Simulation Results and Discussion
5.1 Simulation Environment
We used Visual C++ to simulate and evaluate the performance of LCFT, E-LCFT and the sleep mode operation in the IEEE 802.16e in terms of the percentage of sleep periods and average packet delay. The percentage of sleep periods, which reflects the power consumption of an MSS, is defined as (number of sleep frames) / (number of sleep frames + number of listening frames + number of awake frames). The average packet delay is the average elapsed time from the time that a packet enters the BS to the time that the packet completes its transmission to the MSS. The simulation environment is similar to that in [10]. The duration of an OFDM frame is assumed 5 ms, and the maximal data rate that a BS can offer an MSS is assumed 1600 kbps. That is, the frame length is 1000 bytes. Eight different traffic connections were defined and the parameters of them are described in Table 3 and Table 4. Some parameters were referred from [4] and [10] and we modified part of them to demonstrate the energy efficiency of our proposed schemes in every respect.
Connections A, B, C and D are power saving classes of type I, and connections E, F, G and H are power saving classes of type II. The main difference between connections in each
18 18
performance under different traffic loads. The values of packet size and interval of packet arrival for each packet in type I connections were randomly generated from the ranges specified in Table 3.
Table 3. Parameters of type I connections
Connection A B C D
Type I I I I
Packet size (Bytes) 1~1000 1~1000 1000~2000 1000~2000
Sleep Period (ms) [5, 320] [5, 160] [5, 320] [5, 160]
Interval of packet arrival (ms)
1~350 1~180 1~350 1~180
Table 4. Parameters of type II connections [10]
Connection E F G G
5.2 Simulation Results and Discussion
In Fig. 8, it shows the percentages of sleep periods using the three different schemes (802.16e, LCFT and E-LCFT). The higher percentage of the sleep period is, the longer common free time that an MSS can enter the sleep mode and save more power. The notation A+E means that there are only two connections, A and E. It is a simple traffic environment.
By increasing the number of connections, the traffic environment becomes more complex. We found that if the traffic environment becomes more complex, the percentage of sleep periods will become smaller in all the three schemes. In all cases, both the proposed two schemes performed better than the IEEE 802.16e. In Fig. 8, it shows that the percentages of sleep periods of LCFT and E-LCFT are 14% to 50% and 33% to 68% more than IEEE 802.16e, respectively.
The overhead of the proposed schemes (LCFT, E-LCFT) is that they have longer average delay than the scheme of IEEE 802.16e. The reason is that we delay the listening windows and reduce the number of listening windows. The buffered data in the BS are sent only after the listening windows of type II connections. However, we bounded the delay. The listening windows were postponed by the evaluation with delay constraint of type II connections. For connections of power saving classes of type II, we also guarantee their QoS. For the connections of power saving classes of type I, we wouldn’t make their average delay longer
20 20
Fig. 9 shows the average packet delay in different traffic environments. For type I (T1) connections, the LCFT has 6% to 31% longer average packet delay than the IEEE 802.16e scheme and the E-LCFT has 71% to 77% longer average packet delay than the IEEE 802.16e scheme. For type II (T2) connections, the LCFT has the same average packet delay as the IEEE 802.16e scheme since the LCFT did not modify the sleep mode operation of type II connections. The E-LCFT has 84% to 88% longer average packet delay than the IEEE 802.16e scheme. The IEEE 802.16e scheme achieves the lowest packet delay, because its MSS wakes up more frequently to transmit packets. Nevertheless, the simulation results indicate that all schemes, no matter type I or type II connections all satisfied the QoS requirement in terms of delay constraints specified in Table 4.
Since we can bound the average packet delay under the delay constraint of type II connections in our schemes, here we present another simulation results of the percentage of sleep periods and average packet delay with a tight delay constraint in Fig. 10 and Fig. 11, respectively. The parameters of connections E’, F’, G’, and H’ are the same with connections E, F, G and H respectively, except the delay constraint. We changed from the loose delay constraint of 100 ms to a tight delay constraint of 30 ms. The simulation results still show that the LCFT is 14% to 50% better than IEEE 802.16e and the E-LCFT is 26% to 57% better than IEEE 802.16e in terms of the percentage of sleep periods. The average packet delay of the LCFT is still the same as the IEEE 802.16e scheme, but the average packet delay in the
E-LCFT decreases obviously. This is because the value of delay constraint affects the length of sleep interval in type II connections. For type I connections, the E-LCFT has 38% to 44%
longer average packet delay than the IEEE 802.16e scheme. For type II connections, the E-LCFT has 33% to 67% longer average packet delay than the IEEE 802.16e scheme. If the delay constraint is set smaller, the average packet delay in the E-LCFT will become smaller.
This is because we use the delay constraint to calculate the number of packets that can be grouped into a single frame for transmitting and thus to guarantee the QoS.
The listening windows of type II connections can transmit the MAC SDUs (service data units). If the size of a type II packet is smaller, the unused space in the listening window will be larger. In this situation, we can have higher probability to transmit type I packets within the type II listening window. Then the MSS will have more frames to enter the sleep mode. In Fig.
12, it shows the effect of packet size on the percentage of sleep periods. The A+E and C+E have more sleep periods than the A+G and C+G, respectively, because the packet size in connection E is smaller than that in connection G. The A+E also has a higher percentage of sleep periods than the C+E, because the packet size of A is smaller than that of C.
Note that the proposed two schemes were designed to piggyback the type I connection’s traffic indication message at the type II connection’s traffic indication message. If the number of type II connections is much less than type I connections, the percentage of sleep windows of our schemes will become smaller and the average delay may become longer. In the
22 22
following, we will evaluate this situation. In Fig. 13, there is only one type II connection E in the following cases: A+E, A+B+E and A+B+C+D+E. The average packet delay increases when the number of type I connections increases. This is because the total size of buffered packets is larger than the unused space that a listening window of type II can provide. The BS then has to transmit the unsent packets with another frame(s) and the packet delay becomes longer. In the case of A+B+C+D+E+F, there are two type II connections of equal packet size.
Comparing between A+B+C+D+E and A+B+C+D+E+F, the average packet delay of the latter is smaller than the former, since the latter has more type II connections of equal packet size.
Therefore, the proposed two schemes are suited to environments that allow type II connections to utilize its unused space in a frame to carry type I’s packets.
0.0%
Fig. 8. Percentage of sleep periods under the loose delay constraint (100ms)
0
802.16e T1 802.16e T2 LCFT T1 LCFT T2 E-LCFT T1 E-LCFT T2
24
Fig. 10. Percentage of sleep periods under the tight delay constraint (30ms)
0
802.16e T1 802.16e T2 LCFT T1 LCFT T2 E-LCFT T1 E-LCFT T2
Fig. 11. Average packet delay under the tight delay constraint (30ms)
0.0%
Fig. 12. The effect of packet sizes on percentage of sleep periods
0
802.16e T1 802.16e T2 LCFT T1 LCFT T2 E-LCFT T1 E-LCFT T2
Fig. 13. The effect of number of type II connections on average packet delay
26 26