Received November 15, 2018, accepted December 7, 2018, date of publication December 20, 2018, date of current version January 11, 2019. Digital Object Identifier 10.1109/ACCESS.2018.2887167 MF-TDMA Scheduling Algorithm for Multi-Spot Beam Satellite Systems Based on Co-Channel Interference Evaluation JUAN-MANUEL RODRIGUEZ BEJARANO 1,2 , CARLOS MIGUEL NIETO1 , AND FRANCISCO JAVIER RUIZ PIÑAR1 , (Member, IEEE) 1 Escuela 2 Thales Técnica Superior de Ingenieros de Telecomunicación, Universidad Politécnica de Madrid, 28040 Madrid, Spain Alenia Space, 28760 Madrid, Spain Corresponding author: Juan-Manuel Rodriguez Bejarano (jmrb@alumnos.upm.es) ABSTRACT Very high throughput satellites (VHTS) are the new trend in satellite communications systems in order to satisfy the increasing demand for both higher throughput and ubiquitous connectivity. Implementing an IP interactive network over a VHTS would enable the satellite to compete with terrestrial networks by offering ubiquitous added value services for end-user traffic. Even so, the simulations presented in this paper show that a 30% of throughput can be lost due to co-channel interferences. In existing interactive satellite networks such as REDSAT or AmerHis, the return uplink transmissions are inherently benefited from the MF-TDMA scheduling to mitigate the co-channel interferences present in VHTS. In such networks dynamic resource assignation (DAMA) is executed at each SuperFrame period (<100 ms). Therefore, any feasible optimization on the scheduling shall be performed in real time. In this paper, two parallel problems are faced at the same time: co-channel interference mitigation and real time scheduling. In particular, two novel interference-aware scheduling algorithms for a DVB-RCS2 return uplink are presented. Despite other previous works, the proposed algorithms are designed on the basis to be executed in real time on a fixed SuperFrame period basis. The performance assessment of the proposed algorithms is carried out through system-level simulations based on realistic satellite scenarios and interactive traffic models using DAMA in real time. INDEX TERMS MF-TDMA, satellite, DVB, ICIC, resource scheduling. I. INTRODUCTION VHTS systems are aimed at satisfying the increasing demand of higher throughput while decreasing the cost per transmitted bit. In that sense, multi-spot beam satellites can meet the capacity demands through highly effective resource reuse, dividing the terrestrial coverage in a multi-spot beam network, taking the example of terrestrial LTE networks. A cellular spatial plan can be used in order to provide wide radio coverage, using different spot beams managed by different satellite antennas. The total satellite bandwidth is divided in several channels with equally-large bandwidths that are used in non-adjacent cells. It is usual to talk about colors when a particular bandwidth of a given polarization is used, so every spot is assigned a unique fixed color. One of the main complications when using a multi-spot beam communication system is that all co-channel beams VOLUME 7, 2019 have an effect on the overall Carrier to Interference ratio (C/I) of a particular user. This problem occurs not only in satellite systems but also in other widely deployed systems, such as LTE cellular networks [1]–[3]. In LTE the distance between the co-channel antennas is usually large enough to avoid interferences. Even so, its effect is considered to maximize the overall system capacity. In the particular case of multi-spot beam satellite systems, beams having the same color are physically separated by few hundred kilometers on the ground; yet all users are transmitting to the same satellite in space (around 36.000 km in case of a GEO satellite). When comparing spot distances with the satellite height, it can be seen that user distances to the satellite are approximately the same. This similarity in the user distance is one of the big differences when compared 2169-3536 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. 4391 J.-M. Rodriguez Bejarano et al.: MF-TDMA Scheduling Algorithm for Multi-Spot Beam Satellite Systems II. STATE OF THE ART FIGURE 1. Different Coloring schemes in a multi-spot system. All co-channel spot beams (same color) contribute with interferences in a transmission of a beam depending on the distance. FIGURE 2. Same color beams are physically separated by few hundred kilometers on the ground but when comparing spot distances with the satellite height, it can be seen that user distances to the satellite are approximately the same. with LTE and can lead to a decrease in overall network C/I performance due to co-channel interference. Figure 2 illustrates well how a terminal transmitting in other spot can interfere over others, as part of its signal can be received by other antenna lobe. The work presented in this paper is focused in minimizing the interference effects on interactive satellite networks like AmerHis or REDSAT [4]. In such innovative systems dynamic resource assignation is used (DAMA) [5] for the user services provision at real time. Therefore, timing constraints for allocation are relevant. Particularly the AmerHis-REDSAT family Network Control Centre (NCC) must perform scheduling in real time within a 69.632ms SuperFrame period [6]. In this work, an