3 Sectional Transmission
3.4 Reception Flow
The Sectional Transmission scheme is backward compatible to the IEEE 802.11 MAC, but there is no extra information field (in the ST-MPDU) needed for the identification of the ST-MPDU. Figure 11 shows the reception flow to accomplish the compatibility as we talk above. If the incoming packet is a MPDU, the receiver goes toward the “Path B” or “Path C” [as shown in Figure (11)]. If the incoming packet is a ST-MPDU, the receiver goes toward the “Path A” or “Path C”. The “Path A” means the incoming packet is a ST-MPDU, thus the Header FCS is checking pass. After the checking Header FCS, the receiver determines how many subframe in this ST-MPDU and which point of the ST-MPDU needed to perform CRCRI. Further, the receiver will combine the entire (ex) error-free subframe together and then check the integrity of this ST-MPDU. If the ST-MPDU after combination is perfect then go toward “Path A1” to send an (traditional) ACK frame back to the sender, else if the ST-MPDU after combination is non-perfect then go toward “Path A2” to send a ST-ACK frame back to the sender. And the “Path B” means the incoming packet is a (traditional) MPDU, thus the Header FCS is checking fail and the receiver will send an (traditional) ACK frame back to the sender after the outer (original) FCS is checking pass. Lastly, the “Path C” means the incoming packet could be a ST-MPDU with error in header or a (traditional) MPDU with error, thus the receiver will send nothing.
Figure 11 – Backward compatible ST-MPDU reception flow.
Chapter 4
Numerical Analysis
The analysis of the packet loss probability is based on those given in [3]. In this numerical analysis, we assume that the channel model is uniform distribution and modeling all channels in BER. And the Fragmentation ignores the limitation of the aMaxTransmitMSDULifeTime, defined in section 9.4 of [1].
Although those assumptions are making here, but we believe that our analysis results will show the same trend with the realistic (simulation) results.
4.1. Loss Probability of MSDU/MMPDU
Let
P
b denotes the bit error rate, and the byte error ratePB =1−(1−Pb)8,where PayloadSize is the length of the frame body (payload) and the constants 24 and 4 denote the length of MAC Header and the length of FCS [as shown in figure 6(a)]. In the IEEE 802.11 MAC, it does have a retry limit for every MPDU, if the retransmission times over the retry limit then the sender discard the MPDU instead of the retransmission. LetRdenotes the retransmission times, and the probability of a MPDU successfully received by the receiver at the (re)transmissions is obtained by
∑
MSDU/MMPDU into several MPDU that the length is not greater than aFragmentThreshold. The length of a fragmented MPDU shall be an equal number of octets (bytes) for all fragmented MPDUs except the last, which may be smaller [1].
= P PayloadSiz e aFragmentT hreshold N
P
lferr blk (5)Equation (4) and (5) show the error probability for the fragmented MPDU and the last fragmented MPDU, where N is the number of fragmented MPDU after fragmenting a MSDU/MMPDU.
∑
received by the receiver within R (re)transmissions, and equation (7) shows the probability of the last fragmented MPDU successfully received by the receiver within R (re)transmissions. Lastly, the loss probabilityP
lossfrg(R) of a MSDU/MMPDU, with the Fragmentation, within R (re)trans-missions is obtained by
)
4.2. Loss Probability of ST-MPDU
In the Sectional Transmission, the receiver will ignore the received ST-MPDU if any reception error is found inside the MAC header or header FCS, because of the ST-MPDU format couldn’t be decoded. Just like the Fragmentation, the length of a subframe shall be an equal number of bytes for all subframes except the last, which may be smaller. Let
P
hdr denotes the error probability of the MAC header of the ST-MPDU that is given by is an approximation equation for the error probability of a subframe,[
blk]
k after splitting a MSDU/MMPDU then the mean length of all subframes is given byThus, the probability of i subframes, out of j subframes, are incorrect is given by
i
where the
n
x means the number of the incorrect subframe after x transmissions. For example:R = 3
∑ ∑
Selective Repeat and Multi-copy ARQ under the Sectional Transmission, withinR (re)transmissions is obtained by
∑
4.3. Numerical Results
Figure 12 shows those packet loss probabilities of 12 cases, and those cases could be classified in following four groups:
1) “DCF”: the MPDU without Fragmentation using equation (3).
