61 HPBW 4 4 4 Min. separation 1.05 HPBW \ / POTENTIAL FR ADVANTAGE OF FACTOR OF 2.3 Figure 14. Illustration of the frequency reuse factor increase with DBF due to the closer spacing between adjacent beams (a) Superresolution techniques such as MUSIC, where estimates of the directions of arrival are obtained based on the analysis of the covariance matrix, obtained by the cross-correlation of the signals on each of the radiating elements. The operation of correlation can be performed digitally on board, and the analysis of the covariance matrix can be performed on the ground. The result is the direction of arrival of the signal at the antenna. With MUSIC, high accuracy estimates of the direction of arrival can be obtained. The drawback is that a very acurate calibration of all errors on the analogue front-ends is required and that the covariance matrix analysis has to be performed on ground. Once the direction is known the appropriate beam can be implemented in the digital beamformer on board. (b) Reference-based beamforming technique. This technique obtains a beam that maximizes the signal-to-noise ratio of a signal of known frequency and modulation characteristics, without requiring a knowledge of its direction of arrival. A simplified block diagram is shown in Figure 15. This technique uses an adaptive algorithm and it is based on the property of correlation of the received signal with a locally generated reference.** In the case of an unmodulated carrier, the reference can Figure 15. Simplified block diagram of a reference based beamforming technique for on-board user location be generated with a simple PLL placed at the beam output. After convergence, the weightvectors on the beamformer will generate a beam pattern pointing to the incoming signal and compensating for any phase/amplitude tracking error on the element receiving chains. Therefore, this technique does not require any calibration of the phase and amplitude errors of the analogue frontends. In addition, all of the processing can be performed on board; therefore, a reduced number of satellite to ground links is necessary. Simulation results of this reference-based technique have shown that a very fast convergence is achieved, although a beam pattern with only coarse pointing to the direction of arrival of the incoming signal is obtained. In order to improve the performance, the modified architecture zyxwvu zyxwv 132 J. VENTURA-TRAVESET ET AL. shown in Figure 16 can be used. It consists of the basic architecture of Figure 15 plus a correlator block. When the adaptive algorithm has forced a coarse pointing of the beam, the reference generator will be locked to the incoming signal. Then, the locked reference signal is correlated with each of the signals present at the radiating elements. The result is the optimum weight vector that generates a beam highly matched to the direction of arrival of the incoming signal, and compensating for all phase and amplitude errors that may be present on the receiving chains, without the need for calibration procedures. This technique has been implemented on a digital beamformer breadboard23 with the following results: for an L-band phased array with 18 radiating elements and six degrees total 3 dB beamwidth, pointing errors of less that 0.1 degrees have been measured with correlation based on only 100 samples and with a CIN ratio of -5 dB at the radiating element inputs. (c) Simple direction estimation techniques can also be used, such as direct correlation between signals at radiating elements (i) or comparison of power levels at the output of different adjacent beams (ii). Both techniques are very simple to implement and their associated accuracies will depend on the quality of the signals in terms of CIN ratio (for (i)) and on the number of beams used (for (ii)). In general, their accuracies will never be as good as with superresolution or reference-based techniques, although they may have sufficient accuracy for the specific application. Implementation issues A digital beamformer is conveniently implemented using application specific integrated circuits (ASIC). Very generally, and assuming as an example a frequency multiplexing of carriers, such an ASIC accepts a time multiplexing of sampled complex signals that can correspond to different demultiplexed frequency slots of one or several radiating elements. The ASIC performs the basic beamforming function (complex multiplication by a set of complex coefficients and the addition of the resultant products). The complex coefficients will change function as to the particular time slot