Energy Efficient Adaptive Network Coding Schemes for Satellite Communications
Ala Eddine Gharsellaoui, Samah A. M. Ghanem, Daniele Tarchi, Alessandro Vanelli Coralli
aa r X i v : . [ c s . I T ] A p r Energy Efficient Adaptive Network CodingSchemes for Satellite Communications
Ala Eddine Gharsellaoui , Samah A. M. Ghanem , Daniele Tarchi , andAlessandro Vanelli Coralli Department of Electrical, Electronic and Information Engineering,University of Bologna, Bologna, Italy Huawei R&D Labs, Stockholm, Sweden
Abstract.
In this paper, we propose novel energy efficient adaptive net-work coding and modulation schemes for time variant channels. We eval-uate such schemes under a realistic channel model for open area envi-ronments and Geostationary Earth Orbit (GEO) satellites. Compared tonon-adaptive network coding and adaptive rate efficient network-codedschemes for time variant channels, we show that our proposed schemes,through physical layer awareness can be designed to transmit only if atarget quality of service (QoS) is achieved. As a result, such schemes canprovide remarkable energy savings.
Key words:
Energy Efficiency, Network Coding, Satellite Communica-tions
Network coding is a transmission technique that, by performing algebraic op-erations across transmitted packets rather than relying on packet repetition orreplication, allows to reliably transmit with lower end to end delays in a commu-nication system. Additionally, network coding mechanisms are key enablers toenergy efficient communications. Due to the steady increase in energy consump-tion and energy costs in mobile communication systems, more efficient schemesare required. In particular, with higher reliability obtained via network cod-ing, less re-transmissions are required. Consequently, more energy savings canbe achieved [1]. Moreover, when the network coded schemes are specifically de-signed for enhancing their awareness with respect to the system characteristics,higher performance gains can be achieved in terms of delay, throughput or energyefficiency [2, 3]. One of the most important issue to be considered in satellite com-munications is energy efficiency. In [4], several factors that have a direct impacton energy efficiency of satellite and mobile terminals have been discussed, includ-ing dynamic spectrum access and cross layer design. In [2] and [5], the authorsshow that channel-aware transmission schemes jointly with network coding, canserve to reduce the delay and allow for energy performance gains. In [3], the au-thors propose novel adaptive network coding schemes and show a clear trade-offbetween energy-driven channel-aware schemes, that remain silent when channel
Gharsellaoui, Ghanem, Tarchi, Vanelli Coralli encounter high erasures, and rate-driven channel-aware schemes that chose totransmit more to account for erasures.In this work, various aspects of energy efficiency using network coding andmodulation schemes are proposed. The schemes are evaluated in a realistic satel-lite channel model for open area environments and Geostationary Earth Orbit(GEO) satellites. Simulation results demonstrate clear trade-offs among averagenumber of transmissions, delay, throughput and energy efficiency. We highlightthat adaptation through channel-aware policies allows for silence periods or lesstransmissions which leads to significant energy savings compared to non-adaptivenetwork coding or adaptive network coded schemes that are rate efficient.
