Nt1310 Unit 4

Note that in the above equations, the $R_{sp}(b_j)$, $\forall b_j \in B$, $RB^M(u_i)$, $\forall u_i \in U$, and $RB^S(u_i)$, $\forall u_i \in U$, are unknown variables. The objective function of the above formulation is to maximize the estimated total amount of data, i.e., to maximize the network throughput. The constraint C1 restricts the split data rate $R_{sp}(b_j)$, $\forall b_j \in B^C$, should be less than $b_j$'s input data rate $R_{in}(b_j)$. The C2 demands that the $D^M_p(u_i)$ cannot be larger than the summation of (i) UE $u_i$'s input data volume at MeNB in the upcoming $I_t$, i.e., $R_{agg}^M(u_i) \times I_t$ and (ii) the remaining data located at MeNB $D_r^M(u_i)$. The C3 restricts the $D^S_p(u_i)$ on SeNBs, and the idea is similar

In our simulations, there are one MeNB and one SeNB, and the distance between MeNB and SeNB is 1 km. All UEs are assumed to have dual connectivity capability, and their initial locations are randomly assigned. A UE can connect to the SeNB if the detected signal strength on that SeNB is larger than $-75$ dBm. Each UE is configured to have one bearer, and the traffic type of the bearer is constant bit rate (CBR). For a UE, it may receive data from MeNB, SeNB or both of MeNB and SeNB depending on its location or the decisions of our traffic scheduler. We observe the network throughput (in Mbps) by varing the following four parameters: (i) The numer of UE, say $N_{ue}$. (ii) Per UE's CBR traffic data rate, say $R_{cbr}$, in unit of kbps. (iii) Per UE's random walk speed, say $V_{walk}$, in unit of km/hr. (iv) The transmission power of the SeNB, say $P_{senb}$, in unit of dBm. The proposed traffic scheduler works every 100 TTIs, i.e., we set $I_t=100$ ms. Other system parameters are listed in \Tab{tab:dis_para}. Furthermore, we compare the proposed scheme with the fix traffic scheduling in \cite{132859}. Recall that in \cite{132859}, $x\%$ (resp., $(1-x)\%$) of per UE's traffics will be handled by the MeNB (resp., SeNB). In our simulations, we set $x$ to be 100, 70, 50, 30, and 0 (denoted by F(100), F(70),

We can see that the proposed scheduling scheme (denoted by OUR) can perform the best in all cases. In this simulation, when a UE is located near the boundary of the SeNB, the reported CQI values of the UE will be small. But, when using fixed traffic scheduling, the MeNB still asks the SeNB to relay $1-x\%$ of parckets to that UE, and thus degrades the network throughput. We can see that when $R_{cbr}$ values are lower, the network throughput of F(100) can be better than that of F($x$), where $x=70, 50, 30, 0$. This is because that the MeNB has a larger coverage range, if the MeNB handles most traffics, the network throughput will likely to be better. But, when $R_{cbr}$ becomes larger, F(70) can perform better because that the MeNB is overloaded, and the SeNB can help to increase some extra throughput. Moreover, \Fig{fig:dis_sim_var}(b) shows the result that we vary $P_{senb}$ from 18 to 28 dBm, and fix $N_{ue}=50$, $R_{cbr}=400$, and $V_{walk}=6$. We can see that when $P_{senb}$ becomes larger, the performance gaps between OUR and the fixed scheduling will be smaller. This is because that when using a larger $P_{senb}$, the SeNB has more chances to serve UEs, and thus the network throughput can be improved. Furthermore, we also observe the network throughput by varying $N_{ue}$ and $V_{walk}$. (The results are not shown because of lacking spaces.) The