Satellite System Design Examples for Maximum MIMO Spectral Efficiency in LOS Channels A. Knopp and R.T. Schwarz D. Ogermann, C.A. Hofmann, and B. Lankl Fed. Office of the German Armed Forces (Bundeswehr) Munich Univ. of the German Armed Forces (Bundeswehr) for Information Technology, Dept. for Satellite Institute for Communications Engineering Communications, 56076 Koblenz, Germany 85579 Neubiberg, Germany Email: {andreas1knopp, robert7schwarz}@bundeswehr.org Email: {andreas.knopp, r.schwarz, berthold.lankl}@unibw.de Abstract—MIMO satellite links have recently attracted a high interest with respect to possible link capacity enhancements. In [1] it has been shown that especially in Line-of-Sight (LOS) satellite channels maximum multiplexing gain can be achieved via the construction of orthogonal channels by means of the geometrical arrangement of the ground terminal antennae in relation to the antennae in orbit. Based on these theoretical results, we present practically relevant design and configuration examples for satellite communication systems, involving transparent payloads for the first time. Thus, we significantly extend the results of [1], [2] that have been limited to regenerative payloads. The examples cover multiple-satellite and single-satellite MIMO scenarios. The assets and drawbacks of the applications are investigated, especially highlighting system-inherent design uncertainties. I. I NTRODUCTION Due to their potentially high bandwidth efficiency [3], multiple-input, multiple-output (MIMO) systems are very promising especially with regard to satellite communications. This is mainly for the reasons that on the one hand there is a rising demand for higher data rates, while on the other hand the usable frequency spectrum becomes short of available bandwidth, which has, therefore, become particularly expensive. However, in MIMO satellite communications (SatCom) it is nowadays widely believed that a high MIMO capacity gain can only be achieved in the rich scattering multipath propagation channel [4], but not in the LOS channel due to its high correlation coefficients [5]. Contrarily to this impression, early results of [6] indicated that in principle it is possible to form a LOS channel with maximum MIMO multiplexing gain by smart geometrical arrangements of the antenna elements at both link ends. Despite, as typical examples reveal [4], [5], [7], in SatCom the possibility of enhancing the MIMO spectral efficiency via smart arrangements of the satellite and ground terminal antennae has hardly been realized so far. Recently, in [1] a theoretical derivation was presented that firstly provided a design prescript for satellite links with optimum MIMO capacity. For mathematically manageable results, the analyses are limited to a satellite MIMO system with two satellite antenna elements only, which have to be placed on a single satellite each. Keeping the number of satellites and satellite antennae low seems to be demanded also for economical reasons. The number of ground terminal antennae can be chosen arbitrarily. Finally, the results ended up in tractable prescripts for the positioning of the satellites in the geostationary orbit in relation to the placement of the corresponding ground terminals on earth. Illustrating the optimization procedure, in [1] the theory was applied to a number of practical systems consisting of two satellites and fixed or mobile receivers. Further examples were added in [2] indicating the applicability also in satellite broadcast systems. Extending the examples in [1] and [2] we do not limit ourselves to communications satellites carrying a regenerative payload, but we focus on the practically much more relevant case of transparent satellites only working as wireless relays. We include the cases of a single-satellite MIMO link as well as a multiple-satellite MIMO link covering the most important applications. Based on the design examples, the paper also addresses assets and drawbacks of the different system configurations, and the impact of positioning accuracy. II. T HEORETICAL BASICS AND C ONVENTIONS A. Characteristics of the Satellite MIMO Channel 1) Channel Model: We presume a flat-fading channel model, only taking into account the LOS signal path as the dominating wave propagation mechanism. This is a reasonable assumption because in SatCom multipath signals