Contention Resolution Diversity Slotted ALOHA - IEEE Xplore

0 downloads 0 Views 770KB Size Report
loss ratio conditions (e.g. 17-fold improvement at Packet Loss. Ratio = 2 · 10−2). CRDSA allows to boost the performance of random access (RA) channels in the ...
1408

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 4, APRIL 2007

Contention Resolution Diversity Slotted ALOHA (CRDSA): An Enhanced Random Access Scheme for Satellite Access Packet Networks Enrico Casini, Riccardo De Gaudenzi, Senior Member, IEEE, and Oscar del Rio Herrero

Abstract— In this paper a new multiple access scheme dubbed Contention Resolution Diversity Slotted Aloha (CRDSA) is introduced and its performance and implementation are thoroughly analyzed. The scheme combines diversity transmission of data bursts with efficient interference cancellation techniques. It is shown that CRDSA largely outperforms the classical Slotted Aloha (SA) technique in terms of throughput under equal packet loss ratio conditions (e.g. 17-fold improvement at Packet Loss Ratio = 2 · 10−2 ). CRDSA allows to boost the performance of random access (RA) channels in the return link of interactive satellite networks, making RA very efficient and providing low latency for the transmission of small size sparse packets. Implementation-wise it is shown that the CRDSA technique can be easily integrated in systems equipped with digital burst demodulators. Index Terms— Access control, interference suppression, multiaccess communication, satellite communication, time division multiaccess.

I. I NTRODUCTION ESPITE having been proposed more than 30 years ago, Slotted Aloha (SA) [1], [2] and its slightly enhanced version named Diversity Slotted Aloha (DSA) [3] are today widely used in satellite networks for initial terminal access or short packet transmissions over a shared medium. The current satellite standards for interactive satellite broadband networks like the Digital Video Broadcasting (DVB) Return Channel via Satellite (DVB-RCS) [4] and the Telecommunication Industry Association (TIA) IP over Satellite (IPoS) [5] provide the capability to transmit small packets through a SA Random Access (RA) contention channel. In particular, the IPoS standard exploits the DSA protocol to enhance the RA channel capabilities. The satellite standards also include a capacity reservation mechanism, Demand Assignment Multiple Access (DAMA) [6], for longer packets transmission or for terminals offering a medium to high level of traffic aggregation. However, DAMA response time can be too long for the transmission of short bursts, which is frequent in consumer type of terminals [7]. The family of Carrier Sense Multiple Access (CSMA) [8] protocols cannot be used for satellite networks because their

D

Manuscript received July 10, 2005; revised December 5, 2005, April 30, 2006, and August 3, 2006; accepted August 10, 2006. The associate editor coordinating the review of this paper and approving it for publication was V. Leung. This work was supported by the European Space Agency. The authors are with the European Space Agency, European Space and Technology Center, ESTEC, Keplerlaan 1, 2200 AG, Noordwjik, The Netherlands (e-mail: [email protected]; [email protected]; [email protected]). Digital Object Identifier 10.1109/TWC.2007.05528.

large propagation latency (250 ms for a geostationary satellite) prevents the exploitation of the carrier sensing mechanism. Moving towards consumer type of interactive satellite terminals (ST), the amount of traffic aggregation at the ST will largely decrease and consequently the RA channel utilization potential will increase. In fact, although SA represents today a well established Random Access technique for TDMA satellite networks doubling the maximum throughput compared to the pure Aloha protocol [9], its utilization is typically limited to initial login, capacity request or Medium Access Control (MAC) signalling packets. This is because in practice SA works with very moderate normalized average loading (e.g. 25%) to ensure acceptable packets transmission delay and loss probability [10]. DSA provides better delay and throughput performance than SA under very moderate loading conditions by transmitting twice the same packet in a different TDMA slot, or a different frequency and time slot in case of Multi-Frequency TDMA (MF-TDMA) [3]. However, the throughput difference between Aloha and Slotted Aloha or Diversity Slotted Aloha is limited and quite poor in absolute terms. Another possible improvement of SA is the so called Selective Reject Aloha (SRE) protocol [11], [12]. Its main claimed advantage lies in the SRE capability to achieve throughput performance similar to the SA without the need for STs network synchronization. SRE exploits message subpacketization jointly with selective reject ARQ retransmission for partial packet overlaps occurring in practice avoiding the need for network synchronization. This advantage is however mitigated by the need for extra overhead in the packets. It is therefore pivotal to enhance the satellite RA channel performance in terms of throughput and delay with minimum impact on the existing satellite standards, currently based on MF-TDMA access scheme. The novel Contention Resolution Diversity Slotted Aloha (CRDSA) scheme described in the present paper represents an improved version of the well known SA and DSA schemes. Similarly to DSA, the CRDSA protocol generates two replicas of the same burst (in the following we will call burst the physical layer packet) at random time within a frame instead of only once as in SA. While the driver for DSA is to slightly enhance the SA performance by increasing the probability of packet successful transmission at the expense of increased RA load, CRDSA in addition is designed in a way to resolve most of the DSA packet contentions. Burst collisions are cleared up through a simple yet effective iterative Interference Cancellation (IC)

c 2007 IEEE 1536-1276/07$25.00 

CASINI et al.: CONTENTION RESOLUTION DIVERSITY SLOTTED ALOHA (CRDSA)

approach that uses frame composition information from the replica bursts. The main CRDSA advantages lie in the improved packet loss ratio and reduced packet delivery delay performance versus channel load jointly with a much higher operational throughput compared to SA and DSA.

1409

PK 6

PK 1

PK 3

PK 2

PK 2

PK 3

PK 5

PK 5

PK 4

PK 4

PK 1 PK 6

M slots per RA frame

Nslot.TS RA frame (T F seconds)

II. S YSTEM A SSUMPTIONS As stated in Section I, we consider here the return link of a satellite access network (i.e. link from satellite terminal to the gateway). Although the CRDSA application is not restricted to satellite networks, it appears to be the most natural application of this scheme. For simplicity we assume a bent-pipe satellite payload in which users are connected through the satellite to one gateway providing ground network access. Findings reported in the following sections can also be applied to the case of regenerative systems. In this case, the inbound link demodulator will be located on-board the satellite. The various STs will share satellite inbound resources according to the selected access scheme. Conventional MultiFrequency TDMA (MF-TDMA) will be considered, although for simplicity the following analysis will be focused on one single carrier. STs once registered in the network will keep TDMA slot synchronization through procedures such as the one described in [13] or in [5] with a slot timing random error RA Ts , where Ts is that we assumed bounded by τmax = Nguard RA the TDMA symbol duration, Nguard represents the TDMA slot guard expressed in symbols and Rs = 1/Ts is the ST baud rate. STs transmitted power can be optionally controlled by a power control mechanism based on a closed loop algorithm. III. P ROPOSED R ANDOM ACCESS S CHEME The proposed TDMA frame structure for the RA scheme RA . is shown in Fig. 1. Each RA frame is composed of Mslots A terminal can transmit at most one MAC packet per RA frame. In fact, the terminal will physically send two copies of the same MAC packet (also called "twins" bursts) with exactly the same preamble and payload information bits in two randomly selected slots within the same RA frame (see Fig. 1). The payload shall also contain signalling information concerning the slot position of the corresponding twin burst within the frame. Each burst signalling information points to its twin location and vice versa. It is known that sending the same packet twice slightly improves the probability of transmission success (i.e. no collision) for small MAC loads [3]. The CRDSA scheme key novelty is that the recovered information from a successful packet is exploited to cancel the interference that its twin may generate on another TDMA slot. This approach is iterated to recover most of the frame packets that were initially lost due to collision(s). In the example provided in Fig. 1, packet 2 cannot be initially recovered as both twins have suffered a collision in slot 1 and slot 4 of the frame. However, one copy of packet 3 (in slot 5) has been successfully recovered and its information can be used to cancel the interference caused by packet 3 in slot 4. Then packet 2 can be recovered in slot 4, after removing the interference generated by packet 3. Removing packet 2 in slot 1 allows to recover packet 1 so that also packet 6 can be decoded in slot M. This heuristic explanation

Fig. 1.

