1. A Dynamic Resource Allocation Scheme for Guaranteed Bit Rate Services in OFDMA Networks
Chrysostomos Koutsimanis and G´abor Fodor
Ericsson Research, SE Stockholm, Sweden
ICC 2008 1 1

2. Outline Introduction OFDM Resource Allocation for Narrow-band Services
OFDM Resource Allocation for Elastic Services
Numerical Results
Conclusions 2 2

3. Introduction
Several previous works have considered the problem of resource allocation in multi-cell OFDMA networks
but only a few contributions have explicitly taken into account the elastic nature of data applications
In this work, we consider two kinds of user requirements
Guaranteed bit rate (GBR):
with a minimum and maximum resource block (RB) requirement
Peak bit rate (PBR)
with a fixed RB requirement
We propose a method that balances between maximizing the overall throughput and being feasible in real systems

4. Key Trade-offs Key trade-offs has not been proposed
Coordination between BSs obviously increases the overall system throughput at the expense of backhaul communication and intra-node processing
Limiting the use of some of the sub-carriers reduces inter-cell collisions at the expense of loosing some degree of multi-user diversity in a frequency selective environment
Throughput maximization often leads to unfair allocation of resources which in turn may lead to QoS violations for GBR users

5. Dealing with the Trade-offs (1)
In order to deal with the first and second trade-off above, we distinguish between two time scales
Superframe level resource allocation
for inter-cell resource coordination
an OFDM superframe that consists of a number of consecutive frames
including inter-cell collision avoidance and inter-cell power control based on the longer term channel conditions
Frame level resource allocation
for intra-cell scheduling and resource allocation
allocating sub-carriers and power for the duration of the next scheduling interval (being typically at the millisecond level) based on the instantaneous channel gains of the users

6. Dealing with the Trade-offs (2)
Finally, the third aspect is addressed by allowing a minimum and a maximum number of resource blocks to be associated with each user
Requiring that the number of resource blocks allocated to each user must be between these values

7. Basic Considerations
We first consider narrow-band (PBR requirement) services
The corresponding RB requirement is ni = Nmin for each user i
Environment
The frequency resource is divided into N traffic channels
The time resource is divided into frames and superframes
The OFDMA system consists of
L base stations (BS)
users
Ml denotes the number of users that are connected to BS-l
l(m) denotes the BS that serves User-m
Gm,l is the long term channel gain between User-m and BS-l
ym,n take the value of 1 whenever RB-n is assigned to User-m
Pl,n denote the transmission power employed by BS-l on RB-n

8. Total Number of Bits Carried over RB-n
The long term SINR values experienced by User-i on RB-n
where Ml is the set of users served by BS-l
We assume that the system employs an adaptive modulation and coding scheme (AMC) that is characterized by the link adaptation function fLA
The average number of bits transmitted (during a superframe) for User-i on RB-n can be expressed by
assume the fLA function is given
The total number of bits carried over RB-n in the multi-cell system is

9. Superframe Level Problem Formulation (1)
The resource assignment problem at the superframe level can now be formulated as
finding the Y and P matrices such that the overall multi-cell throughput is maximized

10. Superframe Level Problem Formulation (2)
The key characteristics of the superframe level allocation is that it does not require the instantaneous channel conditions
it does not require inter BS communication at the frame level
Once Y and P are available and assuming that the instantaneous channel conditions are available, it is possible to take advantage of multi-user frequency diversity
Let hm,n denote the instantaneous channel gain between User-m and BS-l(m) on RB-n

11. Frame Level Problem Formulation (1)
When RB-n is assigned to User-i, the number of bits transmitted on that RB becomes
where denotes the instantaneous SINR
Let zi,n be an indicator taking the value of 1 when RB-n is assigned to User-i

12. Frame Level Problem Formulation (2)
The frame level optimization task for each BS-l is formulated as follows
building on the results from the superframe level

13. Resource Block Assignment (1)
We assume that a central entity gathers intercell information and performs intercell channel assignment and power control
according to the proposed Radio Network Controller (RNC) algorithm
Resource Block Assignment
Within the BS loop, it calculates the throughput contribution (Ω) of each served user on each resource block
this calculation takes into account previous allocations to already examined BSs and treats the produced interference as background noise
After calculating Ω for all the users of the particular BS over all resource blocks, a binary linear optimization problem maximizes the sum of these values by assigning RBs to users
under the constraint that a user must not be assigned more than ni RBs

14. Resource Block Assignment (2)
initialize variables
calculate throughput contribution of user m on RB-n
find optimal allocation for BS k

15. Power Control
The power control algorithm attempts to further increase the overall throughput by decreasing the power that is allowed for each resource block
decreasing the interference level
The Jm element of the vector J specifies the index of the dominant interfering base station of User-m
The frame level problem is an integer linear programming problem and can be solved with standard techniques

16. OFDM Resource Allocation for Elastic Services
We now consider elastic services and the corresponding resource block requirement is ni,min = Nmin ≤ ni,max = Nmax for each user i
For elastic services, the third constraint of superframe level allocation takes the form of
The frame level allocation problem for elastic services is similar to that of narrow band services
except that the exact number of resource blocks ni in the second constraint is an output of the superframe level problem solution rather than a predefined value

17. Radio Network Controller Algorithm for Elastic Services

18. Simulation Environment and System Parameters

19. Numerical Results for Narrow Band Services (1)
Total number of users: 7 * 15
The number of required resource blocks per user: nmin = 2
average load per cell in terms of the used resource blocks: 37.5% F SF
SF+F F SF
SF+F

20. Numerical Results for Narrow Band Services (2)
SF+F vs. F
SF+F F SF

21. Numerical Results for Elastic Services
Each session is associated with
nmin = 3

22. Conclusions
We considered a multi-cell OFDMA network that supports narrow band and elastic services
It is appropriate to model these types of services as ones which have an associated minimum and maximum OFDM resource block requirement
We formulated the resource block and power allocation problem as a throughput maximization problem
We proposed a two level approach that separates the inter-cell coordination task appropriate at the OFDM superframe level from intra-cell scheduling appropriate at the frame level
By means of system simulations we found that the two level approach seems feasible in a realistic model
in terms of complexity and execution time