Interference Mitigation for the Reverse-Link of Interactive Satellite Networks

Transcription

1 1 Interference Mitigation for the Reverse-Link of Interactive Satellite Networks M. Debbah, G. Gallinaro, R. Müller, R. Rinaldo, A. Vernucci Abstract This paper presents some of the results of a study aimed at investigating and optimizing possible Interference Mitigation techniques for wideband fixed-service satellite systems. In particular this paper discusses the performance improvements that can be achieved in the user-to-hub link (Reverse-Link, RL) of a multibeam satellite system by using more advanced processing techniques than those implemented in today s receivers. After illustrating the operational scenario taken as reference, the paper illustrates in particular two Interference Mitigation techniques (MMSE and MMSE-SIC) that are considered to be good candidates for the RL. After considering a possible channel estimation strategy, the paper continues on by presenting the obtained simulation results. T I. INTRODUCTION HE growing interest for multimedia applications is encouraging the deployment of fixed telecommunications satellite systems capable of offering high-speed point-to-point links at competitive service fees. To achieve such goal, the next generation of broadband satellite systems shall be designed to offer higher throughputs than what provided by current systems. This will be possible by e.g. exploiting the higher frequency bands allocated to fixed satellite systems (e.g. the K a band), generating a great number of narrowaperture high-gain beams, and re-using the frequency bands as much as feasible. Going that way, system throughput will become more and more affected by the extent of interference occurring among signals that share common band portions (commonly called the intra-system interference). Great aid in that respect is given by the use of highly efficient coding schemes, such as those used in the Reverse Link (RL) of current DVB-RCS satellite systems and those standardized for the Forward-Link (FL) by the DVB-S2 working group. Those measures can be accompanied by Interference & Fading Mitigation Techniques, e.g. Adaptive M. Debbah is with the Institut Eurecom, 2229 Route des Cretes, B.P. 193, Sophia Antipolis Cedex, France. G. Gallinaro and A. Vernucci are with Space Engineering, Via dei Berio, 91, Rome, Italy R. Müller was with FTW GmbH, Tech Gate Vienna, Donau-City-Straße 1/3, 1220 Vienna, Austria. He is now with the Institutt for elektronikk og telekommunikasjon, NTNU, 7491 Trondheim, Norway. R. Rinaldo is with the European Space Agency (ESA), Noordwijk, The Netherland. The results reported in this paper were obtained in the frame of contract Novel Intra- System Interference Mitigation Techniques and Technologies for Next Generation Broadband Satellite Systems awarded by the European Space Agency (ESA) to Space Engineering, which Eurecom and FTW are subcontractors to. Coding and Modulation, which are being studied and optimized with the aim of providing a higher flexibility and improving the overall system efficiency. It should however be noted that the introduction of those techniques becomes more challenging in presence of very efficient coded modulations, that turn into low operating signal-to-noise ratios. In such context, new Interference Mitigation techniques, requiring the receiver to simultaneously process more signals and hence generally identified as Multi Detection (MUD) techniques, have been proposed as a promising solution to further increase system capacity in an interferencelimited system operating under heavy traffic load. In the last decade an impressive amount of theoretical investigations have been carried out in the field of MUD algorithms; in particular efforts have been focussed on CDMA systems for mobile applications [5], while TDMA systems for fixed applications have been considered to a lesser extent. Moreover proposed concepts were often analyzed in hardly realistic scenarios, so proving that only limited effort were devoted to further developing the theoretical background, so as to make it readily applicable to practical systems. On those grounds the European Space Agency (ESA) has promoted an ad-hoc study aimed to devising and assessing efficient MUD techniques, which can be successfully integrated in planned or future broadband systems. This paper reports some of the results of that