Performance assessment of Fade Mitigation Techniques for the GEOCAST IST project with transparent and OBP architectures

Transcription

1 Space Communications 22 (2009) DOI /SC IOS Press Performance assessment of Fade Mitigation Techniques for the GEOCAST IST project with transparent and OBP architectures Laurent Castanet a,, Ana Bolea-Alamañac b,, Michel Bousquet c, Laurent Claverotte d and Ricardo Gutierrez-Galvan d a ElectroMagnetics and Radar Department, ONERA, Toulouse, France b Co-operative Research Lab on Aerospace Communications, TESA, Toulouse, France c Aerospace Electronics and Communications Programmes, Toulouse, SUPAERO, France d Alcatel Alenia Space, Toulouse, France Abstract. The objective of this paper is to discuss the issues related to the use of Fade Mitigation Techniques (FMT) for the GEOCAST satellite system, considering both transparent and regenerative payloads. The main characteristics of the GEOCAST systems are presented first. Baseline link budgets are established both for the transparent return link and for the regenerative configurations. Then the FMTs designed for GEOCAST are presented. Theoretical performance of the FMTs and their practical implementation are described focusing on the FMT control loop design. Finally, simulation results for system performance with emphasis on physical layer, are presented both for transparent and regenerative configurations. Keywords: GEOCAST, propagation impairment, Fade Mitigation Technique, fade estimation, FMT control logic, air interface 1. Introduction GEOCAST (Multicast over geo-stationary EHF satellites) is a project funded by the European Information Society Technology programme (IST) for three years from May 2000 to April 2003 and driven by Alcatel Space (now Alcatel Alenia Space). The main goal of the project is the integration of the satellite into the Global Information Infrastructure to provide multicasting services. The objectives of this project are then to define next generation multicast satellite systems, demonstrate their feasibility and optimize their performance with an emulator realization. The study focus of the project is a system at Ka-band over Europe having both transparent and regenerative payloads and multiple spot beams. The task of ONERA (Office National d Etudes et de Recherches Aérospatiales, French aerospace research * Corresponding author: Laurent Castanet, ElectroMagnetics and Radar Department, ONERA, 2 avenue E. Belin BP 4025, Toulouse Cedex 4, France. Laurent.Castanet@onera.fr. ** Now with ESA/ESTEC. center) in this project deals with studying the influence of the Ka-band propagation channel on the air interface of the GEOCAST system, and provide test results of Fade Mitigation Techniques (FMT) to insure an acceptable service availability for the end user [3,17,18]. The GEOCAST system relies on a Ka-band payload including two types of repeaters: transparent repeaters to provide a star network services and regenerative repeaters to support mesh capabilities. The objective of this paper is to discuss the issues related to the use of FMTs, considering both transparent and regenerative payloads. It also emphasizes the study of main differences from the FMT design point of view, in terms of design drivers, mode of operation and performance. In the first part of the paper, the characteristics of the GEOCAST system are presented and baseline link budgets are established both for the transparent return link and for the regenerative configuration. Section 2 presents a description of the FMTs designed for GEO- CAST. Section 3 includes the FMTs relevant to the GEOCAST system and their theoretical performance assessment through link budget analyses. The practi /09/$ IOS Press and the authors. All rights reserved

2 2 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures cal implementation of these FMTs is described with focus on the FMT control loop design and the simulation tool. Section 4 provides simulation results of the system performance. Real propagation events obtained during the OLYMPUS propagation experiments are introduced in the GEOCAST FMT simulator, and results of the physical layer performance for both transparent and regenerative configurations are presented. 2. Radio layer baseline definition Two payload configurations have been considered in the framework of the GEOCAST project: a regenerative packet switch system and a transparent repeater. The regenerative configuration of the GEOCAST system as described in [5] is compared with the transparent return link system configuration GEOCAST transparent system air interface The GEOCAST transparent system aims at providing multicast services to end users through a star network architecture. Therefore, different kinds of user Earth stations are interconnected through large gateways (larger than the gateways considered in the GEO- CAST regenerative system). The GEOCAST transparent system uses a 4-fold frequency re-use pattern. A total bandwidth of MHz is available in each spot beam. The frequency ranges of GHz and GHz are used for the uplink and the downlink respectively. Uplink and downlink are both linearly polarized, each one using a given polarization plane. Two distinct formats are considered: a MF-TDMA format compliant with the DVB-RCS standard for the return link and a TDM format compliant with the DVB-S standard for the forward link. Three classes of service are proposed to the end-user, with data rates of 144 kb/s for consumers, 384 kbit/s for professional consumers and 2 Mbit/s for corporate communications (see Table 1). The main characteristics of the earth stations of the GEOCAST transparent system are the following: an antenna diameter of 5.5 m and a power amplifier of 120 W for gateways leading to an EIRP at saturation of 60 dbw; an antenna diameter of 0.8 m (consumer) and 1.2 m (corporate) and a power amplifier of 2 W for user earth stations. As far as the satellite is concerned, the main performances are the following: on the forward link, the access scheme is single channel per carrier, and the satellite figure of merit G/T is 19 db/k, on the return link, one channel can be filled with either 192 carriers at 144 kbit/s, or with 72 carriers at 384 kbit/s, or with 13 carriers at 2048 kbit/s; the satellite EIRP at saturation is 48 dbw and the satellite figure of merit is G/T = 15.5 db/k;the satellite transponder bandwidth is MHz. The radio layer of the transparent system is designed to achieve the same required availability as the regenerative packet switch system, that is: 99.5% for 95% of the satellite coverage for consumer Earth stations, and 99.8% for 95% of the coverage for corporate Earth stations. The performance objective of the air interface is to reach a packet error rate PER lower than 10 7 which corresponds to a bit error rate BER lower than for 53 bytes and to a BER lower than for 188 byte packets GEOCAST system link budgets The baseline return link budget are summarized in Table 2. In this link budget, Uplink Power Control (ULPC) adjusts the output power of the transmit Earth station [13] in order to keep the satellite input carrier level constant [9] equal to 125 dbw. This input car- Table 1 Main characteristics of the GEOCAST transparent system physical layer Physical layer Forward link Return link Waveform DVB-S DVB-RCS Coding DVB-S coding rates 6/7 + Reed Solomon (188/204) Data rates 57.6 Mbit/s 144/384/2048 kbit/s Table 2 Transparent system return link budget Return link budget Clear sky Coding rate 6/7 Information data rate 2048 kbit/s Bandwidth per carrier 2 MHz Earth station SSPA output power 0.85/1.7 W Required satellite input power 125 dbw Required E b /N 0 (BER = ) 6dB Uplink margin 0/3 db Downlink margin 1.9 db

3 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures 3 rier level is needed to obtain the optimal satellite operating point resulting from link budget optimisation. Also, in a multicarrier operation, all the carriers are requested to reach the transponder with a similar power in order to avoid carrier suppression and capture effects. Concerning the Earth station ULPC range, the minimum allowed Output Back-Off is 0.5 db to reduce non-linear effects arising when TWTA is operated near saturation. The maximum Earth station transmitted power is then fixed to 1.7 W. The minimum power (when uplink is in clear sky) to reach the required 125 dbw at satellite input is 0.8 W. This gives an ULPC dynamic range of approximately 3 db. Another optimization can increase the uplink dynamic range by reducing the required satellite input power of all carriers. However, this will reduce the downlink margin which is necessary to cope with downlink attenuation. For comparison purpose, the link budget of the GEOCAST regenerative system is given in Table 3. Only uplink is given since it is an On-Board Processed (OBP) payload. The characteristics of the ground segment is similar to the one of the transparent system, a station of type 3 that allows up to 2 W of output power has been chosen for a fair comparison. Note that, in the OBP configuration, 1.75 db of margin is achieved using 600 mw, while in the transparent payload 800 mw are needed in order to meet the 125 dbw of required satellite input power. Also the maximum margin achieved in the regenerative case is 6.9 db. The ULPC dynamic range is therefore 5.2 db although 0.5 db should be deducted from this margin to do a fair comparison since no minimum OBO (Output Back-Off) has been taken into account for the OBP configuration. This larger dynamic range is essentially due to the fact that for the OBP configuration a more robust coding has been chosen: turbo code 2/3 instead of 6/7 used for the transparent payload. This translates in a higher used bandwidth for a similar information data rate: indeed, the OBP payload needs 2.5 MHz per carrier instead of the 2 MHz per carrier for the transparent one. Table 3 Regenerative system uplink budget Frequency GHz Coding rate 2/3 Uplink information data rate 2272 kbit/s SSPA output power 600 mw 2 W Available uplink E b /(N 0 + I 0 ) 6.35dB Required E b /N 0 (BER = ) 4.6dB Uplink margin db In conclusion, when ULPC is used, the performance of the transparent payload is not very different from the OBP case. The only significant difference is the need of taking into account the downlink with its associated margin in the transparent case in order to close the link budget; while in the regenerative case, the signal is restored on-board and the link budgets are therefore independent. 3. Definition of FMTs for the transparent system return link In this section, the FMTs relevant to the GEOCAST system and their theoretical performance are assessed through link budget analyses. Then the practical implementation of these FMTs is described, with focus on the FMT control loop design and the simulation tool Return link budgets with FMTs A preliminary analysis of the return link budget shows that it is clearly governed by the uplink definitely requiring the implementation of powerful FMTs. In this case (as for the GEOCAST regenerative system), the state-of-the art in Ka-band SSPA does not allow power higher than 2 W to be obtained for this kind of user Earth station due to cost considerations. So the implementation of Uplink Power Control (ULPC) only is not sufficient to comply with the required availability and a combination of FMTs, at least ULPC and Adaptive Coding (AC) [12], has to be used. The bandwidth is kept constant to avoid constraints on the Media Access Control (MAC) layer. So, changing the coding rate implies changing the information rate. When no more robust codes are available, the last resource will be the use of Data Rate Reduction (DRR) with spreading sequences in order to keep a constant bandwidth [14]. Table 4 shows the margins obtained by the combined use of these techniques. Note that the uplink margin corresponds only to the ULPC margin. In fact, adaptive coding or data rate reduction do not modify the carrier power and therefore they cannot increase the uplink margin. They change the required Eb/No making easier to close the overall link budget. In this configuration any uplink attenuation exceeding 3 db will degrade the carrier power at satellite input. Table 5 hereafter shows the FMT definition carried out for the GEOCAST regenerative system.