Our focus is on a GEO satellite communications system forward link trans-mission, by considering a mobile terminal with constant speed in a open areaenvironment. In such system, the transmitter performs a Random Linear Net-work Coding (RLNC), which is a Network Coding (NC) scheme that relies oncoding across the packets using random linear coefficients in order to increase thetransmission reliability mimicking the wireless diversity concept. The open areaenvironment is modeled by resorting to the Land Mobile Satellite (LMS) modelin [6], that is one of the most known in the literature. This model is based on ajoint exploitation of a state based and a Loo based distribution [7] that allow toefficiently reproduce the shadowing and fading effects of a forward link satellitechannel under mobile terminal assumptions. In this paper, we capitalize on thecoded/uncoded packet transmission over the Markov model proposed in [2] toanalyze our proposed schemes that rely mainly on channel variation awareness.Each state in the Markov model is represented by the couple ( i, h j ) that standsfor the number i of coded/uncoded packets to be sent and the channel state h j ,whose value varies over time. Therefore, such Markov model can be expressedby a one-step transition probability matrix P , whose size is defined by a finitenumber of states, and its components are defined by using two transition proba-bility components: p ( i,h j ) → ( i − ,h j +1 ) = 1 − P e ( h j ) , and p ( i,h j ) → ( i,h j +1 ) = P e ( h j ) , where P e ( h j ) is the packet erasure probability at the channel state h j for theduration of the packet transmission, and the probability of transitioning fromthe channel state and back to itself equals zero due to channel variation overtime. This means that for a packet of size B bits, and with bit error proba-bility P b ( h j ) at a given channel state h j , the erasure probability is given as, P e ( h j ) = 1 − (1 − P b ( h j )) B . We resort to the approximation of the delay undernetwork-coded transmission provided in [2], where the expected time to deliver N i coded packets is given as: T ( i, h j ) = T d ( N i , h j ) + i X l =1 P N i ( i,h j ) → ( l,h j + Ni ) T ( l, h j + N i +1 ) , (1) nergy Efficient Adaptive Network Coding Schemes for Satellite Scenarios 3 with T d ( N i , h j ) = N i T p + T w , where T p is the duration of one packet, and T w is the waiting time for acknowledgment. P N i ( i,h j ) → ( l,h j + Ni ) , corresponds to thetransition probability between states at the N thi transition of matrix P . Finally,in j + N i + 1 the term +1 appears due to acknowledgment. Fig. 1 illustratesthe Markov chain as proposed in [2]. This model of coded and uncoded packettransmission over time varying channels assumes a finite number of time slotsfor sending a given number of packets. Thus, the model and the delay approxi-mation inherently constraints the number of re-transmissions of packets, but hassufficiently large number of slots for a reliable approximation. , h , h , h , h ... , h , h , h ... , h , h ... AbsorptionState Pe ( h Pe ( h Pe ( h Pe ( h Pe ( h Pe ( h − Pe ( h
0) 1 − Pe ( h
1) 1 − Pe ( h − Pe ( h
1) 1 − Pe ( h
2) 1 − Pe ( h Pe ( h − Pe ( h − Pe ( h − Pe ( hj ) Pe ( h Pe ( h − Pe ( h Fig. 1.
Time Varying Channel Model of 3 Packets Transmission in [2]
The main objective is to propose energy efficient schemes by the exploitation ofthe adaptation of the coded packets transmission to the channel awareness underdifferent levels of algorithm complexities. We discuss three proposed schemes andcompare them to non-adaptive network coding scheme for time varying channelsand to the rate driven adaptive network coded scheme in [2].
This scheme adapts the transmission for achieving the energy efficient, by fol-lowing the observation of the channel erasure; the strategy is to transmit smallbatches of coded packets if the observation of channel erasure is high (applies
Gharsellaoui, Ghanem, Tarchi, Vanelli Coralli to low SNR), and to transmit larger batches of coded packets if erasure is less(applies to high SNR). Through this, the system can reduce transmissions andre-transmissions and save energy. Furthermore, QoS measure has been intro-duced to design such algorithm. In particular, if a bit error probability P b lessthan 10 − is not met , the transmitter will choose to be silent with no transmis-sions. Therefore, much energy savings can be obtained. The following equationillustrates the number of coded packets N i required to be sent at the channelstate h j when i degrees of freedom (dof) are required at the receiver, N i = j + i − X s = j (1 − P e ( h s )) . (2)It’s worth mentioning that N i required to be rounded to nearest decimal orinteger number, because it represents the number of coded packets. Moreover,it is worth to note that the sum is expressed with a shifted start of the state ofmeasurements. This is due to the channel evolution over time, thus, a new roundof transmission/re-transmission is associated to shift in the channel window.When erasures are high (at low SNR) such sum vanishes to zero correspondingto no transmission. However, when erasures are very low (at high SNR) suchsum converges to the transmission of i degrees