study. Reference is here made to a satellite system featuring a multi-star-network topology where a few (and relatively large) GateWay stations (GWs) support two-way communications with a great number of Terminals (UTs) equipped with a relatively small-aperture antenna and transmitting low RF power. More details on the assumed reference scenario are given in Sect. II. MUD techniques can in principle be applied to both the FL and the RL of the system; but here we focus on some of the results obtained for the RL (an overview of techniques applicable to the FL is presented in [1]). The RL access scheme is assumed to be low-rate TDMA, with data rates ranging from few hundreds kbit/s to few Mbit/s. With those choices the MUD techniques proposed herein become applicable to the RL of most current Very Small Aperture Systems (VSATs), i.e. those based on the DVB-RCS access scheme as well as on similar proprietary schemes. In particular two MUD techniques are evaluated here, i.e. the Spatial MMSE (Minimum Mean Square Error) processing

2 2 and the Spatial MMSE-SIC (Successive Interference Cancellation). Both techniques are based on spatial processing which is only feasible in conjunction with multibeam satellite systems a configuration already adopted for wideband mobile systems (e.g. Inmarsat 4, Thuraya), and which will also become typical for all future high-capacity fixed-service satellite systems. It should just be noted that the proposed techniques can also find an application in the mobile-service context [2]. In Section II, the conventional operational scenario taken as reference for performance comparison is illustrated followed, in Section III and Section IV respectively, by the description of two Interference Mitigation techniques (MMSE and MMSE-SIC) that are considered to be good candidates for adoption in the RL of broadband interactive satellite networks. After considering channel estimation (see Section V), the paper continues on by presenting the obtained simulation results (see Sect. VI). Finally conclusions on performed work are drawn in Section VII. II. REFERENCE SCENARIO To test the effectiveness of the proposed techniques, an hypothetical broadband satellite systems providing a European coverage was initially designed. Given the current spectral congestion, operation at K a -band was assumed with a transparent bent-pipe satellite architecture. A preliminary design of the on-board antenna for the user-link was performed, which resulted in the feasibility of covering Europe by means of 88 beams each having an approximately 0.5 deg. beamwidth (Figure 1). Figure 1 - Assumed -link Antenna Coverage A detailed analysis was performed to determine the antenna pattern using the GRASP [3] software tool, given the strong sensitivity of the performance of spatial-processing MUD techniques to antenna patterns. Two different were investigated with regard to the GWs segment. In the first case, we assume 11 GWs in the system each one serving 8 beams. The 8 beams served by the same GW are referred here as a beam cluster. 11 beam clusters are thus present in the system. The cluster configuration used in our simulation is shown in figure 2. In the case, we assume that each GW serves all the 88 beams. Clearly each GW will only use a portion of the band allocated to each beam. To compute system capacity, we considered the following UT parameters: - Symbol Rate: 512 KBaud - Antenna Diameter: 70 cm - HPA Power: 1 W. Figure 2 - Antenna Coverage with 88 Beams Partitioned in 11 Clusters of 8 Beams Taking into account the need to operate the UT High-Power Amplifier (HPA) with some Output Back-Off (OBO), so as to comply with the applicable off-beam emission regulations, the above parameters translate into a maximum UT Effective Isotropically Radiated Power (EIRP) of 41.6 dbw. Table 1 shows the assumed RL satellite repeater parameters. An on-board RF power of less than 1 kw was assumed, as most of the available on-board power will typically be dedicated the FL. HPA Saturated Power 30 W HPA Bandwidth 500 MHz Spot Beamwidth 0.5 deg Antenna Gain (edge of coverage) 47.3 dbi Post-HPA Loss 2.5 db Receiver Noise Figure 2.5 db Total Saturated RF Power 885 W Nominal OBO 4 db Average TWTA db OBO 32.2 % HPA DC Power 870 W Receiver Noise Figure 2.5 db Pre-LNA Loss 1.5 db Table 1 - RL Repeater Characteristics The system capacity was computed, with and without the use of spatial processing, taking the above parameters. For the case without spatial processing, a conventional three-color frequency-reuse scheme was considered. The obtained system capacity has been taken as reference for evaluating the effectiveness of the proposed techniques (see Section VI). With regard to the physical layer, a DVB-RCS-like system enhanced with ACM has been assumed. Also, a more efficient Turbo-Φ coding proposed in [4] has been assumed. In Table

