IEEE COMMUNICATIONS LETTERS, VOL. 16, NO.

4, APRIL 2012 479

Joint Power and Bandwidth Allocation Algorithm with

QoS Support in Heterogeneous Wireless Networks
Jie Miao, Zheng Hu, Kun Yang, Canru Wang, and Hui Tian

Abstract—Heterogeneous wireless networks give rise to a

multi-radio access environment which becomes crucial for load
balancing to avoid network congestion and providing high quality
services to maximize radio resource utilization. To this end,
a joint power and bandwidth allocation with QoS support
algorithm is proposed using convex optimization method in this
letter. The algorithm achieves the goal of the total system capacity
maximization, while efficiently satisfying the minimum rate
constraint of delay-constraint service traffic and proportional ...
fairness of best-effort service traffic. The simulation results show
that the proposed algorithm significantly outperforms the existing
scheme in QoS guarantee.
Index Terms—Heterogeneous wireless networks, power, band-
width, QoS.
...
I. I NTRODUCTION

W ITH heterogeneous wireless networks deployed in the

same area, multi-mode terminals (MMTs) are able to
access multiple heterogeneous wireless networks simultane-
Fig. 1. The multi-radio access scenario in heterogeneous wireless networks.

the main contributions are QoS support consideration on the

ously. For such a multi-radio access system, it is necessary minimum rate of delay-constraint (DC) traffic and fairness
to design a novel scheme among Radio Access Technologies guarantee among best-effort (BE) traffic. And the proposed
(RATs) to efficiently utilize scarce radio resource. However, algorithm implemented in distribute way outperforms in QoS
radio resource allocation, such as power and frequency band- guarantee than the existing one significantly.
width, presents a huge challenge for operators in term of
exploiting the cooperative diversity and further meeting QoS II. S YSTEM M ODEL AND P ROBLEM F ORMULATION
requirements of diverse service traffic.
The scheme proposes here, both RATs and MMTs are
In general multi-radio access resource allocation problem, implemented by the reconfigurable Software Defined Radio
as proposed in [1]–[3], the main focus resides in resource (SDR) technology in heterogeneous wireless networks. Hence,
allocation without traffic class consideration. Dimou et al. [1]
each MMT is able to access multiple RATs simultaneously,
propose a greedy based rate allocation algorithm in parallel whereas it is also capable of operating different frequency
transmission across heterogeneous networks. And Zhu et al. bandwidth through parallel transmission. Under this system
[2] address the problem of rate allocation among realtime assumption, MMTs can access the multiple RATs for their
services terminals sharing multiple heterogeneous access net- data transmission in parallel. Besides, there is no interference
works. The joint power and bandwidth allocation scheme has between them; and perfect synchronization is necessary for
been studied by Chio et al. in [3] without service traffic QoS no inference happening due to different frequency bands
consideration. This letter fills the gap exploiting the uplink ra- operation. Channels experience slow fading such that the
dio resource allocation further to provide an optimal resource
instantaneous Channel State Information (CSI) can be updated
allocation solution with QoS support for different classes periodically with the help of feedback channels. In our system
service traffic. Joint power and frequency bandwidth allocation model, we consider a heterogeneous wireless environment
for different classes traffic will be studied in this letter, and consisting of M active MMTs requesting diverse services and
Manuscript received November 15, 2011. The associate editor coordinating N available RATs as depicted in Fig.1. Assuming that M
the review of this letter and approving it for publication was Y.-D. Lin. MMTs are classified by heterogeneous traffic delay require-
This work was sponsored by the National Natural Science Foundation ments: one class terminals have Delay-Constraint (DC) class
of China (No.60971125), the National Key Program of New Generation of
Broadband Wireless Mobile Communication Networks (No.2011ZX03005- service traffic, defined as DC MMTs [4]. And the other one
004-02) and EU FP7 EVANS (No.269323). class (M − K) MMTs transmit Best-Effort (BE) class service
J. Miao, Z. Hu, C. Wang, and H. Tian are with the Wireless Technology traffic, defined as BE MMTs. The former ones require a
Innovation Institute, Key Laboratory of Universal Wireless Communications,
Ministry of Education, Beijing University of Posts and Telecommunications, minimum constant transmission rate of Rimin (i = 1, 2, . . . , K)
Beijing, China (e-mail: [email protected]). bits constraint, the latter ones need to be guaranteed pro-
K. Yang is with the School of Computer Science and Electronic Engineering portional fairness. Each MMT obtains the resource from the
(CSEE), University of Essex, Wivenhoe Park, Colchester, CO4 3SQ, UK (e-
mail: [email protected]). available access networks. For RAT j (j = 1, 2, . . . , N ) and
Digital Object Identifier 10.1109/LCOMM.2012.030512.112304 MMT i (i = 1, 2, . . . , M ), channel transfer function and total
1089-7798/12$31.00 
c 2011 IEEE
480 IEEE COMMUNICATIONS LETTERS, VOL. 16, NO. 4, APRIL 2012