of freedom almost surely. In this scheme, we propose a self-tracing ANCEF scheme, which improves AN-CEF by adding to the observation of the channel erasure the capability oflooking-forward into the channel erasures if, looking- backwards, the packetsat earlier transmissions are lower than the dof. Thus, the transmission strat-egy of such algorithm is similar to the ANCEF discussed in previous section,where less coded packets will be transmitted adaptively at high erasures, andmore packets will be transmitted adaptively at low erasures. However, there isan amount of coded packets ∆ i needed to be as additional amount of futurere-transmissions to establish all lost dof. Therefore, such ∆ i decreases as wemove towards higher SNR, until it vanishes to zero when the transmission strat-egy of N i = i . Once again, a certain QoS measure needs to be met, otherwise,the transmitter remains silent. The following equation represents the STANCEFtransmission strategy of the number of coded packets N i required to be sent ata certain channel state h j : N i = j + i + ∆ i − X s = j (1 − P e ( h s )) , (3) The acceptable bit error rate acceptable by the ITU ranges between 10 − to 10 − based on the rate and service expected at the mobile terminal A degree of freedom corresponds to the number of linear combinations that arerequired at the receiver to allow decoding the RLNC combined packetsnergy Efficient Adaptive Network Coding Schemes for Satellite Scenarios 5 where ∆ i is the foreseen losses due to self-tracing which is the difference between i dof required at preliminary transmission at state h j ∗ and the number of codedpackets N i adapted to the channel at the same state. It is given by, ∆ i = i − N i ,where ∆ i equals 0 at the initial state of first transmission. However, ∆ i is higheror equal to 0, at h ∗ j corresponds to zero or more coded packets contributed at are-transmission stage one step ahead of its previous transmission. Thus, ∆ i at h ∗ j will contribute to N i at a forward channel state h j ∗ + N i +1 +1 , where the additionof one represents the one step ahead due to ACK after preliminary transmissionand before re-transmission. In this scheme, we integrated adaptive modulation to the ANCEF scheme. Therationale behind this, is that, on the one hand, a higher modulation order m allows for transmitting the same amount of information in shorter packets dueto the concatenation of more bits in the real and imaginary spaces. On theother hand, a higher modulation order is associated with higher energy persymbol, and less energy per bit, i.e. E b /N = E s / ( N ∗ log m ). Thus, a higherbit error probability suggests that higher number of packets need to be sentdue to adaptation. Such trade-off between the packet length and the number ofcoded packets for a given modulation scheme is of particular interest to addresswhen taking into account energy efficiency. ANCMEF transmission strategy ofcoded packets N i m is given by, N i m = j + i ∗ log m − X s = j (1 − P e m ( h s )) , (4)Resorting to the energy efficiency of the proposed scheme, the lower boundon random linear network coding, with N i m ≥ i , i.e. with equality, is used,this was reflected in the sum range, by scaling the degrees of freedom i by afactor log m that unifies the energy per symbol for each modulation scheme. P e m ( h s ) is the erasure probability of that modulation which can be derived as P e m = 1 − (1 − P b m ) B , where P b m is the bit error probability for the givenmodulation order m , and B is the number of bits per packet.Indeed the aim of the scheme is to find the optimal number of coded pack-ets N mi for a given modulation order m to be transmitted/re-transmitted forassuring successful reception of a given number of i dof along the way with en-ergy efficiency; hence when P b m of m -th modulation order is derived, for fairenergy comparison among the different modulation schemes, E s is supposed tobe constant for each modulation scheme. We shall now provide a set of illustrative results that cast further insights to theproposed schemes. Particularly, we focus on a satellite scenario and its related
Gharsellaoui, Ghanem, Tarchi, Vanelli Coralli
LMS channel considering GEO satellite with delay T w equals 0 . sec , openarea, and a mobile terminal with speed of 10 m/s . To construct the simulationenvironment, we consider to transmit 4 coded packets/dof, the maximum batchlength due to channel adaptation of the schemes is constrained to 16 coded pack-ets/dof, and the number of transmission/re-transmission trials is constrained to10. This corresponds to a transition matrix of maximum size equals 401 × E s /N level.The number of bits per packet is considered to be 1000, and since the max-imum number of packets per batch equals 16, the maximum possible batch ofpacket size equals 16000 bits. This number of bits corresponds to the same, thehalf, the one-third, and the one-forth number of samples in BPSK, QPSK, 8PSK,and 16QAM, respectively.First, the average number of coded packets sent back to back for the differentschemes is compared; under different E s /N values and with mobility speedequals 10 m/s , this can be seen in Fig. 2. In general, we can see that the averagenumber of