ܶ
୭୮୲௧݂
௧ݔ
ܶ
୭୮୲௧ିଵ݂
௧ିଵݔ
>ŵŽĚĞ
^ŵŽĚĞ tŝƚŚŽƵƚ ŚLJƐƚĞƌĞƐŝƐ ŵĂƌŐŝŶ
ܶ
୭୮୲௧݂
௧ݔ
∆
>ŵŽĚĞ DŽĚĞĐŚĂŶŐĞ͗
ĂĚĚŝƚŝŽŶĂůƉŽǁĞƌ
ĐŽŶƐƵŵƉƚŝŽŶ EŽŵŽĚĞ
ĐŚĂŶŐĞ ŚLJƐƚĞƌĞƐŝƐ ŵĂƌŐŝŶtŝƚŚ dƌĂĨĨŝĐůŽĂĚŽĨ
ĂƐƵďĂƌĞĂ
Figure 4.7: An example of a case where the traffic load of a subarea is close toTopt. Without hysteresis margin (ΔT), mode change may occur because of changes to Topt and power consumption increases through the mode change, while with ΔT, mode change may not occur and power consumption increase is avoided.
change is needed according to traffic loads. However, frequent mode change may increase the power consumption of the network. Fig. 4.7 shows a scenario in where the traffic load of a subarea is close toTopt. For example, a subareaah was in LC mode at time t−1, and its traffic load Tat−1
h was close to Toptt−1. At next time t, let Toptt < Tat
h. Then the subarea ah will be in SC mode, which means that the mode change occurs in ah and power consumption of network increases beacuse of the additional power consumption caused by this mode change. If this scenario happens frequently, the effectiveness of our proposed scheme will decrease. To avoid this additional power consumption, we introduce hysteresis margin (ΔT). If the traffic load of the subarea falls within the hysteresis margin, the subarea does not change its mode, i.e., there are four cases as below:
• LC mode at t−1. If Tat
h ≤Toptt + ΔT /2, then stay in LC mode att.
• LC mode at t−1. If Tat
h > Toptt + ΔT /2, then change to SC mode at t.
• SC mode at t−1. If Tat
h ≤Toptt −ΔT /2, then change to LC mode at t.
• SC mode at t−1. If Tat
h > Toptt −ΔT /2, then stay in SC mode at t.
Note that ifToptt + ΔT /2> TBBU, then we use TBBU instead of Toptt + ΔT /2.
/ĨĂƐƵďĂƌĞĂǁĂƐ>ŵŽĚĞĂƚ 1 /ĨĂƐƵďĂƌĞĂǁĂƐ^ŵŽĚĞĂƚ 1 ୭୮୲௧
>ŵŽĚĞ ^ŵŽĚĞ
Δ 2⁄
௧
>ŵŽĚĞ ୭୮୲௧ ^ŵŽĚĞ
Δ 2⁄
௧
(a) (b)
Figure 4.8: Hysteresis margin and the probabilities of the modes at t.
Additionally, Fig. 4.8 shows the four cases as mentioned. If a subarea was in LC mode at t −1, the area corresponding to the LC mode (probability of LC mode) is larger than the area corresponding to SC mode, which means the probability of mode change is decreased. On the other hand, if a subarea was in SC mode at t−1, the area correspondiong to the SC mode (probability of SC mode) is larger than the area corresponding to LC mode, which has the same meaning as the case of LC mode. Based on the stochastic method, the expected number of mode changes from 0 tot is formulated in the Appendix of this thesis.
As mentioned above, hysteresis margin can reduce the additional power con-sumption caused by mode changes. However, if the hysteresis margin is too large, then the mode change cannot occur appropriately, resultin in increasing the power consumption through ineffective RRH and BBU power saving. Thus, we investigate this potential tradeoff through simulation evaluation.
50 55 60 65 70 75 80 85 90
0 60 120 180 240 300 360 420 480
E( X )
t
Figure 4.9: The graph of E(X) used in this simulation evaluation.
