The performances of pure PRMA, MD PRMA and RCMA will be compared, all with the same number of resource units parameters, the impact of acknowledgement delays and TDD operation on voice d
Trang 1Multiple Access Protocols for Mobile Communications: GPRS, UMTS and Beyond
Alex Brand, Hamid Aghvami Copyright 2002 John Wiley & Sons Ltd ISBNs: 0-471-49877-7 (Hardback); 0-470-84622-4 (Electronic)
8
MD PRMA ON CODE-TIME-SLOTS
This chapter is concerned with MD PRMA on perfect-collision code-time-slot channels The simple and abstract channel model used, representative for a blocking-limited system,
allows one to consider an arbitrary number of code-slots E per time-slot, without having to
worry about the spreading factor required to meet a certain packet erasure performance In this framework, the scope of investigations can conveniently be extended to two extreme
cases, namely only one code-slot per time-slot, but numerous time-slots N per TDMA
frame, and only one time-slot per frame carrying numerous code-slots In the first case, the CDMA feature is relinquished, and MD PRMA degenerates to pure PRMA In the second case, the TDMA feature is relinquished While this configuration (and in fact also PRMA itself) can simply be viewed as a special case of MD PRMA, it actually corresponds to the Reservation-Code Multiple Access (RCMA) protocol proposed in Reference [35]
As in Chapter 7, only voice-traffic will be considered However, the focus shifts from load-based access control to fixed permission probabilities and backlog-based access control (the latter in the shape of Bayesian broadcast) The performances of pure PRMA,
MD PRMA and RCMA will be compared, all with the same number of resource units
parameters), the impact of acknowledgement delays and TDD operation on voice dropping performance is also studied Furthermore, the code-time-slot channel model is enhanced
to account for multiple access interference (MAI) In this scenario, unlike the perfect-collision case, load-based access control can make sense Therefore, on top of ‘conven-tional’ Bayesian broadcast, a scheme combining Bayesian broadcast with a channel access function is considered
8.1 System Definition and Simulation Approach
8.1.1 System Definition and Choice of Design Parameters
The common thread in this chapter is the consideration of code-time-slots based on the TDMA frame duration specified in Section 5.3, namely the 4.615 ms used in GSM and originally proposed for TD/CDMA However, the focus is not limited to the TD/CDMA
Trang 2but only one ‘code-slot’ per time-slot, and one with E= 64 codes on a single ‘time-slot’ Effectively, the first case represents pure TDMA, where MD PRMA degenerates
to conventional PRMA, and the second case is pure CDMA, for which MD PRMA corresponds to RCMA proposed in Reference [35] Choosing the same frame duration
and the same number of resource units U (namely 64) for all three schemes allows for a
fair comparison of their respective performance
For these three cases, MD PRMA for frequency division duplexing as defined in Section 6.2 is investigated, assuming immediate acknowledgement and using either fixed permission probabilities for voice (again the only traffic considered), or backlog-based
calculated according to the Bayesian algorithm adapted for MD PRMA, as outlined in Subsection 6.5.4 Equation (6.9) is used to carry out the estimation of the arrival rate
end of each time-slot in such a manner that it is available to all mobile stations with full precision before the next time-slot starts
of acknowledgement delays is also studied by varying the parameter x introduced in
Subsection 6.2.6 This parameter determines how many time-slots a terminal must wait for
an acknowledgement following the time-slot in which it sent a packet in contention mode
While waiting, it is not allowed to contend again In the case of Bayesian broadcast, if x >
0 (i.e acknowledgement is not immediate), the Bayesian algorithm needs to be modified,
configuration of resource units, the performance of MD FRMA for TDD with a single switching-point per frame, as specified in Subsection 6.3.3, is assessed From one to eight time-slots per TDMA frame are assumed to be assigned to the uplink direction, where the last case is obviously only of academic interest, since no resources would be available for the downlink in this case