play a secondary role. The power of the LOS signal component outvalues the power of reflections and scattered waves [8]. Denoting the carrier frequency by fc , the frequency-flat MIMO channel is described by its channel transfer matrix H(fc ) = H ∈ CM ×N for a MIMO system consisting of N transmit and M receive antennae. The element [H]mn of the channel transfer matrix at the position mn in equivalent baseband notation is described by the mechanism of free-space propagation according to 2πfc (1) rmn Hmn = amn · exp −j c0 where rmn is the distance between the m-th transmitter (Tx) antenna and the n-th receiver (Rx) antenna. c0 is the speed of light in free space. amn is the complex envelope that is calculated by: amn = c0 · ejϑ0 /(4πfc rmn ) with ϑ0 marking the carrier phase angle at the time of observation. Observing the channel path gain for all M × N pairs of TxRx antenna combinations, they are found to be approximately 978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 1 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings. Authorized licensed use limited to: UNIVERSITAET DER BUNDESWEHR. Downloaded on December 22, 2008 at 11:23 from IEEE Xplore. Restrictions apply. parameter EIRPu (G−T )u Bu Lu EIRPd (G−T )d Bd Ld value 69.3dBW 2.4dBK 5MHz 209.3dB 35.4dBW 16.4dBK 1MHz 207dB explanation uplink effective isotropic radiated power logarithm. ratio of uplink fig. of merit [8] transmission bandwidth for uplink path loss uplink, carrier fc,u = 18GHz downlink effective isotropic radiated power logarithm. ratio of downlink fig. of merit [8] transmission bandwidth for downlink path loss downlink, carrier fc,d = 14GHz TABLE I MIMO CAPACITY CALCULATION PARAMETERS FOR THE EXEMPLARY identical due to the large distance between Tx and Rx, i.e. |amn | ≈ const. ∀ m, n with |.| denoting the absolute value. A further, important property of the LOS MIMO channel is its high spatial correlation. As a novelty compared to previous publications (e.g. [5]), this correlation is used to construct high-capacity channel transfer matrices. 2) Assumptions for the System Parameters: In order to illustrate the capacity gains possible with SatCom MIMO systems, we have to define a set of characteristic figures describing the satellite link, the link budget, and the signal-tonoise ratio (SNR) for the MIMO capacity calculation. Table I summarizes the characteristic figures for the uplink and downlink channels. In general, the transmission channel from the transmitting ground terminal to the satellite is denoted by uplink channel (formula index u) and the corresponding propagation path from the satellite down to the Rx ground station is denoted by downlink channel (formula index d). Using the figures of table I, for the SNRu at the satellite and the SNRd at the Rx on earth ground we obtain SNRu = EIRPu + (G−T )u − Lu − K − Bu = 24dB, (2) SNRd = EIRPd + (G−T )d − Ld − K − Bd = 13.4dB, (3) respectively, where K = 10 log10 (k) is the logarithmic value of the Boltzmann constant k. Analogously, Bu/d = 10 log10 (Bu/d ) denote the logarithmic values of the up- and downlink bandwidth. This way, the transmit and receive antennae are not counted among the propagation channel, they are part of the Tx and the Rx, respectively. Furthermore, it has to be pointed out that in a MIMO system the SNR calculated above is measured at each combination of Tx and Rx antennae that is possible, i.e. each antenna radiates the full EIRPu or EIRPd as it is connected to an own amplifier. B. MIMO Channel Capacity Basically, we have to distinguish between a regenerative payload, where at satellite level the signal is completely regenerated in the baseband, and a satellite with transparent payload acting as an amplifying relay while converting the signal directly from the uplink to the downlink carrier. 1) Regenerative Payload: If the satellites are endowed with a regenerative payload, we treat the uplink and the downlink as two different end-to-end communication links. Consequently, the uplink and the downlink channels provide different channel capacities, where the smaller capacity determines the overall data rate of the system. In the light of the MIMO capacity optimization, it is most important that both, the uplink and the downlink, have to be optimized separately, which is done by means of the respective placement of the antennae [1]. For a MIMO channel without a relay, for a frequencyflat transmission channel the time invariant MIMO spectral efficiency without channel knowledge at the transmitter is calculated according to Telatar’s well-known eq. [3] (4) C = log2 det I M + ρ · HH H . Here, the transmit symbols are realizations of uncorrelated independent, identically distributed (i.i.d.) Gaussian random variables. Furthermore, {.