Documente Academic
Documente Profesional
Documente Cultură
1
II-A Physical layer E E
45 45
degrees degrees
As mentioned in IEEE 802.16-2004 Standard, only time divi-
sion mode (TDD) is supported in Mesh mode. Furthermore, Link A Link A
the MAC layer is assumed to schedule data to multiple access G
d*0.1 G d*0.1
(TDMA) through single carrier channel. Thus, as long as the
bandwidth allocation result is calculated, the frame in each link F
can be built following this result in a simple way of mapping.
Link B Link B
We have the following rules defined in this paper: Rule 1:
A node can’t transmit and receive at the same time. Rule 2: (a) (b)
H H
The relaying data traffic received by one SS can’t be trans-
ferred immediately to its neighbor in the same frame slot. That
is because the SS usually has performance constrain in buffer Figure 1: Region of Interference
writing and reading. Rule 3: Nodes within the transmission
range of an active node are blocked to avoid the interference.
Rule 4: Any two traffic that are not interfering with each other III Mathematical Model
can potentially transmit data packets over the physical channel
Centralized scheduler provides bandwidth allocation scheme
simultaneously.
for each node such that traffic can reach its destination in the
It is believed that interference between concurrent transmis- scheduling period. Usually, the longer scheduling period the
sions from neighboring nodes is one of the most significant lower bandwidth efficiency, so minimizing the scheduling pe-
factors that limit the system throughput and scalability of wire- riod becomes an important object in scheduling algorithm de-
less multi-hop network. Directional antennae at the transmit- sign. In the following, we build an optimization model to study
ters as well as the receivers to minimize transmission range of the minimal scheduling period that can be provided by an cen-
the nodes. According to the Rule 4 the throughput of the mesh tralized scheduler with specified topology and traffic distribu-
work can be improved. tion.
Given an access tree where the nodes are ac-
As mentioned above, all nodes share a wireless channel and
cess points, and the links are bidirectional wireless links be-
communicate on that shared channel. Each node is assumed to
tween neighboring pairs of access points, . All nodes
be equipped with multiple directional antennas[10]. A direc-
in are labelled with an integer and the root node is labelled
tional antenna can transmit (receive) over a small angle (e.g.,
!" #
with . The root node is BS, and the other nodes
45 degrees), centered on the receiver (transmitter), and several are SS. Each node has a specified capacity which is
directional antennas may be used together to cover all direc-
tions. We also assume that there is no interference at angles
$&% '$ % ( *)+,,)-./01
2
the data rate it can support.
, is the neighboring parent nodes of
$ % 34
beyond this beam width, or at distances beyond 10% of the node . With being an access tree, each node has at most one
transmitter receiver link length (see Fig.1).
% 5*)+67*
)89:
;
neighboring parent node, that is, .
)<= % >/%>?@4-A,AB % % C >7D.E
be the neighboring children of node .
Directional transmissions over two different links will in-
FC
Each is given a label , and is the
HI%C
terfere at the two receiving access points if the access points
>GD.E
neighboring child of node . and all its children form the
J G/ represents all the children of node , and J 7KL M N'P OKM HS%C .
are located within the beam of other link, and transmissions
branch of node which is denoted as .
by two or more links will interfere at the same receive node
even if different directional antennas at the node receive these
transmissions. This assumption is justified, for instance, in the
V-ET% 7Kis isthetheuplink
number of hops from root U to node X
V *
W % . C,Q&R
traffic request of node and
case where signals received by all antennas are combined be-
^
shows the interference. The Fig.1(b) also shows a condition shown in the figure.
where node G can safely transmit to node H, since it will not Let represent the number of frame slots to carry all the
interfere with node E’s communication to F. requests to its destination, and the scheduling problem, is to
2
Z Ws[ \ ~xY143srtrsrpu^ (9)
7^ l 4mET7KuKD.E
BS
Constraint (2) represents that the uplink traffic via node to
Fi
E&G/
root node must be transmitted out before the
yFl i , k frame such that these traffic can reach root node after hops.
xi , k The uplink traffic via node includes both the traffic originated
i
at node and the traffic originated by children of and relayed
xN1 ,k yi1,k
N
yi , ik by to root node.
^
i
yi2, k xNNi i,k Equation (3) means that the downlink traffic from root node
xN 2 , k
1
i
Ni
to node should reach node inside the scheduling period .
N Ni
#
i
...... Constraint (4) requires that the total traffic to and from node
N i2 can not exceed the total capacity of node .
......
Y+Y4-trtrsrt^
Constraint (5) means that the uplink traffic sent by node
during frame slots should not be more
Figure 2: Access Tree and its Notions
Y
than what it has received from its neighboring children and
its own originated traffic during the frames. And constraint
^
find a scheme to minimize . This problem can be formulated
Y+Y=43trsrtrp^
(6) requires that the downlink traffic sent by node during
frames should not be more than what it has
in a linear programme problem as follows:
Z Ws[ \
received from its neighboring parent node.
