Documente Academic
Documente Profesional
Documente Cultură
, Golnaz Farhadi
, John Cio
i=1
2
R
b
1
G
i
(1)
Now assume S
1
acts as a hotspot and relays S
2
and S
3
data
to the destination node D. The 2 TVWS channels are used by
each of the slave nodes S
2
and S
3
and the 3 LTE channels
are used for transmission between the hotspot S
1
and the D
node. The total power in this mode is calculated as follows:
P
R
= 3
2
3R
3b
1
G
1
+
i=2,3
2
R
b
1
G
i1
(2)
The ratio of the power consumed in direct mode relative to
relay mode is given by:
P
D
P
R
=
3
i=1
1
Gi
3
G1
+
i=2,3
1
Gi1
, (3)
where comparison is among a) direct:
1
G2
+
1
G3
and b) relay:
2
G1
+
1
G21
+
1
G31
terms. In this scenario, term b) is smaller
than term a) (hence lower transmission power in relay mode)
for two reasons:
1) The channel gain to noise ratios G
ij
over TVWS are in
general much greater than the G
i
s over LTE, because
of the lower TVWS frequencies, and
2) The hotspot S
1
is selected such that it has better channel
gain to the destination node and hence G
1
G
2
, G
3
.
III. SYSTEM MODEL
Assume a cell (e.g. rectangular region) with the BS located
at its center. There are U active users/nodes uniformly dis-
tributed inside the cell and need to communicate with the
BS. Each node u U = {1, 2, ..., U} is selected to be
either a hotspot or slave. The set of hotspots is denoted
by H and the set of slaves is denoted by S, such that
S H = U and S H = . This approach creates a two
layer transmission scheme, where layer 1 consists of the BS
to hotspot connections and layer 2 consists of hotpspot to
slave connections. Other than providing the required data to
the hotspot nodes, themselves, layer 1 connections are used as
wireless back-haul for the slaves of each hotspot. It is assumed
that each node can operate on both cellular 4G and the TVWS
bands with the same cellular access technology.
Notation: Consider K
n
ij
. The second subscript, j, corre-
sponds to the transmitter and the rst subscript, i, corresponds
to the receiver when i = j. The superscript n corresponds to
the operating frequency band. K
n
jj
correpsonds to the case
where the transmitter is the BS and the receiver is node j. For
ease of representation, in most cases we refer to the hotspots
with j and the slaves with i.
Two sets of spectrum resources are available within the cell,
such that the layer 1 and layer 2 connections operate on non-
overlapping spectrum. The rst set consists of M channels of
total width W
1
, and the second set consists of N channels of
total width W
2
. Each set of channels n N and m M
can correspond to different or similar spectrum bands. For
example, one set can correspond to the TVWS bands and the
other set can correspond to the LTE bands or both sets can
correspond to the same but non-overlapping LTE or TVWS
bands. Resource allocation is based on resource block groups
(RBG) of an LTE network [5, Ch.7], however, it can be based
on subcarrier bands of an OFDMA network, as well. The
number of resource blocks (RB) in each RBG depends on the
bandwidth of the LTE frequency band. For every 5MHz band,
13 resource block groups of 360KHz are assumed here. Each
TVWS channel spans 6MHz of bandwidth, but to be consistent
with cellular LTE technology, only 5MHz is used (1MHz
guard band) and resource allocation is based on cellular LTE,
such that there are 13 RBGs per TVWS channel.
Each node i has a required rate R
i
that has to be satised.
If a node is a hotspot, it has to satisfy its own rate requirement
as well as its slaves rate requirements. The average channel
gain to noise power ratio across each RBG n between any
two nodes i and j is denoted by g
n
ij
. Each cluster can use the
set of available RBGs n under the condition that it creates
insignicant interference on other clusters.