packets for all the schemes at low SNR is greater than those athigh SNR due to the higher probability of re-transmission at low SNR. Due toenergy efficiency rule, the average number of packets will be as much small asto commensurate with the higher erasure probability at low E s /N values, sincethey are designed to limit the number of sent packets in case of bad channelconditions i.e., there is no need to spend more energy in such low chance ofdelivery. This can be emphasized looking into the low SNR, where the erasureprobability is high such that energy efficient schemes favor not to send anythingin order to avoid energy wasting; this is the contrary for the ANC benchmarkthat has been designed to achieve the highest possible rates. It is worth tonote that the average number of coded packets in ANCMEF gets larger forhigher E s /N values; this is expected since we aim to adapt the number of sentpackets for achieving target reliability, keeping into account the energy efficiency. nergy Efficient Adaptive Network Coding Schemes for Satellite Scenarios 7 Es/No (dB) A v e r age N u m be r o f P a ck e t s NCANCANCEFSTANCEFANCMEF
Fig. 2.
Average number of sent packets for variable E s /N values and 10 m/s mobilityspeed. However, the maximum constrained batch size in ANCMEF is affected not onlyby E s /N but also by modulation method and energy consumption, so at lowSNR it behaves similarly to the NC, then, going to higher SNR, it selects ahigher modulation order allowing to increase the transmission reliability by theexploitation of higher diversity or modulation order. However, the increase ofnumber of packets of ANCMEF is associated to shorter size packets that allowsfor equivalent energy per symbol for all modulation orders and across all schemes.In Fig. 3, we can see the behavior in terms of transmission delay for theproposed schemes in a GEO satellite scenario. The proposed schemes have thehigher delay for low SNR values; this is due to the fact that the energy efficientadaptive schemes adapt its transmission to small size batches at the low E s /N associated with high erasures. Therefore, the transmission suffers from extrawaiting times for acknowledgment at the end of each short batch. Thus, asillustrated in Fig. 3, the time spent waiting for acknowledgment is very largecompared to the time of delivering the coded packets. After a certain E s /N value, the schemes have similar performance due to the number of packets in eachbatch that has been increased and then converged to the exact dof value of thenon-adaptive NC scheme. ANCMEF is not an exception since a normalization onthe average number of packets with log m matches with the dof for such shorterlength packets. It is indeed straightforward to understand the delay saturationof all schemes to its minimal value at the high SNR.Fig. 4 presents the throughput for the measured schemes NC and ANC withthe proposed ones, i.e., ANCEF, STANCEF and ANCMEF. For low and in-termediate E s /N values, the schemes show remarkable differences but all of Gharsellaoui, Ghanem, Tarchi, Vanelli Coralli
Es/No (dB) D e l a y ( s e c ) NCANCANCEFSTANCEFANCMEF
Fig. 3.
Transmission Delay in a GEO Satellite scenario for variable E s /N values and10 m/s mobility speed. them give less throughput than ANC and NC schemes due to the reduced num-ber of transmitted packets. However, we emphasize that the main aim of theseschemes is to avoid any source of energy consumption that rise due to bad chan-nel conditions, hence, our schemes favors to be silent from transmission, thanconsuming energy by limiting the transmitted packets, and its utilization toadaptive coding techniques allows for reliable transmission and less energy dueto decreased re-transmissions encountered. Thus, achieving less throughput isunderstandable. Furthermore, its worth to observe that at a certain E s /N , allthe schemes converge to maximum saturation throughput independent of mo-bile speed or channel variation. While, ANCEF and ANCMEF almost coincidein the performance behavior but not the reliability, we can see that STANCEFtries to build a trade-off that resonates just in a limited throughput gain dueto its dof one step loss tractability. In fact, such a small gain in throughput isshown to be associated with a small cost in the energy. For medium E s /N theextra complexity due to self tractability and excess transmissions has no gains,therefore, we see that STANCEF throughput performance deteriorate with re-spect to the ANCEF and ANCMEF. Finally, Fig. 5 presents the most importantpart of this study, the total energy consumption that has been calculated usingboth the spectral noise level as N = − dBm and the expected time neededto send 4 coded packets/dofs for each E s /N of each scheme. It is possible tonote that all the proposed schemes allow significant gains in terms of energyconsumption with respect to the benchmarks. Its clear that at the low SNR wecan see STANCEF gains roughly 15 mW att/Hz with respect to the NC scheme.Moreover, it is worth to see that ANCEF and ANCMEF reduce remarkably the nergy Efficient Adaptive Network Coding Schemes for Satellite Scenarios 9 Es/No (dB) T h r oughpu t ( pa ck e t s / s e c ) NCANCANCEFSTANCEFANCMEF
Fig. 4.