Ttend
from time 0 totend. Then,
Ttmatend =
⎛
⎜⎜
⎜⎜
⎜⎜
⎝ TA0 TA1 ... TAtend
⎞
⎟⎟
⎟⎟
⎟⎟
⎠
=
⎛
⎜⎜
⎜⎜
⎜⎜
⎝ Ta0
1 Ta0
2 · · · Ta0
Na
Ta1
1 Ta1
2 · · · Ta1 .. Na
. ... . .. ... Tatend
1 Tatend
2 · · · Tatend
Na
⎞
⎟⎟
⎟⎟
⎟⎟
⎠
, (4.3)
whereTAt is the set of traffic loads in the network service area att,Tat
h is the traffic load of subareaah att, and letTAt followft(x). In the simulation evaluation, we set the parameter E(X) forft(x) as follows:
E(X) = Xmax−Xmin
2 cos
2π tend
t−tend 2
+Xmax+Xmin
2 , (4.4)
where we set Xmax = 80 %, Xmin = 60 %, and tend = 480. The parameter σ for ft(x) is set as 0.05, 0.1, 0.15, and 0.2. Note that theσ is not changed by varying time. Moreover, we consider timeslots and their length, i.e., the interval between t and t + 1 is set to 3 minutes. Fig. 4.9 shows the graph of E(X) which is used in this simulation. We generated 100 random samples of Ttmatend that follow f0(x), f1(x),· · · , ftend(x). The other parameters such as power consumption are the same as the time-homogeneous analysis.
20 40 60 80 100 120 140 160 180 200
0 2 4 6 8 10
Average Numbers
ΔT [%]
LC mode RRHs SC mode OD RRHs SC mode HD RRHs Mode switchBBUs
Figure 4.10: The average number of active RRHs and mode switch for σ= 0.1.
Fig. 4.10 shows the average number of active RRHs and mode switch in a subarea for σ = 0.1. As ΔT increases, the average number of mode changes obviously decreases. However, we can see that the average number of active LC mode RRHs is increases. Thus the average number of active BBUs is also increased.
16.4 16.6 16.8 17 17.2 17.4
ΔT = 6.0%, 17.14kW
Average Power Consumption [kW]
Total power consumption RRHs and BBUs
0.3 0.4 0.5 0.6 0.7 0.8 0.9
0 2 4 6 8 10
ΔT [%]
Mode switch
Figure 4.11: The average power consumption of active RRHs and BBUs, mode switch, and the average total power consumption forσ = 0.1.
switch, and the average total power consumption for σ = 0.1. Here, we set the energy consumption of one time mode switch of a subarea to 3592.8 J, and convert them to the average power consumption of mode switch of all subareas.
As ΔT increases, the power consumption caused by mode switch decreases, but the power consumption of RRHs and BBUs increases. The increase in the power consumption of RRHs and BBUs is more affected by BBUs because the number of active SC mode RRHs decereases but the number of active BBUs increases as shown in Fig. 4.10. From this result, we can see a tradeoff between the power consumption of RRHs and BBUs, and the power consumption caused by mode switches. Therefore, there will be an optimal value of ΔT, and in this case, at ΔT = 6.0 %, the total power consumption is minimized.
16.8 17.2 17.6 18 18.4 18.8
0.05 0.1 0.15 0.2
Average Power Consumption [kW]
σ
TTh = 0%
TTh = 100%
TTh = Topt without ΔT TTh = Topt with ΔT
Figure 4.12: The average total power consumption of the network, simulated for different σ. Range bars present the maximum, average and minimum values of 100 samples.
Fig. 4.12 shows the simulated average total power consumption of the network with TTh = 0 %, 100 %, Topt without ΔT (ΔT = 0), and Topt with optimal ΔT. Regardless ofσ, our proposed scheme saves energy in case of time-varying traffics.
In addition, our proposed scheme with an optimal ΔT can save more power consumption in time-varying traffics.