4 4 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures Table 4 Return link budget with ULPC, AC and DRR for the transparent system (QPSK modulation) Return link budget ULPC AC AC AC AC DRR Coding rate 6/7 4/5 3/4 2/3 1/2 1/3 Information data rate (kbit/s) 2048 kbit/s /398/199 kbit/s Bandwidthpercarrier 2MHz 2MHz 2MHz 2MHz 2MHz 2MHz Nominal GES TWTA power 0.85/1.7 W 1.7 W 1.7 W 1.7 W 1.7 W 1.7 W Req. satellite input power (dbw) 125 dbw dbw Req. E b /N 0 for BER = db 5.3 db 4.6 db 4.0 db 3.2 db 2.5 db Uplink margin /3 db 3 db 3 db 3 db 3 db 3 db Downlink margin 1.9/1.9 db 3.1 db 4.1 db 5.2 db 7.2 db 9.7/12.7/15.7 db Overall margin 1.9/4.7 db 5.9 db 6.9 db 8.0 db 10.0 db 12.5/15.5/18.5 db FMT dynamic range 0/2.8 db 4.0 db 5.0 db 6.1 db 8.1 db 10.6/13.6/16.6 db Table 5 Link budget for the regenerative system with ULPC, AC and DRR (QPSK modulation) Frequency GHz GHz GHz GHz GHz Coding rate 3/4 1/2 1/3 1/3 1/3 Uplink info. data rate 2272 kbit/s 1515 kbit/s 1010 kbit/s 505 kbit/s 252 kbit/s SSPA output power 0.8/2 W 2.0 W 2.0 W 2.0 W 2.0 W Uplink E b /(N 0 + I 0 ) 7.0/11.0 db 12.7 db 14.5 db 17.5 db 20.5 db Req. E b /N 0 ( ) 5.4 db 3.6 db 2.9 db 2.9 db 2.9 db Uplink margin 1.6/5.6 db 9.1 db 11.6 db 14.6 db 17.6 db FMT dynamic range 0/4.0 db 7.5 db 10.0 db 13.0 db 16.0 db Making a direct comparison between the transparent and regenerative configurations is complicated since the physical layer choices are not the same: ATM encapsulation for OBP is used instead of MPEG for transparent configurations leading to a different required E b /N 0. The bandwidths used are 2.3 MHz for OBP instead of 2 MHz for transparent payloads. Nevertheless, it can be inferred from link budgets that the characteristics and the performances of the joint FMT designed for both GEOCAST transparent and regenerative systems are similar. However, two differences have to be pointed out. For the transparent satellite case it is necessary to take into account both uplink and downlink attenuations to close the link budget and to assign the right coding. The throughput depends on both uplink and downlink attenuations. Secondly, a degraded carrier satellite input power in a transparent transponder will suffer an additional degradation due to the carrier suppression effect. To minimize this effect a lower satellite input power is set impacting clear sky throughput FMT control loop for the transparent system return link In the previous sections, the air interface of the GEOCAST transparent system has been described and FMTs have been defined. This Section defines the fade detection scheme and the simulator used to assess physical layer performance. In a transparent configuration, the detection has to be carried out at ground level. As the downlink performance depends on the propagation conditions on the uplink, the uplink and downlink attenuations impact overall link budget in a different way (which mainly depends on the link budget balance, ratio between uplink to downlink carrier-to-noise-ratio). For mitigation purpose, it is therefore necessary to be able to separate uplink and downlink fades, so as to apply mitigation on the relevant link. On the other hand, the detection scheme to be considered in the GEOCAST transparent system has to be centralized, which means that the decision function is implemented in the Network Control Center (NCC). Consequently, each Earth station has to send a request to the NCC in order to be authorized to trigger a FMT. To be able to decide properly about the triggering of a FMT, it is assumed that the NCC collects two different types of information (see Fig. 1). On the one hand, the NCC measures in real time the Carrier to Noise Ratio of the link from User Station (UES) to the NCC and from the Gateway Station (GES) to the NCC, on the other hand, a down-link beacon is available, allowing the measurement of propagation impair-

5 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures 5 Fig. 1. Detection scheme for the GEOCAST bent pipe system. ment on the satellite NCC link. Two main assumptions are considered for this analysis: firstly, the interference level does not vary significantly and can be considered constant, and secondly the transponder operates in the linear zone for the links from UES to NCC and from GES to NCC. The implementation of a FMT control loop requires to perform three main tasks: namely detection, prediction and decision [7,11]. Figure 2 shows the main control loop functions, signal level is estimated, the detection function allows the estimation of attenuation and scintillation components. Following, a short-term prediction is carried out in order to anticipate system reaction time and finally NCC activates a possible FMT taking into account of intermediate margins. Figure 3 presents the FMT simulator developed by ONERA [4] as a part of the European action COST 255 Radiowave propagation modelling for new Sat- Com services at Ku-band and above and upgraded in the framework of GEOCAST. The main characteristics of this FMT control loop simulator are the following: direct use of propagation time series for beacon monitoring simulation (satellite, NCC), introduction of propagation time series in the overall link budget, to produce carrier-to-noiseratio time series (SNIR 1 from UES to NCC and SNIR 2 from GES to NCC) that are introduced in the simulator,

6 6 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures Fig. 2. Main control loop functions. Fig. 3. GEOCAST FMT simulator flow chart. iterations considering link budget evolutions between t and t + δt. Reference [5] describes the control loop for the regenerative system and the optimization results of the main internal parameters of the FMT control loop. 4. Simulation results 4.1. Propagation event introduced in the simulation Figure 4 presents the rain event considered for the radio layer analysis of the GEOCAST system. It has

7 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures 7 Fig. 4. Shower event introduced for the analysis of forward and return uplinks. been collected by Université Catholique de Louvainla-Neuve during the Olympus Propagation EXperiment [15] and is a typical shower event characterized by a maximum attenuation of 15 db at 30 GHz. For both transparent and regenerative systems, the analysis is focused on the uplink, because link budgets are essentially uplink limited. However, all the simulation results concerning the transparent system present the overall link budget with and without FMT, in terms of E b /N 0 time series Return link of the transparent system Figure 5 shows simulation results where the fade event presented in the previous section is introduced on the uplink. For the baseline configuration, shown in Fig. 5a, without FMT, the return overall link is in outage during the main part of the shower fade event and the unavailability period reaches 400 s. With ULPC implementation, shown in Fig. 5b, the dynamic range of 3 db is not sufficient to compensate for the fade event but the unavailability period can be reduced to the main part of the peak event, which corresponds to 200 s. With the combination of FMTs defined in Subsection 2.2, ULPC is applied firstly, followed by AC with DRR (to maintain the assigned carrier bandwidth), and finally by DRR (see Fig. 5c). In Fig. 5c the black line represents the E b /N 0 of the system and the green line is the outage threshold. Note that the outage threshold varies since adaptive coding is used and each different coding rate has a different level of E b /N 0 required to guarantee BER. The blue line represents the coding rate and signifying the lower the rate the more robust the code less power is needed to achieve a given BER. With this combination of FMTs, the shower fade event is completely mitigated and the link remains available any time Comparison of regenerative and transparent configurations Considering same shower fade event both regenerative and transparent systems performance is compared from the FMT implementation point of view. Figure 6 shows such a comparison, where the behavior of each system is presented with ULPC and ULPC + AC + DRR implementations. The FMT definition for the regenerative packet switch system and the corresponding simulation results are given in [5]. As far as the performance with ULPC is concerned, 150 s of outage is obtained for the regenerative system and 200 s for the transparent system. This difference can be explained by the different required E b /N 0 in each case. Note that 6 db (green line) are required in the transparent case since the nominal coding is a turbocode 6/7. For the OBP, however, the coding (Fig. 6a)

8 8 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures (a) (b) Fig. 5. Overall link budgets without FMT (a), with ULPC (b) and with ULPC, AC and DRR (c) for the return link. is a turbo code 2/3 with a required E b /N 0 of 4.6 db. Regarding the combination of ULPC, AC and DRR in Fig. 6b, the shower fade event is fully mitigated in both system configurations. The main conclusion of this comparison is that the performance of the radio layers of both systems are similar with respect to the FMTs designed in the framework of GEOCAST. These similar results are mainly due to the fact that large gateways have been used for the transparent configuration. The link budget is mainly driven by the UES-satellite link. Also, this result is achieved at an expense of increasing complexity in the fade detection scheme for the transparent configuration with respect to the OBP system. The detec-

9 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures 9 (c) Fig. 5. (Continued.) tion scheme is more complex for the transparent system due to the increased number of exchanges required between UES, GES and NCC, whereas only an information from the satellite to the Earth station is required for the regenerative configuration. A downlink beacon signal is needed from the satellite to the NCC in order to be able to separate uplink and downlink fades. The main consequences of this increased complexity of the transparent configuration with respect to the regenerative case is a longer system reaction time due to the exchanges between the system component and increased signaling allowing these exchanges. One way to decrease the complexity of the transparent configuration would be to implement the control loop and the decision in the gateways instead of the NCC. In this case, the exchange of information is only needed between the transmitter and the receiver. But on the other hand, each gateway should be equipped with a beacon receiver. A further analysis shows that the use of a beacon signal may not be necessary when link budget is clearly uplink or downlink driven so alternative fade detection architecture could be proposed with gateway detection without beacon signal. With respect to the OBP alternative, however, the round-trip delay is at least twice longer (i.e., station-satellite-gateway) instead of station-satellite path. 5. Conclusion This paper addresses the study of the influence of the Ka-band propagation channel on the air interface of the GEOCAST system, architecture based on transparent and regenerative payloads and multiple spot beams at Ka-band over Europe. As the current technology does not allow large propagation margins to be implemented in satellite systems at Ka-band, Fade Mitigation Techniques have been considered early in system design to provide the end-user with an acceptable service availability. Regarding FMT design, it has been shown that the use of FMTs enables the system to cope up with important fades and therefore to comply with the required availability. However, the design of these FMT is strongly dependent on the system requirements, e.g., service availability, user minimum data rate, system capacity or interference reduction [6]. ULPC in particular, can limit the interference level in the system (adjacent beam interference) by reducing clear sky power. The use of adaptive waveform FMT, such as adaptive coding or adaptive modulation, is very promising, as it may increase the spectral efficiency of the link in clear sky leading to an improvement in the overall system capacity [8,10]. Actually, the trade-off between user data rate, system capacity and interference reduction,

10 10 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures (a) (b) Fig. 6. Comparison between: return uplink budget for the regenerative system with ULPC (a) and with ULPC and AC (b), and return overall link budget for the transparent system with ULPC (c) and with ULPC AC and DRR (d). and DVB-RCS standard compatibility led to the choice of using Up-Link Power Control (ULPC), Adaptive Coding (AC) and Data Rate Reduction (DRR) to work at a constant bandwidth. Then, it has been demonstrated in GEOCAST that the best compromise results in the design of joint FMTs, relying on a combination of ULPC, AC and DRR, with respect to single mitigation scheme. Finally, the radio layer analysis of GEOCAST shows that events up to 15 db can be completely mitigated with an appropriate combination of

11 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures 11 (c) (d) Fig. 6. (Continued.) ULPC, AC and DRR; therefore, a 99.5% availability is achievable for Ka-band satcom systems in Europe with this form of FMT. The main conclusions of this paper is that the performance of the radio layer of the transparent system is similar to that of the regenerative configuration with respect to FMT. FMTs prove their interest to comply availability requirements and to improve system efficiency with both OBP and transparent systems. With the transparent system configuration, this performance improvement is achieved due to the use of large gateways. The complexity of the on-board processing sys-