}H denotes the complex conjugate transpose. In this equation, the channel path loss for each mn-th combination of the m-th out of M receive antennae and the n-th out of N transmit antennae is included in the corresponding entry [H]mn of the channel transfer matrix. Thus, ρ is the (linear) ratio of the transmit power at each transmit antenna and the noise power at each receive antenna. With eq. (2) and (3) it is calculated (up-/downlink) ρu = 10(SNRu +Lu )/10 , and ρd = 10(SNRd +Ld )/10 . (5) As stated in the introduction, in this paper we aim at a maximum multiplexing gain of each MIMO channel described by the channel matrix H. In the following, w.l.o.g. we use the uplink channel H u to recapitulate the optimization criterion according to [1]. The procedure for the downlink is identical. Presuming M antenna elements in the geostationary orbit and N ground terminal antennae, maximum multiplexing gain is achieved if the eigenvalues γi , i ∈ {1, ..., min(M, N )} of the matrix Q are given equally by γi = |a|2 · max(M, N ), where H Q = H uH H u for M < N, Q = H u H u for M ≥ N (6) In this case we obtain an optimum eigenvalue profile and an orthogonal channel transfer matrix. Then the channel capacity also obtains its optimum value which is given by Cu,opt = min(M, N ) · log2 1 + ρu |a|2 max(M, N ) . (7) 2) Transparent Payload: If we are dealing with a transparent payload instead, the channel comprises both, the uplink as well as the downlink. Furthermore, the satellite payload has to be incorporated as a relay that introduces an amplification and a phase shift to the signal. In the following, we presume N transmit antennae for the uplink ground station, M receive antennae and M transmit antennae at the satellite and, finally, Z receive antennae at the downlink ground terminal. In this case, the MIMO channel capacity for given channel transfer matrices H u ∈ CM ×N and H d ∈ CZ×M is calculated according to [9] H −1 (8) C = log2 det I M + ρu H u H H − ρ H H S u u u u with the matrix S given by S = I M + σu2 /σd2 · F H H H d H dF . (9) The Matrix F ∈ CM ×M describes the transfer matrix of the relay, i.e. the satellite payload, and σu2 and σd2 denote the noise power at the receiver in the up- and downlink, respectively. For 978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 2 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings. Authorized licensed use limited to: UNIVERSITAET DER BUNDESWEHR. Downloaded on December 22, 2008 at 11:23 from IEEE Xplore. Restrictions apply. ρd > 0 and any H u , H d , it becomes optimal with respect to the overall channel capacity if we choose [9] parameter RS F = V d ΛF U H u. RE θT φT θS,m (10) V d and U u are unitary matrices that result from the eigenvalue decompositions of the down- and uplink MIMO channels, i.e. H H H H dH H d = V d Λd V d and H u H u = U u Λu U u . Finally, ΛF is a diagonal matrix that describes the power allocation within the relay. Hence, the matrix F can be considered as a matched filter along the singular vectors of the channel matrices [9]. Now, a special case is obtained if the uplink and downlink channel transfer matrices H u and H d are both optimized for optimal capacity through the geometrical arrangement of the antennae according to [1]. This way, the matrices V d and U u become identity matrices and, consequentially, the MIMO relay in orbit obtains a very simple structure: Its transfer matrix F is solely formed by the diagonal matrix ΛF , which means that there are M independent propagation paths through the satellite payload without any cross-couplings. A second, very important advantage for the system design, which only applies to optimized links with orthogonal channel transfer matrices H d and H u , is revealed if the entries of F = ΛF are further analyzed. As the matrix entry [F ]mm describes the transfer function of the m-th MIMO branch through the satellite payload, it consists of a power amplification and a phase shift. Normally, the satellite