improvement of classical uplink MF-TDMA frequency reuse schemes is proposed for its use in multi-spot satellite networks. In particular two aspects are targeted: enhance overall system throughput of the network by considering the overall network C/I performance and allocate the resources in real time within a given period. This paper presents two novel algorithms for interference aware scheduling in an interactive DVB-RCS2 return uplink [7]. These algorithms increase the overall system throughput while its calculations can be performed in real time, as required for the interactive scheduling at the NCC. 4392 Dynamic Frequency Allocation in Fractional Frequency Reuse has been typically extended for LTE networks [8]. The same approach has been studied in order to be implemented in satellite networks [9]–[11] where the applicability of interference coordination (ICIC) techniques borrowed from the terrestrial communication systems is investigated. In the context of satellite networks, there are recent works that aim to solve the problem of optimal allocation in VHTSs that cannot flexibly change this allocation during operation [12]. In a more relaxed way a quasi-optimal bandwidth allocation for multi-spot MFTDMA satellites is found in [13]. This work is focused in the uplink scheduling and offers good results in the bandwidth optimization. Even so, the target system assumes that slot assignment is static and may be changed once per hour. Therefore, this algorithm expects results within a few minutes. The target of the present paper are systems based on real time dynamic bandwidth allocation, therefore results presented in [13] are not applicable to our current satellite scenario as dynamic allocation (DAMA) requires real time allocation within a SuperFrame period. The work performed in [14] is based on dedicated spotto-user concept and model interference based on each user’s radio propagation property. It offers good timing results but still not enough for real time allocation. In [15] a multi-spot beam model assuming MF-TDMA where the transponder carrier frequencies can be reused is introduced. In this work the frequency reuse is not regular as in currently existing systems, but is optimized through computations jointly for carriers and power to meet the SINR individual constraints of users. A less complex algorithm is proposed in [16], which is suitable for such satellite multi-spot beam networks. This algorithm is based in continuously re-arranging resource block to each user in every beam based and the frequency domain (through carriers of the MF-TDMA assignments table) such as to maximize the C/I . Apparently the main drawback of this algorithm is that results are based on an iterative process of approximation to the best results. Therefore cannot be implemented in a real time system using DAMA as the number of iterations to achieve a good performance is not deterministic. An interesting approach considering a real-time (slot-by-slot) implementation of an interference aware algorithm is performed in [17]. The numerical results, show promising performances regarding system throughput and sub-carriers utilization. Even so, authors indicate that it is still necessary to reduce drastically the computational complexity of the proposed methods before considering a real-time (slot-by-slot) implementation. Two novel algorithms for interference aware scheduling in an interactive DVB-RCS2 return uplink [7] are presented in this paper. Only the DVB-RCS2 SuperFrame at the return uplink has been considered in this work because previous VOLUME 7, 2019 J.-M. Rodriguez Bejarano et al.: MF-TDMA Scheduling Algorithm for Multi-Spot Beam Satellite Systems FIGURE 3. Different Coloring schemes in a multi-spot system. In the figure, the terminal T1 situated in B1 suffers interference by transmissions of terminals T2 and T3, as these terminals are using the same pair (f,t) for their timeslot transmission. This interference is perceived with a different gain depending on the spatial location of the terminals T2 and T3. results [4] have shown that the bottle neck in the addressed systems is in the return uplink. The optimizations done are working both with the DVB-RCS2 MODCODs and the MF-TDMA scheduling in the TBTP2. The algorithms envisaged in the present work are oriented to be implemented in a real satellite system where the MF-TDMA assignations need to be calculated within a SuperFrame period so that they can be considered as real-time optimizations. Therefore, no iterations can be addressed and assignations will be performed based on a time constrained algorithm. Additionally, computation will be simplified in order to permit that the TBTP generation can be implemented in a HW system as the NCC or a satellite on board processor itself. III. SYSTEM MODEL Focusing on a particular spot beam of a multi-spot satellite system, the satellite antenna gain for this beam can be considered constant. Due to the circular nature of the antenna design, the gain pattern outside the studied spot beam is projected in shape of concentric crowns (See Figure 3) with different gain values. As the interfering spot beam is usually large enough to have different gains in different geographical regions, it will be divided in a number of areas that depend on the studied