2) “Fragment”: the MPDU with Fragmentation using equation (8).
3) “SRARQ”: the ST-MPDU (i.e., Sectional Trans-mission is enabled) with Selective Repeat ARQ using equation (13) and
k = 1
.4) “MCARQ”: the ST-MPDU (i.e., Sectional Transmission is enabled) with both Selective Repeat ARQ and Multi-copy ARQ using equation (13) andk =2
And the number with parts per million (ppm) denotes the channel BER. Note that, all charts in this thesis depict the same form like this. For example: case
“100ppm-SRARQ” means the result comes from equation (13) (i.e., the third group shown above) and the channel BER is 100ppm. The aFragmenThreshold in equation (8) is the sum of (Threshold + MAC header + FCS), i.e., Threshold + 24 + 4, where the Threshold is shown in Table 1, and the number of fragmented MPDU (N) in equation (8) and the number of subframe (N) in equation (13) are referred to the same row (N) shown in Table 1, and the Retry Limit (R) is 7 in all cases.
Table 1
Five different thresholds, remainders and numbers of (ST-)MPDU for five different Payload Sizes
5000 1,500 2,500 3,500 4,500
Figure 12 – Loss Probability vs. Payload Size.
We firstly observe the first group (i.e., without Fragmentation and Sectional Transmission) under 3 different channel BERs as shown in figure 12, such 3 curves are the worst performance under there own channel BER due to no Fragmentation and Sectional Transmission. They show that both the Payload Size and the channel BER will significantly increase the loss probability when either the channel BER or the Payload Size is increased. Hence, we can easily imagine that both packet delay and throughput will be rapidly degraded when both channel BER and Payload Size is large. Now, we observe the rest of group at the same time, all of them are obviously improved the loss probability even those loss probabilities show the same result (close to zero) when the channel BER is 50ppm or 100ppm. In those cases of the channel BER equal to 500ppm, case “500ppm-MCARQ” presents the best performance. And the performance between case “500ppm-SRARQ” & “500ppm-Fragment” are very close, but it is easily conclude that both packet delay and throughput should be better in the case “500ppm-SRARQ” due to the overhead in the Sectional Transmission is smaller than that in the Fragmentation.
Chapter 5
Simulation Results
In the previous Chapter, we make both approximation and assumption in the numerical analysis. Now, we should further confirm the performance in more realistic environment. We modify the network simulator (NCTUns 3.0) to perform the Sectional Transmission and Frame-Level ARQ, and there is only one AP and one STA presented in all scenarios those we talk about in this Chapter. The traffic is saturated from AP to STA, and the protocol of Transport Layer could be UDP (User Data Protocol) or TCP (Transport Control Protocol), and the channel model is the uniform distribution BER that only occurs inside the MAC frame (i.e., the PLCP header and Preamble not included). The conditions of the IEEE 802.11 are:
a) PHY rate = 11Mbps.
b) DCF is used.
c) Slot time = 20us.
d) SIFS = 10us.
e) CWmin = 31.
f) CWmax = 1023.
g) No RTS/CTS.
h) Retry Limit = 7.
i) Preamble = 144bits.
j) PLCP header = 48bits.
And the fragment threshold (in Fragmentation), splitting threshold (in Sectional Transmission) and the number (N) of MPDU/ST-MPDU are shown in table 1.