will contain travelling wave tube amplifiers (TWTAs) or solid state power amplifiers (SSPAs) that are driven in saturation. Thus, we presume |[F ]mm | ≈ const ∀ m ∈ {1, ..., M } with good accuracy. Contrarily, each phase angle arg{[F ]mm } will usually be different for the M signal paths because each path converts the carrier frequency from uplink to downlink separately. Even if the same local oscillator is used for all paths, there will be a non-constant phase offset between them. Only in case of orthogonal channels the matrix S remains optimal with respect to the channel capacity, independently from the phase angles arg{[F ]mm }. Of course F can be optimized also for non-orthogonal channels H d and H u , but in this case also the phase shifts become crucial and must be optimized. As a consequence, keeping them constant over time is a very hard requirement for the hardware used. If the antennae in orbit are mounted on different satellites, constant phase shifts cannot be guaranteed at all. Contrarily, this requirement can be omitted completely for orthogonal channels. C. Capacity Optimization for Practical Satellite Scenarios The strategy for the MIMO capacity optimization in SatCom scenarios was derived and thoroughly explained in [1] and [2]. Thus, at this point we only recapitulate the most important results, for a deeper insight into the mathematical derivations we particularly recommend a study of [1]. As we have stated before, we focus on a satellite scenario consisting of only 2 satellite antennae that are either mounted both on the same satellite or on two different satellites. Hence, we deal with an 2 × N system for the uplink and a Z × 2 system for the downlink with N and Z arbitrary. In the δT dT explanation radius of satellite orbit, measured from earth center (42.164km) mean earth radius (6.378km) geographical latitude of ground terminal (ULA center) geographical longitude of ground terminal (ULA center) geographical latitude of m-th satellite antenna, convention for nomenclature: θS,1 < θS,2 tilting angle of earth terminal antenna ULA, measured relative to the connecting line of the satellite antennae, i.e. parallel satellite and ground antennae arrays for δT = 0 antenna element separation within earth terminal ULA TABLE II FORMULA SYMBOLS FOR THE SCENARIO DESCRIPTION previous section it was shown that in all cases, transparent and regenerative satellite payloads, we have to form both an orthogonal uplink as well as an orthogonal downlink channel. Thus, H u and H d must be optimized separately via the antenna arrangement to fulfill a particular, geometrical condition that is outlined exemplarily for the downlink H d subsequently, for the uplink the result is valid accordingly. In [1] it is basically shown that an orthogonal MIMO channel is obtained for M = 2 if the very general condition c0 , rk1 − rk2 + rl2 − rl1 = v · (k − l) Z · fc (11) k, l = 1 . . . Z, v ∈ Z, v Z is fulfilled. Here, rμν , μ, ν ∈ Z basically denotes the geometrical distance between the ν-th satellite antenna and the μ-th ground terminal antenna and v Z means that v is indivisible by Z. The parameter v ∈ Z is a constant that regards the periodicity of the capacity optima in multiples of the wavelength. Equation (11) shows that the antennae on earth ground and on the satellite platform have to be arranged accordingly to ensure very distinct LOS path lengths. Now, the positions of the antennae have to be formulated appropriately to calculate all distances rμν . As usual in SatCom, the principle of the Mercator projection [8] is used to describe the positions of the ground terminals and the satellites. Applying the nomenclature1 provided in table II and the substitutions s2m = cm = 2 RS2 + RE − 2RS RE · cos φT cos (θT − θS,m ) (12) 2RS · [cos δT sin (θT − θS,m ) + (13) sin δT cos (θT − θS,m ) sin φT ] , in [1] the practical optimization criterion for M = 2 satellite antennae for the downlink is re-formulated as c1 c0 c2 = v· − , v ∈ Z, v Z. (14) dT 2s1 2s2 Zfc Here it is presumed that the ground terminal antennae are arranged in form of a uniform linear array (ULA). They can either be mounted in close separation on a single terminal or on separate terminals each. In the latter case, the ULA is formed by Z terminals, but eq. (14) can still be applied (see again [1], [2] for a more detailed discussion of eq. (14)). 