spot beam antenna gain pattern. In order to simplify the problem, it will be assumed that the gain of users situated in a particular area will be the same. Within this approach, the interference generated by a terminal is completely characterized by the area where the terminal is situated. A. INTERFERENCE MATRIX The system model presented permits to anticipate the interference contribution from users transmitting in co-channel beams at a same pair (f , t). This interference value can be calculated at system level and is invariant from the VOLUME 7, 2019 system operation, so it is a value that can be provisioned for each terminal as a parameter. In order to do so, the interference values associated with an area that potentially interferes with the beam under study can be organized in terms of an interference matrix. After the areas’ model is calculated for each interfering beam, an interference matrix can be generated, representing the level of interference perceived by a user situated in a given beam, which is imposed by potential transmissions of other users situated in different areas of other co-channel beams. Formalizing the notation: B is the set of beams of a same colour in the system; A is the set of areas present in each of the beams; Î the interference matrix shown above, where all the interferences are quantified. The Î matrix will therefore have a dimension of dim(B) x dim(A) x dim(B). Let ‘s’ and ‘b’ be beams of B (s, b ∈ B), and let Is,b,a be an element of this matrix with ‘s’ being the beam where the terminal under study is placed, and ‘b’ and ‘a’ being the beam and area where the interfering terminal is situated, respectively. Following the MF-TDMA principles, it is not possible to have interfering terminals in the same beam as the terminal under study, i.e. formally (∀ b, a, Ib,b,a = 0). The rest of the non-zero Is,b,a elements will be calculated by computing the antenna pattern and assigning to each area of each beam the interference value. In order to calculate the interfering terminals’ Interference value it is necessary to take into account the antenna directivity and the situation regarding the studied terminal. In terms of directivity, it is calculated by taking the normalized magnitude of the Complex Directivity in Far Field (CDFF) using the antenna patterns of the different antennas. In order to calculate the co-channel beams contributions, it is necessary to calculate the antenna directivity with respect to these beams, and extract from it the perceived gain of a transmitting user. The following formula is proposed for the previously described directivity computation: beam b b Directivitybeam theta,phi = 20log10 CDFF theta,phi (dB) (1) where ‘b’ is the beam where the interfering terminal is situated at and the pair (theta, phi) are the parameters described in spherical coordinates which will be projected in the earth surface plane. The Gain G of the antenna associated to this Directivity can be directly obtained by multiplying this value by the antenna efficiency ε which typically has values situated between 0.5 and 0.7 [18]. polarity Gtheta,phi = Directivity · ε (2) For the C/I ratio calculation a fixed satellite service (with dual polarization) is assumed where a Root-Sum-Square (RSS) interference model is proposed to measure the relation between the intended signal with respect to the different Interference contributions. In general, the aggregate C/I ratio perceived by the antenna is quantified by the following formula where G denotes the above defined gain perceived by 4393 J.-M. Rodriguez Bejarano et al.: MF-TDMA Scheduling Algorithm for Multi-Spot Beam Satellite Systems the antenna from the different components of its gain pattern. copols (Gθ,ϕ ) (C/I )beams = P P θ,ϕ copolb xpolb ( b=isofreq Gθ,ϕ + b=isofreq Gθ,ϕ ) copol (3) xpol The terminal under study T1 can receive several G contributions from different terminals situated in different beams and different areas. These contributions will be computed as interferences while the Gain from the intended beam will be computed as a Gain. System noise features can also be included in the formula obtaining the SNIR. The SNIR is the commonly known Signal to noise Ratio (SNR) where the interference is included as part of the noise [19]. The interfering beam is divided in areas grouping nearby users. In that way, it is necessary to note that an interfering user can be situated in an area for an intended beam, but can be situated in other totally different area if the beam under study is different. In other words, the user area is a function of its coordinates within the beam and the interfered beam to be evaluated. Applying all the considerations above, the final result will be an interference beam-areas matrix that needs to be calculated for a given particular scenario. The interference evaluation will be based on this pre-computed interference matrix with inter-beam areas of interference. B. INTERFERENCE CONTRIBUTION CALCULATION This work aims to propose an allocation algorithm for MF-TDMA where interferences from co-channel beams are relevant parameters for the transmission allocation. That means that during the allocation process for a given beam, the different interferences contributions from co-channel beams will be evaluated. This section, formalizes the