12 12 L. Castanet et al. / Performance assessment of FMT for the GEOCAST IST project with transparent and OBP architectures tem architecture is transferred to the ground segment in the case of transparent payload configuration. In terms of FMT implementation, the transparent configuration adds additional complexity to the physical layer as opposed to the OBP solution due to the dependency between the up and down links. The absence of on-board processing prevents on-board fade estimation to be carried out, so a more complex detection scheme has to be designed to separate uplink and downlink contributions as well as propagation and interference aspects. In addition, the control logic needs to be more sophisticated for transparent configurations and the roundtrip delays will be in general longer than in the OBP case. This work has been focused essentially on physical layer issues, where analysis and simulation have been carried out to optimise the FMT control loop and to assess performance. However, the implementation of advanced short-term prediction techniques should further improve system performance, especially when large delays, specific to transparent configurations, are involved [2]. On the other hand, FMT implementation clearly impacts signalling and upper layers, in particular the MAC layer through the resource management [16]. Within the framework of the GEOCAST project, the combined role of FMTs considered in the present paper has been analyzed for the design of reliable satellite protocols and are presented in [1]. References [1] F. Arnal, A. Bolea-Alamañac, M. Bousquet, L. Claverotte, L. Dairaine, R. Gutierrez-Galvan and G. Maral, Reliable multicast transport protocols performances in emulated satellite environment taking into account an adaptive physical layer for the GEOCAST system, in: Proceedings, COST International Workshop on Satellite Communications from Fade Mitigation to Service Provision, Noordwijk, Netherlands, 2003, pp [2] A. Bolea-Alamañac, M. Bousquet, L. Castanet and M.M.J.L. Van de Kamp, Impact of the implementation of short-term prediction models in Fade Mitigation Techniques control loops, in: COST International Workshop on Satellite Communications from Fade Mitigation to Service Provision, Noordwijk, Netherlands, 2003, pp [3] L. Castanet, J. Lemorton and M. Bousquet, Fade Mitigation Techniques for new SatCom services at Ku-band and above: a Review, in: 4th Ka-band Utilization Conference, Venice, 1998, pp [4] L. Castanet, D. Mertens and M. Bousquet, Simulation of the performance of a Ka-band VSAT videoconferencing system with uplink power control and data rate reduction to mitigate atmospheric propagation effects, International Journal of Satellite Communications 20(July August) (2002), [5] L. Castanet, J. Lemorton, M. Bousquet and L. Claverotte, A joint Fade Mitigation Technique applied to the regenerative packet switch payload of the GEOCAST system, in: 8th Kaband Utilization Conference, Baveno, Italy, 2002, pp [6] L. Castanet, A. Bolea-Alamañac and M. Bousquet, Review of interference and Fade Mitigation Techniques for Ka and Q/V band satellite communication systems, in: COST International Workshop on Satellite Communications from Fade Mitigation to Service Provision, Noordwijk, Netherlands, 2003, pp [7] COST 255, Radiowave propagation modelling for new SatCom services at Ku-band and above, in: COST 255 Final Report, Chapter 5.3, Impairment Mitigation and Performance Restoration, ESA Publications Division, SP-1252, March [8] R. Gaudenzi and R. Rinaldo, Adaptive coding and modulation for next generation broadband multimedia systems, in: 20th AIAA International Communication Satellite Systems Conference, Montréal, 2002, paper AIAA [9] H. Dodel and C. Riedl, Can ULPC double satellite capacity?, in: 2nd European Conference on Satellite Communications, Liège, Belgium, 1991, pp [10] M. Filip and E. Vilar, Implementation of adaptive modulation as a fade countermeasure, International Journal of Satellite Communications 8(June) (1990), [11] B.C. Gremont, Event detection, control and performance modelling, in: COST International Workshop on Satellite Communications from Fade Mitigation to Service Provision, Noordwijk, Netherlands, 2003, pp [12] B.C. Gremont, A.P. Gallois and S. Bate, Efficient fade compensation for Ka-band VSAT systems, in: 2nd Ka-band Utilization Conference, Florence, Italy, 1996, pp [13] J. Hörle, Up-link power control of satellite Earth stations as a fade countermeasure of 20/30 GHz communications systems, International Journal of Satellite Communications 6(June) (1988), [14] K. Kerschat, O. Koudelka, W. Riedler, M. Tomlinson, C.D. Hughes and J. Hoerle, A variable spread-spectrum fade countermeasure system for the DICE video conference system, in: OLYMPUS Utilization Conference, Sevilla, Spain, 1993, pp [15] OPEX, Reference Book on Attenuation, in: 2nd Workshop of the Olympus Propagation Experimenters (OPEX), ESA- ESTEC-WPP-083, Vol. 1, Noordwijk, The Netherlands, [16] P. Pech, L. Castanet, J. Radzik and M. Bousquet, Insights into an architecture of Ka-band OBP satellite system involving a Fade Mitigation Technique (FMT): challenges, simulation and performance, in: 8th Ka-band Utilization Conference, Baveno, Italy, 2002, pp [17] G. Tartara, Fade countermeasures in millimetre-wave satellite communications: a survey of methods and problems, in: Olympus Utilization Conference, Vienna, Austria, 1989, pp [18] M.J. Willis and B.G. Evans, Fade countermeasures at Ka-band for OLYMPUS, International Journal of Satellite Communications 6(June) (1988),

13 Space Communications 22 (2009) DOI /SC IOS Press A study on SNR estimation algorithms for channel state estimation in Communication Satellite Systems employing Fade Mitigation Techniques Anbazhagan Aroumont a, Laurent Castanet b and Michel Bousquet a a Signal, Communications, Antennas & Navigation Research Group, ISAE Toulouse, Universite de Toulouse, Toulouse, France b Department of Electromagnetism and Radar, ONERA Toulouse, Toulouse, France Abstract. Channel state estimation is one of the key processes carried out at the physical layer of Satellite Systems employing Fade Mitigation Techniques such as Adaptive Coding and Modulation (ACM). An accurate and reliable channel estimate is needed to fully realize the capacity gains accrued by using ACM. In this paper the SNR estimation algorithms, which perform the role of channel state estimators, are analysed from a system point of view to get a quantitative idea on the number of received symbols needed to get a reliable estimate, and the impact of interference noise on it. A DVB type satellite system has been investigated for the study. An improvement over the Maximum Likelihood (ML) estimator using Bayesian principles is suggested and illustrated. Keywords: ACM, SNR estimation, maximum likelihood estimator, Bayesian principles 1. Introduction Capacity gains accrued by using ACM in satellite systems have been extensively studied [1,2]. Current satellite systems look to use ACM to increase system capacity and availability. Channel state estimation is one of the key processes carried out at the physical layer of the satellite systems that use ACM. At any given point of time the system switches to a different modulation and coding scheme, called mode, or retain the current one based on the channel state estimated at that time. Hence the efficiency of the ACM procedure is tied to the accuracy and reliability of the channel estimate. SNR estimation algorithms take up the role of channel estimators in systems employing ACM. In the literature SNR estimation algorithms have been proposed for the AWGN channel with the analysis focussed on the Mean Square Error (MSE) of the estimator, and on the efficiency with respect to the Cramer Rao bound (CRB). Although this is useful to choose the best among several candidates, issues like interference noise or number of symbols required to get a given performance are not addressed by this approach. An analysis of the SNR estimation algorithms from the system point of view may thus seem pertinent. Such an analysis may also provide clues on how to improve the performance of the estimator as is shown later in this study. The reference system on which the SNR estimators have been evaluated is a DVB type broadband interactive satellite system compliant with the DVB-S2 and DVB-RCS air interface standards in the forward and return links respectively. The rest of the paper is organized as follows: Section 2 briefly describes the SNR estimation algorithms and their performances are derived via simulations with the performance criterion being the number of received symbols needed to obtain the SNR estimate for a given error margin. Section 3 provides a system study wherein the SNR estimators are evaluated on a DVB type system. Section 4 details the improvement suggested over the Maximum Likelihood (ML) estimator /09/$ IOS Press and the authors. All rights reserved

14 32 A. Aroumont et al. / SNR estimation algorithms for channel state estimation in Communication Satellite Systems using Bayesian principles. Section 5 provides the conclusions. 2. SNR estimators Several SNR estimation algorithms have been proposed in the literature for the AWGN channel [3 5]. Not all of them are solutions to direct parameter estimation problems, and not all of them are amenable to theoretical treatment like MSE or efficiency with respect to the CRB. For example the BER measurement method, as described in [4], derives the SNR estimate by mapping the measured BER to the SNR for the given mode. Since this procedure is not a direct parameter estimation problem, no analysis like MSE or efficiency was carried out in [4]. In [3] details about the ML algorithm are provided as well as that of many others. The SNORE algorithm as described in [1,2] is nothing but the ML algorithm as is shown later. In [5] an estimation algorithm based on second and fourth order moments of the received symbols has been described, appropriately modified for signal with nonconstant envelope like 16-QAM or 16-APSK as most prominent examples in this respect. A few concepts need mentioning before we choose the algorithms for analysis. They could use either known transmitted symbols for estimation in which case they are called data aided, or receiver decisions on symbols in which case they are called decision directed. The moments based SNR estimator is an example of a non data aided solution. The ML algorithm operating at the output of the matched filter assumes perfect carrier synchronization whereas the moments based estimator does not need this. Analysing the various SNR estimators [3 6] the following details could be inferred: The decision directed versions of the estimators suffer from bias at low SNR. For a given number of symbols the data aided ML estimator is much more efficient than the moments based estimator. The data aided ML estimator is by far the most efficient estimator with respect to the CRB. Considering the criteria of bias and efficiency, the data aided ML estimator is the best estimator among the various algorithms. In the following we analyse the performance of this estimator. Since we are analysing a data aided estimator the modulation scheme assumed for all further analysis is QPSK Maximum likelihood SNR estimator The equations for the ML algorithm as given in [3] are reproduced here for the sake of completeness: [ 1K K 1 k=0 Ŝ ML = (r ] 2 I,km I,k + r Q,k m Q,k ) 1 K 1 K k=0 (m2 I,k + m2 Q,k ), ˆN ML = 1 K 1 [r K I,k 2 + r2 Q,k ] k=0 ŜML K 1 1 [m 2 K I,k + m2 Q,k ], k=0 where S ML is the signal power, N ML is the noise power, K is the number of received symbols used for the estimation, r I and r Q are the in-phase and quadrature components of the received signal, m I,k and m Q,k are the in-phase and quadrature component of the message symbol. Comparing the SNORE estimator equations from [1,2] with the ML equations we see that for MPSK symbols, they are the same since K 1 1 [m 2 K I,k + m2 Q,k ] = 1. k=0 Figure 1 shows the performance of the ML algorithm as a function of number of symbols needed to obtain the estimate with a given error margin delta error() delta such that ( n ) 99%, where error isthe estimation error defined as SNR ŜNR and error() is the n element array containing the estimation error corresponding to each iteration. SNR is the true SNR and ŜNR is the estimated SNR. n is the number of times the SNR is estimated using K symbols and is chosen to be 10 4 to provide reliable results. 3. System study With the performance of the algorithm at hand we could evaluate it on a realistic system to identify whether it satisfies the requirements of such a system from the point of view of channel estimation. The system chosen is a point to point oriented multimedia broadband Interactive Satellite System with highly asymmetric links: a high speed forward link, from the gateway to the users, assumed to be adhering to the

15 A. Aroumont et al. / SNR estimation algorithms for channel state estimation in Communication Satellite Systems 33 Fig. 1. ML algorithm: number of symbols needed to obtain the estimate as a function of SNR for a given error margin delta. DVB-S2 standard [7], and a low data rate bursty return link, from the users to the gateway, assumed to be adhering to the DVB-RCS standard [8]. The satellite coverage is multi-beam with frequency reuse. The operating frequency is in the Ka/Q/V band. The focus of the analysis is on the effects the following constraints have on the estimator s error performance: the availability of symbols for SNR estimation for a given error margin, the noise due to interference which may differ from the Gaussian model, the variation of SNR during the estimation interval due to rain fade slope Forward link Interference considerations Interference in multi-beam satellite systems are due to several sources like adjacent beam, adjacent system etc. The level of the interference varies from point to point in the coverage with the worst case occurring in the centre of the coverage. A brief overview of the various sources of interference in a multi-beam satellite system is provided in [9]. Without going into the details of their origins, their impact on the estimator will be analysed Impact of interference noise on the SNR estimator output In the forward link (DVB-S2), as there is a continuous flow of high speed data with all of the data slots completely filled, the interference sources are always active. So for any user the level of interference does not change with time. In [1] it is claimed that this noise due to interference could be modelled, based on extensive simulations, as a Gaussian process. Taking this proposition as true the AWGN in the forward link is considered to be made up of the receiver thermal noise and the noise due to the interference. This implies that the noise due to interference has no impact on the estimation process. We next analyse the number of pilot symbols offered by DVB-S2 that can be used for the SNR estimation. In DVB-S2 we have 36 pilot symbols once every 16 slots of 90 data symbols. The data slots are transmitted in frames with each frame containing either 64,800 bits or 16,200 bits. Table 1 provides the number of pilot symbols present in one frame, corresponding to various modulation schemes. All of the SNR estimation algorithms have been developed on the basic premise that the SNR to be estimated remains constant or at least the variation is negligible during the estimation interval. If this condition is not satisfied then their performances would break

