2017 3rd IEEE International Conference on Computer and Communications

Performance Analysis of the Statistical Priority-based Multiple Access

Jian Liu, Tao Peng, Qingyi Quan, Lili Cao

Key Laboratory of Universal Wireless Communication, Ministry of Education
Beijing University of Posts and Telecommunications, Beijing, 100876, China
e-mail: [email protected]

Abstract—Statistical Priority-based Multiple Access (SPMA) is CTS packet containing the message indicating resource
the MAC protocol used for Tactical Targeting Network allocated to the sender is sent back. In this way, it is
Technology(TTNT), the next generation of Data Link System guaranteed that different nodes get orthogonal resource.
of the US Army. As a new MAC protocol on testing and to be However the use of RTS/CTS increases the time cost to
used in the near future, SPMA is required to support for second level.
multiple different Classes of Services(COS), provide a much
lower latency for packets with high priority and an extremely
high first transmission success rate for all packets sent out. All
these properties make SPMA a much better protocol for
Aeronautical Ad-hoc Network(AANET) than others. In this
paper we propose a simple but accurate analytical model to
compute the Slot Transmission Probability(STP) in
assumption of saturated input. With the STP, we get the mean
delay, packet loss rate, throughput and other properties of the
system. Simulation implemented support the accuracy of the
analytical model. All results show that SPMA can guarantee
the timeliness and reliability of the highest priority and
maintain the stability of the system throughput.

Keywords-SPMA, MAC protocol, TTNT, COS, AANET, STP

I. INTRODUCTION Figure 1. Aeronautical ad-hoc network.

Ad-hoc Networks are defined to be a kind of network
organized by devices themselves with no center control The MAC protocol used for US army’s next generation
station [1]. ANNET is the kind of Ad-hoc network in which of military communication system is SPMA [8]. In this
the basic node could be aircrafts, ground station, satellite, or article, we mainly analysis the performance of SPMA using
warships. FIG.1 gives a basic structure of ANNET. The high a fixed-point way different from the traditional Markov
mobility of ANNET leads to a frequently changing topology method introduced in [9]. We also implement M/G/1
which makes wireless communications complicated and queuing theory to analysis the contention process for
difficult to be maintained [2]. different classes of packets.
As one of the key technologies in ANNET, MAC The rest of the paper is outlined as follows. In
protocol is used to share the limited channel resource to the Section II we describe the SPMA protocol indetails.In
nodes in network [3]. The MAC protocol used for ANNET Section III, we promoted the concept of STP, and analysis
could generally be classified into three categories. One the performance of SPMA using a fixed-point way and
typical kind is time division multiple access(TDMA). In a M/G/1 queuing theory. In Section IV, simulation results are
traditional TDMA system, nodes use the slots delivered to exhibited and compared with the theoretical analysis. Section
them predefined. A central control station is always needed. V concludes this article. Section VI is the acknowledgement.
There is also many other modified protocols of TDMA [4],
II. PROTOCOL D ESCRIPTION
[5]. In [4] an interference-based distributed TDMA
scheduling algorithm(IDTA) is promoted. The IDTA Statistical priority-based multiple access(SPMA) is de-
protocol does not need a central station and has the ability to signed for spread spectrum systems. By setting different
serve differently to packets of different priorities. However
transport thresholds to packets of different priorities,
TDMA would fail need to create these components,
incorporating the applicable criteria that follow. SPMA assures that emergency messages with high priority
Another MAC protocol used for ANNET is the IEEE to be transported with extremely short delay.
802.11 DCF and the modified versions of it [6], [7]. In [6] a And SPMA could maintain the system in a stable
MAC protocol based on 802.11DCF is illustrated. The status by putting back the packets with low priority when
communication period is divided into two parts, the the traffic suddenly surge. Another principle purpose of
contention part and the transporta- tion part. In contention the SPMA protocol is to achieve a 99 percent first time
part a RTS packet carrying resource request is sent out and a success.