1 A thorough graphical illustration of all parameters has already been provided in [1], [2]. Studying this figure is recommended to facilitate readability. 978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 3 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings. Authorized licensed use limited to: UNIVERSITAET DER BUNDESWEHR. Downloaded on December 22, 2008 at 11:23 from IEEE Xplore. Restrictions apply. III. E XAMPLES FOR P RACTICAL MIMO A PPLICATIONS For all of the scenarios presented in the following, we exemplarily assume the subsequent ground terminal positions: Tx position θT,u φT,u Rx position θT,d φT,d Berlin 13◦ 52◦ Munich 11◦ 48◦ A. Multiple Satellites carrying Regenerative Payloads This first type of application is mentioned here for completeness only. It was extensively studied already in [1], [2]. B. Single Satellite carrying Transparent Payload A MIMO system requiring a single satellite only, is very desirable in practice for economical reasons. An illustration of the scenario involving all the necessary symbols is provided in fig. 1. The most essential finding when optimizing the 2 Besides the universal notations d and δ describing either the up- or the T T downlink or both, we additionally use the indices u and d to distinguish intentionally between up- and downlink if necessary, i.e. dT,u , dT,d , δT,u , δT,d . dS,d = dS,u θS,2 θS,1 MIMO Rx M =2 MIMO Tx Hd F Hu terrestrial network MIMO AFE (uplink) dT,u N =2 Fig. 1. capacity degradation C/Copt [%] D. Sources of Capacity Degradations in Practical Systems In advance of the analyses in III, we briefly summarize the most probable reasons for a capacity degradation encountered in a practical MIMO system that is based on our optimization. 1) The Satellite Station Keeping Box: As it is in practice impossible to maintain the satellite absolutely immobile at its defined position a station keeping box is defined, which represents the maximum permitted values of the satellite excursions. For a capacity-optimized system it is, thus, desirable to guarantee the optimum capacity as long as the satellite stays within this virtual box. Subsequently, we assume typical box dimensions, which are ±37.5km in longitude and latitude and ±17.5km for the eccentricity [8]. We have investigated the effect of satellite drifts in all three dimensions finding significant effects in cases of changes in longitude and latitude with respect to the nominal satellite position. Instead, the eccentricity remains of less relevance for the channel capacity. 2) Miss-Positioning due to Ground Terminal Movement: Naturally, each change of the ground terminal position immediately changes the phase angle relations of the entries within the MIMO channel transfer matrix and, thus, the MIMO channel capacity. In this context the antenna spacing dT is incorporated as well as the tilting angle2 δT because both parameters have an interdependence with regard to the capacity optimization criterion according to eq. (14). Even in a scenario with static ground terminals or anchor stations there are environmental conditions, e.g. wind, that slightly move the antennae. It must be remembered that even minor changes of the antenna positions can significantly impact on the signal phase angles at the Rx due to the very short wavelength in SatCom. Hence, the effect of terminal movement has to be studied depending upon the addressed application. 3) Delay of Signals: Especially multiple-satellite MIMO systems exhibit high delay differences of the signal portions arriving from different satellites at the ground Rx or vice versa. Countermeasures for this phenomenon are not tackled in the course of the subsequent capacity analyses. However, there are system design approaches available in the literature, e.g. [5]. terrestrial network MIMO AFE (downlink) dT,d Z=2 AFE: antenna frontend single satellite and dislocated ground terminal antennae miss-pos. Tx miss-pos. Rx miss-pos. Tx,Rx δT,u = δT,d = 0 θS,1 = 11◦ 59 59.9755 θS,2 = 12◦ 0 0.0245 miss-positioning ΔdT = dT − dT,opt [km] Fig. 2. MIMO channel capacity as a function of the antenna spacing on earth ground, example for dS,u = dS,d = 10m, N = M = Z = 2, v = 1 capacity of such a MIMO system is a straightforward result of eq. (14): Because of the comparatively small antenna spacing dS,u and dS,d at satellite level, the ground station antenna