16 34 A. Aroumont et al. / SNR estimation algorithms for channel state estimation in Communication Satellite Systems Table 1 Number of pilot symbols present in one DVB-S2 frame for different modulation schemes Spectral efficiency (b/s/hz) Number of pilot symbols per frame Frame length = Frame length = 64,800 bits 16,200 bits down. In the context of our analysis, the sources that could perturb the SNR are the fluctuation in the rain attenuation or fade, and the noise due to interference. The fluctuation in the fade is approximated by a linear slope of about 0.6 db/s (for 0.01% of the time) [10], and in the forward link the noise due to interference remains constant at any point. Hence, whether or not the SNR can be considered constant during the estimation period depends essentially on the estimation time. We next compute the worst case estimation duration, i.e., estimation involving maximum number of symbols, for the DA ML estimator. The estimation duration considers only the time needed to acquire the required number of symbols. It does not consider the processing time. The number of symbols chosen for each error margin corresponds to the worst case SNR of 2 db. The results are provided in Table 2. As the symbol rate is assumed to be constant, the estimation time is independent of the data modulation scheme. Since the estimation time is of the order of few milliseconds, the SNR is practically constant during the estimation interval thereby validating the assumption involved in the estimation process. The DVB-S2 standard recommends an error margin of 0.3 db. But from Table 2 it can be seen that an error margin of even 0.2 db is achievable in the forward link using the pilot symbol aided ML Return link The return link confirming to the DVB-RCS protocol handles low speed bursty traffic. The traffic in the return link largely consists of requests, from users handling interactive applications, which is by nature sporadic with low data rate. In DVB-RCS the data transmission is organized as superframes composed of several frames which contain the traffic slots in which the terminals transmit their data. As capacity needed by Table 2 Number of symbols needed and the corresponding estimation time for various error margins Error Number of pilot symbols Estimation margin (db) SNR = 2 db time (ms) , a terminal is dynamically allocated, prior knowledge about the amount of data that would be transmitted during a given period is not available. At best we could only assume a worst case value for the traffic. We assume hereafter that every terminal is allotted at least one traffic slot per superframe. From [11] we take the superframe duration to be 26.5 ms. The DVB-RCS protocol does not have a pilot symbol structure that could be used for channel estimation. Instead we have to rely on the fixed symbol preamble data to perform the channel estimation Impact of interference noise on the SNR estimator output The noise due to interference in the return link poses altogether a different problem when compared to the forward link. As the data traffic is sporadic not all the traffic slots may be filled with data. Hence the interference sources may not be always active at any point and at any given time. So the level of interference experienced by any user may vary from slot to slot which will result in the SNIR varying from slot to slot [2]. If the estimation interval is more than one slot duration then the estimator output is some sort of an average of the SNR values of the individual time slots SNR estimation for the return link Assuming that the preamble sequence preceding the traffic slot is of 48 symbols in length we could have about 2000 symbols in a second for data aided channel estimation. This is just enough to perform the estimation using the ML algorithm with an error margin of 0.5 db. However interference noise and rain fade slope deteriorate the estimator accuracy. The alternatives are that either the estimation has to be done within one traffic slot duration or it should be performed using a combination of data aided and decision directed algorithms so as to make use of the traffic data. The error margin that could be obtained in the return link channel estimation directly affects the possible threshold levels of modulation and coding schemes in a full scale ACM scheme.

17 A. Aroumont et al. / SNR estimation algorithms for channel state estimation in Communication Satellite Systems Improvement over ML algorithm using Bayesian principles The various processes that could affect the SNR of a signal in a satellite system are listed below: Signal power Thermal noise Interference noise Scintillation Atmospheric attenuation. Let us suppose that we are in the ACM regime where the transmitted power remains constant, and that the scintillation component has been filtered out before estimation. In this scenario the SNR as seen by the estimator is affected only by the atmospheric attenuation, mainly rain attenuation, and interference. The increase in thermal noise due to rain attenuation could be absorbed in the rain attenuation component itself. Put in other words, we could say that successive values of SNR differ due to the variation in the noise caused by interference and rain attenuation. In a realistic system, on a time scale of a second or less, the noise due to interference does not change by more than a couple of dbs and the rain attenuation change by about 1 db (rain fade slope). So successive SNR values, with each sample separated from the next by about a second, do not differ by more than a few dbs. Given the previous SNR estimate we could predict to within a few dbs the next value without ever performing any estimation. From a system point of view this is a useful prior knowledge about the channel state that could be used to improve the efficiency of the estimation process. The ML estimation process does not exploit any prior information. However, the estimation process as viewed by the Bayesian approach provides the theoretical framework to exploit any such prior information [12]. The essential principle of the Bayesian approach is to generate the posterior distribution of the parameter of interest from the prior distribution using the data at hand. A particular value is chosen as an estimate from the posterior distribution based on minimizing a suitable cost function [13,15]. The mathematical tool used is the Bayes theorem. To proceed with the Bayesian parameter estimation problem the following complex signal model is assumedasin[3]: r k = S(m i,k + jm q,k ) + (w i,k + w q,k ), (1) where r k is the kth symbol of the received symbol sequence, S is the signal power, m is the message sequence (i and q indicate the in-phase and quadrature components, respectively), and w is i.i.d. sequence of AWGN samples with variance σ 2.LetK be the number of received symbols. The problem is to estimate the ratio S/σ 2 which provides the required SNR estimate. Let S R = σ 2, S = Rσ. (2) Then (1) becomes r k = Rσ(m i,k + jm q,k ) + (w i,k + jw q,k ). (3) Therefore, R, which is the parameter of interest in our case, appears in the signal model and the noise variance has become a nuisance parameter. Now writing (1) explicitly in terms of the real and imaginary components, we have: r i,k = Rσm i,k + w i,k, r q,k = Rσm q,k + w q,k. (4) The posterior distribution of R, given the received symbol sequence {r k }, is given by p(r/{r k }, I) = p(r, σ/{r k }, I)dσ, (5) 0 where I indicates any background information known about the problem. Now the joint distribution of R and σ is given by Bayes theorem as follows: p(r, σ/{r k }, I) = p({r k}/r, σ, I)p(R, σ/i), p({r k }/I) (6) where the first term of the numerator on the right-hand side of (6) is the likelihood function of the received sequence, given R, σ and I, and the second term is the joint prior distribution of R and σ. The denominator is the probability of observing the received symbol sequence conditioned on I. This is simply a normalizing constant in our analysis.

18 36 A. Aroumont et al. / SNR estimation algorithms for channel state estimation in Communication Satellite Systems Using (4) and the fact that the noise is i.i.d. Gaussian, the likelihood function is written as follows: Let p({r k }/R, σ, I) = (2π) N σ 2N [ ( 1 N exp 2σ 2 (r i,k Rσm i,k ) 2 k=1 Q(σ) = + N (r q,k Rσm q,k ) )]. 2 (7) k=1 N (r i,k Rσm i,k ) 2 k=1 + N (r q,k Rσm q,k ) 2. (8) k=1 After some algebraic operations Q is rendered in the following form where ( ( ) Q(σ) = N d 2 1 N 2 ) Re(r N k m k) k=1 ( + N Rσ 1 N 2 Re(r N k k)) m, (9) k=1 d 2 = 1 N (r N i,k 2 + r2 q,k ). k=1 So (7) is rendered as: p({r k }/R, σ, I) ( = (2π) N σ 2N exp Q(σ) 2σ 2 ). (10) Assuming that the parameter of interest and the nuisance parameter are independent, i.e., knowledge about σ does not provide any information about R then their joint distribution could be written as p(r, σ/i) = p(r/i)p(σ/r, I) = p(r/i)p(σ/i). (11) As discussed at the beginning of this section we have a reasonable knowledge about the magnitude of the variation between successive SNR measurements. So, given the previous estimate, the present estimate is bound to lie within a few dbs of it. Put in other words: we have a prior knowledge on the estimate and its variance. Such type of prior knowledge is best captured by the Gaussian distribution which satisfies the principle of Maximum Entropy [12,13], i.e., we are completely ignorant of the noise variance σ using it as a simple scale parameter. Complete ignorance about a scale parameter is best captured by the Jeffrey s prior distribution [13,14]. Based on the above arguments, the prior probability distributions for R and σ are provided below: 1 p(r/i) = exp ( (R R p) 2 ) 2πσp 2 2σp 2, p(σ/i) = 1 σ, (12) where R p is the mean value of the prior distribution (in our case it is the previous SNR estimate) and σ p is the standard deviation of the prior distribution (in our case it is the reasonable knowledge of the variation between successive SNR values). Using (10) and (12) and absorbing all the constants into the proportionality sign (5) is rewritten as p(r/{r k }, I) exp ( (R R p) 2 ) 2σp 2 ( σ (2N+1) exp 0 Q(σ) 2σ 2 ) dσ. (13) There is no closed form solution to (13) and a numerical approximation is required. The approximation problem is being studied and the results would be published in a sequel or addendum to the present article. However to illustrate the utility of the Bayesian approach we will proceed with a simplified parameter estimation problem. To simplify the problem let us assume that the noise variance σ 2 is known. In this case we could normalize (1) by dividing the signal power by the noise variance, dispensing the latter as follows: r k = A(m i,k + jm q,k ) + (w i,k + jw q,k ), (14)

19 A. Aroumont et al. / SNR estimation algorithms for channel state estimation in Communication Satellite Systems 37 where w k is now i.i.d. Gaussian with unit variance and A is the parameter to be estimated given by A 2 = S σ 2. To compare the Bayesian approach with the ML algorithm, we will derive expressions for the estimate using both ML and Bayesian arguments. As before the posterior probability distribution of A is given by p(a/{r k }, I) = p({r k}/a, I)p(A/I). (15) p({r k }/I) 4.1. ML argument In the ML approach there is no concept of prior information, i.e., we are completely ignorant about the parameter to be estimated. In the Bayesian approach a complete ignorance about the parameter is captured by a uniform prior distribution (as A is a location parameter) [12,13]. In that case the posterior distribution given by (15) entirely depends on the likelihood function p({r k } A, I). Hence, it can be seen that the ML algorithm is a special case of the Bayesian approach when there is a complete ignorance about the parameter of interest. So, for the ML method we have p(a/{r k }, I) p({r k }/A, I). (16) As before, using (4) and the fact that the noise is i.i.d. Gaussian with unit variance, the likelihood function is given by p({r k }/A, I) ( ( = (2π) N exp 1 N (r 2 i,k Am i,k ) 2 k=1 + N (r q,k Am q,k ) )). 2 (17) k=1 Taking the logarithm of the likelihood function and finding the value of A that maximises it, we have the ML estimate as ( ) Nk=1 2 Â 2 (r i,k m i,k + r q,k m q,k ) ML = Nk=1 (m 2 i,k + m2 q,k ), (18) where ÂML is the ML estimate of the parameter A.The variance of the posterior distribution is 1/N Bayesian argument In the Bayesian approach we take into account the prior knowledge available about the parameter to be estimated. As discussed earlier a Gaussian prior distribution is assigned to the parameter A with A p as the mean and σ p as the standard deviation. Therefore, we have 1 p(a/i) = exp ( (A A p) 2 ) 2πσp 2 2σp 2. (19) Hence, the posterior distribution of A is given by (absorbing the constants into the proportionality sign) p(a/{r k }, I) ( ( exp 1 N (r 2 i,k Am i,k ) 2 k=1 )) N + (r q,k Am q,k ) 2 k=1 exp ( (A A p) 2 ) 2σp 2. (20) Now for a Gaussian posterior distribution, irrespective of the related cost function, the estimate value is always the same given by the mean, median or mode of the posterior distribution [15]. It is to be noted that for a Gaussian distribution the mean, median, and mode values are all the same. Therefore, by taking the logarithm of the posterior distribution and finding the value of A that maximizes it, we have the Bayesian estimate as ( Nk=1 (r i,k m i,k + r q,k m q,k ) + Ap ) 2 Â 2 B = σp 2 Nk=1 (m 2 i,k + m2 q,k ) + 1, σp 2 (21) where ÂB is the Bayesian estimate of the parameter A. The variance of the posterior distribution is given by var(a) = 1 N + 1. (22) σp 2 Comparing the variances of the posterior distributions obtained from the ML method and the Bayesian

20 38 A. Aroumont et al. / SNR estimation algorithms for channel state estimation in Communication Satellite Systems method, we could infer that, given reasonable prior information, the Bayesian estimator always provides a lower variance for the estimate. To compare the performances of the ML and the Bayesian estimators, simulations have been performed whose results are presented in Figs 2 5. The purpose of the simulations is to show the efficiency of the Bayesian method over the ML method. To this end a simple SNR time series has been used and the normalised mean square error obtained by both ML and the Bayesian estimators for each of these estimation instants has been plotted. For the Bayesian estimation the prior mean is taken to be the previous SNR estimate. The prior variance chosen is indicated in the figures. Figures 2 4 show that a good prior knowledge results in increased estimator efficiency which could be translated into a reduction in the number of symbols required to get an estimate with a required error margin. However, as shown in Fig. 5 when the prior knowledge is erroneous (the prior variance has chosen to be close to zero implying that the present estimate is same as the previous estimate which is not the case in the chosen time series) the Bayesian estimator performs poorly as it has been fed with wrong prior information. Fig. 2. NMSE with N = 32 and prior variance = 2dB.SNRtimeseries:[203012](indBs). Fig. 3. NMSE with N = 64 and prior variance = 2dB.SNRtimeseries:[203012](indBs).