Interference contribution calculation. Let denote V̂b as the MF-TDMA allocation vector where the users of the beam ‘b’ are allocated to transmit at a given SuperFrame. Let be Vib an element of this vector at position i = {fi , ti } of V̂b where ‘f’ is the carrier frequency and ‘t’ is the time interval where to transmit. This position correspond to the slot i of the SuperFrame. Each user ubj at a beam ‘b’ (where j = 1..Ub and Ub is the number of users of beam ‘b’) requesting a transmission will be allocated in the V̂b matrix. If a user is allocated to transmit in the ‘i’ position of the V̂b matrix, then Vib will reference the user ubj ; if not, Vib = 0 referencing this timeslot is still not in use. Each user ubj will have a correspondent location within the beam ‘b’, so it can be assigned to an area. As previously discussed, this area will change depending on the beam under study ‘s’. Therefore, the area will be a function of the user location and beam ‘s’. Formally: aj,b,s = fa (ubj , s) (4) When a user needs transmission resources in a SuperFrame of beam ‘s’, the NCC should assign the best slots, 4394 in order to increase the chances of error-free communication.. Therefore, NCC should assign the slots having the lowest interference in their position i = {fi , ti }. In order to calculate the best position where a user must transmit it is necessary to evaluate the aggregate interference at each ‘i’ slot of the SuperFrame. In order to do so, Isi will reference the potential interference that a user of the beam ‘s’ will perceive from co-channel beams if its transmission is allocated at the position ‘i’. Note that the Isi value will be different each time there is a new transmission affecting the ‘i’ slot position. This is because at the same SuperFrame, a transmission in a Vbi will generate intrinsically an interference in a Vsi (as they both share the position i = {fi , ti }). The Isi value can be quantified thanks to the Î matrix by adding the Isba elements when there is a user interfering in beam ‘b’ to the beam ‘s’ . Formally: XB b b Iis = (5) b=1 Is,b,aj,b,s where vi = uj ∀vbi 6=0 The cumulated Isi will univocally determine the perceived interference in the slot ‘i’ . C. THROUGHPUT CALCULATION In the target system based on DVB-RCS2 [2] uplink, the used MODCOD will determine the maximum number of bytes to be transmitted in this burst. This MODCOD will be determined by the existing C/(N + I ). Considering that for a given user the perceived N will be increased by the perceived interference I and therefore minimizing the margin to transmit a valid signal. The minimum required Es /No depends on the used MODCOD . As reference, the target Es /No values for each MODCOD are taken from the requirements of the FP7 BATS project [7]. These values have been calculated for a sub-set of MODCODs used in the return link targeting a FER = 10−5 and in clear sky conditions. Taking into account the system parameters, the required Es /No can be directly derived from the required C/N by resorting to the following equivalence: C/N = Es /No · fb /(B · log2 M ) (6) where fb is the channel bit rate, B is the channel bandwidth, and M is the number of modulation symbols. In this context the interference level of the C/(N + I ) will be calculated by using the Isi value. Finally, the following Table 1 summarizes the overall endto-end link performances considered in clear sky conditions that are required to transmit data by using each of the available MODCOD. IV. INTERFERENCES CONTRIBUTION IN MF-TDMA The perceived interference is directly related with the available capacity in a certain burst transmission, as it impacts the usable MODCOD. This Section demonstrates how this circumstance impacts the overall system throughput when VOLUME 7, 2019 J.-M. Rodriguez Bejarano et al.: MF-TDMA Scheduling Algorithm for Multi-Spot Beam Satellite Systems TABLE 1. Required C/(N+I) per MODCOD. FIGURE 4. Example of Interference Matrix. The perceived interference from a co-channel user can be quantified depending on the spatial situation within a beam and an area. This matrix represents the interference level for 4 co-channel beams and 3 areas. TABLE 2. System simulation parameters. FIGURE 5. GRASP file representation of one beam using the system parameters. The represented areas show the different interference contribution at the co-channel beams. Pb ≈ Ps /k, k = 4 and N=599∗ 8 bits (see Table 1). 3 Es erfc( ) 2 No Eb Es =k No No s 3 4E s Pb16QAM = erfc( ) 2·4 No Ps16QAM = compared to a single spot scenario where interference is not present. The targeted satellite system will be based in a 19 co-channel beams scenario. Co-channel beams of the same colour share a 36MHz bandwidth that is used to transmit in an MF-TDMA fashion, using 22 carrier frequencies and a 69.632 ms SuperFrame divided in 30 timeslots. These assumptions have been taken from the AmerHisREDSAT family of DVB-RCS regenerative satellite systems [4], [6], [7] and from the [8] project. The system parameters used are summarized in Table 2: The Î matrix has been calculated following the approach expressed in Section IV-A and using data from a real satellite system. Finally, the calculated Isi is used to determine the C/(N + I ) and required Es /No . The different Isi values have been calculated by using a GRASP file