spacing dT,u and dT,d has to be chosen large. If for example a spacing of dS,u = dS,d = 10m is used at satellite level, which already marks a very large spacing, the minimum antenna element spacings at ground level are calculated dT,u,opt (v = 1) = 32.1km for the uplink and dT,d,opt (v = 1) = 40.9km for the downlink if the remaining degrees of freedom are chosen as stated in the caption of fig. 2. One has to note, that dT,opt (v = 1) even further increases for decreasing dS . This scenario is applicable in practice for example if a number of anchor stations is available at several locations of a particular territory. Such a system design is desirable in military applications because a set of dislocated ground stations increases the overall link availability as the probability of simultaneous attacks is lower than for compact conglomerates. A terrestrial backbone network is needed for the connection of the antenna front ends to a common Tx / Rx. Otherwise, this application is very advantageous with respect to the impact of a miss-positioning of the antennae on earth ground as well as the station keeping maneuvers of the satellite. As it can be observed from fig. 2 again, even a miss-positioning of ΔdT = dT − dT,opt in the range of 978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 4 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings. Authorized licensed use limited to: UNIVERSITAET DER BUNDESWEHR. Downloaded on December 22, 2008 at 11:23 from IEEE Xplore. Restrictions apply. C. Multiple Satellites carrying Transparent Payloads In most practical cases very compact ground terminals are required. This means that the MIMO antenna elements have to be placed close to each other. As a straightforward result of eq. (14) the MIMO antennae on satellite level now have to be displaced very far in order to obtain the optimal MIMO spectral efficiency. Consequentially, a single satellite platform is no longer sufficient. Instead, the M = 2 antennae have to be placed on two separate platforms as illustrated in fig. 3. Because it is not possible to illuminate different ΔθS M =2 θS2 θS1 F MIMO receiver Hd MIMO transmitter Hu N =2 dT,d MIMO AFE (uplink) dT,u Fig. 3. MIMO AFE (downlink) Z=3 AFE: antenna frontend two satellites and compact ground terminal antennae satellites simultaneously with a typical base station antenna (the satellites are presumed to be separated ΔθS ≥ 0.5◦ in latitude), we assume two parallel single-input, single-output (SISO) uplinks to the satellites. These uplinks have to be designed such that the available data rate is sufficiently high to support the MIMO downlink. The scenario was also proposed in [5], but not optimized for the channel capacity. Again, we calculate the channel capacity according to eq. (8) because we still assume transparent payloads. The uplink channel transfer matrix H u in eq. (8) has nonzero values only on its main diagonal, but the downlink channel transfer matrix H d describes a real MIMO channel that is optimized with respect to the multiplexing gain. At this point it is useful to further discuss the interdependence of the free parameters in eq. (14), which builds the basis for the construction of orthogonal channels. Those are mainly the antenna spacing dT,d at the ground station, the angular separation of the satellites ΔθS = |θS,1 − θS,2 |, and the tilting angle δT,d that provides the rotational shift of the ground antenna array in comparison to the antenna array in orbit (see also table II). fc,d = 14GHz, M = Z = 2, v = 1 satellite displacement dT,d,opt [m] kilometers only slightly degrades the capacity for a satellite with transparent payload according to eq. (8). The satellite station keeping has almost no impact. This effect becomes severe in case of multiple satellites only. Further crucial parameters, instead, remain the tilting angles δT,u and δT,v of the ground terminal antenna arrays. In our example, those are optimally matched to the antenna spacing in fig. 2. The impact of deviations in δT from the optimum is highlighted in the context of the next subsection. ΔθS = 0.5◦ ΔθS = 1◦ ΔθS = 2◦ ΔθS = 5◦ tilting angle δT,d [◦ ] Fig. 4. interdependence of degrees of freedom for the capacity optimization The curves in fig. 4 show possible combinations of the four parameters that lead to an orthogonal channel matrix H d for the downlink. The optimal