21 A. Aroumont et al. / SNR estimation algorithms for channel state estimation in Communication Satellite Systems 39 Fig. 4. NMSE with N = 32 and prior variance = 4dB.SNRtimeseries:[203012](indBs). Fig. 5. NMSE with N = 32 and prior variance 0.SNRtimeseries:[203012](indBs). This stresses the point that the prior information has to be reasonable. The number of symbols used reflects the fact that the estimation process is a simplified one (noise power is known). In a realistic scenario the number of symbols needed might be in the range of several hundreds if not thousands. It could be reasonably assumed that the number of symbols needed by the Bayesian estimator would be less when compared to the ML estimator. The exact nature of the savings would be clear when (13) has been solved. 5. Conclusions In this paper we have analysed the suitability of contemporary SNR estimation algorithms as channel state estimators for a DVB type satellite system compliant to the DVB-S2/DVB-RCS standards. In particular, we have analysed the ML algorithm from a system point of view wherein the number of symbols available from the system is compared to the symbol requirements of the estimator in performing the estimation to a given accuracy. The impact of interference and that of rain fade slope on the estimator performance has been investigated for both forward and return link scenarios. It is evident from the analysis that the ML algorithm is perfectly suitable for the channel estimation process in the forward ink whereas in the return link the problem of interference coupled with the scarcity of symbols significantly degrades its performance. To this end we have shown the utility of prior knowledge in channel estimation through the Bayesian approach. Although the full scale Bayesian parameter estimation problem has not been solved, the simplified problem neverthe-

22 40 A. Aroumont et al. / SNR estimation algorithms for channel state estimation in Communication Satellite Systems less shows the efficacy of the Bayesian method compared to the ML method. Acknowledgement This work was carried out under the framework of the European Union funded Satellite Network of Excellence (SATNEX). References [1] S. Cioni, R. De Gaudenzi and R. Rinaldo, Adaptive coding and modulation for the forward link of broadband satellite networks, in: Proc. of the IEEE Globecom Conference, San Francisco, CA, Vol. 6, 1 5 December, 2003, pp [2] S. Cioni, R. De Gaudenzi and R. Rinaldo, Adaptive coding and modulation for the reverse link of broadband satellite networks, in: Proc. of the IEEE Globecom Conference, Dallas, TX, Vol. 2, 29 November 3 December, 2004, pp [3] D.R. Pauluzzi and N.C. Beaulieu, A comparison of SNR estimation techniques for the AWGN channel, IEEE Trans. Commun. 48(10) (2000), [4] N. Celandroni, E. Ferro and F. Potorti, Quality estimation of PSK modulated signals, IEEE Commun. Magazine July (1997), [5] W. Gappmair and O. Koudelka, Moment-based SNR estimation of signals with non-constant envelope, in: Proc. of the Advanced Satellite Mobile Systems Conference, Herrsching, Germany, May 2006, pp [6] A. Aroumont, A. Bolea-Alamanac, L. Castanet, M. Bousquet, S. Cioni and G.E. Corazza, Performance of channel quality estimation algorithms for Fade Mitigation Techniques with Ka/Q/V band Satellite Systems, in: Ninth International Workshop on Signal Processing for Space Communication, Noordwijk, The Netherlands, September [7] ETSI EN V1.1.1 ( ) Digital Video Broadcasting (DVB); Second generation framing structure, channel coding and modulation systems for Broadcasting, Interactive Services, News Gathering and other broadband satellite applications. [8] ETSI EN V1.3.1 ( ) Digital Video Broadcasting (DVB); Interaction channel for satellite distribution systems. [9] L. Castanet, A. Bolea-Alamanac and M. Bousquet, Interference and fade mitigation techniques for Ka and Q/V band satellite communication systems, in: COST Workshop, Noordwijk, The Netherlands, May [10] L. Castanet and M.V. de Kamp, Modelling the dynamic properties of the propagation channel, COST 280: Propagation impairment mitigation for millimeter wave radio systems, pm5-040, [11] ETSI TR V1.3.1 ( ) Digital Video Broadcasting (DVB); Interaction channel for Satellite Distribution Systems; Guidelines for the use of EN [12] E.T. Jaynes, Probability Theory The Logic of Science, Cambridge University Press, Cambridge, UK, [13] D.S. Sivia, Data Analysis A Bayesian Tutorial, 2nd edn, Oxford University Press, Oxford, UK, [14] H. Jeffreys, Theory of Probability, 3rd edn, Oxford University Press, Oxford, UK, [15] H.L. Van Trees, Detection, Estimation, and Modulation Theory, Part I, Wiley Interscience, New York, NY, 2001.

23 Space Communications 22 (2009) DOI /SC IOS Press New variant of the deployable ring-shaped space antenna reflector E. Medzmariashvili, Sh. Tserodze, V. Gogilashvili, A. Sarchimelia, K. Chkhikvadze, N. Siradze, N. Tsignadze, M. Sanikidze, M. Nikoladze and G. Datunashvili Georgian Institute for Space Constructions, Kostava Str. 68b, Tbilisi 0171, Republic of Georgia Abstract. Despite that theoretical, design and experimental studies connected with the creation of large deployable antenna reflectors have already been carried out for quite a long time, this area still remains of lively interest and has good application prospects. An example is the program of the European Space Agency (ESA) on the creation of a qualified variant of a 15 m deployable space offset reflector. It should be said that by the time the ESA adopted a decision on the implementation of this program, a deployable space reflector 7 m in diameter had already been created in Georgia and, in 1999, this space reflector was successfully tested aboard the orbital station Mir according to the joint Georgian Russian program. Georgian Institute for Space Constructions, continues developing new design principles for large space deployable systems (Transformable Space and Ground Structures (in Russian), A Monograph, Georgia Germany Liechtenstein, 1995). In the paper we describe the invention of an absolutely new design of a space antenna for symmetric and nonsymmetric radio telescopes, which has been developed by the team of Georgian designers headed by Professor E. Medzmariashvili. The novelty consists in that new engineering and technological effects we obtained by synthesis of ring-shaped and ribumbrella type systems. In particular, the central hinge that connects the main levers of the pantograph is arranged asymmetrically. As a result of this asymmetrical arrangement, during deployment the consoles (folded along the package longitudinal axis and connecting the adjacent sections) move in radial directions from the center to the periphery and, simultaneously, deploy in an umbrella-like manner. Keywords: Deployable reflector, transformation, radial rib, supporting ring, central interface 1. LDR general scheme This part of the report deals with structural analysis of the large deployable reflector and contains a description of the developed model. A general scheme of attachment of the large deployable reflector to the satellite is shown in Fig. 1. The connection between the satellite and the reflector is effected, through the truss type console element mounted on the ring, by means of a foldable leg. Figure 2 shows two side views (of the short and long axes) and a top view of the reflector. The stowed reflector configuration is shown in Fig. 3. The reflector structure consists of the following main parts: The central interface with radial and complementary ribs The supporting ring The system of consoles The deployment stabilization and the stiffening systems Central interface with radial and complementary ribs The central interface consists mainly of a drum and a supporting cone and is designed to attach radial ribs to the drum so that they could be wound onto it. 24 radial ribs are installed between the central interface and the supporting ring (Fig. 4). In order to decrease the radial rib mass, the peripheral part of each rib is bifurcated in the following manner: the upper long bifurcation branch forms the contour to which the reflecting mesh is attached, while the end of the lower short bifurcation branch is fixed by means of a rod to the middle of the long bifurcation. As a result, the rib end acquires a triangular configuration /09/$ IOS Press and the authors. All rights reserved

24 42 E. Medzmariashvili et al. / New variant of the deployable ring-shaped space antenna reflector Supporting ring Stiffening system A Console deployment stabilization system B-B A-A Console B B 3996 mm Mesh Ring 2800 mm Complementary ribs A Mesh Central interface Fig. 1. Reflector-to satellite attachment mm Console Supporting ring Radial rib Short branch A-A Rib 780 mm Central interface Fig. 3. Reflector transport package mm Fixed strut Deployment stabilization system Central interface B-B Fig. 2. A top view and two side views of the large deployable reflector. Long branch Fixed strut Cone Short branch Central interface Ring Drum Console Radial ribs Mesh To minimize the reflecting mesh pillow effect, radial ribs are used as supports for the system of complementary ribs which serve as transverse links. Each complementary rib is attached to two neighboring radial ribs, thereby forming additional contours for mesh attachment. The quantity of such contours determines a level of admissible systematic deviations of the reflecting surface. In the transport configuration, the radial ribs and the system of complementary ribs with the mesh attached to them are wound onto the central drum and put into the cavity of the stowed supporting ring (Fig. 4) Supporting ring assembly The supporting ring is designed to deploy the reflector, to stretch out the reflecting surface mesh by System of complementary ribs Fig. 4. Central interface with radial and complementary ribs. tensioning the system of radial ribs, and to keep them in the tensioned state throughout the reflector service life. The ring is a 24-hedral multi-component mechanism which, when stowed, forms a truncated pyramid whose small base is at the top and large base at the

25 E. Medzmariashvili et al. / New variant of the deployable ring-shaped space antenna reflector 43 bottom. During deployment, the ring, when in the intermediate-deployed configuration, forms a prismatic surface, whereas, when in the fully deployed position, it forms a truncated pyramid whose large base is at the top (Fig. 5). The faces of the polyhedron are composed of two adjacent lever systems which form a double pantograph (Fig. 6). During deployment, the plane of each face acquires a slope due to an asymmetrical arrangement of points of intersection of pantograph levers. In that case, in the deployed configuration the plane of each one of 24 faces is inclined at an angle of 41 to the pyramid base plane. The supporting ring structure is conditionally divided into eight fragments whose quantity corresponds to the quantity of electromechanical drives. A fragment consists of three cells, each of them including the main lever pair; a rhomb lever system hinge-connected with this pair and a console element (Figs 7 and 10). The neighboring cells of the ring are interconnected via mobile rollers and rigidly fixed brackets to which Stowed state (pyramid) the respective ends of the main levers and the rhomb system levers are hinge-connected. While the ring is deploying, the rollers travel along the guides which pass through the brackets but cannot travel relative to them along the roller axis. To make all the sections work simultaneously during ring deployment, each sliding roller is equipped with locker-synchronizers (Fig. 8). The synchronization mechanism consists of mutually tangent oval elements which are rigidly fixed in the Console guides One cell of the ring fragment Slide rollers Rigidly fixed bracket Fig. 7. A fragment of the ring in the intermediate deployed configuration. Intermediate state (prism) Deployed state (pyramid) Fig. 5. Three positions of the supporting ring at deployment. A The upper Synchronizer support A Main lever pair Console h h 1 2 Cross-section in stowed state A-A Cross-section in deployedstate 41 View B View A A Pyramid base plane Rhomb lever system B The lower synchronizer support Fig. 6. A ring cell. Fig. 8. The upper and lower locker-synchronizers.

