Step 6: A sensor node executes the
“comparing procedure” to obtain neighboring information such as weighted values, source ID and the working time of neighbors in its sensing coverage. Jump to the next step.
Step 7: First, calculate the weights (Wv) of a sensor node
W
v= w1N
v+ w2E
v, (3)w
1+ w2=1, Nv≤10 andE
v≤10 where w1, w2are weighting parameters.N
v and Ev represent respectively the degree of neighbors and the residual energy level of node v. A sensor node broadcasts a compare message with the weighted value obtained by eq.(3).To compare the weighted values with neighbors, a node determines whether it is large as a clusterhead node to afford sensing tasks in the working interval. When the working interval is up, clusterhead nodes return to step 1. If equal, then a sensor node should jump to the next step. If small, then it becomes a member node. In meanwhile, it checks whether there are at least k neighboring clusterhead nodes. If true, it jumps to step 3; otherwise, it return to step 1.
Step 8: To compare the ID no., if it has the largest ID no. as a clusterhead node to afford sensing tasks in working interval. When the working interval is up, clusterhead nodes return to step 1. If small, then it becomes a member node. In meanwhile, it checks whether there are at least k neighboring clusterhead nodes. If true, it jumps to step 3;
otherwise, it return to step 1.
Figure 30. Clusterhead election procedure.
As regards to the complexity, let n, d and
k denote total number of nodes, the restricted
degrees of neighbors and the number of neighboring clusterhead nodes as an off-duty eligibility rule, respectively. Our algorithm tries to collect neighboring information has a com- plexity of O(d). The complexity of determining an off-duty eligibility rule has a cost of O(k). So the complexity of our algorithm has a cost of O(ndk) for overall nodes. Due to constant k and d, our complexity approaches closely about O(n).Therefore, the complexity of the cluster-based algorithm is superior to general location-based algorithms as O(nd log d) which need to sort the data of all angles of neighboring nodes.
4 Simulations
4.1 Simulation Metrics and Environment An important metrics is the “data delivery lifetime” which is defined as the time from initial deployment to all sensor nodes die off. In addition, we define metrics for the performance of data success ratio as follows:
E Ratio D S
Data uccess (4)
where D denotes the amount of data from sensor nodes that is successfully received at the control center. E denotes the number of events generated by the source in data delivery lifetime.
This algorithm does not stop until all sensor nodes die off. We deploy (54, 63, 72, 81) nodes uniformly within a fixed square space (18m x 18m). Numbers of nodes in a sensing coverage of a sensor node may have 4 to 7. Each node with the transmission range of 10m has a sensing radius of 3m. To calculate the data success ratio, we give a sensing event per 10 seconds that randomly occurs by means of an exponential distribution at each cell with (3mx3m) area.
Each node has 50 Joules energy. We analyze various values of k, w1, w2 and deployment nodes to observe the effect on data delivery lifetime and data success ratio. Also, we compare the cluster-based approach and the probe-based approach by measuring the
“data delivery lifetime” until data success ratio drops below 75% or all sensor nodes die off.
4.2 Simulation Results
From Figures 31 to 33, we observe that when k is set at 2 the data delivery lifetime is the highest. That is because k=2 has less number of clustering execution than the others. Hence, sensor nodes have no additional energy to consume in clustering operation. We can see the higher the node density in deployment, sensor nodes may have higher probability to be put to sleep. In addition, a sensor node with higher residual energy may have higher possibility of being a candidate of a clusterhead node. Then, the sleep time of redundant nodes becomes long that may save energy consumption. The data delivery lifetime of the cluster-based algorithm is superior to the maximum idling lifetime when all nodes stay idle.
From Figures 34 to 36, based on eq.(4) we observe the data success ratio is between 63% and 80%. When k is set to be 3 and w1
is set at 0.8, the data success ratio is higher.
That is which the sensing area may be covered by at least three clusterhead nodes.
This increases the flexibility of design in different requirements of data success ratio.
However, there is higher possibility of message loss when node density is high, so that the data success ratio may degrade.
Therefore, by carefully choosing a suitable k and w1, one can obtain the desired data success ratio.