being processed. Important issues are the quantization of the input signal samples and of the complex coefficients of the beamformer. The number of bits necessary to quantize the signals is related to their total dynamic range: the quantization noise will be kept well below the local level of thermal noise in order not to degrade the total CIN ratio, and the full range value will be above the maximum expected signal value in order to avoid clipping of the signals, which would degrade the performance. The number of bits required on the beamformer complex coefficients will be such that the associated quantization sidelobes on the antenna radiation pattern are kept below a certain specified value. When spatial nulling of cochannel interference is considered, the depth of the achievable nulls, and hence the level of suppression of the interference, is related to the quantization of the complex coefficients. As an implementation example, a digital beamformer ASIC has been developed under ESA contract.23 This ASIC accepts a time multiplex of 8-bit sampled complex signals with a maximum total rate of 32 MHz and performs the beamforming operation (complex multiplication and accumulation) with 11bit complex coefficients, providing a 16-bit beam output. This ASIC has been manufactured using a 1 pm standard cell CMOS technology, and its power consumption at maximum 32 MHz multiplexed rate is 0.75 W. Using this data and considering an application with a total feeder link bandwidth of 25 MHz and an array antenna with 100 radiating elements, the total digital beamforming function would consume around 60 W, irrespective of the number of beams. In addition, ESA is at present developing an on-board oriented dual TxIRX DBFN ASIC" to be integrated in a demonstration payload processor unit (PPU) (mentioned in Section 5.1), which should prove the concept of the digital transparent payload being considered herein. Finally, a similar activity related to the ASIC implementation of a narrowband DBF has recently been initiated,I9 which includes the integration of several of those basic DBF ASICs into a multichip module (MCM) package (see Section 5.2). zyxwvuts zyxwvutsrq zy zyxwvutsr &xp--b--, zyxwvutsrq . ,. .............................................. AVERAGE Figure 16. Reference based beamforming for on-board user location including a correlator block zyxw zyxwvutsrqp 133 KEY PAYLOAD TECHNOLOGIES 7. EUROPEAN DIGITAL TECHNOLOGY Reference 24 provides an excellent review on the problems of VLSI for DSP in space and on the ESA activities performed in this field. We intend here, for the sake of completeness, to highlight the basic points. Though complex digital implementations are nowadays very common in commercial applications, satellite payloads have traditionally used predominantly analogue devices. The main problems arise with the qualification process of the digital technologies and the lack of experience with space flight digital units. The European Space Agency (ESA) has for several years pursued the development of digital ASICs for on-board digital devices: as has been mentioned, for instance, the developments, under ESA contracts, of digital demultiplexers and DBF ASICs (see Sections 5.1 and 5.2, respectively). This effort in ASIC design and manufacturing is complemented by other ESA efforts in producing European digital flight qualified technologies from several different foundries. In summary, major parameters of ASICs that need to be considered for their inclusion in digital payloads in space are - . (a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (1) Table 111. Forecast of European radiation protected digital technology for a satellite payload to fly in 1998-2000 Feature size: Maximum number of gates (per ASIC): Power (FWlgatelMHz): 0.8 pm 150,000 3.0 shows the forecasts for European technology16 availability in view of an hypothetical flight in 1998-2000. 