TDMA frame structure for the Random Access channel.

clearly shows the pivotal role of IC jointly with DSA and twin location signalling for effective interference resolution in CRDSA. In the following sections, the CRDSA processing will be generalized, detailed and formalized. A. RA Channel Description As described in the previous section, each RA frame is RA . Each TDMA composed by a fixed amount of slots Mslots RA slot of duration Nslot symbols can allocate one RA burst RA composed of Npre acquisition preamble symbols, followed by RA Npay payload symbols. The guard-time is required in practice to compensate for the incoming TDMA burst timing errors. RA We will assume that Nguard guard symbols are allocated in RA RA RA RA + Npre + Npay . The RA the slot so that Nslot = Nguard RA frame duration in symbols then corresponds to Nframe = RA RA RA Nslot Mslots and TF = Nframe Ts . Let us now describe the signals characterizing the RA channel behavior. We refer to a generic RA frame, thus for notation simplicity we drop the dependency on the frame index. We also represent discrete signals samples at symbol distance and we assume for notation simplicity without loss of generality that relative STs burst delays are occurring in integer multiples of the symbol period itself. Let’s now consider the generic discrete burst signal samples array s[i, n] generated by ST # i in slot # n. s[i, n] is composed by a preamble sub-array spre [i], a payload subarray spay [i, n] and an empty guard time sub-array sguard so that: RA Nslot

s[i, n] = spre [i] =

    PTx [i] [spre [i], spay [i, n], sguard ], (1)   c1 [i], c2 [i] . . . cNpre RA [i] , (2) RA Nguard

sguard

=

spay [i, n] =

   [0, 0, . . . 0] 1 √ [dp,1 [i, n] + jdq,1 [i, n] . . . 2  RA [i, n] + jdq,N RA [i, n] . . . dp,Npay pay

(3)

(4)

where cl [i] is the l-th symbol of the preamble binary (±1) BPSK modulated sequence (in the more general case the preamble can also be QPSK modulated) and dp,l [i, n] and dq,l [i, n] are the l-th in-phase and quadrature binary (±1) payload symbols, respectively. The payload dependency on the slot n is due to the twin burst signalling information relation to the current burst location. Assuming that the delay, amplitude and phase of the received signal at the gateway remains constant over a TDMA slot, the received signal samples can

1410

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 4, APRIL 2007

be written as: r[n] =

N ST

δ[i, n]L[i, n]s[i, n]z −D[i,n]

i=1

· exp {j (φ[i, n] + Δω[i, n] t[n])} + w[n] (5) where NST represents the total number of registered STs, δ[i, n] is 1 if the i-th terminal is active in slot # n and 0 otherwise, L[i, n] < 1 represents the signal attenuation, RA is the differential TDMA ST slot delay 0 ≤ D[i, n] ≤ Nguard −D[i,n] is the delay operator shifting towards the in symbols, z right the array s[i, n] by D[i, n] positions (symbols), φ[i, n] and Δω[i, n] represent the carrier phase and frequency offset respectively, t[n] is the time corresponding to the start of RA elements each slot n and w[n] is a complex array of Nslot representing a circular symmetric white Gaussian noise with 2 . We also define the received preamble for user # variance σw i in slot # n as the following sub-array derived from r[n]:   r pre [n, i] = rD[i,n]+1 , rD[i,n]+2 , . . . rD[i,n]+Npre RA .

(6)

In a similar way we introduce the received burst payload subarray as:  r pay [n, i] = rD[i,n]+Npre RA +1 , rD[i,n]+N RA +2 , . . . pre  (7) . . . rD[i,n]+Npre RA +N RA . pay For our analysis we will assume that the power of the ST burst received in slot # n from user # i is given by PRx [i, n] = L2 [i, n]PTx [i]. PRx [i, n] is modelled by a lognormal r.v. which is characterized by a mean power P Rx [i] and lognormally distributed with standard deviation in dB σPRx [i] . We also define the received burst signal amplitude at the gateway as   ARx [i, n] = PRx [i, n] = L[i, n] PTx [i]. The lognormal received power distribution well approximates the combined effects of the time variant atmospheric propagation, open loop power control errors (if applicable), ST EIRP and satellite receive antenna gain variations. The

number of packets present NST δ[i, n]. in slot # n corresponds to N [n] = i=1 B. Burst Preamble Issues As we have seen in Section III, compared to known solutions the performance boost of the CRDSA protocol is achieved thanks to the implementation on the gateway burst demodulator of a contention resolution capability. In fact, if one of the twin bursts transmitted in the frame is successfully decoded, the information about the replica burst location allows resolving the possible generated collisions by exploiting interference cancellation (IC) techniques. IC techniques have been largely investigated for Code Division Multiple Access (CDMA) [14] but, at the authors knowledge, have never been proposed in a TDMA SA context. One of the main issues in applying IC techniques to TDMA SA is related to the need for accurate channel estimation for the burst where collision(s) occur. In fact, collisions in a (satellite) TDMA multiple access channel are typically destructive. This is particularly true in satellite networks whereby the near-far effect is typically very limited.

Contention resolution is achieved in CRDSA by means of IC which requires good channel estimation for colliding burst(s) removal. As detailed in Section III-C, while carrier frequency, amplitude and timing estimation for IC can be derived from the twin "clean" burst, carrier phase has to be estimated on the slot where collision(s) occurs. This is because the phase in practical broadband systems is time variant also from slot to slot. This key problem has been solved here by exploiting the burst preamble which is now individually "signed" by a pseudo-random binary sequence spre randomly selected among the available code family by each active ST for each burst in each frame. Both frame replicas (twins) of the same burst use the same preamble code. This approach does not require a centralized preamble code assignment thus allowing to maintain the random access nature of the proposed scheme. In this way the preamble can be used for carrier phase estimation also in case of multiple collisions which normally are destructive for channel estimation and payload decoding. The preamble "signature" sequence is randomly selected out of a set belonging to a known family of size SPR . The family RA of signature sequences of length Npre shall provide good auto and cross-correlation properties and having a family size RA . Due to comparable to the preamble length i.e. SPR  Npre the TDMA slots random timing offset, cross-correlation with colliding bursts will also involve the first symbols of the payload. It is important to note that despite using a pseudo-random sequence, the proposed CRDSA scheme significantly differs from a code-time two-dimensional Aloha scheme (e.g. Spread Aloha). CRDSA is basically a TDMA access scheme that uses the information from the successfully decoded packets to cancel the interference their replicas may generate on other slots. We are able to cancel the interference because we know the data symbols of the interference from the successfully decoded packets. In CRDSA, a pseudo-random sequence has been applied only to the preamble (and not the payload data field) for the solely purpose of deriving the carrier phase information of the un-clean burst. In a code-time two-dimensional Aloha scheme such as Spread Aloha, packets are sent only once (no replicas are sent) and a spreading code is applied over the whole packet (preamble and payload) with a certain spreading factor. It is important to observe that as the preamble sequences are randomly selected by the ST modulator from a finite family of SPR codes, there is a non zero probability of reusing the same preamble sequence within the same slot from one or more STs simultaneously transmitting a burst. However, due to the good off-peak autocorrelation property of the preamble sequence, the event will be catastrophic only when the two ST bursts reusing the same preamble sequence are arriving with an absolute differential delay of less than 1 symbol. Assuming that the differential delay among STs is uniformly distributed between [0, Dmax ] the probability of preamble collision in slot n can be computed as1 (see Appendix I): RA G·Mslots −1