Lastly, the simulation time is 15 second in each case, and we determine the statistics with the data of the last 10 seconds due to the unstable link may occur in the first 5 seconds. In this Chapter we only consider three groups (“DCF”,
“Fragment” & “SRARQ”), and the depicting form is the same as the previous Chapter. Thanks to network simulator “NCTUns”, all figures in this Chapter show the realistic statistic measured from the real TCP/IP protocol stack of the Linux kernel, those figures with an even figure number are the result under TCP and those figures with an odd figure number are the result under UDP.
5.1. Loss Probability
Firstly, we observe those results are shown in figure 12 and figure 13, they are very close on each other due to the UDP is an unidirectional flow (unlike the TCP is a bidirectional flow). Because the numerical analysis in the previous Chapter only considers the data frame lost (i.e., no model the ACK frame lost), that is why those results shown in figure 13 are worse than that in figure 12 about several percent of loss probability except the group “Fragment”. The case
“500ppm-Fragment” in figure 13 is up to 13% worse than that in figure 12 due to the Fragmentation needs an ACK frame in each fragmented MPDU separately (i.e., much more ACK frames are needed). No doubt the biggest difference, between numerical analysis and simulation, will be occurred in those cases of the Fragmentation. Although there is not the same result shown in figure 12 and 13, they still have the same trend. Now, we observe those cases of the first group shown in figure 13, i.e., “DCF”, such 3 curves show that the loss probability will still significantly increase when either the channel BER or the Payload Size is increased (the same as those numerical results), and they are still the worst under there own channel BER. Both case “500ppm-Fragment” and “500ppm-SRARQ”
(in figure 13) are confirmed that they greatly improve the loss probability and the loss probability is close to zero when the channel BER is 50ppm or 100ppm.
In fact, the case “500ppm-SRARQ” totally outperforms the case
“500ppm-Fragment” within UDP (as shown in figure 13).
Each packet sent by TCP to be acknowledged is needed, due to the TCP performs the positive acknowledgement, and both the flow control & the congestion control are implemented in TCP. Thus, the result of loss probability within TCP is more unpredictable as compared with that within UDP. But in figure 14 those loss probabilities still show the same trend (as figure 13), and both Fragmentation & Sectional Transmission are also confirmed that they greatly improve the loss probability within TCP.
5000 1,500 2,500 3,500 4,500
Figure 13 – Loss Probability vs. Payload Size in UDP.
5000 1,500 2,500 3,500 4,500
0.1
Figure 14 – Loss Probability vs. Payload Size in TCP.
5.2. Packet Delay
Figure 15 shows the packet delay in 10 cases (with 3 groups and 4 channel BERs). The first we observe the case “0ppm-DCF”, the result of such case is the best performance in this chart. Because the channel BER in case “0ppm-DCF” is zero (perfect channel), and it has the smallest overhead (as compared with Fragmentation and Sectional Transmission). In those cases of group “DCF”, the packet delay will rapidly increase when the channel BER increases, and such effect will be worse and worse when the Payload Size is larger and larger.
Hence, the curve of “100ppm-DCF” is lay higher than that of “50ppm-DCF”, and the curve of “50ppm-DCF” is lay higher than that of “0ppm-DCF”. The second, we observe those results of Fragmentation, the curve of
“0ppm-Fragment” is lay at the lowest position and the curve of
“100ppm-Fragment” is lay higher than that of “50ppm-Fragment”. The third, we observe those results of group “SRARQ”, the curve of “0ppm-SRARQ” is lay at the lowest position and the rest of curves is lay higher when their channel BER is higher. Now, we observe three groups at the same time, those results of group
“SRARQ” provide amazing performance. The curve of case “0ppm-SRARQ” is very close to that of case “0ppm-DCF”, it means that the overhead of Sectional Transmission only introduce a little packet delay. Although, the packet delay still increases when the channel BER is worse and worse, the packet delay increases slowly even the case “500ppm-SRARQ” performs an acceptable packet delay. In the other hand, the packet delay of Fragmentation is better than the group “DCF” if and only if both the channel BER is greater than 50ppm and the Payload Size is greater than 1500 bytes. According to the figure 16 we could make the same conclusion as we made in figure 15, it means that those results in TCP have the same trend as those in UDP.