values dT,d,opt are calculated for the case v = 1 and, thus, larger but still optimal values dT,d can be calculated by choosing v > 1, v ∈ Z, v Z. Hence, we observe three major effects:3 a. a larger satellite displacement requires a smaller antenna spacing of the terminal on earth b. an increased rotational shift of the antenna arrays on earth and in orbit requires an increased antenna spacing of either the ground terminal or the satellite antennae4 c. an optimal antenna spacing on earth ground occurs with a periodicity in v (see also the discussions in [1], [2]) Unfortunately, in practice the choice of the actual parameter combination is limited by realizability. It further determines if the system will be resistant against capacity degradations caused by one of the error sources mentioned in subsection II-D. An increased tilting angle δT,d , for example, requires a more accurate positioning of the antenna array. This is observed from fig. 4 again, because the slope of the curves increases for higher values of δT,d . Also the accuracy of dT,d is important. If the capacity is drawn as a function of the miss-positioning compared to the optimum antenna spacing, a very similar curve as provided in fig. 2 will be obtained. But as a crucial difference to the single-satellite case, in the current scenario the ground terminal antenna spacing must be adjusted with a typical accuracy in the range of few meters (no longer kilometers) to avoid severe capacity degradations. In general, if two satellites are used it can be summarized that the farther the satellites are displaced, the more exactly the ground terminal antennae have to be separated by the theoretically optimal value of dT,d and vice versa. This fact was already discussed and proven by simulations in [1], [2] in more detail. 3 Of course, if available, for a MIMO uplink involving two separate satellites, these effects are found identically (see also examples in [1], [2]). 4 Basically, for any configuration there always exists a value δ T,d for which no optimal solution is possible because Hd generally becomes rank-deficient and the left hand side of eq. (14) turns to zero. Then, dT,d,opt = ∞ would be required theoretically. In the current example this occurs for δT,d = 90◦ . 978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 5 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings. Authorized licensed use limited to: UNIVERSITAET DER BUNDESWEHR. Downloaded on December 22, 2008 at 11:23 from IEEE Xplore. Restrictions apply. C [Bit/s/Hz] 0.9Copt = 11.6Bit/s 0.9Copt = 10.9Bit/s 0.9Copt = 9.5Bit/s δT,d = 40◦ θS,1 = −1◦ θS,2 = 1◦ } Δθ S = 2◦ v · dT,d,opt [m] practically feasible while choosing v as well as δT as small as possible for the construction of orthogonal channels 3. higher numbers of ground terminal antennae are advantageous for stable channel capacities and high receive power To finally provide a precise example for a possible system design, we could choose the following set of parameters: N = M = 2, Z = 4, δT,d = 40◦ , ΔθS = 1◦ ({θS,1 , θS2 } = {−0.5◦ , 0.5◦ }), v = 10 ⇒ dT,d,opt (v = 10) = 4.31m. In order to guarantee that the MIMO channel capacity degrades less than 5% down from its optimum, we would have to adjust dT,d with an accuracy of dT,d,opt (v = 10) ± 0.72m while preserving a deviation of less than ±5◦ for δT,d compared to its specified value. Such an accuracy can actually be maintained in practice. IV. C ONCLUSION Fig. 5. worst-case capacity degradation as a function of the antenna spacing on earth, two satellites performing station keeping maneuvers are considered Finally, the mobility of the satellites caused by their independent movements within the station keeping box was examined. It turned out that the capacity degradation, which is encountered due to varying phase angles of the receive signals, is strongly linked to the actual antenna spacing dT,d = v · dT,d,opt (v = 1). Therefore, the stability of the channel capacity depends upon the choice of v. This fact is illustrated by fig. 5, which shows the channel capacity as a function of the antenna spacing of the ground terminal. The abscissa comprises discrete values that are obtained by increasing v ∈ Z. Each ordinate value C(v · dT,d,opt ) represents the worst-case capacity value that is found if the channel capacity is calculated for all possible pairs of latitude