Figures 37 to 39 illustrate the data delivery lifetime which cluster-based algorithm compares with probe-based algorithm at the termination of algorithms until data success ratio drops below 75% or all sensor nodes die off. We see that the data delivery lifetime of the cluster-based algorithm that could be approximately double that of the probe-based algorithm is best when k is set to be 2 for different number of nodes. When w1is set to be 0.8 or node density is higher, the data delivery lifetime is best. However, the data delivery lifetime at k=4 could be worst because there is much energy overhead while clustering.
Hence, to avoid energy consumption the designated number of cluster- head nodes may be set at k=2 or 3 and keep higher data success ratio.
Our algorithm provides the flexibility of adjusting the weighting factors and
k
parameter according to the system needs.Data delivery lifetime (k=2)
0
no. of nodes
time(sec.)
w1=0.2 w1=0.5 w1=0.8
Figure 31. Data delivery lifetime (k=2).
Data delivery lifetime (k=3)
0
no. of nodes
time(sec.)
w1=0.2 w1=0.5 w1=0.8
Figure 32. Data delivery lifetime (k=3).
Data delivery lifetime:(k=4)
0
no. of nodes
time(sec.)
w1=0.2 w1=0.5 w1=0.8
Figure 33. Data delivery lifetime (k=4).
Data su cc ess ratio(k=2 )
Figure 34. Data success ratio (k=2).
Da tasu cc ess ra tio (k =3 )
Figure 35. Data success ratio (k=3).
Da tasuc c ess ra tio(k =4 )
Figure 36. Data success ratio (k=4).
Datad eliv e rylifetime (w1 =0 .2 )
Figure 37. Comparison of data delivery lifetime (w1=0.2).
Datad eliv erylifetime(w1 =0 .5)
0
Figure 38. Comparison of data delivery lifetime (w1=0.5).
Datad eliv erylifetime(w1 =0 .8)
0
Figure 39. Comparison of data delivery lifetime (w1=0.8).
VII. The CLLA Scheme
Just before sending a packet, the MAC layer must know whether the channel is
good enough to guarantee a successful
transmission in specified PHY mode (coding and modulation schemes). The transmitting station assumes that packets can be received correctly within the channel coherence time.Then, the CLLA scheme determines the most appropriate access mechanism (basic, RTS/CTS or fragmentation mechanisms) according to the wireless channel condition to achieve the highest throughput. To do so, the CLLA scheme performs a simple table lookup, using the most up-to-date system status as the index.
Given a target packet error rate (PER), the CLLA scheme uses a pre-established PHY mode table indexed by the system status triplet that consists of the PHY mode, data payload length and RSL. The table is established as follows.
The PER is set according to the lowest system efficiency required. For instance, in IEEE 802.11b standard with MPDU payload length of 1000 bytes, the PER should be less than 8%. The PER of a frame can be
where Pb_Predenotes the bit error rate (BER) of the PLCP preamble transmitted at 1 Mbps data rate with DBPSK. Pb_Hdr denotes the BER of the PLCP header transmitted at 2 Mbps data rate with DQPSK modulation scheme. PMPDUdenotes the BER of the MPDU transmitted using a certain modulation scheme. N denotes the payload length.
Hence, given the PER and N, PMPDUcan be obtained. Note that different forms of modulation have different curves of theoretical bit error rates versus Eb/No. And then the SNR can be derived accordingly.
Given SNR, the required RSL can be obtained by
RSL[dBm]= Noise Floor[dBm]+ SNR[dB].
1. Channel Information
A mobile station may receive the beacon, broadcast, ACK, CTS and data frames in the BSS of an IEEE 802.11 WLAN, so the mobile station can record channel information such as the received frame time and the RSL that can be calculated according to those frames coming from the AP.
2. Frame Transmission
A mobile station calculates time duration between the last received frame time and current time. If the duration is less than the pseudo channel coherence time, the channel status is assumed to be not changed. The mobile station can use the payload length and the last received frame’s RSL as an index to select a proper PHY mode by a simple table lookup. Otherwise, the channel status may have changed, and the station needs to get the current channel information.
Here, the RTS/CTS mechanism can be used to get current channel status.
When a station transmits a data frame and does not receive the ACK frame in a specific period of time, timeout occurs. The timeout can be classified as CTS timeout and ACK timeout.