Ppre−coll (G) =



Pint (i| G)

i=1 1 In Section V-B it is shown that the preamble collision probability is typically lower than MAC packet loss ratio.

CASINI et al.: CONTENTION RESOLUTION DIVERSITY SLOTTED ALOHA (CRDSA)

⎤ ⎡ i  i ·⎣ · pj · (1 − p)i−j ⎦ j j=1 RA G·Mslots −1

=



  Pint (i| G) · 1 − (1 − p)i

(8)

i=1

where G is the normalized MAC load measured in packets RA per slot, G · Mslots is the MAC load measured in packets per RA frame (assumed to be an integer), p is the probability that the same sequence and differential delay of the useful burst are chosen by an interfering burst (i.e. p = 1/ [SPR · Dmax ]) and Pint (i| G) is the probability that i interfering packets are present in the same slot n (see Appendix II). The normalized MAC load G does not take into account the replicas; this normalization has been chosen in order to facilitate comparison with other access schemes (e.g. SA). Instead the physical channel load takes into account the CRDSA physical layer burst replicas, therefore a normalized MAC load G=0.5 corresponds to a physical channel load of 1 in RA ) also CRDSA. It is remarked that for CRDSA, (G · Mslots represents the maximum number of physical packets that can be present over one slot as both copies of the same MAC packet cannot be sent over the same slot. It is important to observe that with the proposed CRDSA scheme, the gateway TDMA demodulator is used in a "conventional" manner i.e. it successfully operates when there is a clean burst. In this situation the TDMA demodulator is capable to recover the timing and carrier frequency offset due to the offset caused by STs relative synchronization errors. The key differences is that max for the same RA frame the burst demodulator is used Niter times due to the proposed successive cancellation algorithms that allows to increase the throughput. This iterative burst demodulator exploitation is made possible by the fact that the input frame complex signal samples are initially stored to allow for successive burst "cleaning". This process is described in the following section. C. Interference Cancellation Algorithm As stated in Section IV-B, the CRDSA burst demodulator will store in memory the baseband samples corresponding to a full RA frame duration in order to perform an iterative decoding process. The CRDSA demodulator iteration counter is set initially to Niter = 1. At each iteration, the demodulator will perform the following steps: 1. Demodulation and decoding of clean bursts: By clean bursts we mean those bursts for which the signal, noise and interference level allow to achieve preamble recognition and payload decoding (e.g. packet 3 in slot 5 in Fig. 1). (a) In this step the gateway burst demodulator shall search in parallel for each slot of the full frame for all the SPR possible burst preambles. This can be efficiently achieved by implementing a bank of preamble sequence acquisition units similar to the one used in CDMA packet networks [15]. Typically the preamble time search region is limited to the guard time period around the nominal preamble location in each frame slot. Once the presence of one or more preamble sequences are recognized in the slot by the multi preamble searcher, the burst demodulator will estimate, as for a conventional one, the burst

1411

channel parameters (clock timing, carrier frequency and phase) and attempt to decode the payload content. If the preamble is detected and the burst payload Code Redundancy Check (CRC) verification is successfully passed, then the recovered burst is declared as "clean". At this point the operation of a conventional burst (D)SA demodulator will terminate. We assume that there are Nrecov (Niter ) bursts recovered at the current iteration. (b) When a burst is successfully decoded it can be fully regenerated at complex baseband level by re-encoding and modulating the decoded useful bits multiplexed with the current burst slot location signalling bits. In the twin burst regeneration the slot nr where the "replica" of the burst was transmitted (e.g. packet 3 in slot 4 in Fig. 1) is derived from the burst payload signaling information bits. Furthermore, the acquisition preamble binary signature sequence and its timing are extracted from the burst demodulator preamble code correlator and timing estimation unit respectively. (c) The twin burst signalling information is protected by the same FEC of the useful payload bits thus it is correctly recovered when the CRC check is positive. Detected clean burst(s) twin(s) location within the frame is stored and used for next step jointly with their amplitude and clock information derived from the clean burst detection. The phase information extracted from the clean burst cannot be used for the twin burst being the carrier phase typically uncorrelated from burst-toburst because of the local oscillator instabilities. 2. Contention Resolution Algorithms: Following the previous step, the CRDSA demodulator will now process the slots where the replica burst of the "clean" bursts were transmitted and which have not been already detected in the previous step (i.e. step 1-(a)) of the current iteration (e.g. packet 3 in slot 4 in Fig. 1). So the demodulator will now operate on the slots where collision(s) occurred (i.e. whereby more than one burst were simultaneously transmitted and destructively interacting). The CRDSA algorithm aims at post-processing the stored frame samples to resolve contention in some of the slots where collisions occurred. Let us assume that in the current frame and iteration Niter , the set of bursts identified by index q = [q1 , q2 , . . . qNrecov (Niter ) ]2 corresponding to STs i = [i1 , i2 , . . . iNrecov (Niter ) ] have been successfully decoded in slots n = [n1 , n2 , . . . nNrecov (Niter ) ]. We also assume that the replicas of bursts q are located (according to clean packet signaling information) in slots nr = [nr1 , nr2 , . . . nrNrecov (Niter ) ] belonging to the same frame. (a) We assume that the successfully detected clean bursts q from STs i in slots n provides an accurate estimate of the signal amplitude AˆRx [ik , nk ], k = 1, 2, . . . Nrecov (Niter )  k , nk ], k = and of the angular carrier frequency offset Δω[i 1, 2, . . . Nrecov (Niter ). The carrier phase between two successive bursts sent from the same ST is typically uncorrelated mainly because of the fast phase noise components of the ST. Received burst frequency, timing and amplitude coming from the same ST can instead be assumed almost constant within  k , nk ], A[i ˆ k , nr ]  A[i ˆ k , nk ]  k , nr ]  Δω[i a frame thus Δω[i k k 2 As described in the previous paragraph the set q contains only those clean bursts for which their replicas have not been detected yet at iteration Niter .