26 44 E. Medzmariashvili et al. / New variant of the deployable ring-shaped space antenna reflector end parts of each main lever. Oval elements are hinged by means of a bracket to the sliding rollers and impart geometrical stability to the system. The system of rhombs not only contributes to increasing ring deployment stability, but also locks the ring in the deployed position. For this, the upper and lower points of the rhombs are provided with racks along which telescopic tubes with a ratchet-and-pawl gear mechanism travel while the ring is deploying (Figs 7 and 9). Ratchet-and-pawl gear mechanisms may perform free movement only towards the point of intersection of the main lever pair, while their travel in the opposite direction is locked. Deployment of the ring is carried out by the deployment mechanism (RDM) (Fig. 10). The RDM consists of an electric motor with a reduction gear, a rope system with guide rollers, and a compensation spring with a power sensor. Electric motors with reduction gears are installed on the lower end of every third guide. Altogether, on the ring there are 8 drives which work in parallel but independently of each other. The main requirement for the ring having the form of a truncated pyramid is its high stability. In the existing ring reflectors, the loss of ring shape stability increases because of the presence of consoles Rack Telescopic tube Pawl Fig. 9. Telescopic tube with a ratchet-and-pawl gear mechanism. to which the peripheral elliptic contour of the reflecting mesh is attached. Under the action of this contour, in the consoles there arise bending forces which in turn produce torsional forces in the ring. Due to the long arms of the consoles, torsional moments arising at console-ring junctions have high values. All this leads to additional deformation and enhances the risk of ring stability loss. We can get rid of this negative effect if we increase the cross-section area of the structure elements, but this will lead to an increase of the reflector mass. In our new variant we have succeeded in eliminating the above-mentioned disadvantages Consoles Folding consoles are installed along the ring contour (Fig. 11) and, when unfolded, they form a supporting frame for the peripheral part of the reflector, which is approximately 45% of the total reflector surface. Structurally, a folding console is a light-weight truss of triangular cross-section, one of whose rods is used as a guide along which the sliding rollers of the ring pantograph travel (Fig. 12). To the lower end of the console truss is hinge-connected the lower (short) bifurcation branch of a radial rib, while the upper (long) bifurcation branch is connected to the respective place on the truss. To the peripheral part of the console is fixed a narrow plate whose upper contour continues the upper contour of the respective radial rib. The upper contours of the main rib and the plate form a single paraboloid generatrix. Thus each console element combines two main functions, namely: its upper part completes the con- Fig. 10. A ring fragment and the deployment mechanism. Fig. 11. The system of consoles (for clearness, only one rib attachment is shown).

27 E. Medzmariashvili et al. / New variant of the deployable ring-shaped space antenna reflector 45 Central interface Truss type console Stiffening system A-A A A Fig. 12. A truss type console. tour of a radial rib and forms an elliptic contour of the reflecting surface, while its lower part serves as a guide for ring hinges, thereby providing (together with locker-synchronizers) geometrical stability and synchronous deployment of the ring The deployment stabilization and the stiffening systems The functions of deployment stabilization and the stiffening systems partially overlap each other. The deployment stabilization mechanism is designed to prevent the ring from rotation during deployment of ribs. The deployment stabilization system consists of three rods which are arranged along the package longitudinal axis in the stowed state, but, as soon as deployment begins, they pass into a radial position at an angle of 120 to each other (Figs 13 and 14; radial ribs are not shown). The stiffening system increases torsional stiffness of the whole system and consists of three pairs of ropes whose one ends are fixed to the lower end of the console element, while the other ends are fastened to the central interface at the opposite points so that ropes can be wound onto special drums (Fig. 13). After the reflector is deployed, the ropes together with the central interface form a triangular stiff system located in the plane perpendicular to the reflector axis. 2. Deployment process analysis 2.1. Mathematical model of the reflector deployment process Based on the proposed structural configuration, a mathematical model of the reflector deployment Deployment stabilization system Fig. 13. The deployment stabilization and the stiffening systems. Stowed state Central interface Intermediate deployed configuration Console Radial rib Deployment stabilization system Fig. 14. Ring rotation hold-down system. process has been developed so as to perform calculations for the stress-strained state of reflector structure elements during deployment both in the ideal case and in the case of various emergencies. In the design model we use the Cartesian coordinate system whose origin lies at the geometrical center of the reflector. The X- and the Y -axes lie in the ring plane, the Z-axis is directed along the rotation axis of the drum onto which ribs are wound (Fig. 15). In developing the reflector model, we made the following assumptions: Pantograph levers and other rod elements of the structure are considered as elementary rods connecting the respective structure components

28 46 E. Medzmariashvili et al. / New variant of the deployable ring-shaped space antenna reflector Fig. 15. The global coordinate system. working in compression, tension and bending (torsion is neglected). The reflector center is assumed to be a fixed point at which the drum with ribs wound on it is located (rib thickness is neglected). The total structure mass is assumed to be concentrated at the design nodes. With these assumptions taken into account, we developed a nonlinear finite-element model of structure deployment (Figs 16, 17 and 18), where the reflector model has the following components: A ring consisting of a double pantograph whose levers consist of parts of different length. Ribs each of which is approximated by successively connected rod elements arranged along the upper and lower rib edges. A fixed strut connecting the upper and lower rib edges. A ring rod and three guides, which prevent the ring from rotation. The model consists of 898 nodes and 4096 elements. Differential equations of motion are derived and a difference scheme is written for them, which is solved by means of step-by-step integration. As a result, the timedependent trajectories and coordinates of all reflector nodes are obtained and thus we can determine forces acting in the elements and to obtain animation. Fig. 16. The mathematical deployment model. Fig. 17. The mathematical deployment model (showing a ring fragment and one rib) Conclusions of deployment process analysis The results of deployment process analysis are shown given in Figs The first 4 5 seconds the self-deployment process occurs at the expense of bending stiffness of ribs (stage I) after that, the motors get into operation and impart acceleration to the system (stage II) the system acts quicker than the rope gets wound and hence Fig. 18. The full deployment stage; for clearness, only three ribs are shown.

29 E. Medzmariashvili et al. / New variant of the deployable ring-shaped space antenna reflector 47 Fig. 19. The deployment radius of upper and lower points of the supporting ring as a function of time. Fig. 22. The deployment stages of a rib (a top view). Fig. 20. Rib tensioning as a function of time. Fig. 23. The reflector deployment stages; for clearness, only three ribs are shown. Fig. 21. Transverse force acting on one of the three guides as a function of time. the cable sags so that for some time the structure deploys by inertia while the motors haul in the sagging (stage III) after some time, the system gets gradually decelerated by friction, and the motors again get into operation and operate at a sufficiently small power (stage IV) the last (stage V) is the tensioning of ribs, which, simultaneously, are put in a vertical position. Fig. 24. The rib part jammed on the drum. The model envisages various kinds of emergency situations that might occur during deployment, for instance: a rib part might get stuck on the drum Fig. 24, two or more ribs may get jammed and so on.

30 48 E. Medzmariashvili et al. / New variant of the deployable ring-shaped space antenna reflector References [1] W. Beitz and K.-H. Küttner, Dubbel: Handbook of Mechanical Engineering, Springer-Verlag, London, [2] E. Medzmariashvili, Transformable Space and Ground Structures (in Russian), A Monograph, Georgia Germany Liechtenstein, [3] E. Medzmariasvili, Sh. Tserodze, L. Datashvili, K. Chkhikvadze and A. Sarchimelia, Analytical investigation of the surface geometry of an offset reflector, Problems of Applied Mechanics 4(5) (2001),

31 Space Communications 22 (2009) DOI /SC IOS Press Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks Eleni Noussi, Boris Grémont and Misha Filip Microwave Telecommunication Systems Research Group, University of Portsmouth, Department of Electronic and Computer Engineering, Anglesea Road, Portsmouth, PO1 3DJ, UK Tel.: , Fax: s: {boris.gremont, Abstract. This paper addresses the integration of a Fade Mitigation Technique (FMT) within a centrally managed MF-TDMA satellite network in the context of guaranteed QoS delivery. Burst Length Control (BLC) is introduced as a suitable FMT and the methodology of its design and implementation within the existing DVB-RCS standard is described and simulated. The proposed technique provides compensation for rain attenuation at the expense of capacity utilisation. Within any fixed partition of the MF-TDMA space, by considering the space-time properties of rain attenuation, suitable geographical multiplexing of rainattenuated and non-attenuated links allows the optimisation of the channel capacity utilisation while still exploiting the statistical multiplexing of the sources on the return channels. Keywords: Burst Length Control, Bandwidth on Demand, Quality of Service, rainfield attenuation, resource management, geographical multiplexing, link availability 1. Introduction A consequence of the need to accommodate higher network capacities is to migrate to higher frequency bands namely Ka band (27 40 GHz) and V band (40 75 GHz). This trend is also justified by the relatively large segments of frequency spectrum required for supporting the high data rates required to deliver advanced multi-class multimedia services [16]. A major drawback is significant rain attenuation that increases rapidly with increasing microwave frequency [17,26,32]. Rain can cause serious signal quality degradations on individual earth-space communication links and this situation will have a major impact on the link availability and the effective throughput of a network, hence the implementation of Fade Countermeasures becomes necessary and is of great importance to system designers. Studies in the past have developed novel Medium Access Control protocols for Multi-Frequency TDMA satellite channels [18,24,25,33], and dynamic resource allocation algorithms in an MF-TDMA basis for DVB-RCS [21], to maximize system throughput. Pech et al. [29], have conducted one of the very few studies that have taken into account fading conditions. Although the dynamic resource allocation in MF-TDMA systems is of great interest, most studies have not considered the implementation of satellite Fade Mitigation Techniques, necessary for such services and even when FMTs were introduced in resource management protocols, work focused on a single link analysis, lacking a global approach that would consider a large number of links simultaneously affected by rain. Celandroni et al. have described the concepts and implementation details of a complete fade countermeasure system based on adaptive TDMA, with demand assignment of capacity [7]. Association of an up-link power control feature with the bit and coding rate variation gives the system an interesting ability to cope with fade conditions [3 6,8]. These studies are of great significance as an adaptation of such modelling to the case of MF-TDMA for DVB-RCS is actually essential. Figure 1 presents the basic DVB-RCS network architecture [11]. It consists of the following elements: (i) GEO Ka-Band satellite payload, using the /09/$ IOS Press and the authors. All rights reserved

