Problem Formulation
The proposed RBA optimisation problem is based on maximising the sum-SINR of the users within each macro cell, while ensuring the effective interference experienced is at its minimum. The optimisation problem can be formulated as:
max a NRB X r=1 ¯ Km X k a(m)k,r γk,r(m), ∀m ∈ [1, 2, · · · , M] , s.t. γk,r(m)= g (m) k,r ρ (m) k,r M X d= 1, d∈Tm j∈[1,2,··· ,¯Kd], a(d)j,r=1 g(d)k,rρ(d)j,r + (σk,r(m))2 , Km X k=1 a(m)k,r ≤ 1, ∀r, m, a(m)k,r ∈ {0, 1}, ρ(m)k,r ≥ 0, if a(m)k,r = 1, ∀m, k, (5.4.12) where a is the RBA solution to the given optimisation problem, and a = [a(m)k,r , ∀ m, k, r].
The problem above is a constrained non-linear optimisation problem, which is very complex to solve. The authors in [53, 54] stated that the calculation of the SINR values of each user is not possible, since the SINR cannot be calculated without first allocating RBs. The authors sought another approach that eliminated the use of interference estimation for the users on each RB. A centralised strategy that assigned the users into different clusters using interference weights was proposed, and each cluster was assigned a RB to maximise the throughput. Although the strategy aims to avoid interference, the clusterisation strategy does not take into account the full interference from the neighbouring cells on individual users simultaneously. Subsequently, the RBs are allocated to each cluster to maximise the sum-SNR of the users in the cluster. The centralised clustering approach using interference weights and the maximisation of the sum-SNR over the clusters does not effectively mitigate the interference in the network, it also does not seek to maximise the sum-SINR in each macro cell sector and does not solve the challenges of the centralised approach as explained in Section 5.2.1. For this reason, a step-by-step algorithm under the distributed RBA approach that maximises the SINR of the users in each macro cell sector, is proposed to solve the problem in (5.4.12).
Problem Solution
The first and foremost objective of the proposed distributed RBA solution based on maximising the sum-SINR within each macro cell, is obtained using a distributed approach to allocate RBs within a cellular system, that would significantly reduce the overhead resources and time required for jointly allocating RBs to the users. Secondly, the proposed distributed RBA strategy aims to choose the RBs that maximise the sum- SINR of the users within the given macro cell sector, while avoiding the reception of high interference from neighbouring macro cells on the same RB. In order to achieve a distributed RBA based on interference avoidance, and obtain the perceived SINR of each user on every given RB in each macro cell, it is important to have pre-knowledge of the RBs already allocated in the neighbouring cells.
RBA time (t1)
RBA Overhead Transmit assigned RB
info to neighbouring cells, (exchange time, e2)
RBA Overhead
Transmit assigned RB info to neighbouring cells, (exchange time, e2)
Transmit assigned RB info to neighbouring cells, (exchange time, eδ)
. . . . RBA Overhead Start RBA macro cell sector, l = 1. . .
Perform distributed RBA in macro cell site, (w = 1). Perform distributed RBA in macro cell site, (w = 2).
Perform distributed RBA in macro cell site, (w = W).
macro cell sector, l = δ.
. .
Perform distributed RBA in macro cell site, (w = 1). Perform distributed RBA in macro cell site, (w = 2).
Perform distributed RBA in macro cell site, (w = W).
RBA time (t2) RBA time (tδ) macro cell sector, l = 2. . .
Perform distributed RBA in macro cell site, (w = 1). Perform distributed RBA in macro cell site, (w = 2).
Perform distributed RBA in macro cell site, (w = W).
Figure 5.6: Proposed distributed or de-centralised RBA strategy for HomoNets.
Without this knowledge, the interference to each user cannot be properly taken into consideration during the RBA process. It is easy to see why this metric for RBA proves
to be a challenge and is almost impossible to achieve, if the interfered RBs are not yet known. The definition of the variables used in the flow charts and algorithms can be found in Table 5.2.
Obtain the channel and user location information on each macro cell site w. Set l
= 0. Start
End Find the SINR for all users across
all the RBs in macro cell site w, sector l, w = 1: W.
Assign RBs to the macro cell users to maximise the sum-SINR
in each cell macro sector.
For all W MC sites, set l = l + 1.
Is l = δ?
Find the ICI from already assigned macro cell sectors in
set Tm, m = (w-1)*δ + l, to all
users in macro cell site w, sector l, across all RBs.
No
Yes
Figure 5.7: Flow chart of the proposed distributed RBA strategy for W macro cell (MC) sites and δ sectors per cell site.
Algorithm 1
Figs. 5.6 and 5.7 present the proposed distributed RBA strategy using a round robin, sector-by-sector approach, as described below:
with antenna index l = 1, for all macro cell sites w = 1, 2, · · · , W. The perceived SINR for the users on each RB is estimated while the interference received is considered as zero, since the users in other neighbouring cell sectors with macro cell sector index l = 2 and 3 have not been assigned RBs. Then using the well-known Hungarian method [72], the RBs are assigned to all the users to maximise the sum-SINR within the macro cell sector.
Step 2: The assigned RB information are then passed to the neighbouring macro cell sites of neighbouring and interfering BSs in the set Tm, where Tm is the set of neighbouring
and interfering BSs to the m-th macro cell sector.
Step 3: Now the macro cell sectors with antenna index l = 2 finds the SINR at each user on each RB, while considering the interferers on the already assigned RBs in neighbouring macro cell sectors with sector index l = 1, as shown in ‘Block B’ in Fig. 5.8. The Hungarian method is used to assign the RBs to all users to maximise the sum-SINR within each macro cell sector.
Step 4: The assigned RBs information are then passed to the neighbouring macro cell sites of neighbouring and interfering BSs in the set Tm.
Step 5: Now the macro cell sites with macro cell sectors with antenna index l = 3 finds the SINR at each user on each RB, while considering the interferers on the already assigned RBs in neighbouring macro cell sectors with sector index l = 1 and 2, as shown in ‘Block C’ in Fig. 5.8. The Hungarian method is used to assign the RBs to all the users to maximise the sum-SINR within the macro cell sector.
Step 6: The assigned RBs information are then passed to the neighbouring macro cell sites of neighbouring and interfering BSs in the set Tm.
Note that as the users enter and leave the network, the proposed RBA strategy enables RBs to be assigned to new entrants while ensuring that the interference within the network is managed. This continuous process leads to a distributed RBA strategy fit for a SON.