1412

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 4, APRIL 2007

and τˆ[ik , nrk ]  τˆ[ik , nk ] for k = 1, 2, . . . Nrecov (Niter ). For notation simplicity in the following the carrier frequency offset will be dropped. (b) The amplitude information of the replica burst slot nrk , k = 1, 2, . . . Nrecov (Niter ) can be accurately estimated from the twin "clean" burst which has been sucˆ , nr ]  A[i ˆ k , nk ]  cessfully detected in slot nk as: A[i   ∗ T  k k 1  sˆpay [ik , nk ] , where the complex array RA r pay [ik , nk ] · Npay sˆpay [ik , nk ] represents the estimated payload transmitted symbols as derived by re-encoding at the CRDSA demodulator the decoded bits, the operator T indicates array transposition and ∗ the complex conjugate. Having assumed a correct decoding of the payload encoded bits it follows that sˆpay [ik , nk ] = spay [ik , nk ]. (c) For each replica burst slot nrk , k = 1, 2, . . . Nrecov (Niter ), the carrier phase information corresponding to this slot for user ik can be derived by correlating the stored slot nrk soft samples rpre [ik , nrk ] with the user ik preamble seRA r ˆ quence pre symbols as: φ[ik , nk ]   spre [ik ] of length N T where we assumed that the arg r pre [ik , nrk ] · s∗pre [ik ] burst timing offset error is negligible. This timing estimate can be based on the successfully detected packet from user ik detected in slot nk . (d) Colliding burst from user ik in slot nrk can now be removed by IC (e.g. in Fig. 1 packet 3 can be removed from slot 4):

A B and Ppd correspond respectively to the probability where Ppd that the twins A and B of the same packet are successfully decoded. By symmetry, the two probabilities are equal i.e., A B A = Ppd . An upper bound for the probability Ppd can be Ppd recursively derived as follows: RA G·Mslots −1

A (Niter | Ppd

G) ≤

PalA (G)

+



Pint (i|G)

i=1

 B i · Ppd (Niter − 1| G)

(11)

where PalA (G) represents the probability that the packet is alone in the slot (or "clean" i.e. no other interfering burst is arriving in the same slot) and Pint (i|G) represents the probability that the useful burst is colliding with i interfering B represents the probability that the bursts on the same slot, Ppd replicas of the interfering packets are successfully decoded and RA G · Mslots − 1 represents the maximum number of interfering packets that can be present in one slot. By recalling that A B = Ppd we get the following recursive equation: Ppd RA G·Mslots −1

A Ppd (Niter |

G)



PalA (G)

+



Pint (i|G)

i=1

 A i · Ppd (Niter − 1| G)

(12)

A with an initial value of Ppd (0) = 0. The derivation of Pint and A P is provided in Appendix II. The first iteration corresponds al r r ˆ k , nk ] r[nk , Niter + 1]  r[nk , Niter ] − A[i    to the case of no interference cancellation and yields the same ˆ k , nr ] + Δω[i  k , nk ] t[nrk ] performance of DSA [3]. The second iteration corresponds · exp j φ[i k   to one interference cancellation iteration and so forth until (9) · sˆpre [ik ], sˆpay [ik , nk ] max iterations are accomplished. Equation (12) represents an Niter upper bound as the probability of occurrence of the so called (e) Increase the iteration counter as: Niter = Niter + 1. (f) Having introduced the CRDSA demodulator maximum "loops" has not been considered. A loop can occur when, in the max max , if Niter = Niter then stop, else recursive process of decoding a packet, we find an interfering number of iterations Niter packet which is either a replica of the packet of interest or go to step 1-(a). a replica of any of the previously found interfering packets. In these situations a loop (or deadlock) occurs thus making D. Analytic CRDSA Throughput Upper Bound Derivation the recursive contention resolution process impossible. For In this section, a bound for the throughput of the CRDSA example, in the frame shown in Fig. 1, a loop occurs between access scheme is derived. All the subsequent results are ex- packets 4 and 5, i.e. to recover packet 4 in slot M RA − 1, we slots pressed as a function of the normalized MAC load G measured need to cancel the interference from packet 5, but to cancel the in packets per slot. It should be recalled that the CRDSA and interference from packet 5 we need the information from its DSA physical layer load is double compared to SA because of twin copy which is in slot M RA − 3. Unfortunately, packet 5 slots the twin burst transmission. The following assumptions have in that slot has an interfering packet which is the twin copy of been made for the derivation: a) The probability of preamble packet 4, so a loop occurs between them and they cannot be be collision is assumed to be negligible (this hypothesis will recovered. The situations whereby occurrence of loops takes be validated in Section V-B); b) The MAC load per frame, place can be very diverse (e.g. different number of packets can RA ), is assumed to be be involved in the loop) and thus cannot be easily modelled in measured in packets per frame (G · Mslots an integer; c) The demodulator can perform a maximum of the analytical model. Therefore, the current analytical model max Niter iterations. represents an upper bound for the performance of CRDSA. The throughput T at the iteration Niter for a load G In Section V the tightness of the analytical throughput upper measured in packets/slot, can be computed as: T (Niter| G) = bound for the region of interest will be demonstrated. G Ppd (Niter | G), where Ppd (Niter | G) = P { packet successfully decoded at iteration Niter | load =G}. The probability IV. I MPLEMENTATION C ONSIDERATIONS Ppd can be derived as: The definition of the location signalling information inside   A (Niter | G) Ppd (Niter | G) = 1 − 1 − Ppd the MAC packets will be specific to each satellite air interface   B standard. In general, a field of one byte (i.e. a RA frame of · 1 − Ppd (Niter | G)   up to 256 slots) would be sufficient. In the case of traffic 2 A (10) bursts containing one or more ATM cells the overhead shall (Niter | G) = 1 − 1 − Ppd

CASINI et al.: CONTENTION RESOLUTION DIVERSITY SLOTTED ALOHA (CRDSA)

be ≤ 1.8% and for the case of a traffic bursts containing one or more MPEG2 cells the overhead shall be ≤ 0.5%. Unused existing fields could also be redefined for this purpose yielding 0% overhead. When transmitting very short bursts (e.g. 14Byte long bursts such as SYNC bursts in the case of a DVBRCS system), the overhead due to the replica pointer may become non-negligible (e.g. 1Byte/14Byte, which is about 7% overhead). In the following, a possible way to apply CRDSA for a DVB-RCS type of system is shortly discussed. First, for logging into the system the current Slotted Aloha mode proposed by the standard is used. Once the ST has been logged into the satellite network and the ST timing error has been reduced within the acceptable boundaries trough standard ST synchronization procedures, the random access channel can operate in CRDSA mode. RA shall be selected to transmit small size or infrequent packets while DA shall be selected when the ST buffer exceeds a certain threshold. A short functional description of the modulator and demodulator operations is presented in Fig. 2. Functions not present in a standard (D)SA burst modulator/demodulator will be indicated in italic in the text and in the block diagram. A. CRDSA Modulator Operation The CRDSA modulator depicted in the upper part of Fig. 2 represents a simple variant of a classical MF-TDMA burst modulator. In fact, incoming information packets are first buffered and then segmented into fixed size MAC packets. MAC packets are then encoded, modulated and located into two random slots (ni and nri ) of a generic frame at a certain carrier frequency ω[ik ] according to the ST burst time plan (TBTP) that is broadcasted to all STs through the satellite downlink signalling channel [4]. It should be remarked that for a given number of useful payload bits, the overall CRDSA ST burst duration is more than doubled compared to a conventional burst modulator because of the transmission of the twin burst within the same frame plus the small twin location signalling overhead. For each frame the mapping of the ST RA twin bursts onto the TBTP is randomly generated and signaled in the burst payload field reserved bits (twin burst signalling information generator and multiplexer). The burst accommodates a preamble composed of a pseudo-random (e.g. Gold) sequence followed by the complex payload symbols. As detailed in Section III-B, the preamble pseudo-random sequence is the same for the twin bursts and it is randomly selected by the ST among the SPR available preamble sequences. This is a key difference with respect to conventional (D)SA whereby the same preamble sequence is used by all STs. The payload symbols are the results of the coding and modulation of the useful information bits and twin location signalling information. B. CRDSA Demodulator Operation The CRDSA demodulator is depicted in the lower part of Fig. 2. The local oscillator converts the input (MF-TDMA) signal coming from the RF front-end to a low IF frequency. The ADC then converts the low-IF input analogue signal in