978-1-5090-6352-9/17/$31.00 ©2017 IEEE 30

 III. PROTOCOL MODELING
With the increasing of offered load, the system will get
into a saturated status which means that there is always
packet being on service. By putting off the transmission of
packets with low priority, the throughput of the system
maintains in a stable state. The probability that a node has a
packet sent out in an arbitrary slot should be a fixed value.
We define the probability as slot transmission
probability(STP). In the following path we model the
system to get the STP and other properties. The mean
assumption is that the input of the system is saturated and
Figure 2. SPMA protocol achitecture. channel condition is perfect. First we define the following
parameters of the back-off procedure:
δ : The length of a slot.
m : The largest back-off stage.
Wi : The length of back-off window on back-off
stage i. At the beginning of back-off stage i, a integer n
is chosen randomly in [1, Wi] . After n slot, the channel
occupancy will be detected again. If the channel is still
busy and no packets with higher priority come, the back-
off stage will increase by one.
K : The largest back-off times.
A. Calculation of STP
We assume the load of packets with priority i follows
Figure 3. Waveform of SPMA. the Poisson distribution with λi. Then the probability that
there has k packets of priority i come in time period t is:
A. Architecture and the SPMA Protocol
e Oit ˜ (Oi t )k (1)
The general architecture of SPMA protocol is illustrated gi (k , t )
in FIG.2. As showed in the figure, when a packet comes, it k!
entries the queue of its priority. If there is no packets with
higher priority existed, Channel Occupancy statistic will be To packet with priority i, if packets with higher priority
come during its back-off interval, the back-off will be can-
compared with the predefined threshold corresponding to
celed. With (1) we could get probability that the back-off is
the priority queue to determine whether to send this packet
not interrupted in time period t as:
or take a random back-off.
­ i 1 i 1

°– g j (0, t ) –e
O jt
B. Waveform Used by SPMA ,i t1
fi (t ) ®j 0 j 0
The waveform used by SPMA is a combination °1, i 0 (2)
of frequency-hopping and time-hopping. Once the Channel ¯
Occupancy allows the packet to be sent, the packet will be In(2), we assume that priority 0 is the highest priority.
parsed into a polarity of sub blocks hopped in time and The back-off of priority 0 would never be interrupted. If no
frequency [10]. Turbo code is used to do channel coding. packets with higher priority come during the process period,
FIG.3 shows the case in which sub blocks from different the mean service time of a packet with priority i should be:
users are transmitted simultaneously.
­m a i 1 K m i 1 i 1

°¦ baJ i –– e  ¦ bmJ ia –– e c j (– e  Ocbm ) a  m , i t 1

a  Oc b j O b
C. Key Parameters for SPMA °a 1 j 1 c 0 a m 1 j 1 c 0 c 0
Xi ®m
The following parameters affect the performance of the ° bJa 
K (3)
SPMA system. ¦
°̄ a 1 a i ¦
a m 1
bmJ ia ,i 0
(1) Thresholds for different priorities. The protocol
deter- mines whether to send or not by comparing the
thresholds with Channel Occupancy statistic. In(3), bh stands for the average slots need to wait in each
(2) Length of the statistical sliding window. System gets back-off. γi is the probability that the Channel Occupancy is
the parameter Channel Occupancy by count the packets higher than the threshold of priority i. The probability that
received in the sliding window. the packet is sent out after j times back-off is show as(4).

31
 In FIG.5 we show plots of Γ(PSTP ) versus Pout. The
intersection point stands for the STP of the system. FIG.6
W WLPH shows the STP we get through the fixed-point equation(10)
Figure 4. Statistical sliding window. and the result from simulations. The simulation result has
tiny difference with the theoretical analysis result. The
thresholds set for priority 0 , 1, 2, 3 are 22, 18, 14, 10
­ respectively, which could assure that the packets sent out be
°1  J ,j 0
° i recovered from the sub blocks received with the probability
° j j
(4) higher than 99 percent . Threshold setting will be discussed
\ (ji ) ®J i – fi (ba )(1  J i ),1 d j d m in the following part.
° a1
° j m
°J i – fi (ba ) fi (bm ) (1  J i ), m d j d K
j m

¯ a1

With(4), we get the probability that the packet on service

could be sent out without interruption of packets with higher
priority is: K
(i )
psent ¦\
j 0
(i )
j
(5)
pmax

We define Ui Oi X i and U
i 0
Ui . According to ¦
M/G/1 queuing theory, the probability that the priority of the
packet on service equals i is as follow:
Ui iO ˜ Xi
Pi
U P max

¦O
Figure 5. Plot of Γ(PSTP ) versus Pout
j ˜Xj (6)
j 0
According to(3),(5) and (6), we finally get the STP as :
P max
1 (7)
PSTP ¦ p ˜p
i 0
i
(i )
sent
Xi
In FIG.4 WT is the statistical sliding window. When a
packet get on service at time moment t, the Channel
Occupancy statistic will be compared with the threshold of
this packet. Here we express the STP as Pout. We assume
that the traffic load is same to all the nodes in the network.
Then we get that the distribution of the total number of
packets, transported in the sliding window WT , N follows
Binomial distribution as following:

P( N k ) Cnk ( Pout
k
˜ (1  Pout )nk (8)
Figure 6. STP in analysis and simulation.