5000 1,500 2,500 3,500 4,500
UDP - Packet Delay (us)
0ppm-DCF
Figure 15 – Packet Delay vs. Payload Size in UDP.
5000 1,500 2,500 3,500 4,500
0.5
TCP - Packet Delay (us)
0ppm-DCF
Figure 16 – Packet Delay vs. Payload Size in TCP.
5.3. Throughput
Figure 17 shows the throughput in 11 cases (with 3 groups and 4 channel BERs). The case “0ppm-DCF” performs the best throughput in all cases, due to both the lowest overhead and the perfect channel (BER=0) that it has. The throughput of group “DCF” will speedy degrades when the channel BER is increase, but it slowly degrades in the group “SRARQ”. And the curve of case
“0ppm-SRARQ” is very close to the curve of case “0ppm-DCF”, it means that the overhead of Sectional Transmission lightly degrades the throughput. Now we observe the group “Fragment”, they perform terrible throughput even the case “0ppm-Fragment” still worse than the case “100ppm-SRARQ” (due to the Fragmentation introduces a large overhead). According to the figure 18 we could make the same conclusion as we made in figure 17, it means that those results in TCP have the same trend as those in UDP.
5000 1,500 2,500 3,500 4,500
Figure 17 – Throughput vs. Payload Size in UDP.
5000 1,500 2,500 3,500 4,500
1
Figure 18 – Throughput vs. Payload Size in TCP.
Chapter 6 Conclusion
In this thesis, we propose the Sectional Trans-mission scheme that is a backward compatible and easy to be implemented scheme, it improves the packet loss, packet delay and throughput at the same time when the IEEE 802.11 is working on noisy channel. The novel Frame-Level ARQ (Selective Repeat ARQ and Multi-copy ARQ) is also introduced to improve the retransmission performance, and the performance is confirmed via both numerical analysis and simulation.
References
[1] IEEE Std 802.11-1999 (R2003), Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications, ANSI/IEEE Std 802.11, 2003.
[2] S. Lin, D. J. Costello, Jr., and M. J. Miller, “Automatic-repeat-request error-control schemes”, IEEE Communications Magazine, vol. 22, no. 12, Dec. 1984.
[3] S. Choi, Y. Choi, and I. Lee, “IEEE 802.11 MAC-Level FEC Scheme with Retransmission Combining”, IEEE Transactions on Wireless Commu-nications, vol. 5, no. 1, pp. 203-211, Jan. 2006.
[4] H.-L. Wang, J. Miao, and J. M. Chang, “An Enhanced IEEE 802.11 Retransmission Scheme”, Wireless Communications and Networking, 2003. WCNC 2003. 2003 IEEE, vol. 1, pp. 66-71, Mar. 2003.
[5] R. D. Stuart, “An Insert System for Use with Feed-back Communication Links”, IEEE Transactions on Communications, vol. 11, pp. 142-143, Mar. 1963.
[6] A. Annamalai, and V. K. Bhargava, “Analysis and Optimization of Adaptive Multicopy Transmission ARQ Protocols for Time-Varying Channels”, IEEE Trans. on Communications, vol. 46, no. 10, pp. 1356-1368, Oct. 1998.
[7] S.Y. Wang, The GUI User Manual for the NCTUns 3.0 Network Simulator and Emulator. Network and System Laboratory, Department of Computer Science, National Chiao Tung University, Mar. 2006.
[8] S.Y. Wang, C.H. Huang, C.C. Lin, C.L. Chou, and K.C. Liao, The Protocol Developer Manual for the NCTUns 3.0 Network Simulator and Emulator. Network and System Laboratory, Department of Computer Science, National Chiao Tung University, Mar.
2006.