1413

digital samples for further processing. The digital demultiplexer is able to separate the multi-carrier MF-TDMA input signal in different individual carriers. The frame baseband digital complex samples corresponding to the different MFTDMA carriers separated by the digital demultiplexer are then stored in a digital memory encompassing at least 2 times the full RA frame duration. In this way once a full frame is stored the iterative CRDSA burst processing can be performed at higher speed while next frame’s samples are being collected in the remaining available memory. The overall demodulation processing delay is therefore bounded to less than one frame. The frame digital samples stored are read in a sequential order corresponding to the individual MF-TDMA slots. For each slot of the frame the multiple preamble searcher looks in parallel for the possible SPR preambles presence in order to detect the presence and the start time of the bursts. The burst demodulator performs an initial burst carrier frequency and phase estimation based on the preamble. The same unit also extracts the symbol timing and performs symbol matched filtering to deliver the burst I-Q baseband samples at symbol rate. Typically, a more accurate phase estimation is performed by processing the payload symbols to enhance the initial preamble-based phase estimate. The payload Forward Error Correcting (FEC) decoder extracts the burst payload useful information bits as well as the information about the location of the "twin" burst within the frame. The useful and signalling bits are then separated by a demultiplexer. The estimated burst information bits coming out from the FEC decoder after multiplexing with the current burst slot location signalling bits are re-encoded by the payload FEC encoding block to provide the twin burst coded bits information to the replica burst generation. The encoded payload bits together with the information provided by the preamble searcher, the payload burst channel estimate (frequency, amplitude, clock) and the twin preamble phase estimator are used by the twin burst regeneration unit to produce the replica burst in the right slot location. The twin burst interference cancellation unit iteratively removes possible burst collisions to allow DSA performance boost through the contention resolution process. V. N UMERICAL R ESULTS A CRDSA system simulator has been developed where a number of independent traffic generators are generating traffic bursts and driving the TDMA CRDSA modulator. Each modulator burst carrier phase is a r.v. uniformly distributed over [0, 2π) and constant over the burst. All modulators are affected by independent amplitude fluctuations with programmable statistics and adding up after a delay corresponding to the ST to gateway propagation delay (about 250 ms for the GEO satellite considered in the following). When not stated otherwise, the ST transmitted power lognormal r.v. is assumed to have a mean and standard deviation of 0 dB. Complex AWGN noise samples addition follows. The noisy TDMA signals are then entering the CRDSA demodulator for further processing. To properly assess the CRDSA scheme performance, the channel estimation for the useful "clean" burst demodulator is considered ideal. In all simulations, a RA frame size of 100 slots has been assumed. Longer RA frames do not provide any further significant performance

1414

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 4, APRIL 2007

TRAFFIC PACKETS QUEUE

MAC PACKET SEGMENT

BUFFER

PAYLOAD FEC CODING & MODULATOR

MUX

TWIN BURST SIGNALLING INFORMAT GENERATION

BASEBAND SRRC FILTERS

BURST MUX

VARIABLE PREAMBLE GENERATION

CARRIER LOCAL OSCILLATOR

TWINS BURST MODULATOR CONTROLLER

MULTIPLE ACCESS CHANNEL

USEFUL PAYLOAD BITS

AWGN FEC DECODER ADC

DIGITAL DEMUX

FRAME SAMPLES MEMORY

P/S CONVER TER

BURST DEMODULATOR

DEMUX

CURRENT SLOT SIGNALLING BITS TWIN SLOT SIGNALLING BITS

FEC ENCODING PAYLOAD CHANNEL ESTIMATE

CARRIER LOCAL OSCILLATOR

MULTI PREAMBLE SEARCHER TWIN BURST INTERFERENCE CANCELLATION PROCESSOR

TWIN PREAMBLE PHASE ESTIMATOR

TWIN BURST PREAMBLE SEQUENCE

TWIN BURST REGENERATION

TWIN BURST PHASE ESTIMATE

BURST DEMODULATOR CONTROLLER

Fig. 2. CRDSA modem functional block diagram. In dark grey the new blocks compared to a conventional TDMA burst demodulator. In light grey the modified ones.

improvement. A MAC packet corresponds to 1 ATM cell. For the simulations in Section V-A, Poisson and Web Client traffic sources have been used. These models are considered representative of the return link traffic generated from small terminals with low level of traffic aggregation. The Web Client traffic source has been derived from [16], but assuming an exponential packet length distribution with a mean of 40 bytes. The performances achieved are so similar between both traffic sources that no distinction has been made between them in the results obtained. For the simulations in Sections V-B and V-C, Poisson traffic sources have been assumed.

A. Results Assuming Ideal Channel Estimation for Interference Cancellation Performance results from Figs. 3 and 4 have been obtained in the following way. For all studied random access schemes (SA, DSA and CRDSA), we have assumed an open loop transmission scheme. Therefore, no congestion control or retransmission mechanisms have been applied, and packets are sent by the transmitters as they arrive after a randomized time. This is the reason why after a certain system loading, all the schemes saturate and performance (throughput) starts to collapse (i.e. too many collisions occur on the channel).

CASINI et al.: CONTENTION RESOLUTION DIVERSITY SLOTTED ALOHA (CRDSA)

1415 0

0.7

10

Simulated CRDSA Analytical CRDSA Simulated Slotted−Aloha 0.6

Niter=16 −1

10

0.5

Niter=6 Packet Loss Ratio PLR

Throughput

Niter=3 0.4 Niter=2 0.3

o −2

X

10

o

o X

X

Niter=1 −3

10

0.2

T

T

T

CRDSA SA DSA

0.1 −4

10

−3

10

0

−2

−1

10

10

0

10

Normalized load [G]

0

0.1

0.2

0.3

0.4 0.5 0.6 Normalized Load [G]

0.7

0.8

0.9

1

Fig. 3. Simulated (circles) and analytical (dotted line) results for the CRDSA throughput (ideal channel estimation for IC) versus the normalized channel loading for Niter = 1, 2, 3, 6, 16. Slotted Aloha (SA) (continuous line) and Diversity Slotted Aloha Niter = 1 performance is also reported for comparison.