The link variable y
ij
denotes if there is a link between the
nodes i and j. For the case where i = j, y
ij
is equal to 1 if j
1. Cluster
nodes
2. Resource allocation
among slaves
End
Initialize w
Reduce w
ij
for i
that requires
largest P
ij
3. Power
and rate
requirements
met
5. Rate
requirements
met
4. Resource allocation
among hotspots
Reduce w
jj
for j
that requires
largest P
jj
w
w
yes
no
no
yes
Fig. 3: Proposed iterative clustering & resource allocation algorithm
transmits directly to i and it is 0 otherwise. The term y
jj
= 1
indicates that node j is directly connected to the BS (i.e. j is
a hotspot) and y
jj
= 0, indicates that j is a slave. The U U
matrix Y contains all such variables. The RBG variable x
n
ij
indicates if RBG n is being used by the i, j link or not. The
parameter x
n
ij
is equal to 1 if RBG n is being used by the i, j
link and it is 0 otherwise. The U U max(M, N) matrix X
contains all such variables. The power variable p
n
ij
indicates
the power allocated to the nth RBG on the i, j link. The
U U max(M, N) matrix P contains all such variables.
IV. ITERATIVE HOTSPOT-SLAVE SELECTION AND
RESOURCE ALLOCATION ALGORITHM
The proposed algorithm consists of ve main steps as
summarized by the owchart in gure 3. The darker colored
shapes represent the steps that are taken at the BS and the
lighter shapes represent steps that can be taken at each hotspot.
Step-1: Clustering Nodes into Hotposts and Slaves
The rst step consists of clustering the nodes into groups
by selecting hotspots and the corresponding slaves. The BS
performs such clustering based on the average channel gains
g
ij
s between the nodes and the nodes and the BS. The BS
then sends a signal to the hotspots informing them about their
slaves. Clustering can be based on maximizing the spectral
efciency between the nodes or minimizing the transmission
power in hotspot-slave mode. The optimization based on
minimizing transmission power according to the generalized
version of the equations in section II, is as follows:
minimize
Y
U
j=1
y
jj
w
jj
g
jj
+
i=j
y
ij
w
jj
g
jj
+
i=j
y
ij
w
ij
g
ij
(4)
subject to
N
j=1
y
ij
== 1, i (5)
y
ij
y
jj
, i, j (6)
y
ij
= {0, 1} , i, j (7)
where the rst and second terms in the objective correspond to
the transmission power from a hotspot to the BS carrying the
rate of the hotspot and its slaves. The third term corresponds
to the transmission power from the slaves to the hotspot.
Parameter w
ij
[0, 1] is the weight given to link i j.
The w
ij
s are initialized to 1 and can be reduced based on the
outcome of the constraint checks performed in steps 3 and 5 of
the algorithm. Reducing w
ij
in the clustering objective means
assuming a lower channel gain between the i and j nodes
and hence trying to separate the two nodes when clustering.
Constraint (5) suggests that each node i can have only one
transmitter j. Constraint (6) forces node i to be connected to
hotspot j only if j is a hotspot. If j is a hotspot (y
jj
= 1),
then y
ij
can equal 1, as well, but if y
jj
is equal to zero (j is
not a hotspot), y
ij
cannot equal 1. Constraint (7) enforces y
ij
to take on binary values.
This is an integer linear programming problem. In order
to convert this problem into a general linear programming
problem, constraint (7) is relaxed to y
ij
0, i, j. Hence the
problem can be solved using interior point methods, which
have polynomial-time complexity [6]. Before moving on to
the next step, the clustering conguration result has to pass a
slave-hotspot check. If a slave is found that has a better average
channel gain to a hotspot that is different from the hotspot that
it has been assigned to, its corresponding slave-hotspot weight
will be reduced and then reclustered.
Step-2: Resource Allocation Among Slaves
Denote the set of slaves associated with hotspot j as S
j
obtained in the clustering step of the algorithm, i.e., S
j
=
{i|y
ij
= 1}. In the second step of the algorithm, each hotspot
allocates resource block groups to its slaves based on an
OFDMA-like approach with rate and power constraints. In this
approach, each RBG can be used by only one slave. Resource
allocation in this step is similar to the optimal resource
allocation optimization performed in OFDMA systems [7].
The difference here is allocating RBGs instead of subcarriers.