32 14 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks Fig. 1. DVB-RCS network architecture. 20/30 GHz band for earth-space communications. We will focus on a bent-pipe system. (ii) The Return Channel Satellite Terminals (RCSTs) can frequency hop across the whole MF-TDMA space and their burst transmission rate is Mbps. (iii) The Network Control Centre (NCC) manages in real-time the communications of the whole MF-TDMA satellite system. MF-TDMA allows a group of Return Channel Satellite Terminals to communicate with a Gateway using a set of carrier frequencies. In order to guarantee a Quality of Service, it is important that the NCC gets to know the actual traffic needs of each of the active RCSTs of the network. Therefore each user station needs to monitor and/or estimate its specific traffic requirements that are then communicated to the NCC. If resources are available, the NCC will generate a new Terminal Burst Time Plan (TBTP) at a superframe/frame level accommodating the needs of all its active stations, so that each RCST learns what timeslots have been assigned to it. Although originally designed to provide bandwidth on demand, this mechanism can also be used to drive a Fade Mitigation Technique. Allocated traffic can be varied not only depending on the RCSTs traffic requirement, but it can also be used to carry extra time slots for the purposes of fade mitigation when a link is under rain conditions [14]. This paper proposes the integration of a Fade Mitigation Technique that is particularly suitable for power limited systems. It is simple to implement within a centrally managed MF-TDMA satellite network and able to provide compensation for rain attenuation within a large dynamic range, at the expense of capacity utilisation. A resource allocation algorithm is applied to a simple MF-TDMA scenario employing suitable multiplexing of rain-uncorrelated traffic streams. Section 2 introduces Burst Length Control as a Rain Fade Countermeasure. Some background on Bandwidth-on-Demand over DVB-RCS is provided in Section 3. Details regarding the integration of the proposed FMT within DVB-RCS and the simulation model are included in Section 4. Performance results are reported in Section 5. Section 6 concludes the paper. 2. Burst length control as a rain fade countermeasure This section will introduce the principles of the proposed FMT. Considering an adaptive MF-TDMA system with a built-in fade countermeasure, a variable portion of MF-TDMA burst plan will be allocated to carry additional FMT timeslots whose role is to compensate for the attenuation suffered by a set of links undergoing rain. In particular, when a traffic burst is subject to fading, the burst will be allocated extra time slots in which the original traffic burst can be expanded. This time expansion results in an increase in average power (or energy/bit) of the signal and so it counteracts the effect of the fade [10]. If the duration of S i traffic slots (emanating from one single source labelled i) is spread by a factor H i

33 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks 15 Fig. 2. Example of (i) contiguous bit (ii) data block repetition. then the number of extra FMT slots required by the ith return connection is: N i FMT = S i (H i 1) FMT slots. (1) By using spreading, the total power margin M i is increased to: M i = M i log 10 H i db, (2) where M0 i denotes the clear-sky power margin while the second term is the extra power gain brought by the FMT. We note that when we spread by an additional factor of two the FMT produces an increase in average transmitted power of 3 db. This technique known as Burst Length Control (BLC) was described in [1]. BLC can be implemented in two ways. For the first approach, the duration of the transmitted bits and corresponding symbols can be increased by a factor of H i (adaptive modem scenario), or equivalently, each bit can be played H i times and passed consecutively to the modulator (operating at fixed symbol rate). The second approach is to contiguously repeat packets of traffic slots by the spreading factor H i asshowninfig.2. Comparison between the two techniques showed that the first technique is less greedy in required FMT slots whereas the second approach only offers power gains for odd numbers of block bit repetition. In a previous study of a typical Ka band return channel (see [27] for details), QPSK with rate 7/8 convolutional coding was chosen as the system baseline (clear-sky) modulation/coding scheme providing maximum user throughput at a BER better than This system leaves a positive clear-sky link margin of M0 i = 2.73 db with a clear-sky CNRi 0 of db Hz. In order to deploy in real-time the BLC mitigation technique, the system needs to determine the required spreading factor H i for each active connection. This in turn will depend on the combined impact of up- and down-link attenuation on the overall CNR on the return link. This will also depend on the clear-sky CNR i 0 and the desired/agreed maximum allowed BER of the return link, i.e., in general: H i = f(ber max, CNR) = g(a i, BER max ), (3) where A i CNR 0 CNR i denotes the total drop in CNR at the Hub/NCC station for link i with respect to clear-sky conditions and H i 1. The total number of FMT slots in terms of A i is NFMT i = S i [g(a i, BER req ) 1]. Combining Eqs (2) and (3), we can express the power fade margin as a function of the drop in CNR. This has been evaluated in [27] and is shown in Fig. 3 with S i set to unity. Therefore Fig. 3 shows the number of required FMT slots per traffic slot for a typical return channel. It shows that every time the burst duration is doubled in time, the FMT offers an extra 3-dB attenuation protection. For example, spreading by a factor of 2 (i.e., using 1 FMT slot per traffic timeslot) will make the system capable of coping with attenuations that are up to 3 db higher than the fixed margin of the system; a spreading factor of 4 (i.e., 3 FMT slots per traffic timeslot) will provide with an additional 3 db attenuation protection and so on. We call H i the BLC spread factor or equivalently the BLC multiplication factor. The latter term emphasises that H i will increase the traffic demand of the source by a factor H i. We will see later how increasing the capacity requests for the RSCT affected by rain can be achieved within DVB-RCS. Finally, the availability of a typical return link, A % i, needs to be achieved by the BLC mitigation technique. In our context, we need to make sure that the range of the FMT multiplication is

34 16 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks Fig. 3. Number of required FMT slots per traffic slot vs. attenuation protection. large enough to cope for the long-term drop in overall CNR on the return channel. As we will see later on, the maximum BLC multiplication factor offered by the system under consideration is 16, i.e., from Fig. 3, the FMT will be able to cope for attenuations of up to 15 db (15 FMT slots per traffic slot). This will result in a 99.92% link availability (providing 1.92% improvement, when compared with the 98% availability achieved without BLC). Before proceeding to implementation details of the proposed technique, the following section will provide us with some background on DVB-RCS and the builtin mechanisms that will be used to drive the proposed FMT. 3. Bandwidth on Demand over DVB-RCS background [11] Most new generation Ka-band satellite systems are designed to provide low-cost telecommunication services to hundreds of users. In order to maximise the system capacity, frequencies and time slots can be allocated dynamically so as to exploit statistical multiplexing of the sources. Bandwidth on Demand (BoD) can be employed to support the maximum amount of users possible. On the billing side BoD may enable users to pay only for the capacity they utilise [30]. The timeslots of the return link are organised and numbered so that the network is able to allocate them to individual active RCSTs. Figure 4 shows how the global return link capacity may be segmented amongst a group of RCSTs; the network will then manage several superframe identifiers, SF_IDs, (i.e., separate sets of carrier frequencies). Without loss of generality, it is assumed that these frequency sets are fixed, i.e., superframes/frames have a fixed bandwidth (otherwise a change in a superframe s bandwidth would affect the whole arrangement and all RCST groups would then have to be notified and have their superframe bandwidth adjusted appropriately via additional service signalling). For each superframe of a given SF_ID, allocation of timeslots is communicated to the RCSTs via the Terminal Burst Time Plan table. An RCST is thereon allowed to transmit data bursts in the timeslots that have been allocated to it. As shown in Fig. 4, the consecutive superframes of a given SF_ID are time-contiguous. Each occurrence of a superframe is labelled with a number called

35 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks 17 Fig. 4. Capacity segmentation and timeslot allocation: an example. SF_counter. These two superframes of this particular SF_ID 1 are of the same composition and duration, unless a notification that a change should be applied is provided; this can occur at the boundary between two superframes (between two consecutive SF_counters of the SF_ID), so that the next superframe is of different composition and/or duration. This notification involves the TBTP and the Superframe, Frame and Timeslot Composition Tables (SCT, FCT, TCT). The DVB-RCS standard specifies that these service information tables will be transmitted at least every 10 seconds. Alternatively the tables can be transmitted synchronously at the exact beginning of every superframe. In fact, if required, the TBTP can even be updated at the beginning of every frame. These tables can therefore be transmitted as frequently as required to provide detailed network information, transmission parameters and timeslot properties (such as symbol rate, code rate, payload content). This is of importance as it is directly related to the achievable update rates when deploying FMTs for the compensation of rain attenuation. A superframe is composed of frames, themselves composed of timeslots. In a superframe, frames are numbered from 0 (lowest frequency, first in time) to N (highest frequency, last in time), ordered in time then in frequency (F_nb represents the frame number) and can take the values 0 N 31. A frame is composed of timeslots that may span over several carrier frequencies. In a frame, timeslots are numbered from 0 (lowest frequency, first in time) to M (highest frequency, last in time), ordered in time then in frequency. The number of slots in a frame can be in the range 0 M 2047 [11]. In our scenario, the bandwidth and duration of superframes, frames and timeslots are fixed providing burst data rate of 2048 kbps and a typical MF-TDMA frame lasts for 24 ms. The frame includes 4 carriers of 128 traffic slots each (i.e., 512 slots per frame) with rate granularity of 16 kbps (i.e., the minimum rate that can be achieved over the period of a frame by the allocation of just one timeslot). Due to the system s BoD capability a user station can vary the average bit rate over each frame according to its traffic needs by being allocated 1, 2, 3,... up to 128 slots/frame leading to average bit rates of 16, 32, 48,... up to 2048 kbps. Each terminal may only transmit on any single frequency (or channel) at a given time; it is not allowed to transmit data on more than one carrier at a time in order to minimise the power output requirement and reduce the hardware complexity of terminals. The RCST processes the TBTP message received from the NCC, to extract the assignment count and timeslot allocations for its next uplink transmissions. The synchronisation burst and the optional prefix attached to S-ATM traffic bursts contain the Satellite Access Control (SAC) field composed of signalling information added by the RCSTs for the purpose of requesting capacity for the session. The Request subfield within the SAC field accommodates the capacity requests from the terminals. There are two main mechanisms available to the RCST for issuing data traffic requests to the NCC (see

36 18 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks Fig. 5. Traffic request SAC Fields for OBR and IBR. Fig. 5). The mini-slot Out-of-Band Request (OBR) is a small burst that can be periodically assigned to loggedon RCSTs. It carries control and management information from the RCSTs to the NCC and it is also used for maintaining RCST synchronisation. As an alternative to the round-robin approach, OBR may also be used in contention-based mode. The other method called the Prefix or In-Band Request (IBR) is based on an optional prefix attached to ATM traffic bursts. The prefix carries control and management sub-field information (see Fig. 5). Obviously, the use of IBR signalling allows for the reduction in the traffic load over the OBR reservation channels as IBR is readily available. IBR has in fact the great advantage of being collision-free. Hence a terminal that has been assigned traffic slot(s) should preferably use IBR. In this way the OBR slots can be shared by terminals that do not yet have any assigned traffic slot. The waiting time to send a reservation will also be reduced due to the availability of both IBR and OBR slots [23]. 4. Integration of FMTs within DVB-RCS Here, the integration of the proposed FMT within the existing DVB-RCS standard will be described and details of the model used to simulate it will be discussed BLC concept The overall proposed architecture for DVB-RCS with Burst Length Control as a rain fade mitigation technique is depicted in Fig. 6. The traffic bursts from the RCSTs are independently monitored by the NCC so that it can determine (see Eq. (3)) the required BLC multiplication factor H i.for this, the NCC measures the received SNR from all the bursts of all active RCSTs SNR estimator The operation can be achieved using a non-data aided SNR estimation technique at the NCC. To improve further the estimation, estimates from multiple bursts for the same connection can be averaged together to reduce the estimation error. One main concern is that the amount of traffic (average number of symbols per unit time) from a RSCT can be very sporadic and small which would have an impact on the accuracy of the burst SNR estimator. We have therefore simulated the ML-RxDA estimator applied to QPSK

37 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks 19 Fig. 6. Architecture of BLC DVB-RCS. Fig. 7. Estimated and true SNR. described in [28] (Maximum-Likelihood Data-Aided estimator using an estimate of the transmitted data from receiver decisions) with the addition of a correction for the bias of the estimator at low SNRs. As a worst case we have assumed that only one traffic slot is processed every second. Figure 7 shows the simulation results. For a typical Ka band 15 db event, the estimated SNR remains within 1 db of the true SNR. This shows the appropriateness of the technique even for very low traffic Simulation model Depending on the current need of the source, the RCST issues normal traffic requests to the NCC via