In the following two sections, when more than one code-slot is considered, these slots are assumed to be mutually orthogonal, which means that MAI is ignored If dedicated channels were used, the system would exhibit hard-blocking, but owing to the PRMA element, it features soft-blocking or soft-capacity In Section 8.4, on the other hand, MAI
is accounted for in the manner specified therein, in order to assess the impact of the loss of orthogonality on access control In this case, depending on the quality of service requirements, we are dealing with an interference-limited system; that is, excessive packet
erasure may prevent all U resource units from being used In the terminology used in
it is rectangular interleaving over the length of a voice frame, which in turn is carried
on four bursts (see Subsection 6.2.4) In this case, request bursts sent in contention mode are dedicated signalling bursts, transmitted on a single code-time-slot By contrast, when interleaving is not applied, they carry not only signalling, but also user data, namely the same amount as carried by information bursts
a small value of 4.615 ms, which is equal to the length of a single TDMA frame In the
the impact of interleaving and dedicated request bursts, the basic scheme is also operated
Trang 38.1 SYSTEM DEFINITION AND SIMULATION APPROACH 313
Table 8.1 Parameters relevant for the physical layer, protocol operation and
traffic models
Dropping Delay Threshold Dmax 4.615 ms (no interleaving)
18.462 ms (with interleaving) Mean Talk Gap Duration Dgap 1.74 s (or 3 s)
Mean Talk Spurt Duration Dspurt 1.41 s (or 3 s)
the next subsection, all parameters mentioned so far are summarised in Table 8.1
8.1.2 Simulation Approach, Traffic Parameters and Performance
Measures
As in the previous chapter, the only traffic considered in the following is packet-voice traffic, using the two-state voice model specified in Section 5.5 Two different parameter
Reference [56] is used This is to establish a link with Chapter 9, where mixed voice and data traffic is considered, and parameters from Reference [56] are used for both voice and Web browsing traffic
The system load is determined by the number of conversations M simultaneously
multiplexing can easily be calculated using Equation (6.1) In Section 8.4, where MAI is
Each simulation-run with fixed M covers 1000 s conversation time Where required,
8.1.3 Analysis of MD PRMA
Pure and modified PRMA systems were analysed for instance in References [135,143,144, 149,150,268,269] Most of these articles provide a full Markov analysis, some an equi-librium point analysis (EPA) Due to the dimension of the state space with the here considered design parameters, a full Markov analysis is rather challenging In Refer-ence [61], we provided an EPA for MD PRMA, which expanded on the EPA for PRMA provided in Reference [143] and adopted a few elements of Reference [149] In certain
Trang 4scenarios, we found EPA to be satisfactory, in others not In the following, we focus on protocol performance assessment through simulation studies
Performances 8.2.1 Simulation Results, No Interleaving
there would not be much benefit in implementing adaptive access control However, while
simula-tions are heavily affected by the instance in time in which the system first experienced congestion Once caught in a congested equilibrium point, it is almost certain that the system remains in this state for the remainder of the simulation run and, from then on,
1.0E-8
1.0E-6
1.0E-4
1.0E-2
1.0E+0
Simultaneous conversations M
Pdrop
p = 0.1 p = 0.2
p = 0.3 p = 0.4
p = 0.5 p = 0.6
p = 0.7 Bayes
MD PRMA, N = 8, E = 8
Dmax= 4.615 ms
Figure 8.1 Simulated MD PRMA performance, overview
Trang 58.2 COMPARISON OF PRMA, MD PRMA AND RCMA PERFORMANCES 315
the dropping probability is close to one For values of M for which stability problems
Figures 8.1 and 8.2 A better performance measure in such cases would be the so-called First Exit Time (FET) proposed in Reference [194] for slotted ALOHA and applied to PRMA in Reference [149] The FET is the average first exit time into the unsafe region (i.e a system state beyond the unstable equilibrium point, see Figure 3.6) starting from
an initially empty channel or system
(that is, the FET is much larger than the duration of an individual simulation-run) BB on the other hand allows for stable operation at high load while ensuring low packet dropping
of M considered.