representing the target system and using system parameters like the ones used in [7]. The antenna patterns (GRASP) have been obtained by using the parameters shown in TABLE 2 and the propagation ITU models like ITU-R P.618, ITU-R P.839, ITU-R P.837. A beam representation is shown in Figure 4. Assuming the maximum throughput, which implies all users are to transmit using the maximum MODCOD (16-QAM with a 5/6 correction) the Es /No can be translated to a Bit Error Rate (BER) by using the following relations (7), (8), (9) and (10) where it is assumed that VOLUME 7, 2019 Pp16QAM = 1 − (1 − Pb16QAM )N (7) (8) (9) (10) A simulation has been performed following a 50 times Montecarlo analysis [20] where users’ traffic requests are modeled using a Rate Based Dynamic Capacity (RBDC) request pattern following a normal distribution requesting an average of 15 timeslots per SuperFrame with a deviation of 5 as the average of a SuperFrame time period. The simulation tool used has been MATLAB. The results represented in the Figure 5 represent the system throughput versus the requested load. Eff . Throughput (in%) = [1 − Pp16QAM ] Request Load (in %) (11) Results are compared with the maximum throughput, assuming this maximum is when all the users transmit with the maximum MODCOD. The simulation results demonstrate that system throughput is less than the maximum supported by the system without interference. In particular, it can be inferred that at full channel load, around 30-35% of throughput is wasted due to the interferences produced in co-channel beams, and causing users to lose the full assigned 4395 J.-M. Rodriguez Bejarano et al.: MF-TDMA Scheduling Algorithm for Multi-Spot Beam Satellite Systems Algorithm 1 Allocation Based on Min Interference 1: n=0 2: WHILE Br > 0 && ∃i/Vis = 0 3: Determine the best ‘i’ 4: Calculate Bi as Bi = fB (Iis ) 5: Vis = usj 6: IF Br ≤ Bi 7: Br = 0 8: ELSE 9: Br = Br − Bi 10: n++ 11: End FIGURE 6. Effects of co-channel interference in a pseudo-random MF-TDMA allocation. The simulation show how effective throughput of the system decrease due to the co-channel interferences. timeslot in their transmission. Even so, this situation could be mitigated if a more robust MODCOD is assigned. V. INTERFERENCE MITIGATION IN MF-TDMA This Section presents different algorithms that aim at improving the user transmissions by calculating the most suitable MF-TDMA slot where a user may transmit, thus reducing the co-channel interferences. In Section IV-A and IV-B it was presented how it is possible to calculate the interfering terminals overall interference value, by taking into account the antenna directivity and the spatial location w.r.t. the studied terminal. This section present two different algorithms that enhance the system performance by taking into account the information contained in the Î interference matrix. Let Br be the total number of Bytes requested to transmit by user usj in the SuperFrame where the allocation takes place, and Bi the number of bytes that is possible to transmit in the timeslot with the lowest interference level at the moment of the allocation. The N timeslots of a user will be allocated sequentially and updated depending on the variation of Bi . Note that once the one slot is allocated Br needs also to be updated. Therefore, let us define Br as the peak rate subscribed by a user and n=1..N the number of allocations to be performed and the following allocation algorithm: In the algorithm the Bi value will be calculated each time, based on the timeslot ‘i’ selected and the relation between C/(N+I) and used MODCOD shown in Table 1. The best ‘i’ can be calculated by using the argmin function and therefore, allocating the user request into the slot position where the accumulated Isi interference is minimal. Note that once a user is allocated in a slot to transmit, users already allocated to use the same (f,t) pair will perceive an additional interference produced by the already allocated user. In this sense, calculating the ‘i’ slot based on the Isi is greedy, as it allocates resources for a user without taking into 4396 account what happens to users situated in other beams. In that context, even though the allocated user experiences minimum interference in the selected slot, other users using the same slots in other beams can have their performance degraded. To cope with this circumstance, it is possible to transform the Isi function to not only care about finding the minimum interference in the beam under study but looking for the overall minimum interference in all the beams. In that way the selected ‘i’ slot would be the one that minimizes the interference for all users already allocated in the remaining beams. The I′i value can be quantified thanks to the Î matrix by adding the Isba elements for all the beams of the system (not only the beam under study). That is taking into account that s = 1..B and therefore note that the I′i value does not depend on ‘s’. XB XB XB s I ′i = (12) b=1 Is,b,aj,b,s = s=1 s=1 Ii ∀vsi 6=0 ∀vbi 6=0 ∀vsi 6=0 where vbi = ubj and I′i will therefore represent the aggregated interference, at the moment of the study, in timeslot ‘i’, caused by the transmission of the other beams and taking into account all the beams. In the same way, the ’greedy’ calculation can be used to schedule users but taking into account that now the best ‘i’ is selected by applying the aggregated interference, at the moment of the study, in timeslot ‘i’, caused by the transmission of the other beams without taking care what happens to the users of other beams. Formally this two ways of determining the best ‘i’ are: (