38 20 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks either the IBR or OBR signalling channels. Upon receiving the capacity requests from each active connection, the DAMA then evaluates the total number of required slots by multiplying each traffic request S i by the BLC multiplication factor H i that has been determined from the SNR estimator subsystem. The DAMA then runs its resource allocation algorithm to generate the new TBTP table. This is then broadcasted on the forward channel so that each RCST can switch to its new frequency/time positions within the next MF- TDMA space including the fade mitigation technique. Note that BLC does not require any additional specific signalling. The technique is therefore efficient in terms of spectrum utilisation. Another important point is that, provided that the request can be serviced, the RCST can work out what BLC spreading factor should apply, without requiring extra signalling, using the following procedures: (i) the RCST issues a request to the NCC for S i traffic slots. This request is processed by the NCC that applies the BLC multiplication factor from the measured SNR. If the DAMA algorithm can service the request, it allocates S tot = S i H i traffic slots to the RCST. By comparing S i and S tot, the RCST can calculate the required BLC expansion factor it has to use over the next allocation period. However, the RCST will not be able to calculate the BLC spreading factor itself in all cases. If the request cannot be immediately serviced and some of it needs to be queued, the RCST will not be able to work out the BLC factor. In such cases, the timeslot type (i.e., payload content) explicitly defined for each timeslot in the TCT sent to the terminals will help the station know exactly how many normal traffic and FMT slots have been assigned to it. We assume that the update rate of the FMT is one second. This update rate is very important when integrating an FMT within DVB-RCS, as it must be able to track the variations of the rain attenuation process. Therefore the source traffic will have to be queued for one second at the source. The source will emit a traffic request and the new allocation will be implemented in the next FMT interval (1 second later). An important issue is that the MF-TDMA should be able to support the active connections including their FMT needs. If the MF-TDMA capacity allocated to flows of the same class is limited and fixed the DAMA algorithm may not be able to grant all the requests. If the agreed QoS or service level agreement allows it, some connections may be blocked until resources become available. More likely the requests may be queued, in which case, the limited capacity in the presence of rain will result in possible delays. To control this delay, it is important to make sure the sources do not generate too much traffic (including FMT). We consider as a baseline case that N connections with peak traffic rate of 128 kbps share a subspace of the entire MF-TDMA space. In order to show how the system can be dimensioned, we assume that the DAMA controller will group N traffic requests so that only one of those requires full FMT support while all other connections operate at maximum bit rate. We consider that each connection transmits at a peak rate much lower that the burst rate of the MF-TDMA system. In our system, if they operated at burst data rate of 2 Mbps, they would require 5333 traffic slots per second. If the peak rate of each terminal is reduced to 128 kbps, they only need 333 slots per second. Furthermore if one of these terminals is attenuated and requires maximum FMT protection it will need 16 times as many slots (i.e., traffic and FMT slots) therefore a fully attenuated return link would require 5328 slots per second. We note that this single attenuated link would occupy a very large portion of one MF-TDMA channel (5333 slots/s). This simplifies greatly the DAMA algorithm and means that our FMT is introduced with no extra queuing since the traffic plus FMT can be supported within one TBTP allocation period. Assuming that we also multiplex 16 return links in clear-sky conditions with one link experiencing maximum rain attenuation, the total traffic of this network would be slots/s. Bearing in mind that any MF- TDMA channel can only have 5333 slots per second, this means that we could fit our 17 links within two MF-TDMA frequency channels (half of a frame width which has four channels). Obviously, when the faded link returns to clear-sky conditions, much capacity is freed since no FMT slots are required and, assuming all connections are rain free, we can support at least 32 individual 128 kbps connections in our two MF-TDMA channels. In this simplified scenario the DAMA algorithm should (i) partition the MF-TDMA in areas of width of two MF-TDMA channels. Multiplex different traffic sources by (ii) choosing one request requiring FMT support and (iii) multiplexing it with a variable number of unaffected return links so as to occupy as best as possible the two MF-TDMA channels of the subspace. The described FMT offers a guaranteed quality of service for all served stations. The parameters chosen to be monitored in this work were the BER (as an indirect way of measuring quality of service), the delay of communication, and the link availability. As it

39 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks 21 will be seen, outage will only occur when the dynamic range of the FMT is exceeded. Additional connections may be brought in whenever rain conditions over the footprint are good enough or are improving. By partitioning the entire MF-TDMA space accordingly and applying appropriate multiplexing (see (i) (iii) above), the proposed FMT can be applied to a large number of connections. However, as this assignment strategy is clearly at the price of a variable number of supportable connections and perfect statistical multiplexing may not always be possible for the entire network, ultimately, some connections may need to be queued or dropped, which will be acceptable as long as the QoS or service level agreement allows it. Figure 8 shows an example of rain cells over areas with active and nonactive users inside the coverage area of a satellite. It becomes apparent that the total number of required FMT slots depends on: (i) users density and location (the greater the concentration of users the greater the number of required FMT slots when it rains); (ii) space and time characteristics of rain (for rain over large areas, more FMT slots are required); (iii) magnitude of the rain attenuation; (iv) actual traffic characteristics of the user stations; (v) agreed QoS parameters. Furthermore, depending on the hour of the day or the month, the rain conditions may be quite severe resulting in a much more extensive need of FMT slots. This would be detrimental if severe fades occurred at times when user traffic is high. In order to study the impact of correlated rain attenuation on the satellite footprint, we have developed a rainfield stochastic simulator described in the Appendix. A typical snapshot of the synthesised rainfield is shown in Fig. 9. In this particular case, there is a raincell in the lower right quadrant with a small peak attenuation of about 2.5 db. Obviously the DAMA algorithm should avoid multiplexing together requests emanating from rainy regions because this would generate a large amount of total traffic (TRF and FMT slots). A uniform density of users has been assumed, so that suitable raindecorrelated traffic streams can be (geographically) multiplexed. The following section will present the performance results of the algorithm. 5. FMT resource management performance results We focus on the transport of aggregate Internet traffic that is known to exhibit self-similarity with longrange dependence and high levels of burstiness [2,9, 19,20,22,31,34]. In our simulations we assumed one connection per terminal, each generating a stream of traffic ATM cells with a Pareto distribution with Peak Cell Rate equal to 128 kbps. The shape parameter of the distribution, α, wassetto1.2. Fig. 8. Active and non-active users in the footprint in the presence of rain.

40 22 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks Fig. 9. Example of synthesised rainfield. For the resource allocation phase, a bin-packing algorithm (first fit heaviest to lightest) has been implemented. This has been applied to all the traffic requests which are cumulated over periods of 1 second. As mentioned above, only one out of N stations experiences rain. Figure 10 shows the variations of the carrier-tonoise ratio for the rain-affected 30/20 GHz return link. The link is attenuated by rain for about 10 minutes and the CNR drop slightly exceeds 16 db. Figure 10 also compares the variable BER of the link with and without FMT. Only by employing the FMT, the link can maintain the BER below the maximum BER threshold of 10 8 for most of the time. However when the total drop in CNR exceeds the 15 db dynamic range of the FMT (see Fig. 3), the BER grows to a large value. This corresponds to an outage that will ultimately contribute towards the long-term unavailability of the return link. Figure 11 shows the corresponding number of FMT slots per normal traffic slot for the link undergoing rain attenuation. This shows results including the DAMA allocation process. We see that at time 1360 seconds, the FMT saturates to its maximum value. The FMT is not able to offer enough protection and hence the link enters an outage state. We see that for most of the time, the link does not need any fade mitigation. This implies that channel capacity should be used by other links as long as they can fit within the available total bandwidth. As it has been discussed earlier, during our simulations, the DAMA controller grouped traffic requests so that only one required FMT support. Figure 12 illustrates the MF-TDMA subspace occupancy during this multiplexing. As it can be realised, even though one of the links suffered from rain during some period of time (around t = 1360 s) the throughput utilisation remained pretty much unaffected and was generally kept at an average of 99.9%. The number of multiplexed connections within our subspace of two channels is illustrated in Fig. 13. This number varies significantly in time in order to achieve the best possible utilisation. In fact, this number was impressively high since not all sources happened to transmit at their peak rates simultaneously as it had

41 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks 23 Fig. 10. Variations of CNR and BER (with and without FMT) during a rain event.

42 24 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks Fig. 11. Variations of the number of FMT slots per traffic slot during rain event for one return link. been assumed in the previous worst case calculations that considered maximum transmission rates for all sources as well as a perfect allocation process. This is of great significance and it clearly demonstrates the advantages of statistical multiplexing even in the case of a rainy link. It should be repeated that this mechanism ensures that all connections are served every TBTP allocation period without the introduction of any additional delay. 6. Conclusions This paper has discussed the integration of a Fade Mitigation Technique within a centrally managed MF- TDMA satellite network in the context of guaranteed QoS delivery. The objective is to determine a way of achieving the integration without requiring any modification of the existing DVB-RCS standard. Burst Length Control is proposed as a suitable FMT. This technique is relatively simple to implement and it provides a large dynamic range that can be arbitrarily chosen depending on requirements. This is not the case of other techniques like adaptive coding/modulation where the dynamic range is limited to a maximum value. For BLC, outage will only occur when the dynamic range of the FMT is exceeded and by limiting the capacity of individual connections, BLC can be introduced without inducing any additional delay. Burst Length Control provides compensation for rain attenuation at the expense of useful traffic throughput. This is encapsulated by the derivation of the required number of FMT time slots as a function of the total attenuation on the return link. The appropriate ex-

43 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks 25 Fig. 12. MF-TDMA subspace utilisation during multiplexing of one affected link with a variable number of unaffected links. isting mechanisms defined in the ETSI standard were extracted and examined in order to achieve the sought integration of FMTs within DVB-RCS. We indicate that in-band requests (IBR) and possibly out-of-band requests (OBR) can be used for issuing normal traffic requests. In this paper, we have considered that the burst time plan has an update rate of 1 second as this is sufficient to track the variations of rain attenuation. The Network Control Centre then centrally detects the fades by applying the ML-RxDA SNR estimation algorithm with a bias correction to all incoming bursts. By simulation, this detector was shown to achieve sufficient accuracy despite the possibly sporadic and scarce traffic from the remote terminals. The measured SNRs are then used by the Network Control Centre to calculate the BLC multiplication factor that is applied to the IBR/OBR traffic requests by the NCC prior to running the resource allocation algorithm. The DAMA resource allocation algorithm was applied to a simple MF-TDMA scenario for which geographical multiplexing has been applied. In particular, the DAMA algorithm groups one rain-attenuated link with a variable number of un-attenuated links so as to maximise the channel capacity within a fixed subspace/partition of the MF-TDMA channel. This requires the dynamic selection of appropriate links based on the outputs of a space-time rain attenuation field simulator. This selection can be performed without knowing the actual locations of the RCSTs simply by sorting the FMT multiplied traffic order from largest to smallest. We have finally quantified the achieved channel capacity utilisation and the variable number of 128 kbps connections that can be serviced within the MF- TDMA subspace. Our results indicate that an average utilisation of 99.9% can be achieved while the av-

44 26 E. Noussi et al. / Integration of Fade Mitigation within centrally managed MF-TDMA/DVB-RCS networks Fig. 13. Connections multiplexed in time to occupy the available MF-TDMA bandwidth. erage number of connections is around 55. The proposed FMT offers a guaranteed quality of service for all served stations. The BER, the delay of communication and the link availability have been monitored; as it was seen, by limiting the capacity of individual connections, BLC was introduced without inducing any additional queuing delay and both traffic and FMT were supported within one TBTP allocation period. As long as the dynamic range of the FMT is not exceeded, the BER will be kept below the maximum threshold and no outage will occur, guaranteeing the long-term availability of the link. Current work includes extension of the algorithm into a multi-class scenario with differentiated services involving traffic streams with different QoS levels (for example different delay tolerance). The basis is the same but further investigation is required to control the increased queuing delays induced due to differing priorities and limited capacity allocated to flows of different classes, especially in the presence of rain. More detailed simulations and analysis will be required to investigate/solve issues regarding the support of multi-class services, the impact of BLC on Quality of Service, Connection Admission Control and Service Level Agreements. The full details and implications on QoS, CAC and SLA will clearly require further investigations into alternative assignment strategies. This will require a detailed analysis of the whole MF-TDMA system in the presence of rainfield attenuation. Future work will involve a detailed study of the impact of correlated rain attenuation on the satellite footprint. Acknowledgements The authors would like to thank the reviewers for their insightful and useful comments.