MoIP Interference-resisted Wireless MAC Designb91038/OP641.pdf · 2007. 11. 30. · MoIP...
Transcript of MoIP Interference-resisted Wireless MAC Designb91038/OP641.pdf · 2007. 11. 30. · MoIP...
MoIP
Interference-resisted
Wireless MAC Design
2007/11/30
Speaker: Yu-Hsiang Lei
Advisor: Hung-Yu Wei
2
Problem Overview
• For the 802.11 wireless system
– Contention based medium access
• No Qos mechanism avaiblable
– Multimedia connections require more bandwidth
• VoIP, video streaming
• Soft bandwidth guarantee
3
System Graph
MoIP nodes
(Interference-Resistant)
Challenge: provide QoS for MoIP nodes Other nodes
(normal)
4
Periodical CW Resetting
backoff
CW=CWmin
High QoS High QoS
T
backoff
CW=CWmin
backoff
CW=CWmin
High QoS
Throughput
Throughput
Time
MoIP node
Time
normal node
5
Periodical CW Resetting (cont.)
• To get a better performance, we can periodically allocate
resource for MoIP nodes
• Define a period T. After each period, we set the contention
window (CW) size of the interference-resistant nodes be
initial value (CWmin)
– Stations with smaller CW will have higher chance to access
the channel
6
To Implement Our Idea…
• We should first check if our idea would work
• Mathematical modeling
– Do the modeling for the system
• Methodology: Markov chain
• Deduce the collision and transmission rate
• Simulation
– Using NS2 and try to get the performance
• Implement the CW resetting module for the nodes
• Investigate the throughput and channel access probability for
the MoIP and normal nodes
• Using NOAH routing protocol to avoid relay
7
Markov Chain Model
• 802.11 back-off mechanism
• Let m be the maximum back-off stage
– CWmax=2m * CWmin
• Define two random variables
– b(t) : back-off stage of stations at instants t
• range from 0 ~ m
– s(t) : time slot of stations at instants t
• Assumption: period is the maximum back-off stage plus 1
• range from 1 to m+1
8
Mathematical modeling
0,1 1,1 2,1 … m-
1,1
m,1
0,2 1,2 2,2 … m-
1,2
m,2
0,3 1,3 2,3 … m-
1,3
m,3
…
0,m 1,m 2,m … m-
1,m
m,m
0,m
+1
1,m
+1
2,m
+1… m-
1,m+1
m,m+
1
1
1-p
1-p
1-p
p
MoIP node normal node
0,1 1,1 2,1 … m-
1,1
m,1
0,2 1,2 2,2 … m-
1,2
m,2
0,3 1,3 2,3 … m-
1,3
m,3
…
0,m 1,m 2,m … m-
1,m
m,
m
0,m
+1
1,m
+1
2,m
+1… m-
1,m+1
m,m+
1
1-p
1-p
1-p
p
9
Solve Markov Chain
To solve the steady state probability of this Markov Chain,we adopt the method from [2].
Consider an irreducible Markov Chain with one setof ergodic states and with transition matrix −→p . Letβj , j = 0, 1, · be a set of nonnegative numbers. If thevalue of βj satisfy
πj =∞∑
k=0
πkpkj , j = 0, 1, . . . , (1)
and∞∑
j=0
πj = 1. (2)
then βj , j = 0, 1, . . . , are the stationary probabilitiesof Markov chain.
10
Solve Markov Chain (cont.)
� We first solve the Markov chain of MoIP node� By the observation from the Markov chain, we can
make a guess as follows:
πi,j =(1− p)pi
m+ 1, 0 ≤ i ≤ j − 2, 2 ≤ j ≤ m+ 1
πi,i+1 =pi
m+ 1
� By checking the conditions (1) and (2), we can makesure that our guesses are true
11
Transmission & collision probabilities
0,m+1 1,m+1 2,m+1 … m-1,m+1 m,m+1
0,m 1,m 2,m … m-1,m
0,3 1,3 2,3
0,2 1,2
0,1
… … …
1-p
1-p
1-p
1-p
1-p 1-p 1-p
p
pp
pp p
↓↓↓↓mm-1…0
sgo
1
sg1
2
sg2 sgm-1 sgm
12
Transmission & collision probabilities
� The stationary probability of sgi , P (sgi), is the sumof ith column.
P (sg0) =m(1− pg)
m+ 1+
1
m+ 1
P (sg1) =(m− 1)(1− pg)pg
m+ 1+
pg
m+ 1
P (sg2) =(m− 2)(1− pg)p
2g
m+ 1+
p2g
m+ 1...
P (sgm−1) =
(1− pg)pm−1g
m+ 1+pm−1
m+ 1
P (sgm) =pmg
m+ 1
13
Transmission & collision probabilities
� The distribution of Bg conditioning on back-off stagei and time slot t
Pr{Bg = b |s = i} =1
2iCWmin
, for b = 0, 1, 2, · · · , 2iCWmin − 1.