i = argmini Iis greedy allocation
(13) Best i fair allocation i = argmini Ii′ Note that for the fair allocation determination the complexity of the calculation is bigger than the one used in the ‘greedy’ allocation context and for a real implementation context a trade-off between the possible system enhancement and the resources utilization must be done. The results of both algorithms (’greedy’ and ’fair’ allocations) will be presented in the next Section by applying the same system constraints as the presented in Section VI. VOLUME 7, 2019 J.-M. Rodriguez Bejarano et al.: MF-TDMA Scheduling Algorithm for Multi-Spot Beam Satellite Systems FIGURE 7. Co-channel interference effects with different allocation algorithms. In this figure all the presented simulation are compared. Interference aware allocations show a clear benefit on the overall system throughput. A simulation study has been performed, testing two different allocation algorithms (’greedy’ and ’fair’ allocations) and comparing their performances with the already presented random allocation of Section IV-C. The considered system parameters are the same as in the previous simulation. The results of the simulation are presented in Figure 7. Results for both allocation algorithms present a considerable performance improvement in the system throughput when compared with a pseudo-random allocation which is not aware of co-channel beams interference. VI. PERFORMANCE ANALYSYS Results presented in Figure 7 show that the amount of improvement depends on channel occupation ratio. Both algorithms present significant differences right from a nominal allocation starting around 10% of channel occupation. Maximum differences can be seen when the channel is fully occupied and does not accept more traffic. In that context the designed algorithms achieve around a +20% more of system throughput when compared with the pseudo-random allocation. If we focus in comparing the ‘fair’ and ‘greedy’ algorithms it is possible to notice that both achieve similar performances independently of the channel occupation. Both algorithms demonstrate to be equally efficient when user requests load is below 90% and both algorithms present the same performance when the channel is fully occupied. Every allocation algorithm introduces an overhead on the allocation processing time that could not be negligible for a real system. In a DVB-RCS network the SuperFrame defines the period where allocation takes place. The RRC would therefore to construct a TBTP table within this period. The computational time complexity overhead included by this channel optimization can be calculated by using the Bachmann–Landau notation [21]. In a practical implementation the RRC algorithm on ‘fair’ allocation needs to calculate all the Isi values of the matrix. VOLUME 7, 2019 FIGURE 8. Complexity of different algorithms implementation depending on TBTP size. The results on the selected system model show a difference in complexity of 1.5 order. To do that, it is necessary to iterate the through all the beams of the system. In particular, if we focus in (11), the algorithm will use two different loops. The outer loop will be executed B-1 times. In the ‘fair’ algorithm the inner loop executes additional B-1 times, but not in the ‘greedy’ one. The time complexities for ‘fair’ (14) and ‘greedy’ (15) are: C = O((B − 1)2 · f · t) (14) C = O((B − 1) · f · t) (15) Both algorithms will introduce an additional overhead for sorting equivalent to O(f · t · log2 (f · t)) [22]. Results presented in Figure 8 in combination with the results of Figure 7 show that the ‘greedy’ algorithm is the candidate for the proposed scenario. For the sake of uplink resource efficiency, the superframe duration is usually envisaged to be in the order of a few tens or hundreds of milliseconds. In the ETSI TR 101 790 standard [23] it is recommended to keep the super-frame duration within the range 20 ms to 750 ms. More restringing is the ETSI TS 102 429-1 standard [6] that put this value into 69ms. These values can be applicable to the DVB-RCS2 evolution as there not particular recommendations on it. Focusing in the State of the Art algorithms, most of them study the overall system C/I performances and link budgets but do not enter in traffic throughput considerations. These algorithms also do not take into account realtime constraints. There are only a few publications targeting results on throughput for interference coordination in MF-TDMA [13] and [17] that provide execution time results. The contribution presented in [13] states that they can obtain a throughput difference of only 1% from the optimum. Supposing that with ideal interference coordination the system can achieve a 100% of throughput, this algorithm can achieve the 99%. Even so, these results need a computation time of 60s, which is not feasible for the standards [6], [23]. The more relevant result of this work presented in [17] is to obtain an improvement of a +49% of throughput over the 4397 J.