3 3 2, the physical-layer modes and the corresponding required E s /(N o +I o ) values for successful decoding are shown. An additional 1.5 db margin has been taken for all modes, but the most protected ones, as a safeguard against possible errors in ACM operation. For standard DVB-RCS systems, that only use QPSK modulation, some QPSK simulations have been performed so as to get indications on the possible advantages, if any, of the proposed techniques. Modulation Coding Rate Efficiency (bps/hz) Req. Es/No (db) QPSK 1/ QPSK 1/ QPSK 3/ QPSK 2/ QPSK 3/ QPSK 6/ PSK 2/ PSK 3/ PSK 4/ APSK 3/ APSK 5/ APSK 9/ Table 2 - RL ACM Modes and Corresponding Required Es/No III. THE SPATIAL MMSE TECHNIQUE The first evaluated technique was spatial MMSE. According to this technique, the GW jointly processes all cochannel carriers received from the beams allocated to the GW (e.g. 8 beams in the first beam-cluster configuration or all the 88 beams in the second beam-cluster configuration). Figure 3 reports a model of the channel from the GW s point of view. The channel can be mathematically modeled as a vector channel. In particular the signals received by the GW from all beams can be represented as a column vector, y. Similarly, the signals transmitted by the UTs, in each given time slot, can be represented as a column vector x. For the time being, we will assume a single active UT per beam at each given instant. We can then write the following equation, relating the vector of received signals, y, to the vector of transmitted signals, x: y = G B A x + σ where: - σ represents the noise (plus external cluster interference) power floor at each of the beam chain receivers at the GW (not necessarily equal for all the chains); - A is a diagonal matrix expressing the up-link complex channel gain (it does also take into account possible up-link fading); - B is the antenna beamforming gain matrix. The element b kj, of such matrix expresses the complex gain of beam k towards UT j; - G is a diagonal matrix whose diagonal element g k is the complex gain of the link from the receiver input of the onboard beam k to the GW. Such matrix takes into account the possible differential amplification and phase shifts of such different paths from each satellite up-link beam input to the GW processor. The product GBA will be hereinafter indicated as H, referred to as the Matrix. It can be shown that the optimal linear MMSE matrix filter, M, at the GW side is: M= H H [(Σ + H H H )] -1 where Σ = E{σσ Η }is the covariance matrix of the noise. #1 #2 #K h 11 h 21 hk1 h12 h22 hk2 h1k h2k hkk Beam #1 Beam #2 Thermal Noise + External Interference Rec. 2 Beam #K Thermal Noise + External Interference Rec. K Figure 3 - RL Model Thermal Noise + External Interference Rec. 1 y 1 y 2 y K Spatial Proc. IV. THE SPATIAL MMSE-SIC TECHNIQUE The Spatial MMSE-SIC is an enhancement of the Spatial MMSE technique, where a multi-step processing approach is followed. Each step is composed of a spatial MMSE filtering (as discussed in Section III), followed by the demodulation and decoding of the UT having the highest Signal-to- Interference + Noise + Ratio (SINR). If decoding is successful, the contribution of such signal to the received signal vector is cancelled, otherwise the process is stopped. A pictorial view of the MMSE-SIC scheme is shown in Figure 4. At each step, to cancel the signal, in case of successful decoding the estimated channel matrix is used. In particular the column h i of the estimated channel matrix H is used for that cancellation. After each cancellation step a new MMSE spatial filter is recomputed as now the channel matrix will have all zero