3. Finally, the ASICs contribution to the overall mass and power of the satellite payload have been estimated in References 11 and 16. In the former, it is stated that PCBs with board areas of 20 cm X 15 cm can accommodate up to 50 ASICs and/or a maximum of 5 W of total power dissipation. A mass of approximately 5.4 kg has been considered for a unit with 9 such PCBs and their associated power supply. These estimates consider a payload implementation using ‘conventional’ PCB technology. It can be anticipated that for the payload concept being considered, the number of ASICs that will be required will be of the order of several hundred with high numbers of interconnections, and power dissipation of the order of a few hundred watts. A first attempt to reduce the power consumption is via the use of CMOS technologies based on 3.3 V operation. With respect to the mass, it becomes clear that the use of packaging methods based on the use of thin film multi-chip module (MCM) techniques will become an essential issue. An improvement by more than a factor of two, in terms of mass, can be achieved through the use of MCM technology based on conventional Kovar packaging. A further improvement of about 20 per cent is expected via the use of aluminium alloy packages. To this end, an ESA activity is at present considering the study and demonstration of representative MCM packaging technique^^^ for their use within digital payloads. zy zyxwvuts integration capabilities radiation hardness packaging power dissipation delaykpeed characteristics possible use of embedded blocks and macrocells library richness and maturity variations with supply voltage and temperature maximum pad/pin count available maximum die size qualification status design tools maturity. The European Space Agency process for ‘capability domain approval’ is rigorous. There are several European foundries that are well placed for the space qualification of their most advanced bulk and SOS CMOS technologies and some of their basic technologies have already been qualified. Some preliminary analysis and considerations of the European technologies required for a mobile personal communications payload in 1998-2000 have been performed under ESA contract.16 The main points are summarized as follows: The long time scale of space hardware design developments creates a 3 to 4 year delay with respect to the state-of-the-art of radiation protected technologies. Accordingly, a digital hardware scheduled for flight in 1998-2000 will have to be designed with radiation protected technologies of circa 1994-1996. For radiation protected technologies, Table 111 8. SUMMARY This paper has presented a technology overview of the different subsystems of a payload concept with on-board digital signal processing operations, including narrowband digital beamforming and digital multiplexing and demultiplexing blocks. This payload concept is adequate for the provision of universal mobile telecommunication Services via ICO/MEO or GEO satellites. Key design aspects, implementation issues and future improvement trends have been discussed for each one of the repeater subsystems with emphasis on the payload processor unit (PPU) which includes the DSP repeater blocks. The development of such a payload architecture 134 zyxwvutsrqpo zyxwvu is in line with present and planned activities of ESA for ME0 and GEO orbit systems, which have also been implicitly discussed in the paper. PPU RF Rx APPENDIX: GLOSSARY OF ACRONYMS AWGN ASIC AWGN BFN BW C/I C/N CDMA ClT CMOS DBF DBFN DC DEBS DSP ECL EMS EoC ESA ESTEC FDMA FFT Additive white Gaussian noise channel Application specific integrated circuit Additive white Gaussian noise Beamforming network Bandwidth Camer level relative to interference level Carrier level relative to noise level Code division multiple access Chirp Fourier transform Complementary metal oxide semiconductor Digital beamforming Digital beamforming network Direct current Dynamically efficient biasing scheme Digital signal processing Emitter-coupled logic European mobile satellite Edge of coverage European Space Agency European Space Research and Technology Centre Frequencykime division multiple access Focused array-fed reflector (antenna) Frequency division multiple access Fast Fourier transform Geostationary Earth orbit Heterojunction bipolar transistor High electron-mobility transistor Highly inclined elliptical orbit High power amplifier Inverse chirp Fourier transform Low earth orbit L-band land mobile (payload) Low noise amplifier Medium altitude global satellite system Medium altitude Earth orbit Metal semiconductor field effect transistor Microwave integrated circuit Monolithic microwave integrated circuit Narrowband digital beamforming Narrowband time division multiple access Noise power ratio Pseudomorphic multijunction high SAW SDR SNR SOS-CMOS TDM TDMA Tx UMTS electron-mobility transistor Payload processing unit Radio frequency Receive(r ) Surface acoustic wave Signal to distortion ratio Signal to noise ratio Semiconductor-oxidesemiconductor CMOS Time division multiplex Time division multiple access Transmit(ter) Universal mobile telecommunication system zyxw zyxwv REFERENCES 1. J. Benedicto, P. Rinous, I. Roberts, A. Roederer and I. Stojkovic, ‘Geostationary payload concepts for personal satellite communications’, IMSC‘93, Pasadena, June 1993. 