In(8), n = WT *nodeNum and nodeN um is the total node B. Transmission Success Rate
number in the network. According to(8), we get γi as (9). In SPMA, once a packet is sent out, the priority of
Ri each packet will be meaningless. All sub blocks in the
J i 1  ¦ P( N k ), 0 d i d Pmax channel have the same rate to collide with others. The
k 0 (9) number of sub blocks transported on one frequency in
each second has the mean value as
Ri is threshold set for priority i. (8) and (9) show Nb ˜ nodeNum ˜ Pout ˈNf is the number of frequency
that γi is an expression of Pout while PSTP is an expression O
G ˜Nf
of γi . It is now natural to expect that the equilibrium points. We assume the length a sub block is Tb and all
behavior of the system will be characterized by the solutions
of the following fixed-point equation: the sub blocks sent out are distributed in Poison. The
probability that a sub block have no collision with others
Pout *( PSTP ) (10) is Pbs = e−2Tb λ. According to the Turbo coding theory, at

32
least half of the total sub blocks should be collision free D. Mean Delay
torecover the original packet. We get the probability that a The priority queues could be analyzed as a M/G/1 pre-
packet sent out could be recovered as: emptive priority queue shown in FIG.8. New packet finds
Nb
its location in the M/G/1 queue according to its priority and
Ps ¦C
i Nb /2
i
P (1  Pbs ) Nb i
i
Nb bs
(11)
generation time.
The mean delay of a packet is defined as the time period
since the time it get into its priority queue to being captured
The thresholds setting need to assure a 99 first by the receiving terminal. The mean delay of a packet with
success. We simulated the case in which that all the priority i is E(T (i)) expressed as(13).
packets have the same priority. And get the threshold which
guarantees a first success higher than 99 as the threshold for E (Ts(i ) ) E ( Di  X i  Ii  Tp  Tt ) (13)
the highest priority. FIG.7 shows the success rate versus Di contains service time the packet needs for packets
threshold. The parameters used are shown in Section IV. As already in the queue with higher or the same priority.
we can see in FIG.7, 22 and 15 should be the threshold set Residual service time when the packet arrive is also a part of
for the highest priority when the length of statistical sliding Di. According to the theory of M/G/1 priority queue we get
window is 60 and 40 respectively. The thresholds of other mean as below:
priorities should not exceed the threshold of the highest
priority.
¦
i
j 0
U j E ( Aj ) (14)
E ( Di ) i 1
(1  ¦ U j )(1  ¦ U j )
i
j 0
j 0

In(14), E(Ai) is the mean residual service time for a

packet with priority i expressed as E( Aj ) E( X i2 ) / 2E( X i ) .
E(Xi) is expressed as (3). Ii is the sum of interruptions.
And Ii contains two parts, the sum of the processing
times of packets that interrupt the packet we are
looking at , and the sum of processing times of the
packets that arrive during periods in which the packet
is already interrupted. Hence we get:
i 1 i 1
E ( Ai ) E ( X i )¦ O j E ( X j )  E ( Ai )¦ O j E ( X j ) (15)
j 0 j 0
According to (15) we have:
E( X i )
Figure 7. Plot of Ps vs Threshold. E ( X i )  E ( Ai ) (16)
1 ¦ j 0 U j
i 1

Tp is time period from the beginning of the first sub block

to the end of the last one. We assume that a packet is break
into N b sub blocks and all the sub blocks are distributed in
a Poisson way, then we have:
( Nb  1) ˜ G
Tp  Tb (17)
Pout ˜ Nb
In (17), Tb is the length of a sub block. When Nb is big,
Figure 8. M/G/1 queuing system. we have Tp | G / PSTP , Tt is the propagation time which is
merely related to the distance between two nodes. According
to (3),(14), (16),(17), we could get the final expression of
C. Throughput of the System mean delay.
The throughput of the system could be expressed as (12).
IV. S IMULATION AND ANALYSIS
NodeNum ˜ PSTP ˜ Ps ˜ L
S (12) In the following part, we simulated the SPMA protocol
G
on Matlab 2015(a). There are packets of four priorities, 0 to
In(12), Ps is the probability that the packet could be 3. Packets of priority i arrive in Poisson process with λi and
recovered from the sub blocks received. L is the average λ0 : λ1 : λ2 : λ3 = 1 : 2 : 3 : 4. The thresholds are 22, 18, 14,
packet length. As we have already learned that Ps is higher 10 respectively to the four priorities.In Table I are some
than 99 percent. We have : S | PSTP ˜ L other parameters used for the simulations.
G
FIG.9 shows the throughput of the system with different