Throughput in a GEO Satellite scenario for variable E s /N values and 10 m/s mobility speed. energy consumption by gaining up to roughly 20 mW att/Hz with respect to theNC scheme; such an amount might seem to be small, however, this representsvery high figure in a large scale system with multiple receivers. Furthermore,at the moderate E s /N we can see that due to the shorter sizes of the pack-ets length used for higher modulation orders, the ANCMEF continues to havehighest energy efficiency with further gains. Finally, at the high E s /N beyond10 dB , is associated with an increase in the transmitted message which necessar-ily increases the overall system energy consumption. This paper addresses energy efficient adaptive network coding schemes for landsatellite mobile with time varying channel. We proposed three novel adaptivephysical layer aware schemes for coded packet transmission over LMS channel.Those schemes compensate for the lost degrees of freedom by tracking the packeterasures over time. The novelty of such schemes is expressed by their remarkableenergy savings due to adaption to a set of various factors such as channel qual-ity, that inherently adapts to the mobile speed, and thus allows due to smartsilence/transmission periods to significant energy savings. Finally, we emphasizethat, at SNR values high enough for reliable transmission, the schemes can beswitched off to allow for a reduction in the processing power at the transmitterside.
Es/No (dB) E ne r g y ( m W / H z ) NCANCANCEFSTANCEFANCMEF
Fig. 5.
Energy consumed in a GEO Satellite scenario for variable E s /N values and10 m/s mobility speed. References
1. Fragouli, C., Widmer, J., Boudec, J.Y.L.: A network coding approach to energyefficient broadcasting: From theory to practice. In: Proc. of 25th IEEE InternationalConference on Computer Communications (IEEE INFOCOM 2006). (April 2006)1–112. Ghanem, S.A.M.: Network coding mechanisms for Ka-band satellite time varyingchannel. In: Proc. of 2013 IEEE 24th Annual International Symposium on Per-sonal, Indoor, and Mobile Radio Communications (PIMRC), London, UK (Septem-ber 2013) 3291–32963. Gharsellaoui, A., Ghanem, S.A.M., Tarchi, D., Vanelli-Coralli, A.: Adaptive networkcoding schemes for satellite communications. In: Proc. of 8th Advanced SatelliteMultimedia Systems Conference and the 14th Signal Processing for Space Commu-nications Workshop (ASMS/SPSC 2016), Palma De Mallorca, Spain (September2016) accepted.4. Alagoz, F., Gur, G.: Energy efficiency and satellite networking: A holistic overview.Proceedings of the IEEE (11) (November 2011) 1954–19795. Lucani, D.E., Kliewer, J.: On the delay and energy performance in coded two-hopline networks with bursty erasures. In: Proc. of 2011 8th International Symposiumon Wireless Communication Systems (ISWCS). (November 2011) 502–5066. Fontan, F.P., Vazquez-Castro, M., Cabado, C.E., Garcia, J.P., Kubista, E.: Statis-tical modeling of the LMS channel. IEEE Transactions on Vehicular Technology (6) (November 2001) 1549–15677. Loo, C.: A statistical model for a land mobile satellite link. IEEE Transactions onVehicular Technology34