The randomization time corresponds to one frame (i.e. a slot is chosen randomly within one frame). The performance parameter used in Fig. 3 is throughput (measured in useful packets received per slot) vs. load (measured in useful packets transmitted per slot). One slot can carry one data packet, that corresponds to one ATM cell in our simulations, for all schemes. In Fig. 3, the throughput of the CRDSA protocol has been first simulated versus the normalized MAC load for a variable number of maximum iterations in the contention resmax = 1, 2, 3, 6, 16) and initially under olution process (Niter the assumption of perfect channel estimation for IC. Computer simulation results are compared in Fig. 3 to the analytical throughput upper bound calculation reported in Section IIID. As expected, for the reasons explained in Section III-D, simulation results are slightly below the bound in particular for normalized loads close to the maximum throughput and when the number of iterations is large. This corresponds to situations when "loops" occurrence probability is non negligible. It is interesting to observe that there is a diminishing return advantage increasing the number of CRDSA demodulator max max . The choice of Niter = 10 appears to achieve iterations Niter most of the CRDSA recursive algorithm potential gain. The results obtained for SA in our simulations correspond to those reported in [2]. To remark that while standard SA reaches its throughput peak of 0.36 for a normalized MAC load G = 1, CRDSA achieves a peak throughput of about 0.52 for a normalized MAC load of 0.65. In addition, CRDSA achieves a linear throughout (with almost no packet losses) up to a normalized channel load of 0.4, while SA achieves a similar behavior only up to 0.1 load. Let us now define the MAC packet loss ratio as: being G the normalized load and PLRMAC (G) = 1 − T (G) G T the normalized MAC throughput. Fig. 4 reports the MAC packet loss ratio (PLRMAC ) for the SA, DSA and CRDSA cases. In this figure we can see that in open loop conditions (i.e. no re-transmissions applied to any scheme), CRDSA has a much lower PLR on all loads, therefore it is a much more

Fig. 4. MAC Packet Loss Ratio for CRDSA (continuous line), DSA (dashed) and SA (dashed dot) with ideal CRDSA channel estimation for IC. Different markers (O,X,T) correspond respectively to PLRMAC = 2 · 10−2 , 10−2 , 10−3 for each random access technique.

reliable RA scheme that SA and DSA. The results show that the CRDSA channel can be loaded 50 times more that the SA channel if we want to achieve a PLRMAC = 10−3 , 26 times more to achieve PLRMAC = 10−2 , and 17 times more if we want to achieve a PLRMAC = 2 · 10−2 . Alternatively, we could say that for a given load G = 0.35, the SA losses (PLRSA MAC (0.35) = 0.3) are 15 times higher that the CRDSA −2 ). losses (PLRCRDSA MAC (0.35) = 2 · 10 In Fig. 5, a re-transmission scheme has been introduced to the CRDSA technique for those packets experiencing collisions and that cannot be recovered by means of interference cancellation techniques at the gateway. It is expected that the losses around the nominal operational point in CRDSA will be very low (e.g. below 0.02). As a consequence, the volume of retransmitted packets on the channel is also expected to be very modest without affecting the average channel load. In order to increase the probability of success for the retransmitted packets and considering the low volume of those packets on the channel, a double retransmission mechanism is suggested for satellite applications, i.e. retransmitted packets are sent twice on two different RA frames. This solution will approximately provide the performance loss of a two-packet retransmission mechanisms and the delay performance of one packet retransmission for the retransmitted packets. Numerical results corresponding to this retransmission mechanism are presented in Fig. 5. The packet loss ratio is significantly improved with a double retransmission mechanism (by more than 3 orders of magnitude at G = 0.35 load and more than 2 orders of magnitude at G = 0.4). The average delay is relatively not affected as the amount of re-transmitted packets represents a small percentage of the total transmitted traffic on the channel (see delay PDF distribution in Fig. 6). B. Results on Burst Preamble Destructive Collision Probability In Fig. 7, the estimation of the analytical formula for the probability of preamble collision Equation (8) is compared with the results of the simulation. This exercise has been done

1416

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 4, APRIL 2007 −1

0.014

10

analytical S PR=31 analytical S PR=63 simulated results

0.012 −2

Probability of preamble collision [Ppre−col l ]

Packet Loss Ratio

10

−3

10

−4

10

0.01

0.008

0.006

0.004

−5

10

0.002

2 re−TX 1 re−TX 0 re−TX

−6

10

0.3

0.32

0.34

0.36

0.38 Normalised Load

0.4

0.42

0.44

0

0.46

Fig. 5. Simulated CRDSA packet loss ratio with double, single and no re-transmissions.

0

0.2

0.4

0.6 Normalized load [G]

0.8

1

Fig. 7. Analytic and simulation results for preamble collision probability for SPR = 31, 63 , Dmax = 5 and Es /N0 = 6 dB. 0.6

0.12 Delay PDF at 0.4 normalized load

0.55

SNR = 8 dB SNR = 6 dB

0.5

0.1

SNR = 5 dB

0.45

0.08 Throughput

0.4

0.06

0.35

0.3

0.25

0.04

0.2

0.02

0.1 0.1

0 0.2

ideal ch. est. SNR = 5 dB SNR = 6 dB SNR = 8 dB

0.15

0.4

0.6

0.8

1

1.2

1.4

0.2

0.3

0.4

0.5 0.6 Normalized load [G]

0.7

0.8

0.9

1

1.6

Delay (s)

Fig. 6. Simulated CRDSA delay PDF distribution for normalized load G = 0.4 with a double retransmission mechanism.

assuming a set of Gold sequence of length SPR = 31 and Dmax = 5. As we can see, for a load G = 0.4, Ppre−coll = 5 · 10−3 . This probability is one order of magnitude lower than the PLRMAC for the same load, thus it is considered to be acceptable. The probability of preamble collision be could halved by using twice the number of codes (i.e. SPR = 63) as shown on the same Fig. 7.

C. Results Assuming Non-Ideal Channel Estimation for Interference Cancellation Next experiment takes into account the non ideal channel estimation effects in solving the channel contentions in the CRDSA burst demodulator. For this simulation the presence RA = 5 symbols, a preamble of of a burst guard time of Nguard RA RA = 424 symlength Npre = 31 symbols and a payload of Npay bols (corresponding to 1 ATM cell with a convolutional code

Fig. 8. Simulated ideal channel estimation (continuous line) and real channel estimation for IC (dashed dot line) results for the CRDSA throughput versus the normalized channel loading for Niter = 10 and Es /N0 = 5, 6, 8 dB.

rate r = 1/2 [1338 , 1718] and QPSK modulation3) have been chosen. For each burst and its duplicate in case of CRSDA, the BPSK preamble sequences are randomly extracted out of the SPR Gold sequences available. For deciding if the packet is correctly received we look if the decoder has correctly decoded the payload burst data. The performance of the chosen code on a AWGN channel are such that BER=10−5 is achieved when Es /N0 = 4.2 dB. Channel estimation for IC is performed according to the algorithms described in Section III-C and we recall that the carrier phase is randomly generated for each burst of the frame even if the bursts are coming from the same ST. The simulated throughput results versus the normalized load for various Es /N0 values and Niter = 10 are reported in Fig. 8 and compared to the case of CRDSA with ideal channel estimation. It is apparent that the imperfect 3 We assume that this physical layer configuration represents a worst-case for the channel estimation in the burst demodulator. The preamble length of 31 symbols is also a worst-case for channel estimation as for this coding rate and modulation scheme typically a longer preamble (≥ 48) is required to provide acceptable burst synchronization performance.