noise power spectral density are denoted as Hij and Nij , Algorithm 1 Joint Power and Bandwidth Allocation Algo-
respectively. Let MMT i throughput be rithm
→
−
N 1: Initialize α and ξ , and estimate average BER ηij , ∀i, j
 gij pij
ri = (1 − ηij )βj xij log2 (1 + ) (1) 2: if k = 0 then
j=1
xij 3: Initialize x0ij , p0ij , μ0i νi0 and ωi0
4: else
where ηij is the average Bit Error Rate (BER) from MMT i
5: Calculate xk+1 using gradient projection method.
to RAT j, βj (0 ≤ βj ≤ 1) represents the system efficiency  ij +
k+1 k ∂L
which can be guaranteed by RAT j to all MMTs [3], xij is 6: xij = xij + α ∂x ij
, ∀i, j
the frequency bandwidth allocated to MMT i from RAT j, pij  j
+
(1−ηij )β 1
is the transmission power of MMT i, and gij is the channel 7: Determine pij = xij μi ln 2 − gij ,∀i, j
gain. 8: if Iteration reaches the convergence precision (condi-
Thus, we have tion) of pij and xij or the maximum iteration number
then
ri ≥ Rimin , i = 1, 2, . . . , K (2) 9: Transmit data packet to the RAT(s) using xk+1 and
ij
rK+1 : rK+2 : . . . : rM = γK+1 : γK+2 : . . . : γM (3) pk+1
ij
10: else
where Rimin (i
= 1, 2, . . . , K) is the minimum rate requirement 11: Update μk+1i , νik+1 , ωik+1 and λk+1
j usingxk+1 k+1
ij , pij
for DC MMT i and γi (i = K + 1, K + 2, . . . , M ) is the information.
proportional fairness factors for BE MMTs. 12: k ←k+1
Therefore, the heterogeneous networks system capacity 13: end if
maximization with QoS support problem can be formulated 14: end if
as follows:
M 
 N
gij pij
max R(x, p) = max (1 − ηij )βj xij log2 (1 + ) for only one RAT of some MMTs available scenario. For
i j
xij
the optimal solution of capacity maximum problem (4), the
(4) Lagrangian formulation is given as below:
subject to:
M
 
M 
N
L(xij , pij ; λj , μi , νi , ωi ) = (1 − ηij )βj xij log 2 (1+
xij ≤ Xj , ∀j (5)
i=1 j=1
i=1
gij pij N M M N K
N
 )+ λj (Xj − xij ) + μi (Pi − pij ) + νi
xij
pij ≤ Pi , ∀i (6) j=1 i=1 i=1 j=1 i=1

j=1 
N
gij pij  M N
βj xij log2 (1 + ) − Rimin + ωi βj x1i
N
 xij
gij pij j=1 i=K+1 j=1
βj xij log2 (1 + ) ≥ Rimin , ∀i ∈ 1, 2, . . . , K (7)
γK+1  gij pij 
N
xij g1j p1j
j=1 log2 (1 + )− βj xij log2 (1 + )
x1j γi j=1 xij
ri γi
= , ∀i ∈ K + 1, K + 2, . . . , M (8) (10)
rK+1 γK+1
xij , pij ≥ 0, ∀i, j (9) where λj , μi , νi and ωi are nonnegative Lagrange multipliers
for the constraints.
where Xj is the total system bandwidth of RAT j, and Pi
The relation between bandwidth and power allocation can
is the maximum available power of MMT i. Note that the
be obtained as
objective function in (5) is concave with respect to {x, p}.  +
This means that an optimal solution can be derived using (1 − ηij )βj 1
pij = xij − (11)
convex optimization method, where a local maximum is also μi ln 2 gij
the global maximum. There is a total system bandwidth and
power constraint due to resource finiteness from inequations where [z]+ = max{z, 0} and βj is defined as follows:
(6) and (7). The other two inequations (8) and (9) constraint ⎧
⎪
⎪(1 + νi ) βj , if i=1,.. . ,K
conditions, which are equivalent to (3) and (4), denote that ⎨ M
QoS constraint for heterogeneous services traffic. βj = 1 + k=K+2 ωk βj , if i=K+1 (12)
⎪
⎪ 
⎩ 1 − ω γK+1 β , if i=K+2,. . . ,M
i γi j
III. O PTIMAL J OINT R ESOURCE A LLOCATION
A LGORITHM According to the convex analysis, the gradient projection
A. Joint Power and Bandwidth Allocation Analysis method is used to approach to the optimal solution, which is
proved to be feasible if the step sizes are properly chosen. So,
This letter proposes a convex optimization method based
best-response method to update the bandwidth is as follows.
joint power and bandwidth allocation with QoS support al-  +
gorithm. In our model, it is assumed that all the MMTs k+1 k ∂L
xij = xij + α , ∀ i, j (13)
have more than one RAT transceiver. And, it is also effective ∂xij
MIAO et al.: JOINT POWER AND BANDWIDTH ALLOCATION ALGORITHM WITH QOS SUPPORT IN HETEROGENEOUS WIRELESS NETWORKS 481