4 4 values in the column corresponding to the cancelled user 1. It shall also be observed that, at each step, the MMSE filtering shall only compute a single user. This implies that the received signal vector Y shall only be multiplied by a row of the MMSE matrix M (i.e. m i ). Received Signal Y MMSE Filter Signal Y1 m 0 Y Subtract First h 0 x 0 Decode 1 MMSE Filter Signal Y2 m 1 Y 1 Subtract Second h 1 x 1 MMSE Filter Figure 4 - MMSE-SIC Concept Decode 2 m 2 Y 2 Subtract Third h 2 x 2 Decode 3 It shall be observed that in a context where ACM is adopted (as considered in this paper), decoding order based on SINR may not be optimum, as it may happen that a signal with higher SINR might be incorrectly decoded whilst a lower SINR signal, using a more protected ACM mode, could instead be correctly decoded. At this regard our simulation assumes perfect SINR estimation at the GW side which will thus always assign UTs the most appropriate ACM mode. The effects of a more realistic assumption on SINR estimation as well as the delay incurring between SINR estimation and actual UT scheduling is subject to future investigation. V. CHANNEL ESTIMATION To estimate the Matrix H at the GW, knowledge of the preambles transmitted by the UTs can be exploited. We assume that a different preamble per beam is used. Also we assume that UTs in the different beams of the same cluster are frame-synchronous. This is actually in line with current operational practice of DVB-RCS, where timing errors of only few symbols are expected in steady-state. For simplicity we will assume in the following a number of beams equal to K and one UT per beam. Matrix H will thus be a K x K square matrix. To estimate the channel matrix H Hadamard sequence preambles would be ideal as, due to orthogonality, they do not enhance noise in linear estimators and simplify computations. Unfortunately, given the fact that UTs are not perfectly 1 The method described here is well-known to be not the most computationally efficient one. It is used here for explanatory purposes only. The new MMSE filter can be easily computed based on the previous one making use of the matrix inversion lemma. symbol synchronous and that they will transmit with some frequency error, Hadamard sequences cannot be used, as their orthogonality gets lost in such an environment. We will thus assume the use of random preamble sequences (one per beam). This would lead to some correlation in the estimation error of each channel matrix element. Such correlation is however quite negligible for sequences of practical lengths. Let c i (t) be the preamble sequence used by UT#i. The sequence is generated starting from a sequence c k, given the pulse waveform g(t), as: c ( t) = i L c k = 1 ik g( t kt ) In the following, we will assume that c ik may have complex values equal to ±0.707±j Due to timing errors, frequency errors and unknown signal phases, the actual preamble as received on-board is: c t τ )exp( j2πf t + ϕ ) i ( i i i Sampling the above sequences would produce a matrix C having on each row the sampled representation of the i-th sequence above. Matrix C will have K rows, as the number of beams and UTs and a number of column, N which is related to the preamble length, L 2. The matrix of received signal samples Y is then: Y = H C + N Post-multiplying the matrix Y by the Moore-Penrose pseudo-inverse of matrix C, indicated with C +, we get: YC + =H + NC + where the pseudo-inverse of matrix C is: C + =C H (CC H ) -1 Hence, apart for the noise term, we are able to get a good estimate of the matrix H. A problem with the above approach is that matrix C is not completely known, even if the preamble sequences are known, due to fact that UTs are not perfectly synchronous and therefore the timing and frequency errors of each UT have to be reflected in matrix C. It shall be observed that a constant phase rotation of each single UT signal would not be a problem, even if the rotation is different for each UT. In fact, this would be equivalent to a phase-only change of the element of the matrix A which does not impact the LMMSE filter. So, for computing the matrix C, we need to now, in addition to the elements c ik, also the τ i and f i. The timing τ i can be computed extracting the symbol timing of each preamble. At this regard it is sufficient to recover at each GW receiver (one per processed beam) only the symbol timing for the UT actually belonging to that beam (which is typically the strongest one). Such UT will be referred to here as the main UT for that beam. However, these timings, as well as the sequence of data received at each GW receiver, would 2 Assuming oversampling by two, it would be N=2L.