Determining the Maximum Allowed Transmission Power:
In order to prevent interference to neighboring clusters, each
cluster determines the maximum power that it can transmit on
each RBG. This ensures that no cluster is creating signicant
interference on others. Let T
n
j
be the maximum transmission
power that cluster j (i.e., the cluster whose hotspot is node j)
can transmit on RBG n. Interference is considered negligible if
it is on the order of noise power N
0
. Considering that the noise
power N
0
is embedded in the parameter g
ij
, the maximum
transmission power over each RBG n for each cluster j is
dened as:
T
n
j
=
max
kK
g
n
lk
, (8)
where (0, 1] is the noise proportionality factor. The
parameters l and K are dened based on uplink or downlink
operation:
l = j and K = S S
j
, downlink
l S
j
and K = H{j} , uplink
If downlink communication (hotspot to slave) is considered,
l = j and interference is measured based on the channel gain
between hotspot j and the slaves of all other clusters. If uplink
communication (slave to hotspot) is considered, l S
j
and
interference on each RBG is calculated based on the channel
gain between all the j cluster slaves, S
j
, and all other hotspots.
Allocating RBGs and Transmission Powers: After deter-
mining the maximum transmission power on each RBG, and
having knowledge of the rate requirement of each slave, each
hotspot solves the following optimization problem:
minimize
P,X
iSj
N
n=1
p
n
ij
(9)
subject to
N
n=1
x
n
ij
b log(1 +
p
n
ij
g
n
ij
x
n
ij
) R
i
, i S
j
(10)
p
n
ij
T
n
j
, n, i S
j
(11)
iSj
x
n
ij
1, n (12)
p
n
ij
0, n, i S
j
(13)
x
n
ij
= {0, 1} n, i S
j
(14)
The objective of the algorithm is to minimize the total
transmission power over all the available RBGs subject to
minimum rate requirements and the maximum transmission
powers over each RBG. Constraints (10) and (11) determine
the rate and power constraints, respectively. Constraint (13)
enforces power to be positive and constraint (14) forces the
channel selection variables x
n
ij
to take on binary values. This
is a mixed integer programming problem with a combinatorial
solution, hence, the integer constraint is relaxed and solved via
dual decomposition. The dual variable updates are performed
using the Ellipsoid method with polynomial-time complexity
[8]. As discussed in [8], discontinuity in power allocation at
the obtained optimum dual point can occur. In this situation,
the RBG assignment found at the optimum dual point is used
and then power is allocated to satisfy the rate requirements.
Step-3: Cluster Rate and Power Constraint Check
This step performs a check on two constraints: 1) The
optimization performed at each hotspot is feasible. That is each
cluster has satised the rate requirement of each of its nodes
based on the maximum allowed transmission powers T
n
j
. 2)
The transmission power of each node is less than the FCC
mandated transmission power on TVWS channels. If either
constraint is not met, all the nodes need to be reclustered.
In order to recluster the nodes with the hope of the new
clusters satisfying their constraints, the problematic clusters
have to reduce the number of their slaves. This is performed
by reducing the weights w
ij
of the i j links that need too
much power, in the clustering problem (4). The hotspots j
whose constraints are not met, select the slave i that requires
the highest transmission power and reduces its weight w
ij
. It
then noties the BS with its new decision. The BS, collects the
updated weights, and solves the clustering problem (4) again
by returning to step 1 of the algorithm.
Step-4: Resource Allocation Among Hotspots
When the constraints at all clusters are met, the BS allocates
its resources among the selected hotspots j H. The resource
allocation in this step is similar to step 3 resource allocation,
with a slight difference in the rate requirements.
minimize
P,X
jH
M
m=1
p
m
jj
(15)
subject to
M
m=1
x
m
jj
b log(1 +
p
m
jj
g
m
jj
x
m
jj
) R
j
+
iSj
R
i
, j H
(16)
p
m
jj
Z
m
, m, j H (17)
jH
x
m
jj
1, m (18)
p
m
jj
0, m, j H (19)
x
m
jj
= {0, 1} m, j H, (20)
Constraint (16) denotes that the rate hotspot j obtains has to
be greater than the sum of it own rate requirement and its
slaves rates. The parameter Z
m
in constraint (17) indicates
the maximum power that can be transmitted on RBG m.
Step-5: BS Rate and Power Constraint Check
In this step the feasibility of the resource allocation per-
formed in step 4 is checked to make sure that all the hotspots
receive the required resources to satisfy their cluster rate
requirements. If the problem in step 4 is not feasible, that is,
there are not enough resources available, the BS reclusters the
nodes after updating the w
jj
values. The BS selects the hotspot
j