-M. Rodriguez Bejarano et al.: MF-TDMA Scheduling Algorithm for Multi-Spot Beam Satellite Systems TABLE 3. Allocation techniques comparison. non-optimal scenario. The mean resolution time to achieve these results is 40s. Again this is not feasible for [6], [23] standards. In order to compare the ‘greedy’ algorithm execution time with State of the Art algorithms, a simulation have been performed measuring the resolution time. The simulation has been performed implementing the ‘greedy’ algorithm in C++ and running it in a Linux environment. To obtain the average execution time, 100 iterations have been considered. Results show that the computation time for the scenario presented in the paper is 25ms using a 3.6GHz processor. In order to compare the ‘greedy’ algorithm execution time with State of the Art algorithms a simulation using the system parameters of [13] and [17] has been performed. For the sake of fairness, the simulation has been run in computation platforms equivalent to the ones used by the previous references. Results can be seen in Table 3. Even the results are not optimal (>80% of throughput compared to the 99% of [13] and +20% of increase compared to the +49% of [17]), the resolution time is fixed and therefore the algorithm can run in a real time system. The multibeam MF-TDMA multiple access optimisation proposed in this paper could be also used in the HTS system with hybrid transponders defined in [12], providing a complementary improvement. VII. CONCLUSIONS In this work an improvement of classical Multi-frequency time-division multiple access (MF-TDMA) frequency reuse allocation algorithm for a DVB-RCS2 return uplink has been proposed. The target system is based on interactive bandwidth on demand services like the AmerHis and REDSAT satellite systems. In such systems the TBTP allocation must be done in real time within a SuperFrame period. Simulations reveal that the presented algorithms increase the overall system throughput while not introducing relevant 4398 processing overhead at the NCC. In particular the novelties presented by this paper are the following: In the first place, it has been demonstrated through simulations that in a multi-spot beam satellite system the interference produced by co-channel beams can decrease the system throughput down by 30%. In the second place, two different MF-TDMA-based allocation algorithms have been presented obtaining +20% more throughput. Both algorithms have similar performances but the ‘fair’ one is computationally is much more complex. The ‘greedy’ algorithm evidences an enhanced behavior compared to a classical MF-TDMA allocation, and demonstrates that by including this new non-intensive processing scheduling task at the NCC, multi-spot satellite systems can increase their throughput. Furthermore, executing times of those algorithms are compliant with the time constraints suggested by ETSI TS 102 429-1 [6] & ETSI TR 101 790 [23]. ACKNOWLEDGEMENTS Special thanks to Dr. Sanchez (Thales Alenia Space) and R. Pinto (European Space Agency) for the review and support on this paper. REFERENCES [1] D. Wu, Q. Wu, Y. Xu, J. Jing, and Z. Qin, ‘‘QoE-based distributed multichannel allocation in 5G heterogeneous cellular networks: A matchingcoalitional game solution,’’ IEEE Access, vol. 5, pp. 61–71, 2016. [2] N. H. Mahmood, K. I. Pedersen, and P. Mogensen, ‘‘Interference aware inter-cell rank coordination for 5G systems,’’ IEEE Access, vol. 5, pp. 2339–2350, 2017. [3] B. Soret, K. I. Pedersen, N. T. K. Jørgensen, and V. Fernández-López, ‘‘Interference coordination for dense wireless networks,’’ IEEE Commun. Mag., vol. 53, no. 1, pp. 102–109, Jan. 2015. [4] A. Yun et al., ‘‘AmerHis next generation global IP services in the space,’’ in Proc. ASMS-SPSC, Cagliari, Italy, Sep. 2010, pp. 169–176, doi: 10.1109/ASMS-SPSC.2010.5586849. [5] B. de la Cuesta, L. Albiol, J. M. Aguiar, C. Baladrón, B. Carro, and A. Sánchez-Esguevillas, ‘‘Innovative DAMA algorithm for multimedia DVB-RCS system,’’ EURASIP J. Wireless Commun. Netw., vol. 2013, no. 1, p. 14, 2013. VOLUME 7, 2019 J.-M. Rodriguez Bejarano et al.: MF-TDMA Scheduling Algorithm for Multi-Spot Beam Satellite Systems [6] Satellite Earth Stations and Systems (SES); Broadband Satellite Multimedia (BSM); Regenerative Satellite Mesh—B (RSM-B); DVB-S/DVBRCS Family for Regenerative Satellites; Part 1: System Overview, document ETSI TS 102 429-1 V1.1.1, Oct. 2006. [7] Second Generation Digital Video Broadcasting (DVB)—Return Channel Via Satellite Interactive Satellite System (RCS2), document ETSI EN 301 545 V1.2.1, Apr. 2014. [8] (2015). BATS (Broadband Access Via integrated Terrestrial & Satellite Systems); European Union 7th Framework Programme. [Online]. Available: http://www.batsproject.eu/ [9] R. Suffritti et al., ‘‘On interference management techniques in broadband satellite systems,’’ in Proc. Ka Conf., Salerno, Italy, 2014, pp. 887–891. [10] N. Privitera et al., ‘‘Interference management strategies for forward and return link in high throughput satellite systems,’’ in Proc. Eur. Conf. Netw. Commun. (EUCNC), Bologna, Italy, Jun. 2014, pp. 1–3. [11] S. Dimitrov et al., ‘‘Radio resource management for forward and return links in high throughput satellite systems,’’ in Proc. 20th Ka Broadband Commun. Conf., Salerno, Italy, 2014, pp. 1–3. [12] K. Kaneko, H. Nishiyama, N. Kato, A. Miura, and M. Toyoshima, ‘‘Construction of a flexibility analysis model for flexible high-throughput satellite communication systems with a digital channelizer,’’ IEEE Trans. Veh. Technol., vol. 67, no. 3, pp. 2097–2107, Mar. 2018. [13] S. Alouf, E. Altman, J. Galtier, J.