positions {θS,1 , θS,2 }, which can occur due to the satellite movements. Of course, only those latitude values falling within the station keeping boxes of the satellites have been taken into account. A zero but constant inclination and eccentricity was presumed for each satellite because both parameters have a negligible impact on the channel capacity for the actual setup. Especially a varying inclination hardly impacts on the channel capacity because the tilting angle was chosen δT,d = 0 and, thus, the phase angle relations within H d are widely independent of the inclination. Contrarily, for nonzero δT,d the inclination becomes increasingly relevant. From fig. 5 we finally observe that the station keeping maneuvers do not cause a significant capacity degradation as long as the number of receive antenna elements is high and, most important, the values of parameter v remain low. In other words, if a large antenna spacing is chosen on earth ground, the system becomes sensitive with respect to even slight changes of the satellite positions. For high v this even means that the capacity probably degrades down to its minimum (keyhole) capacity, as the curves indicate. Based on our investigations, the following design hints for a system with optimum MIMO capacity can be provided, which at the same time is resistant against degradations due to misspositioning of both, the satellites and the ground terminals: 1. the satellite displacement should usually be chosen small 2. the antenna spacing of the ground terminal should be We optimized the MIMO channel capacity for satellite systems that utilize transparent communication payloads. For that purpose, in a first step it was necessary to exploit the channel correlation to construct orthogonal and, thus, optimal MIMO channel transfer matrices for both, the up- and the downlink channel of the satellite link. Extending results of [1], such orthogonal channels have been realized by an adequate positioning of all transmitter and receiver antenna elements. Secondly, the optimum transfer function for the MIMO relay in the geostationary orbit had to be chosen. Basically, this was achieved applying the very general, mathematical results of [9]. In the course of this, we have shown that only for orthogonal MIMO channels the system design for the satellite payload becomes practically feasible without causing capacity degradations due to phase uncertainties within the MIMO system branches. This further outlines the importance of orthogonal MIMO channels. In practice, for the construction of capacity-optimal MIMO channel transfer matrices, the user is provided with several degrees of freedom for the system design. The interdependence and optimization of the most important design parameters was illustrated, also highlighting practical limitations and capacity-degrading error sources. R EFERENCES [1] R.T. Schwarz et al., Optimum-Capacity MIMO Satellite Link for Fixed and Mobile Services, Proc. 2008 ITG Workshop on Smart Antennas (WSA’08), Feb. 2008, pp. 209-216. [2] R.T. Schwarz et al., Optimum-Capacity MIMO Satellite Broadcast System: Conceptual Design for LOS Channels, Proc. Advanced Satellite Mobile Systems Conference (ASMS’08), Aug. 2008. [3] E. Telatar et al., Capacity of multi-antenna gaussian channels. AT&TBell Technical Memorandum, 1995. [4] G.E. Corazza (ed.), Digital Satellite Communications, Springer Series on Information Technology, 2007. [5] F. Yamashita et al., Broadband multiple satellite MIMO system. Proc. VTC-2005-Fall, pp. 2632-2636. [6] P.F. Driessen and G.J. Foschini, On the Capacity Formula for Multiple Input-Multiple Output Wireless Channels: A Geometric Interpretation, IEEE Trans. on Communications, vol. 47, No. 2, Feb. 1999, pp. 173-176. [7] Cheon-In Oh et al., Analysis of the Rain fading channel and the system applying MIMO. Proc. ISCIT 2006, pp. 507-510. [8] G. Maral et al., Satellite Communications Systems. Wiley&Sons, 2006. [9] X. Tang and Y. Hua, Optimal Design of Non-Regenerative MIMO Wireless Relays, IEEE Transactions on Wireless Communications, vol. 6, no. 4, Apr. 2007, pp. 1398-1407. 978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 6 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings. Authorized licensed use limited to: UNIVERSITAET DER BUNDESWEHR. Downloaded on December 22, 2008 at 11:23 from IEEE Xplore. Restrictions apply.