1) CTS Timeout: The RTS and CTS
frames are always transmitted using the most robust modulation scheme. The CTS timeout occurs only during the deep fading, which is the worst channel condition. The transmitting station can do nothing but retransmit the RTS frame.2) ACK Timeout: Error may occur during
the transmission of the DATA frame or the ACK frame. If the DATA frame contains error, it is necessary to update the channel information and adjust the pseudo coherence time immediately. If the ACK frame contains error, the transmitting station can do nothing but retransmit the frame. How an AP knows whether the transmission error occurs in the ACK or the DATA frame is determined by a retry counter for the ACK frame. If the transmission error occurs in the DATA frame, the retry counter will increaseby one, otherwise the counter will remain unchanged.
3. Pseudo Coherence Time Adaptation
The pseudo coherence time is not only used to determine whether to trigger the RTS/CTS mechanism, but it also used to decide the frame size. The value of the pseudo coherence time can be any of 100 ms, 20 ms, 10 ms, 5 ms, 2.5 ms and 1 ms. The default pseudo coherence time is 100 ms.Each mobile station records and uses the number of successful transmission and the number of failed transmission to adjust the pseudo coherence time. In CLLA scheme, the pseudo coherence time will be adjusted if the packet error rate or the number of transmitted packets is larger or less than a threshold.
4. Fragmentation Mechanism
It is always more appropriate to lower the transmission rate rather than to fragment a packet, because every fragment is sent and acknowledged individually, thus the MAC overhead increases linearly with the number of fragments. Throughput improvement due to fragmentation is only observable in a narrow SNR range.
In CLLA scheme, the fragmentation mechanism is usually not used in the first transmission of a frame. It uses the basic mechanism when the wireless channel is in good condition. The transmission time of the frame may be larger than channel coherence time, which will increase the probability of transmission failure. If the transmission error occurred and the error count reaches a threshold, the mobile station then uses the fragmentation mechanism. In order to reduce the probability of transmission error, the mobile station fragments the packet according to the data rate and frame length to fit the pseudo coherence time using a table lookup. This can increase the probability of successful transmission and improve the throughput. Table 8 shows the table for IEEE 802.11b.
TABLE 8
PSEUDOCOHERENCETIME VS. FRAGMENTSIZE INBYTE AT
DIFFERENTDATARATES OFIEEE 802.11b Pseudo
Co_Time
1 Mbps 2 Mbps 5.5 Mbps 11 Mbps
100 ms 2312 2312 2312 2312
20 ms 2312 2312 2312 2312
10 ms 1000 1500 2312 2312
5.0 ms 500 1000 2312 2312
2.5 ms 100 100 1000 2312
1.5 ms 100 100 100 500
1.0 ms 100 100 100 100
5. Simulation Study
We evaluate the performance of the CLLA, RBAR, ARF and other fixed-rate (FR) schemes using the GloMoSim simulator. In the simulation, the following scenario is used to evaluate the throughput when the IEEE 802.11b and 802.11g environments are deployed. We assume a 200×200 m2 region. An AP is fixed in the center. In a WLAN with a Rayleigh faded channel, the AP transmits Beacon frames periodically. The number of mobile stations varies from 5 to 30. Every mobile station follows the
random waypoint
mobility model. The speed of a station varies from 1 m/sec to 20 m/sec. Mobile stations transmit frames of 1500 bytes and 100,000 frames to the AP. An ON/OFF traffic model is used in each mobile station. In this traffic model, the OFF time duration is drawn from Pareto distribution with parameters=0.7, and
ON time duration is drawn from Weibull distribution with parameters k=0.9 and b=90.During the ON period, the inter-arrival time is governed by another Weibull distribution with parameters k=0.5 and b=4.5.
Simulation results are shown in Figures 40 and 41 for the IEEE 802.11b and IEEE 802.11g, respectively. Note that the CLLA has the best performance compared with the others whatever in IEEE 802.11b or IEEE 802.11g environments. The RBAR outperforms the ARF, because RBAR can get channel information more quickly than ARF scheme. It is interesting that the throughputs of FR-11 and FR-05 in Figure 1(a), as well as FR-54, FR-24 and FR-12 in Figure 2(a), outperform the ARF scheme. As
depicted in Figures 1(b) and 2(b), the packet loss rates of the fixed rate schemes are larger than those of link adaptation schemes, because when the mobile station moves far away from the AP, the fixed-rate schemes can not perform link adaptation in environment with high channel error rate.
(a)
(b)
Figure 40. Performance comparisons of the IEEE 802.11b.
(a)
(b)
Figure 41. Performance comparisons of the IEEE 802.11g.