33
node numbers. As we can see in the figure, throughput goes
up as the traffic load growing and then maintains at a stable
level. Yet the figure does not show it , we can imagine that
the line of nodeNum = 5 will goes up to about 19Mbps
with the load growing. The theoretical saturated throughput
get from (12) is shown in Table II, and it is almost the same
as the simulated result shown in FIG.9. FIG.10 shows the
packet loss rate of the system when the node number is 15. A
packet is thought to be lost when its back-off time exceeds K
or the queue of its priority is filled at the moment it is
generated. We can see that the loss rate of packets with
priority 0 keeps nearly 0 until the traffic load goes up to 104
packets/s. Packets of priority 3 are completed blocked when
the traffic load is about 1000 packets/s.
The mean delay of packets sent out is shown in Fig.11.
As shown in the figure, mean delay of priority 3 and priority
4 goes up to a peak and then falls down to zero gradually.
The traffic load corresponding to the peak point is the load at Fig.ure 10. Packet loss rate of the system.
where the loss rate of packets with the same priority goes up
to nearly 100 percent in FIG.10. The mean delay of packets
with priority 0 keeps nearly 40 slots while the load is blow
14000 packets/s . In our simulation the largest value of mean
delay of priority 0 happens when the load is 50000 packets/s
as 374 slots, about 7.48 milliseconds.
TABLE I. PARAMETERS USED FOR
SIMULATION.
Parameters Values
Packet Lenght(L) 1000bits
Number of sub blocks from one packet(Nb) 32
Number of frequency points(Nf ) 30
Lenght of sub block(Lb) 100bits
Length of each priority queue(Lq ) 100
Minimum back-off window(W) 32
Channel transmission rate(Vchannel) 10e6bps

TABLE II: SATURATED THROUGHPUT

Figure 11. Mean delay.
nodeNum 5 15 25 35 45
Throughput(Mbps) 18.925 19.008 19.552 20.097 20.493

V. CONCLUSION
In the paper, we present a fixed-point way to analyze the
SPMA MAC protocol. By solving the fixed-point equation,
we get STP the probability that one node have a packet sent
out in an arbitrary slot. With the STP, we get the expression
of system throughput. We also analyzed the mean delay of
packets with different priority using of M/G/1 queuing
theory. Relationship between threshold setting and packet
transmission success rate is also analyzed in this paper. Both
analysis and simulation results show that SPMA could
provide extremely short delay to packets with the highest
priority and a first success higher than 99 percent for all
packets sent out. The throughput of system could also
maintains in a stable level by using SPMA.
ACKNOWLEDGEMENT
This work is supported by the National Natural Science
Foundation of China (No.61631004) and the National Sci-
Figure 9. Throughput of the system.
ence and Technology Major Project of China under Grant
2016ZX03001017.

34
 REFERENCES COMMUNICATIONS CONFERENCE, 2012-MILCOM 2012. IEEE,
2012: 1-6.
[6] Xu D, Zhang H, Zheng B, et al. “A priority differentiated and multi-
[1] Wu J, Stojmenovic I. “Ad hoc networks[J]. COMPUTER-IEEE channel MAC protocol for airborne networks[C]//Communication
COM- PUTER SOCIETY-, 2004, 37(2): 29-31. Soft- ware and Networks (ICCSN), 2016 8th IEEE International
[2] Rak J, Jonsson M, Vinel A. A taxonomy of challenges to resilient Conference on. IEEE, 2016: 64-70.
message dissemination in VANETs[C]//Telecommunications [7] Li J, Zhou Y, Lamont L, et al. “Packet delay in UAV wireless
Network Strategy and Planning Symposium (Networks), 2016 17th networks under non-saturated traffic and channel fading conditions[J].
International. IEEE, 2016: 127-132. Wireless personal communications, 2013, 72(2): 1105-1123.
[3] Böhm A. Delay-sensitive wireless communication for cooperative [8] Clark S M, Hoback K A, Zogg S J F. “Statistical priority-based
driving applications[D]. Halmstad University Press, 2013. multiple access system and method: U.S. Patent 7,680,077[P]. 2010-
[4] Li J, Gong E, Sun Z, et al. “An interference-based distributed TD- 3-16.
MA scheduling algorithm for aeronautical ad hoc [9] Bianchi G. “Performance analysis of the IEEE 802.11 distributed
networks[C]//Cyber- Enabled Distributed Computing and Knowledge coor- dination function[J]. IEEE Journal on selected areas in
Discovery (CyberC), 2013 International Conference on.IEEE, 2013: communications, 2000, 18(3): 535-547.
453-460. [10] Zogg S J, Clark S M, Haendel R S, et al. “Waveform for virtually
[5] Ripplinger D, Narula-Tam A, Szeto K. Scheduling vs. random access simultaneous transmission and multiple receptions system and method:
in frequency hopped airborne networks[C]//MILITARY U.S. Patent 7,830,781[P]. 2010-11-9.

35