Performance Evaluation for Load-Aware COMIC

In this section, we evaluate our proposed algorithm by simulations. We consider various BS placements to show that our algorithm can approach the lower bound of activated BSs in different deployment strategies. We examine non-uniform traffic load distributions to show that our proposed algorithm can adapt to network traffic

Table 2.1: Simulation parameters of COMIC for downlink.

Parameter Value

channel model path loss+shadowing

shadowing standard deviation σΨ 6 dB minimum received power Pmin -100 dBm

outage probability pout 0.1

path loss exponent α 4

network size 30 Km by 30 Km

K (reference propagation distance = 1m) 10⁻¹

N0 (noise spectral density) -174 dBm/Hz

system bandwidth 10 MHz

subcarrier spacing 15 KHz

interference margin 3 dB

load. We also compare our proposed algorithm with another load-aware cell activation algorithm in various traffic load conditions to show the improvement on energy-saving capability. Finally, we will show the effectiveness on power saving by the deployed small cells.

Network Energy Consumption under Non-Uniform Traffic Load Distribu-tions

We examine the performance of COMIC for saving network operational energy. We evaluate the performance of COMIC for non-uniform traffic load distributions to show that COMIC can adaptively activate BS according to network traffic load conditions.

We also compare COMIC with another load-aware cell activation algorithm and ex-amine how much network power consumption is due to link-level power consumption.

In the simulation, we consider an OFDMA downlink system based on 3G-LTE [44].

The total bandwidth of a BS is 10 MHz and the subcarrier spacing is 15 KHz. Assume that subcarriers are dynamically allocated among users and power is equally allocated among subcarriers [45]. Users are associated to active BSs with highest SINR [14].

Given the average SINR of an MS-BS link Γ, the average spectral efficiency with path

-5 0 5 10 15 20 25 30 35

Figure 2-5: The activated BSs (green) by COMIC with a hot spot area at the center in the network. Only 56 out of 368 BSs (gray) are activated. In the network, 5000 users (blue) are uniformly distributed in the hot spot region and 8000 users (blue) are uniformly distributed in the other region.

loss and log-normal shadowing effects can be approximated by [46]:

C¯ ≈ ln

where ξ is a constant _ln(10)¹⁰ . Note that the shadowing standard deviation σΨ is ex-pressed in dB. To mitigate the potential interference from neighboring cells, we set an interference margin for the MS-BS link budget in the simulations. The network simulation parameters are given in Table 2.1.

We first evaluate the performance of COMIC during the low traffic load period.

Due to the non-uniform user population, hot spot area may be present even in the low traffic load period. Here we consider a network of 368 BSs grid-deployed in the 30 Km by 30 Km region with a hot spot region of 10 Km by 10 Km at the center.

The population density is 50 active users per square kilometer in the hot spot area and 10 active users per square kilometer in the other region [47] and each generates an average data rate 10 Kbps during the low traffic load period. A typical BS power profile Pc = 1000 W and η = 0.5 [29] is given.

The BSs activated by COMIC are shown in Figure 2-5. From the figure, we can

user data rate(bps) networkpowerconsumption(dBm) COMIC(total power)

COMIC(static power) Zhou(total power) Zhou(static power)

10⁴ 10⁵

80 81 82 83 84 85 86

Figure 2-6: The minimal network power consumption as a function of average user data rate. The static part of network power consumption is also shown. The performance is compared with the algorithm by Zhou et al.

observe that the cell size is determined by the traffic load demands around the hot spot area and the cell size is determined by the coverage requirement in other low traffic regions. From the figure, we can observe that COMIC can adaptively activate BSs according to the traffic load conditions and greatly reduce the number of activated BSs in the low traffic regions.

We then compare our algorithm with another load-aware cell activation algorithm [14] when traffic load varies. In the simulations, the BS deployment and active user distribution in Figure 2-5 are considered. The maximum transmit power 46 dBm is given and the performance is averaged over 20 random active user distributions.

The total network power consumption as a function of user traffic rates is shown in Figure 2-6. Also, the static part of network power consumption which is independent of traffic loading is also shown. The performance of COMIC is compared with that of the power saving algorithm in [14] proposed by Zhou et al. In the figure, we can observe that the static power consumption will dominate the network power consumption which verifies our simplifications on BS power consumption. Besides, compared with the power saving algorithm, COMIC can gain more than 1.5 dB

9.8 dB

Figure 2-7: The minimal network power consumption for maintaining network coverage as a function of the number of deployed small cells in the network.

performance for low user traffic rates. That is, approximately 44.5% more power will be consumed by the power saving algorithm than COMIC. Even for higher user traffic rates, COMIC can still maintain lower network power consumption than the power saving algorithm. The reason is that the proposed COMIC algorithm is not only load-aware but also topology-aware.

Network Energy Saving by Network densification with Small Cells

Finally, we examine the impacts of the number of small cells in the network. A ques-tion of interest is whether the network power consumpques-tion can be further reduced with the deployment of small cells during the low traffic load period. In the simu-lations, 100 macro cells and a certain number of small cells are randomly deployed in the network. To facilitate our simulations, we assume the traffic is low enough for us to omit its effects during the low traffic load period. The network perfor-mance is averaged over 20 randomly generated networks. The hardware parameters are: Pc,macro = 1000 W, ηmacro = 0.5, Pc,small = 10 W, and ηsmall = 0.1. The other simulation parameters are given in Table 2.1.

We examine the network power consumption as a function of the number of small cells deployed in the network. Figure 2-7 shows the minimal network power

con-number of deployed small cells

averagenumberofactivatedBSs

Region I Region II Region III macro cells

small cells

10¹ 10² 10³ 10⁴ 10⁵

10⁰ 10¹ 10² 10³

Figure 2-8: The average number of activated BSs for minimizing network power consump-tion while maintaining network coverage as a funcconsump-tion of the number of deployed small cells in the network.

sumption for network coverage preservation as a function of the number of small cells deployed in the network. In the figure, we can observe that the network power con-sumption can be further reduced if we jointly take the macro cells and small cells into COMIC activation processes even though only a few small cells are deployed in the network. In general, more power can be conserved when more small cells are deployed in the network. From the figure, compared with the performance of only activating macro cells, almost 10 dB performance gain can be achieved if we jointly activate macro cells and small cells. Besides, compared with the lower bound of minimal net-work power consumption, our scheme only incurs approximately 2.6 dB performance gap.

The number of macro cells and small cells activated in the entire network is shown in Figure 2-8. In the figure, the number of activated BSs can be divided into three regions. In Region I, macro cells and small cells are jointly activated for minimizing the network power consumption of the entire network. In Region II, deploying more small cells can effectively improve the performance of network power consumption.

Activating macro cells is no more energy-efficient. In Region III, the number of activated small cells is almost the same such that deploying more small cells cannot

significantly reduce more power consumption. From the observation in the figure, the simple transition behavior can be easily applied by a network operator to evaluate the effectiveness of deploying more small cells on network power consumption reduction.

Furthermore, to achieve an energy-efficient green cellular network in the low load period, it can be observed that maintaining network coverage only with macro BSs may not always be the optimal strategy especially when a large number of small cells are deployed in the network.

2.4 Joint Uplink and Downlink Green Network