-F. Lalande, and C. Touati, ‘‘Quasioptimal bandwidth allocation for multi-spot MFTDMA satellites,’’ in Proc. IEEE 24th Annu. Joint Conf. IEEE Comput. Commun. Soc., Miami, FL, USA, vol. 1, Mar. 2005, pp. 560–571. [14] K. Kiatmanaroj, C. Artigues, L. Houssin, and F. Messine, ‘‘Hybrid discrete-continuous optimization for the frequency assignment problem in satellite communication system,’’ IFAC Proc. Volumes, vol. 45, no. 6, pp. 1419–1424, 2012. [15] J. E. Barceló-Lladó, M. A. Vazquez-Castro, J. Lei, and A. Hjorungnes, ‘‘Distributed power and carrier allocation in multibeam satellite uplink with individual SINR constraints,’’ in Proc. IEEE Global Telecommun. Conf. (GLOBECOM), Honolulu, HI, USA, Nov. 2009, pp. 1–6. [16] U. Y. Ng, A. Kyrgiazos, and B. Evans, ‘‘Interference coordination for the return link of a multibeam satellite system,’’ in Proc. Adv. Satell. Multimedia Syst. Conf. (ASMS) 13th Signal Process. Space Commun. Workshop (SPSC), Sep. 2014, pp. 366–373. [17] Y. Couble et al., ‘‘Interference-aware frame optimization for the return link of a multi-beam satellite,’’ in Proc. IEEE Int. Conf. Commun. (ICC), Paris, France, May 2017, pp. 1–6. [18] P. R. Akbar, H. Saito, M. Zhang, J. Hirokawa, and M. Ando, ‘‘Parallel-plate slot array antenna for deployable SAR antenna onboard small satellite,’’ IEEE Trans. Antennas Propag., vol. 64, no. 5, pp. 1661–1671, May 2016. [19] M. Haenggi, J. Andrews, F. Baccelli, O. Dousse, and M. Franceschetti, ‘‘Stochastic geometry and random graphs for the analysis and design of wireless networks,’’ IEEE J. Sel. Areas Commun., vol. 27, no. 7, pp. 1029–1046, Sep. 2009. [20] D. P. Kroese, T. Brereton, T. Taimre, and Z. I. Botev, ‘‘Why the Monte Carlo method is so important today,’’ Wires Comput. Statist., vol. 6, no. 6, pp. 386–392, 2014. [21] P. Bachmann, Analytic Number Theory, (in German). Leipzig, Germany: Teubner, 1894. [22] R. Sedgewick, Algorithms in C++ Parts 1–4: Fundamentals, Data Structure, Sorting, Searching, 3rd ed. London, U.K.: Pearson, 1998. [23] Digital Video Broadcasting (DVB); Interaction Channel for Satellite Distribution Systems; Guidelines for the Use of EN 301 790, document ETSI TR 101 790 V1.4.1, Jul. 2009. VOLUME 7, 2019 JUAN-MANUEL RODRIGUEZ BEJARANO was born in Madrid, Spain, in 1981. He received the M.S. degree in telecommunications engineering from the Universidad Politécnica de Madrid (UPM), in 2008, where he is currently pursuing the Ph.D. degree. His Ph.D. research is in satellite communications and networking. Since 2007, he has been with Thales Alenia Space working in several research projects and has participated in the design of novel satellite payloads and systems such as REDSAT or AmerHis. He has authored some book chapters and more than 10 research papers and has served as a reviewer in some satellite-related conferences. He has been an Active Member on the DVB-RCS2 standardization activities, contributing to all its standards (LLS, HLS, and System specification) and being responsible for the ETSI TS 101 545-3 Dynamic Connection Protocol specification. He has also participated in other ETSI specifications such as the ETSI TR 103 272. Mr. Rodriguez Bejarano was a recipient of the 1st Award to the best paper titled Analysis of the Convergence between DVB-SH and ETSI SDR at the International Conference on Advances in Satellite and Space Communications, in 2009. CARLOS MIGUEL NIETO was born in Madrid, Spain, in 1962. He received the degree in telecommunication engineering from UPM, in 1986, and the Ph.D. degree in telecommunication engineering from UPM, in 1991, working on the application of formal techniques to the performance evaluation of communication systems. He has been Professor with the Department of Telematic Systems Engineering, Universidad Politécnica de Madrid (UPM), since 1988. Since 1988, he has been teaching in computer science and communication networks with the Faculty of Electrical Engineering, UPM. Since 1986, he has been involved in R&D activities related to the following topics: software design, digital transmission, and satellite data networks; contributing to the DVB working group involved in QoS standardization of DVB-S2/DVB-RCS2 networks. Dr. Miguel received the Best Ph.D. Thesis Award from UPM, in 1992. FRANCISCO JAVIER RUIZ PIÑAR (M’96) received the degree in telecommunication engineering from the Universidad Politécnica de Madrid (UPM), Spain, in 1990, and the Ph.D. degree in telecommunication engineering from UPM, in 1994, with a thesis on access algorithms in satellite networks. Since 1998, he has been an Associate Professor with ETSI Telecomunicación, UPM. His research interests include software defined networking and satellite networksand virtualization technologies applied to teaching. He has participated in several national and international projects related to these topics. Dr. Ruiz received the best Ph.D. Thesis Award on Access Networks, in 1995, awarded by the Spanish Colegio Oficial de Ingenieros de Telecomunicación. 4399