� The average back-off counter in back-off stage i andtime slot t is
E[Bg|s = i] =2iCWmin − 1
2,
14
Transmission & collision probabilities
� The average back-off counter is computed as
E[Bg] =
m∑
i=0
E[Bg |s = sgi ] · P (sgi)
=
m−1∑
i=0
2iCWmin − 1
2· [(m − i)(1 − pg)p
ig
m + 1+
pig
m + 1] +
pmg
m + 1
2mCW − 1
2
=1
2(m + 1)
m−1∑
i=0
[(2iCW − 1)(m − i)(1 − pg) + 1]pig +(2pg)
mCW − pmg2(m + 1)
=1
2(m + 1)
m−1∑
i=0
[CW ·m(1 − pg)(2pg)i − CW (1 − pg)i · (2pg)
i −m(1 − pg)pig + (1 − pg)i · p
ig ]
+(2pg)
mCW − pmg2(m + 1)
=1 − pg2(m + 1)
{CW ·m1 − (2pg)
m
1 − 2pg− CW [
2pg [1 − (2pg)m−1]
(1 − 2pg)2+(m − 1)(2pg)
m
1 − 2pg] −m
1 − pmg1 − pg
+ [pg(1 − p
m−1g )
(1 − pg)2−(m − 1)pmg1 − pg
]} +(2pg)
mCW − pmg2(m + 1)
.
15
Transmission & collision probabilities
� τg, τn[3,4]
τg =1
E[Bg] + 1
τn =2(1− 2pn)
(1− 2pn)(CW + 1) + pnCW (1− (2pn)m),
� pg, pn
pg = 1− (1− τg)ng−1 × (1− τn)
nn−1,
pn = 1− (1− τn)nn−1 × (1− τg)
ng−1,
� Solve these four equations by numericalmethod.
16
Computation of throughput
• Transmission cycle
thg =Pg
TI + TC + TS
thn =Pn
TI + TC + TS
17
Computation of Pg and Pn
� Pg = sug ×L, Pn = sun ×L, where
sug = Pr{the transmitting node is a greedy node|number of transmitting node = 1}
=ng · τg(1− τg)ng−1(1− τn)nn
ng · τg(1− τg)ng−1(1− τn)nn + nn · τn(1− τn)nn−1(1− τg)ng
sun = Pr{the transmitting node is a normal node|number of transmitting node = 1}
=nn · τn(1− τn)
nn−1(1− τg)ng
ng · τg(1− τg)ng−1(1− τn)nn + nn · τn(1− τn)nn−1(1− τg)ng
� sug + sun = 1
18
Computation of Ts
DIFS
DATA
ACK
SIFS
� Consider the basic mode (without RTS/CTS)
TS = TDIFS+TPHY+TMAC+TDATA+TSIFS+π+TPHY+TACK+π
19
Computation of Tc
SIFS
DIFS
DATA
ACK
� Collision occurs if ACK is not received
TCsingle = TDIFS + TPHY + TMAC + TDATA + π.
� So the average collision time during a transmissioncycle can be expressed as
TC = E[NC ] · TCsingle
20
Computation of Tc
� Pc, collision probability
Pc = Pr{number of transmitting nodes ≥ 2|number of transmitting nodes ≥ 1}
=1− Pr{number of transmitting nodes = 0} − Pr{number of transmitting nodes = 1}
Pr{number of transmitting nodes ≥ 1}
=1− (1− τg)
ng (1− τn)nn − [ng · τg(1− τg)
ng−1(1− τn)nn + nn · τn(1− τn)
nn−1(1− τg)ng ]
1− (1− τg)ng (1− τn)nn
� We also know that
Pr{NC = i} = P ic · (1− Pc)
� So
E[NC ] =∞∑
i=0
i · Pr{NC = i}
=Pc
1− Pc
21
Computation of TI
� From the transmission cycle diagram, we know that
TI = (E[NC ] + 1) · σE[NS ]
� The distribution of NS is
Pr{NS = i} = [(1−τg)ng (1−τn)
nn ]i[1−(1−τg)ng(1−τn)
nn ]
� Therefore,
E[NS ] =∞∑
i=0
iPr{NS = i}
=(1− τg)
ng (1− τn)nn
1− (1− τg)ng (1− τn)nn
22
Channel access probability
Pksg =Pr{number of tx for greedy node i in a period = k}
=m−k∑
i=0
m−k−i∑
j=0
m!
k! · i! · j! · (m−k− i− j)!Pksg · (P
−1sg)i · (sun(1−Pc))
j ·Pm−k−i−jc
m∑
k=1
Pksg =Pr{number of tx for greedy node i in a period≥ 1}
= 1−P0sg
= 1−m∑
i=0
m−i∑
j=0
m!