One could argue that the performance of BB could be met by choosing a semi-adaptive
be extended to a mixed traffic scenario, possibly with unknown traffic statistics, whereas
BB adapts automatically to different traffic mixes Furthermore, it would also require
argument in its favour In view of the very small complexity of BB, this advantage is of
no relevance in practice, though
Similar considerations apply in the case of pure PRMA In fact, looking at Figure 8.2,
1.0E-8
1.0E-6
1.0E-4
1.0E-2
1.0E+0
Simultaneous conversations M
p = 0.05
p = 0.07
p = 0.1
p = 0.15
p = 0.2
p = 0.3
p = 0.4 Bayes
PRMA, N = 64, E = 1
Dmax= 4.615 ms
Figure 8.2 Simulated PRMA performance, overview
Trang 61.0E-6 1.0E-4 1.0E-2
1.0E+0
60 70 80 90 100 110 120 130 140 150
Simultaneous conversations M
Pdrop
p = 0.7
p = 0.8
p = 0.9
p = 0.95
p = 0.98
p = 1.0 Bayes
RCMA, N = 1, E = 64
Dmax = 4.615 ms
Figure 8.3 Simulated RCMA performance, overview Finally, with RCMA, the situation is slightly different, as illustrated in Figure 8.3 Note
the first packet in a spurt is dropped, such that there will be significant dropping
of the respective curves On the other hand, even if there is a temporary accumulation of contending terminals, they will normally be able to choose between numerous code-slots available for contention, such that the collision risk is small Therefore, stability is not
access dynamically, e.g through Bayesian broadcast, which is also shown in the figure This is very much in contrast to pure PRMA (and to a lesser extent to MD PRMA), where
number of successive collisions During these collision slots, C will grow further, and
is bound to become unstable To complete the discussion of the results for RCMA, with
8.2.2 Performance Comparison and Impact of Interleaving
In Reference [35], it is claimed that ‘RCMA is superior to PRMA in terms of system capacity even when a median size of code set is used’ To come to this conclusion, the authors of Reference [35] applied a frequency reuse factor of seven to PRMA, which may be considered conservative, but is probably not completely unrealistic At the same time however, and curiously enough, the authors spent not a single word on where the
‘median number of codes’ should come from and what kind of bandwidth or spreading factor would be required to support the corresponding number of simultaneous users For reasons outlined in detail in Sections 3.2 and 5.1, we have no intention of stepping onto a field full of mines by trying to assess the spectral efficiency of TDMA, hybrid
Trang 78.3 DETAILED ASSESSMENT OF MD PRMA AND MD FRMA PERFORMANCES 317
1.0E-8
1.0E-6
1.0E-4
1.0E-2
1.0E+0
Simultaneous conversations M
PRMA
MD PRMA RCMA PRMA, interleaving
MD PRMA, interleaving RCMA, interleaving
Bayesian broadcast
Dmax= 4.615 ms (no interleaving)
Dmax= 18.462 ms (with interleaving)
Figure 8.4 PRMA, MD PRMA, and RCMA with Bayesian broadcast
CDMA/TDMA, and CDMA systems operating with PRMA, MD PRMA, and RCMA respectively Here, the focus is exclusively on the efficiency of the multiple access proto-cols as such From this point of view, the only fair comparison appears to be one based on
an equal number of resource units, equal frame duration, and assuming a perfect collision channel for individual units (whether these be code, time, or code-time-slots)
dropping ratio at low load in the case of RCMA, which is due to the single contention
the excellent agreement between the performance of these three schemes can even be
For completeness, as we did already in Section 1.4, we point again at Reference [17], where CDMA, TDMA and hybrid systems are compared from a packet queuing perspective
Performances 8.3.1 Impact of Acknowledgement Delays on MD PRMA
Performance
It was explained in Section 3.6 why immediate acknowledgements are conceptually impossible with PRMA protocols using frequency division for duplexing In
Subsection 6.2.6 a parameter x was introduced to model acknowledgement delays (see
Trang 8also Subsection 8.1.1) With increasing x, one would expect increased Pdrop, since unsuccessfully contending terminals spend extra time to get a reservation On top of
that, a value of x greater than zero can also have a negative impact on the accuracy of the
backlog estimation through the Bayesian algorithm, even though appropriately enhanced
to cope with this situation This is illustrated in Figures 8.5 and 8.6 for the basic MD
Figure 8.6 compares the performance of Bayesian broadcast with that of perfect backlog estimation (in Chapter 7 referred to as known-backlog-based access control, KBAC) for
selected values of x In both cases, as expected, voice dropping increases with increasing
suffers much more with increasing x.