(1) parent node. So . For any SS node
b A DA ced
, there is no downlink traffic unless the SS node has
%ji ^ l ET7K
received some traffic from its parent nodes, and all uplink traf-
k Rf+g3h Z %7[ \ V %ml k V n .'143srtrtrtuA fic should be sent before frame to guarantee the
Q&R c
k V W*%ml k V W n
III-A Scheduling Algorithm
\ ] vC O [ \ fwR npo3q h %ji The linear programme problem can be solved by many com-
Q&R Z %G[ \
mercial tools and solution to the mathematical model can pro-
]%GC [ \
'143srtrsrpuAxy v C O u>z14rtrsrs v O 6A vide uplink traffic
YwY43trsrtrp^
and downlink traffic of node in
M N'O/M M N'O/M
(3)
#
each frame . The solution does not specify
Z %G[ \{l k ] %7C [ \ l ] vC O [ \ l k Z NeO | [ \I} # 7
% [ \ M N'O/M %G[ \
that during the mini-slots for each frame slot, how to arrange
Z
C,Q&R C6Q&R ]XC downlink traffic such that no
,
C &
Q R
the uplink traffic and
xy~srtrsrpu' &>T43trsrtrp % 6Av C O (4) interference occurs. In the following, we propose an algorithm
%7to[ \ its mini-slots based on
\ \
k Z %G[ k fwR kM N&O/M Z NO | [ } V % above principle. In the algorithm, represents the DKD.E mini-
to allocate the traffic of each node
\ \
(5)
Y.
k kM N'O/M ] %7C [ } k fwR ] vC O [ 7
% [ \ 7
% [ \ Algorithm
Input: 1
,
Algorithm 1: Time Slot
Z u] u#u%7[ \ *Y;43srtrtrsu^
Allocation
Q'R C6Q&R Q&R Output: Mini slots assignment .D143trsrtrp# for each *Y .
Y;srtrtrtu^u43trsrtrp.x v C O (6)
%G[ \ %7[ \
1. For each Yw Y24-trtrsrt^ , get Z u]8C
begin
] %7C [ \ ~*Y } ET7/ L% '>T%G[ \ 43srtrsrp % A %7[ \ L{K+D } #A
x4-trtrsrpuA{>T43srtrsrp % sDt W y.D43srtrtrtu#
2. L ¡~* >w143A
Z %G[ \ ~Y9
"^ l ET7/'4-trsrtrpA
(7)
(8) /*Begin from BS node.*/
3
0
BS
1 2 0<−−−>1
0<−−−>2
3 4 5
1<−−−>3
2<−−−>4
6 7
Link
2<−−−>5 downlink traffic
uplink traffic
8
3<−−−>6
3<−−−>7
Figure 3: Simulation Topology
7<−−−>8
Select D?9 , if ] C
% %7[ \S¢
% G[ \ b £% C , ]8%7C [ \ y]8%7C [ \ , ¤4-. % ¡ % =D* ,
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Frame
3.
If ]8C
if >
} % ,, gogo toto step
step 3, otherwise >
3.
,
%
For )C u>z143srtrtrt , D?¥
% %m¦ n
4.
if Z n
[ \ ¢ , n [ \ b 7Kp %7[ \ y,)8p Z n [ \ Z n [ \ ¤4-
nW y n ¤sD*-u % §W % ¤sD* , go to step 4. W 0<−−−>1
5.
$ % ¡ :W , set
= . If , select L0 , and
>z14 , go to step 3.
0<−−−>2
1<−−−>3
end
2<−−−>4
uplink traffic
Link
downlink traffic
2<−−−>5
3<−−−>7
In this section, we use the network shown as Fig.3 as an exam-
ple to compare the scheduling scheme based on the proposed 7<−−−>8
model with FIFO queue. FIFO queue serves the request ac- 12 12.1 12.2 12.3 12.4 12.5 12.6
Minslot of the 13th Frame
12.7 12.8 12.9 13
4
80
Proposed Mode der centralized scheduling in minimal time slots and saves time
FIFO Mode
70 zone for distributed scheduling as much as possible. Compar-
ing with FIFO serving mode, the scheduling scheme based on
60
our proposed model can save half of the time slots under differ-
50
ent traffic load and Concurrence Rate is 3 times better. Simu-
Frame Slots
40
lation results also show that our model has better compatibility
for various traffic distributions.
30
Priority difference between the traffic generated by each
20 node is not considered in this paper. Though Best Effort is
the only service type supported by Mesh Mode defined in pro-
10
tocol, it is still possible to classify different service priorities
0
0.5 1.0 1.5 2.0
in priority/class field in the Mesh Connection Identifier (CID)
Traffic(Unit)
construction. Enhancement of the optimization model to sup-
port multi-priority traffic and local scheduling policy of SSs is
Figure 7: Frame Slots v.s. Traffic Load
our future work.
1:1 column in Fig.6). Fig.4 shows that many links are active si-
multaneously in the same frame slot. But in FIFO mode, only References
one link is active in most of the scheduling time for it serves [1] WiMAX Forum. Business case models for fixed broadband wire-
each node in sequence. In the following, we define the con- less access based on WiMAX technology and the 802.16 stan-
cept of Concurrence Rate, such that the concurrency degree of dard. Whitepaper, October 2004. http://www.wimaxforum.org.
Y+*Y¨©43trsrtrp^ ª \
different scheduling schemes can be measured.
Given a frame slot , is the number [2] IEEE 802.16 Standard-Local and Metropolitan Area Networks-
c
of links that are active for downlink or uplink traffic. Then Part 16. IEEE Draft P802.16/D3-2001
Fig.4 is ¬XX® . Use frame slot 13 as an example, the frame struc- [4] P. Gupta, P.R. Kumar, The capacity of wireless networks, IEEE
ture of each link is shown in Fig. 5. Totally, 7 links(¯;°
Transactions on Information Theory 46 (2) (2000) 388-404.
43¥¯;°±s4;¯;°²~*9¯;°³¬*9¯;°±´¥¯;°¶µ8µ¯;°²· ) [5] Pabst, Walke, et al, “Relay-Based Deployment Concepts for