2. P. Rastrilla, I. Stojkovic, J. Benedicto and P. Rinous, ‘Mediumaltitude payload concepts for personal satellite communications’, A I M 15th International Communications Satellite Systems Conference, San Diego, USA, FebruarylMarch 1994. 3. J. Ventura-Traveset, I. Stojkovic, F. Coromina and J. Benedicto, ‘A technology review for future satellite personal communication payloads’, A I M 15th International Communications Satellite Systems Conference, San Diego, U.S.A., FebruarylMarch 1994. 4. J. Benedicto, G. Solari and I. Stojkovic, ‘Regional and global personal communication satellite systems’, PlMRC‘93, Yokohama, Japan, September 1993. 5 . R. Le Normand, E. El-Shirbini, J. Neron, J. P. Marre, B. Vidal Saint Andre, R. Coirault and E. Rammos, ‘A versatile array fed reflector antenna’, IEEE AP-S Int. Symposium Digest, 1988. 6. R. Lo Forti, T.Jones and A. Roederer, ‘Performance evalu- zyxwvuts F/TDMA FAFR zyxw zyx zyx J. VENTURA-TRAVESET ET AL. GEO HBT HEMT HE0 HPA IClT LEO LLM LNA MAGSS ME0 MESFET MIC MMIC NB-DBF NB-TDMA NPR PM-HEMT ation of a near field fed double curvature reflector’, IEEE AP-S Int. Symposium Digest, 1990. 7. S . Stirland, J. Joshi, A. Roederer and R. Graham, ‘Active and semi-active reflector antenna’, IEEE AP-S Int. Symposium Digest, 1991. 8. A. Roederer, ‘Semi-active satellite antennas’, Invited paper Proc. JINA, 1992. 9. Matra Marconi Space UK limited, ‘Transmitlreceive module’ ESA Contract 10418/931NLIDS, currently ongoing ESA contract. 10. DASA, ‘Final report on technologies for LNA’ ESA Contract 92901901NLIUS. 11. British Aerospace, Alcatel Espace, Frobe and AME, ‘Study and development of digital beamforming techniques’, ESA Contract 10381/93/NLIN, currently ongoing ESA contract. 12. F. J. Lake and A. W. Wishart, ‘An efficient architecture for digital multiplexing and demultiplexing using a Hilbert transform network’, Proc. Second Int. Workshop on DSP Techniques applied to Space Communications, 1990, ESA WPP-019. 13. ANT Bosch Telecom, ‘Final report of VLSI development of a digital demultiplexer’,ESA Contract: 9178190INLIUS(SC), 1994. 14. British Aerospace Ltd., ‘Final report of study on digital beamforming networks’, ESA Contract: 8087/88/NLlJG, July 1990. 15. H . Gockler and H.Eyssele, ‘Study of on-board digital FDMdemultiplexing for mobile SCPC satellite communications (part II)’, European Transactions on Telecommunications, 3, (l), 7-30 (1992). 16. British Aerospace Ltd., ‘Final report of study of applicability of different on board routing and processing techniques to a mobile satellite system’, ESA Contract: 89721901NLIRE. November 1992. 17. ANT & British Aerospace, ‘Design of a 16-channel frequency multiplexer (FMUX) ASIC‘, Rider to ESA Contract 103811 93lNLlJV. z zyxwvutsrqponm zyxwvutsr zyxwvutsr KEY PAYLOAD TECHNOLOGIES UK, ‘ASIC development for use in digital transmultiplexers’, Future ESA Contract. 19. Matra Marconi Space UK, ‘ASIC development for use in digital beamforming networks’, Future ESA Contract. 20. P. M. Bakken and A. Ronnekleiv, ‘SAW-basedchirp Fourier transform and its application to analogue on board signal processing’, International Journal of Satellite Communi18. Matra Marconi Space cations, 7 , 283-293 (1989). 135 network’, ESA Contract 87141901NL (final report in preparation). 24. M.Hollreiser, ‘ W I for DSP in space’, Fourth ESA Workshop on DSP TechniquesApplied to Space Communications, London, 26-28 September 1994. 25. British Aerospace, ‘Demonstration of representative MCM packaging techniques’, Rider-2 to ESA Contract 103811931 NLIJV, currently ongoing ESA contract. 26. V. Ringset, P. M. Bakken, A. Ronnekleiv, E. Olsen and G. Bjornstrom, ‘SAW technology for multicarrier demodulation in advanced payloads’, Space Communications, 7 , zyxwvutsr zyxwvutsrqp zy 21. Delab and Frobe Radio, ‘Developments related to SAW applications for satellites, work-order no. 2: mounting on SAW chirp lines’, ESA Contract 9107lNLIDS. 22. M. A. Lagunas, L. Corbera, P. Remiro, G. Vazquez and J. Fernandez, ‘Final report on digital array processing: comparative analysis of high resolution DOA estimation and time reference beamforming’, ESA Contract P.O. No.112052, Polytechnic of Catalonia, July 1991. 23. ERA Technology, ‘Breadboarding of a digital beamforming (4-6), 521-529 (1990). 27. D. R. McElroy, ‘The FEP communications system’, 12th International Comm. Satellite Systems Conference, AIAA , Arlington, VA, March 1988. 28. G. Bjomstrom, ‘Digital payloads: enhanced performance through signal processing’, ESA Journal, 17, (1993).