CASINI et al.: CONTENTION RESOLUTION DIVERSITY SLOTTED ALOHA (CRDSA) 14

12

amplitude estimation 10

PDF

8

6 phase estimation error [rad] 4

2

0

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Fig. 9. Simulated CRDSA preamble amplitude and carrier phase estimation RA error for Niter = 10, Es /N0 = 6 dB, G=0.4 and Nguard = 5.

1417

to observe the scatter diagram of the baseband signal in the absence of AWGN (Es /N0 = ∞) for a specific slot of the max TDMA frame when multiple collisions occur for Niter =1 max (see Fig. 10(a)) and for Niter = 10 (see Fig. 10(b)). The remarkable cleaning of the baseband QPSK signal scatter diagram thanks to the recursive IC approach is apparent. The impact of imperfect power control or ST power unbalance on the CRDSA protocol performance is shown in Fig. 11 for Es /N0 = 6 dB. In the simulations it is assumed that the ST power is lognormally distributed with 0 dB mean and standard deviation σPrx = 1, 2, 3 dB. We observe that the presence of other STs received power unbalance at the gateway burst demodulator results in a modest degradation of throughput performance that is more evident in the maximum throughput region, G ∈ [0.5−0.8]. Nevertheless, for the nominal operating region, G ≤ 0.4, the loss with respect to the ideal case is almost negligible when σPrx ≤ 2 dB. VI. C ONCLUSION

channel estimation is causing a slight performance degradation which is more noticeable (about 5 %) for normalized load in excess of 0.45 which are not used in practice. For a realistic operational normalized load G = 0.35 corresponding to PLRMAC = 0.02, the impact of the imperfect channel estimation on the throughput is negligible4 . To better understand the causes of degradation it is useful to analyze the key channel parameter estimation errors. Fig. 9 shows the preamble-based carrier phase estimator PDF used for cancelling the colliding burst when Es /N0 = 6 dB and G = 0.4. For this reason the PDF of the carrier phase estimation error is evaluated on the preamble of bursts that have experienced at least one collision. The estimator phase error is a combination of several factors such as the AWGN noise and the colliding bursts preamble cross-correlation effect. The cross-correlation impact increases with the amount of packets N [n] present in slot n. It is also interesting to observe the estimated amplitude error distribution. Following Section IIIC step 2, amplitude estimation takes place on the whole burst payload using the FEC-based regenerated channel symbols. Thus it is expected the amplitude estimation is accurate and very close to a data-aided amplitude estimator. This intuition is confirmed by the simulation results of Fig. 9. Compared to the phase estimator operating on a shorter preamble sequence RA = 31 symbols, the amplitude estimator is of length Npre further advantaged by the fact that it operates on "clean" burst so in the absence of collisions. It is apparent that the main contributor to the contention resolution channel estimation error is the carrier phase5 . To better understand the capability of the proposed CRDSA contention resolution demodulator processing it is interesting 4 Simulation runs indicated that even simulating realistic time variant over the burst ST phase noise process according to the DVB-RCS mask [13], no appreciable impact on the CRDSA performance are experienced for Rs ≥ 128 Kbaud. 5 It can be shown that the equivalent signal to interference ratio caused by the carrier phase of the burst to be removed can be approximated by SIR  1/σ2 ˆ , Δφˆ being the burst phase estimate error. This contribution Δφ is normally negligible compared to the AWGN impairment when no phase noise is present.

In this paper the so-called Contention Resolution Diversity Slotted Aloha (CRDSA) scheme has been introduced and in depth analyzed. It is shown that CRDSA provides a major boost in performance compared to known random access (RA) techniques currently used by TDMA satellite systems such as SA and DSA. By simple modification of the modulator MAC scheme, CRDSA allows to achieve a 17 and 4.5 fold throughput increase compared to SA and DSA respectively for a MAC packet loss ratio of 2 %. Higher gains are achieved at lower MAC PLR (e.g. 50-fold at PLR=10−3 ). An upper bound for the CRDSA throughput has been derived and successfully compared to simulation results. Channel estimation for interference cancellation is obtained by exploiting the burst preamble which is now "signed" by sequences belonging to a family of quasi-orthogonal sequences. Simulation results showed that for realistic preamble lengths used in today TDMA systems, the channel estimation error impact is negligible for realistic operating conditions. Also the probability of preamble collision was analytically derived and results compared to simulation findings. It is shown that this major performance enhancement over known techniques can be achieved with a minor overhead increase. It is demonstrated that the CRDSA contention scheme can be practically operated at normalized loads of up to 0.4, thus allowing the transmission of small/medium sized bursty packets without allocating resources as often done in today satellite broadband networks. The impact of received power errors among STs was found to be fully acceptable. A possible packet retransmission mechanism for lost packets has been also described making the random access channel highly reliable (P LR = 10−4 ) with little impact on the overall channel performance in terms of throughput and delay. A PPENDIX I P ROBABILITY OF P REAMBLE C OLLISION D ERIVATION The probability of preamble collision is the probability that a packet suffers from the collision with other packets that have the same preamble sequence and the same time delay. Considering i packets interfering with the reference packet,

1418

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 4, APRIL 2007

Scatter diagram received Symbols, N

=1

Scatter diagram received Symbols, N

iter

=10

iter

4

1 0.8

3

0.6 2 0.4 0.2

imag

imag

1 0

0 −0.2

−1

−0.4 −2 −0.6 −3

−0.8

−4 −4

Fig. 10.

−3

−2

−1

0

1

2

3

−1 −1

4

(a) Niter = 1

(b) Niter = 10

0.5

1

Simulated CRDSA demodulator scatter diagram for Es /N0 = ∞ dB and G = 0.4 with real channel estimation for IC.

0.55

A PPENDIX II P ROBABILITY T HAT i B URSTS ARE C OLLIDING

Ideal Power Control

0.5

0.45 σPw r = 1 dB

0.4 Throughput

0

real

0.6

σPw r = 2 dB 0.35 σPw r = 3 dB 0.3

0.25

0.2

0.15

0.1 0.1

−0.5

real

0.2

0.3

0.4

0.5 0.6 Normalized load [G]

0.7

0.8

0.9

1

Fig. 11. Simulated ideal channel estimation (continuous line) and real channel estimation (dashed dot line) results for the CRDSA throughput versus the normalized channel loading for Niter = 10 and various STs power error standard deviations.

the probability that j of them choose the same code  and the same differential delay value is P(j out of i) = ji · pj · (1 − p)i−j where p = 1/(SPR · Dmax ). Hence, considering that the probability of having i interfering packets in the same slot n is Pint (i| G) (see Appendix II), the probability of collision is easily calculated as: RA G·Mslots −1

Ppre−coll (G)

=



Pint (i| G)

⎤ ⎡ i=1 i  i ·⎣ · pj · (1 − p)i−j ⎦ j j=1 RA G·Mslots −1

=



  Pint (i| G) · 1 − (1 − p)i(13)

i=1 RA where G is the load measured in packets per slot and G·Mslots is the number of packets per RA frame and it is assumed to be an integer number.