Allocated to MMT1 by RAT1 efficiency (βj = 1, j = 1, 2) and wireless link BER ηij (0 <
ηij < 1) is with normal distribution in [10−3 , 1] for simplicity.
1.6 Allocated to MMT1 by RAT2
Allocated to MMT2 by RAT1
1.4 Allocated to MMT2 by RAT2 In the simulation, it is assumed that there are 3 MMTs with
1.2
Allocated to MMT3 by RAT1 maximum power constraint of 20 mW which are randomly
distributed in overlapped coverage region. 1 DC (R1min =
Allocated to MMT3 by RAT2
Rate (Mbps)

1
1M bps) terminal and 2 BE (γ2 : γ3 = 1 : 1) terminals
0.8 access multiple wireless networks simultaneously. As example
0.6 for finding an optimal solution is provided in Fig.2. The
proposed algorithm resource allocation converge to a unique
0.4
optimal data rate solution, which are jointly determined by
0.2 power and bandwidth convergent values. According to the
0
optimal allocation strategies, it is observed that both BE
0 500 1000 1500 2000 2500 3000 MMTs obtain the same amount of transmission rate under
Iteration (k)
the proportional fairness resource allocation principle. As a
Fig. 2. The convergent data rate solution of the proposed joint power and result, radio resource allocation algorithm with QoS support
bandwidth allocation algorithm. is feasible and can efficiently converge to the global optimal
solution.
Then, we set 10 DC (Rimin = 1M bps, i = 1, 2, . . . , 10)
1.05

1 MMTs and 10 BE MMTs with equal fairness in the sim-

Optimal Scheme for BE MMTs
ulation and vary the SNR for MMTs. Proportional fairness
FI

0.95
Method in [3]
0.9 between MMTs is validated by 
Fairness Index (FI) function
2
i=K+1 ri )
( M
0.85 [5] as follows, F I(ri ) = (M−K) M r2
, where ri is the
10 12 14 16 18 20 i=K+1 i
SNR (dBm) transmission
K
rate of MMT i. And we define ADR(ri , Rimin ) =
i=1 (ri −Ri )
min
20 "

Rmin
as Additional Datarate Ratio (ADR) for DC
i
15 MMTs to weight the degree of satisfaction. To compare the
ADR

degree of satisfaction which the resource allocation scheme

10 affects, both FI for BE MMTs and ADR for DC MMTs are
Optimal Scheme for RT MMTs

5"
Method in [3] presented in Fig.3, which are improved more than 12% and
10 12 14 16 18 20 3% on average respectively. The simulation result shows that
SNR (dBm)
the algorithm outperforms than the existing algorithm [3] in
Fig. 3. The FI and ADR comparison between the proposed algorithm and QoS guaranteeing. The proposed algorithm doesn’t only tend
the method in [3]. to allocate resource to guarantee DC traffic QoS requirements,
fairness is also considered among BE MMTs.
From iterations, we can solve the optimal radio resource
allocation with QoS support problem, which maximizes het- V. C ONCLUSION
erogeneous networks total capacity. A joint power and bandwidth allocation algorithm is pro-
posed in this letter by using convex optimization method
and tackles the QoS support optimization problem of radio
B. Joint Power and Bandwidth Allocation with QoS Support
resource allocation in heterogeneous wireless networks. The
Algorithm
algorithm provides a global optimal solution and accom-
Note that the rij value is determined by both MMT and modates heterogeneous service traffic QoS differentiation in
its connected RATs in Algorithm 1. The proposed algorithm multi-radio access scenario.
realizes the joint power and bandwidth allocation for MMTs
supporting heterogeneous services traffic QoS. Hence, this R EFERENCES
algorithm is implemented by each wireless connection from
[1] K. Dimou, R. Aguero, M. Bortnik, et al., “Generic link layer: a solution
each terminal to all access networks in a distribute decision for multi-radio transmission diversity in communication networks beyond
making manner. 3G,” in Proc. 2005 IEEE Veh. Technol. Conf., pp. 1672–1676.
[2] X. Zhu, P. Agrawal, J. P. Singh, et al., “Distributed rate allocation policies
for multihomed video streaming over heterogeneous access networks,”
IV. S IMULATION R ESULTS IEEE Trans. Multimedia, vol. 11, no. 4, pp. 752–764, 2009.
For simulation, numerical experiments are conducted using [3] Y. Choi, H. Kim, S. Han, and Y. Han, “Joint resource allocation for
parallel multi-radio access in heterogeneous wireless networks,” IEEE
MATLAB. In simulation condition, it is considered that multi- Trans. Wireless Commun. vol. 9, no. 11, pp. 3324–3329, 2010.
radio access scenario where an LTE base station and an IEEE [4] M. Tao, Y. Liang, and F. Zhang, “Resource allocation for delay differenti-
802.11b access point offer access link service to MMTs in ated traffic in multiuser OFDM systems,” IEEE Trans. Wireless Commun.,
vol. 7, no. 6, pp. 2190–2201, 2008.
the overlapping coverage area. And it is assumed that both [5] R. Jain, D. Chiu, and W. Hawe, “A quantitative measure of fariness and
have total frequency bandwidth of 20 MHz and 16.6 MHz discrimination for resource allocation in shared computer systems,” DEC
for service data, respectively. The channels are with the same Research Report TR-301, 1984.