i! · j! · (m− i− j)!(P−1sg )
i · (sun(1−Pc))j · Pm−i−jc
23
Parameters adopted both in simulation &
analysis
Table 1: values of parameterPHY header(including preamble): TPHY 192µ secMAC header: TMAC 224µ secchannel bit rate 1Mbpspropagation delay: π 1µ seclength of a time slot: σ 20µ secduration of SIFS: TSIFS 10µ secduration of DIFS: TDIFS 50µ secduration of ACK: TACK 304 µ secpayload 1000bytesCWmin 31CWmax 1023
24
Results - Modeling
• transmission probability of MoIP and normal node
25
Results - Modeling
• Collision probability of MoIP and normal node
26
Simulation environment
• NS-2
– 100 x 100 grid space
– CBR traffic with small interval for saturation
condition
27
Results – throughput
• Throughput for MoIP and normal nodes
– Node number of both MoIP and normal nodes are the same
28
Results – throughput
• Throughput for MoIP and normal nodes
– Node number of both MoIP and normal nodes are not the
same
Number of node(normal) = 2 Number of node(MoIP) = 2
29
Result - channel access probability
• Prob(number of tx in one period >= 1)
– Node number of both MoIP and normal nodes are the same
0
0.2
0.4
0.6
0.8
1
1 2 3 4 5 10 15 20
number of node
probability analytical-greedy
analytical-normal
simulation-greedy
simulation-normal
30
Result - channel access probability
• Prob(number of tx in one period >= 1)
– Node number of both MoIP and normal nodes are not the
same (number of MoIP node = 2)
0
0.2
0.4
0.6
0.8
1
1 2 3 4 5 6 10 15 20
number of node
probability analytical-greedy
analytical-normal
simulation-greedy
simulation-normal
31
Effect of CWmin - throughput
0
50
100
150
200
250
2 3 4 5 10 15 20
number of node
thro
ughput
(Kbps)
greedy(CW=16)
normal(CW=16)
greedy(CW=32)
normal(CW=32)
greedy(CW=8)
normal(CW=8)
32
Effect of CWmin - throughput
• The gap between MoIP node and normal node with
different initial contention window size (CWmin)
0
50
100
150
8 16 32 64 128
initial contention window size
(CWmin)
thro
ughput (K
bps)
greedy(node = 5)
normal(node = 5)
greedy(node = 10)
normal(node = 10)
33
Effect of CWmin – channel access probability
0
0.2
0.4
0.6
0.8
1
1 2 3 4 5 10 15
number of node
probability
simulation-greedy(16)
simulation-normal(16)
analytical-greedy(16)
analytical-normao(16)
simulation-greedy(32)
simulation-normal(32)
analytical-greedy(32)
analytical-normal(32)
34
Possible Future Work
• Simulation enhancement
– Asynchronous transmission
– VBR traffic
• Such as video(H.264) or voice transmission
• Fairness issue
CW=CWmin
backoff
High QoS High QoS
backoff backoffTr Tn
CW=CWmin CW=CWmin
T
High QoS
35
Possible Future Work
• Integrate with “Adaptive Video Frame Prioritization”
otherdata
APP-layer
MAC-layer
PHY-layer
36
Conclusion
• This method coheres with our expectation that MoIP node
has higher probability to access the channel than normal
node and higher throughput
• We propose a analytical model to approximately model
this mechanism
• The analytical results and simulation results can roughly
match
37
References
• [1] G. Bianchi, "Performance analysis of the IEEE 802.11 distributed
coordinationfunction.“, IEEE Journal on Selected Areas in
Communications, vol. 18, 2000, pp. 535-547.
• [2] Randolph Nelson. Probability, Stochastic Process, and Queueing
Theory : the mathematics of computer performance modeling. Springer-
Verlag, 1995, pp. 348- 349
• [3] Yu-Liang Kuo, Chi-Hung Lu, Eric Hsiao-Kuang Wu, and Gen-Huey
Chen, “Performance Analysis of enhanced distributed coordination
function in IEEE 802.11e,” Proceedings of the IEEE Vehicular
Technology Conference Fall (VTC) Fall, vol. 3, Orlando U.S., pp.1412-
1416, October 2003
• [4] Y.-L. Kuo, C.-H. Lu, Eric H.-K. Wu, and G.-H. Chen, “An admission
control strategy for differentiated services in IEEE 802.11” Proceedings
of the IEEE Global Communications Conference (GLOBECOM), vol. 2,
San Francisco, December 2003, pp. 707-712
• [5] S. Mangold, C. Sunghyun, G. R. Hiertz, O. Klein, and B. Walke, “Analysis
of IEE 802.11e for QoS support in wireless LANs,” IEEE Wireless
Communications, vol. 10, pp. 40-50, 2003.
38
Thanks