We need to derive the probabilities Pint (i| G) and PalA defined in Sections III-B and III-D. First, we derive the probability that the packet pk (represented by the two replicas B pA k and pk ) is present on a given slot Sn from the set of RA Mslots slots available in one frame. It should be recalled that each packet is transmitted twice on a given frame but never over the same slot. Thus the probability can be computed as follows:   A   / Sn P {pk ∈ Sn } = P pA k ∈ Sn + P p k ∈   2 ·P pB (14) k ∈ Sn = RA Mslots assuming that slots are selected randomly with uniform distribution and never selected twice the same slot. Then, the probability that i interfering packets are present on a given slot Sn can be derived as a binomial where i packets are RA − 1 − i packets present in slot Sn and the remaining G · Mslots are not present in slot Sn :

 RA G · Mslots −1 i Pint (i| G) = [P {pk ∈ Sn }] i RA G·Mslots −1−i

· [1 − P {pk ∈ Sn }]

(15)

Finally the probability that the copy A of a packet pk is alone on a given slot Sn is equivalent to having zero interfering packets in slot Sn : RA G·Mslots −1

PalA (G) = Pint (0| G) = [1 − P {pk ∈ Sn }]

. (16)

R EFERENCES [1] L. G. Roberts, “ALOHA packet systems with and without slots and capture,” ARPANET System Note 8 (NIC11290), June 1972. [2] N. Abramson, “The throughput of packet broadcasting channels,” IEEE Trans. Commun., vol. 25, pp. 117–128, Jan. 1977. [3] G. L. Choudhury and S. S. Rappaport, “Diversity ALOHA - A random access scheme for satellite communications,” IEEE Trans. Commun., vol. 31, pp. 450–457, Mar. 1983. [4] Digital Video Broadcasting (DVB); Interaction Channel for Satellite Distribution Systems, European Telecommunication Standardisation Institute (ETSI) EN 301 790 V1.4.1 (2005-09).

CASINI et al.: CONTENTION RESOLUTION DIVERSITY SLOTTED ALOHA (CRDSA)

[5] IP Over Satellite, Telecommunication Industry Association TIA-1008, Oct. 2003. [6] T. Le-Ngoc and J. I. Mohammed, “Combined free/demand assignment multiple access (CFDAMA) protocols for packet satellite communications; Universal personal communications,” in Proc. 2nd International Conf. Pers. Commun., Oct. 1993, vol. 2, pp. 824–828. [7] O. del Rio Herrero, G. Foti, and G. Gallinaro, “Spread-spectrum techniques for the provision of packet access on the reverse link of next-generation broadband multimedia satellite systems,” IEEE J. Select. Areas Commun., vol. 22, no. 3, pp. 574–583, Apr. 2004. [8] S. S. Lam, “A carrier sense multiple access protocol for local networks,” Computer Networks, vol. 4, pp. 21–32, Feb. 1980. [9] N. Abramson, “The ALOHA system - Another alternative for computer communications,” in Proc. AFIPS Conf., Nov. 1970, vol. 37, pp. 281– 285. [10] D. Raychaudhuri and K. Joseph, "Channel access protocols for Ku-band VSAT networks: A comparative evaluation,” IEEE Commun. Mag., vol. 26, no. 5, pp. 34–44, May 1998. [11] D. Raychaudhuri, “ALOHA with multipacket messages and ARQ-type retransmission protocols-throughput analysis,” IEEE Trans. Commun., vol. 32, no. 2, pp. 148–154, Feb. 1984. [12] D. Raychaudhuri, “Stability, throughput, and delay of asysnchronous selective reject ALOHA,” IEEE Trans. Commun., vol. 35, no. 7, pp. 767–772, July 1987. [13] Digital Video Broadcasting (DVB); Interaction Channel for Satellite Distribution Systems; Guidelines for the use of EN 301 790, European Telecommunication Standardisation Institute (ETSI) TR 101 790 V1.2.1 (2003-01). [14] P. Patel and J. Holtzman, “Analysis of a simple successive interference cancellation scheme in a DS/CDMA system,” IEEE J. Select. Areas Commun., vol. 12 , no. 5, pp. 796–807, June 1994. [15] R. De Gaudenzi, F. Giannetti, and M. Luise, “Signal recognition and signature code acquisition in CDMA receivers for mobile communications,” IEEE Trans. Veh. Technol., vol. 47, no. 1, pp. 196–208, Feb. 1998. [16] Universal Mobile Telecommunications System (UMTS); Selection Procedures for the Choice of Radio Transmission Technologies of the UMTS, European Telecommunication Standardisation Institute (ETSI) TR 101 112 V3.2.0 (1998-04) Enrico Casini was born in Figline Valdarno, Italy, on March 13th 1976. He received the Laura Degree (summa cum Laude) from the University of Florence, Florence, Italy in 2002. In 2001 he joined the ESA’s Research and Technology Center (ESTEC), Noordwijk, The Netherlands. During 2001-2003 he was a trainee investigating advanced synchronisation techniques for space communications and interference mitigation techniques. Since 2003 is working for the RF Payload and System Division as communication system engineer. His research interests include the characterization of non-linear satellite channel, advanced modem design, physical layer system simulation and the analysis of the satellite payloads.

1419

Riccardo De Gaudenzi (M’89-SM’97) was born in Italy in 1960. He received his Doctor Engineer degree (cum Laude) in electronic engineering from the University of Pisa, Italy in 1985 and the PhD from the Technical University of Delft, The Netherlands in 1999. From 1986 to 1988 he was with the European Space Agency (ESA), Stations and Communications Engineering Department, Darmstadt (Germany) where he was involved in satellite telecommunication ground systems design and testing. In particular, he followed the development of two new ESA’s satellite tracking systems. In 1988, he joined ESA’s Research and Technology Centre (ESTEC), Noordwijk, The Netherlands where in 2000 he has been appointed head of the Communication Systems Section and since 2005 he is Head of the RF Payload and Systems Division. The division is responsible for the definition and development of advanced satellite system, subsystems and technologies for telecommunications, navigation and earth observation applications. In 1996 he spent one year with Qualcomm Inc., San Diego USA, in the Globalstar LEO project system group under an ESA fellowship. His current interest is mainly related with efficient digital modulation and access techniques for fixed and mobile satellite services, synchronization topics, adaptive interference mitigation techniques and communication systems simulation techniques. From 2001 to 2005 he has been serving as Associate Editor for CDMA and Synchronization for the IEEE T RANSACTIONS ON C OMMUNICATIONS. He is co-recipient of the VTS Jack Neubauer Best System Paper Award from the IEEE Vehicular Technology Society. Oscar del Rio Herrero was born in Barcelona, Spain, in 1971. He received the B.E. degree in Telecommunications and the M.E. degree in Electronics from the University Ramon Llull, Barcelona, Spain, in 1992 and 1994, respectively. He received a post-graduate degree in Space Science and Technology with emphasis in Satellite Communications from the Space Studies Institute of Catalonia (IEEC), Barcelona, Spain, in 1995. He joined ESA’s Research and Technology Center (ESTEC), Noordwijk, The Netherlands, in 1996. In 1996 and 1997 he worked as a Radio-navigation System Engineer in the preparation of the Galileo programme. Since 1998, he has worked as a Communications Systems Engineer in the Electrical Systems Department. His current research interests include high-performance on-board processors, packet access and resource management schemes and IP inter-working for future broadband satellite systems.