The reason is as follows: at low load, the backlog is close to zero in most time-slots,
time-slot period will normally result in a few successive collisions, if the number of available
algorithm will need longer to adapt
1.0E-8
1.0E-6
1.0E-4
1.0E-2
Simultaneous conversations M
x = 6
x = 5
x = 3
x = 2
x = 1
x = 0
MD PRMA, N = 8, E = 8
x : Forbidden time-slots after contention
Bayesian broadcast
Dmax= 4.615 ms
x = 4
Figure 8.5 Impact of acknowledgement delays on Bayesian broadcast
Trang 98.3 DETAILED ASSESSMENT OF MD PRMA AND MD FRMA PERFORMANCES 319
1.0E-8
1.0E-6
1.0E-4
1.0E-2
Simultaneous conversations M
Pdrop
KBAC, x = 6 KBAC, x = 5 KBAC, x = 3 KBAC, x = 0
MD PRMA, N = 8, E = 8
Dmax= 4.615 ms
x : Forbidden time-slots after contention
Bayes, x = 6 Bayes, x = 5 Bayes, x = 3 Bayes, x = 0
Figure 8.6 Backlog estimation (Bayes) vs known backlog (KBAC)
Even worse, it may be deceived by time-slots with numerous idle and success C-slots lying in-between the ‘collision-time-slots’ (due to other terminals accessing the system
in time-slots not affected by collisions) This will further delay the tracking of the real backlog, resulting in packet dropping for those terminals caught in the ‘collision slots’
only are collisions repeated every four time-slots, but also the A[t] patterns will exhibit
in terms of backlog may coincide regularly with ‘bad slots’ in terms of low A[t] values.
This could probably be described as ‘resonant behaviour’ or ‘local catastrophes’ The
considerable dropping, while those MS never caught in a ‘bad slot’ experience very moderate dropping
Figure 8.7 compares the impact of acknowledgement delays on the performance when using fixed permission probabilities with that when using Bayesian broadcast The
uniquely due to mobile stations not getting permission to send contention packets in this
BB, for the reasons just discussed
Trang 101.0E-5
1.0E-3
1.0E-1
Simultaneous Conversations M
p = 0.2, x = 0
p = 0.3, x = 0
p = 0.5, x = 5
p = 0.5, x = 0 Bayes, x = 5 Bayes, x = 3 Bayes, x = 1 Bayes, x = 0
MD PRMA, N = 8, E = 8
x : Forbidden time-slots after contention
Dmax= 4.615 ms
Figure 8.7 Bayesian broadcast vs fixed permission probabilities
of ‘bad slots’ However, this would defeat the purpose of broadcast control, which is
to stabilise the protocol and ensure efficient operation for all possible traffic scenarios,
Results shown in Figure 8.8 serve several purposes They allow the assessment of the impact of:
(assuming immediate acknowledgement);
• the added traffic in the case of interleaving (I/L) due to dedicated request bursts and on average two additional bursts per voice spurt due to rounding up the spurt duration to
an integer number of voice frames (again assuming immediate acknowledgement); and
• inherent delay of acknowledgements (by an average of slightly less than N/2
time-slots) in the case of MD FRMA (again with interleaving)
Note that MD FRMA is considered here with eight uplink time-slots per frame, to compare all schemes with an equal number of uplink time-slots This effectively means using MD FRMA in FDD mode, which would in practice not leave any time to signal acknowledgements for the entire frame before the subsequent frame starts