5 5 need to be compensated for by taking into account differences in propagation time through the different on-board and GW repeater chains. In practice, for small data-rate carriers (as it typically occurs for the RL, where symbol rates are at most a few MBaud), the delay spread is negligible, as it is usually less than 1 ns (and it can be calibrated, so that the unknown residual delay is less than 100 ps) 3 and no further compensation is required. The frequency error of each UT signal could also be estimated at the GW by extracting the frequency information of each preamble. Assuming perfectly coherent on-board repeaters and GW receiver chains, it is sufficient to only recover at each GW receiver the frequency information of the main UT of the beam connected to that receiver. VI. SIMULATION RESULTS A. MMSE Processor The system was simulated for different settings for the frequency-reuse strategy, in particular, assuming a uniform frequency reuse strategy, for the following alternatives: - full frequency reuse. In this hypothesis, the same color is used over all beams (both intra-cluster and inter-cluster); - an ad-hoc frequency plan avoiding to reuse the same frequency slot in adjacent beams belonging to different clusters. Obtained results for availability and total throughput are shown in Table 3. Figure 5 also shows the usage of ACM mode in the case where the ad-hoc frequency plan is selected. The results were obtained by a dynamic link-budget simulation where a large number of UTs are randomly activated in each beam. The ITU-R Rec. P.618 model [6] has been considered for rain fading computation of UTs and GWs. The availability figure reported in the table is the average over space (i.e. over all the UTs in the coverage) and time of the link-availability figure obtained from the simulation. The low availability obtained for the MMSE case with multiple clusters and full frequency-reuse is related to interference at the cluster border which cannot be mitigated. In fact, using a single cluster of 88 beams, or using an ad-hoc frequency plan avoiding to assign the same frequency slot to adjacent beams belonging to different clusters, would significantly improve the availability and throughput (see the relevant columns in the table). Finally it has to be noted that the performance with MMSE may be improved increasing the on-board power, whilst this does not happen for the conventional system with 3-color frequency-reuse. For this last case, in fact, the up-link was definitely the bottleneck. Hence performance cannot be improved increasing the on-board power. With MMSE, actually, less power per carrier is available on the down-link 3 In case the delta propagation times needs to be known with high precision it would be sufficient to transmit a single pilot PN signal from a calibration station and cross-correlate the output of each GW receiver with such PN signal. as the total system bandwidth is increased (three times for full frequency reuse or 1.84 times with the ad-hoc frequency plan) and hence a larger number of carriers can be accommodated. In such a case the up-link is not the only bottleneck as downlink thermal noise may also be a limiting factor. This is shown by simulations where on-board power has been increased by a factor of three. Whilst we did not get any throughput or availability improvement for the conventional case, we got the improvements documented in Table 4 for the MMSE cases (only the full frequency reuse cases are shown). Availab. Throughput M M S E 11 Clusters of 8 beams. Full Freq. Reuse 1 Cluster of 88 beams. Full Freq. Reuse 11 Clusters of 8 beams. Ad-Hoc Freq. Reuse 71.6 % 21.3 Gbit/s % 32.0 Gbit/s % 26.9 Gbit/s Conventional with 3-Colors Freq. Reuse % 19.2 Gbit/s Table 3 - Performance with Full Frequency-reuse and Nominal On-board Power (uniform traffic) Limiting the system to the use of QPSK produced the results shown in Table 5 for the MMSE case with full frequency-reuse and multiple clusters. As reference for the conventional (three-color frequency-reuse) case we got a throughput and availability of about 15.1 Gbit/s and 99.86%, respectively, for both considered cases of on-board powerlevel 30% 25% 20% 15% 10% 5% 0% QPSK_TURBO-1/2 QPSK_TURBO-1/3 QPSK_TURBO-2/3 QPSK_TURBO-3/5 QPSK_TURBO-6/7 QPSK_TURBO-3/4 8PSK_TURBO-3/4 8PSK_TURBO-2/3 8PSK_TURBO-4/5 16APSK_TURBO-3/4 Figure 5 ACM mode usage for the case of ad-hoc frequency plan 11 Clusters of 8 Beams 1 Cluster of 88 Beams Availability 74.84% 94.84% Throughput 24.1 Gbit/s 38.2 Gbit/s Table 4 - MMSE performance with 4.77 db more on-board power. Full frequency reuse Finally it is interesting to know that, with MMSE processing, UTs at beam edge typically have better performance than UTs at beam center (if inter-cluster interference is not a problem) as power from multiple beams may be optimally combined (see figure 6).

6 6 MMSE Full Freq Reuse On-board Power Normal db Availability 71.6 % 74.84% Throughput Gbit/s 23.4 Gbit/s Table 5 - MMSE Performance with 11 clusters of 8 beams each and Full Frequency-reuse. QPSK modulation modes only Figure 6 Throughput map with MMSE processing and full frequency reuse (Single 88 beam cluster) db more power on board. Results were so far obtained assuming an ideal knowledge of the channel matrix H at the GW. Assuming channel matrix estimation based on burst preambles, as illustrated in Section V, we obtained the results shown in Table 6. Estimation through UWs 64 symb. 128 symb. Availability 67.19% 69.34% Throughput 19.6 Gbit/s Gbit/s Table 6 - Performance with Matrix Estimation Errors (Full Frequency-reuse with multiple clusters) B. MMSE-SIC Processor The scheme performance will depend on the decoding order. In system simulations we have assumed that the UTs are decoded according to their SINR. Table 7 shows the achievable performances in ideal channel estimation conditions. The performance with MMSE only and with a conventional processing with three-color frequencyreuse can be seen in Table 3. The simulations were done in a uniform traffic-per-beam hypothesis. Also, it was assumed that the on-board power is the same for the MMSE-SIC and conventional cases, notwithstanding that a three times more bandwidth is available in MMSE based cases. MMSE-SIC Availability 93.3% Throughput 33.4 Gbit/s Table 7 - Performance of MMSE-SIC Assuming Full Frequencyreuse (with 11 Clusters of 8 Beams) 64 Symbols 48 Symbols Availability 89.34% 87.99% Throughput 30.9 Gbit/s 30.1 Gbit/s Table 8 - Performance of MMSE-SIC with Different Estimation Errors The advantages of MMSE-SIC over just MMSE are evident. With channel estimation performed over finite preamble sequences in accordance with the proposed strategy, the performances will be slightly impaired as shown in Table 8. Anyway a good improvement is obtained even with channel estimation over preambles as short as 48 symbols. Finally, Table 9 shows the performance of MMSE-SIC with ideal channel estimation in the case of a single cluster of 88 beams, instead of 11 separate clusters of 8 beams each. MMSE-SIC Availability 99.90% Throughput 55.6 Gbit/s Table 9 - Performance of MMSE-SIC with Full Frequency-reuse and Uniform Traffic for a Single Cluster of 88 Beams VII. CONCLUSIONS The results presented herein demonstrate that suitable spatial receiver processing techniques can provide valuable improvements in terms of system capacity when applied to the RL of a wideband high-capacity multibeam satellite system, without greatly affecting the UT complexity. As a matter of fact all the processing burden occurs at the GW, with a relatively modest complexity increase at the current technology status. Such improvements turn into the possibility of implementing RLs that, for a given total system capacity, consume less bandwidth, thus alleviating the general problem of spectral resources scarcity. It was also shown that the required knowledge of the channel characteristics may be obtained by suitable processing of preambles, which are anyway present in TDMA systems. The resulting errors in channel estimation, with typical preamble sequence lengths, only slightly reduce the potentially achievable performance improvements. Further work will be needed to refine the proposed concepts, with particular regard to optimizing packet scheduling algorithms in general, and to assessing the effects of the decoding order on MMSE-SIC performance. REFERENCES [1] G. Gallinaro, et al., Perspectives Of Adopting Inteference Mitigation Techniques In The Context Of Broadband Multimedia Satellite Systems, 23 rd AIAA International Conference on Satellite Communication, September 2005, Rome, Italy. [2] M. Neri, M. Casadei, A. Vanelli-Coralli, and G.E. Corazza, Multiuser detection for S-UMTS and GMR-1 Mobile Systems, 9th International Symposium on Spread Spectrum Techniques and Applications, August 28-31, [3] General Reflector and Antenna Farm Analysis Software (GRASP) SW from TICRA, Læderstræde 34, DK-1201 Copenhagen K, Denmark, [4] S. Benedetto, R. Garello, G. Montorsi; C Berrou, C. Douillard et al, MHOMS: High-Speed ACM Modem for Satellite Applications, IEEE Wireless Comm., April [5] S. Verdú, Multiuser Detection, Cambridge Univ. Press, [6] ITU-R Rec. P.618-8, Propagation Data and Prediction Methods Required for the Design of Earth-Space Telecommunication Systems, April 2003