The general problem faced in the field of Wireless Multimedia Sensor Networks (WMSNs) is congestion. The most common method in the area of WMSNs to minimize congestion is traffic control. Quality Of Service (QOS) is widely used in WMSNs to guarantee preferential service for critical applications by controlling end-to-end delay, reducing data loss and by providing adequate bandwidth. The present work is on Probabilistic QOS Aware Congestion Control (PQACC) which employs probabilistic method based congestion prediction and priority based data transmission rate adjustment, where inelastic real-time traffic and elastic non-real-time traffic are treated separately. Using the present PQACC approach, average throughput, average source-to-sink delay and average packet loss probability are improved by 9%, 10.33% and 16.03% compared to EWPBRC and achieved 5.97%, 7.05% and 11.69% improvement compared to FEWPBRC. Simulation result reveals that, congestion is effectively predicted, controlled and provides necessary level of QOS in terms of delay, throughput and packet loss, hence making this approach possible in mission critical applications.
Recent advances and development in the field of sensors and availability of economical hardware such as microphone and Complementary Metal Oxide Semiconductor (CMOS) camera allow the emerging of WMSNs. Wireless multimedia sensor network is a collection of a huge number of wireless sensors outfitted with devices which are capable of sensing and retrieving images, audio/video streams and scalar data [
WMSNs face problems due to resource-constrained nature of sensor nodes such as limited energy, processing capacity, bandwidth and small memory. In WMSNs applications, the tiny size of sensors, consumption of power due to heavy volume of traffic and complex computations lead to congestion which subsequently diminishes QOS of applications. Hence the main objective of WMSNs is to develop algorithms that prolong the lifetime of the network and also assures to provide QOS necessities insisted by the applications [
The present study is based on the following considerations: a probability based congestion prediction; a mixture of three parameters to attain accurate determination of congestion; service discrimination among different kinds of data; an algorithm for rate adjustment based on probability based congestion prediction, priority of the source data and location of source sensors.
Congestion in WSN may occur because of two reasons [
Research community has carried out a lot of effort in developing solutions for congestion control in WSN and WMSNs. Congestion Detection and Avoidance (CODA) protocol [
CODA [
The literature review shows that the probability based approach that considers a combination of buffer occupancy, incoming and outgoing data rates for congestion prediction is not proposed so far, by any priority aware congestion control protocol. The present work uses a dynamic threshold value for predicting congestion compared to other existing works. It considers buffer occupancy as one of the parameter to predict congestion. It uses piggybacking concept to propagate feedback about congestion that assures guaranteed delivery of packets.
The present work is motivated from the limitations of EWPBRC [
In the present work, a probability based approach that considers three parameters namely buffer occupancy, incoming and outgoing data rates is used for detecting congestion intensity. Moreover during rate adjustment, preferential treatment is given to sensor nodes based on the class of data generated and their geographical locations.
CDU unit is used to predict congestion in each sensor node. In the present work, CDU unit computes incoming and outgoing data rate and buffer occupancy of each sensor node to assess the congestion status using probabilistic method. The value of congestion status may be a negative or positive number. In each predetermined periodic time interval, the outgoing data rate of the children nodes is computed by each parent node in addition
to its own traffic. The priority of sensor nodes differ from one another since these may be fitted with a variety of sensors. Hence each parent node must consider the priority of its children in determining the outgoing data rate of the child nodes. In WMSN, sometimes sensor nodes are purposely deployed in different geographical locations based on their importance, so that sensor nodes may have dissimilar priorities. RAU unit determines the new outgoing data rate by considering the current congestion status, priority of source data and priority based on location of sensor node. The new outgoing data rate is forwarded to CNU unit to notify about the new rate to child nodes. For efficient utilization of network energy, in the present work, CNU unit uses implicit method for communicating notification about congestion. Upon receiving congestion notification each sensor node adjusts its current outgoing data rate according to the value of congestion status.
The PCPA predicts the congestion intensity in each sensor node with probabilistic approach using dynamic threshold index value on buffer capacity and buffer occupancy of that node. The value of threshold index varies from time to time based on incoming and outgoing data rates of sensor node and remaining space in the buffer.
At time “t + 1”for node “i”, buffer occupancy,
where
At time
if
where
Equation (2) is applicable when there is any dataflow in the node or if incoming data rate is higher than outgoing data rate. The threshold index,
The possibility for congestion at node “i” is determined by comparing the buffer occupancy with the threshold index. The results show that
1) if
2) if
Hence the probability of congestion occurrence for node “i” is given by
The dynamic variable
Status of buffer | Incoming data rate | Outgoing data rate | Threshold index |
---|---|---|---|
Empty-no packet | 4 | 2 | |
6 packets | 4 | 2 |
where
The chance for a node to be in the vicinity of node “i” is Ci which is the coverage range of node “i” and it has a value of 1.
The probability for node “i” to have a neighbour set size of “φ” is determined by
The probability expected by node “i” to receive packets from its neighbour is
The intensity of congestion at a node is calculated from Equations (3) and (6) and it is propagated as feedback in backward direction to indicate the inception of congestion. The level of congestion is predicted by the value of
1) if
2) if
The PCPA is denoted as (P1), for which the algorithm is given below:
Probability Based Congestion Prediction Algorithm (P1)Initially
set ‘capbuf’ to maximum capacity of the buffer
set ‘m’ to the set of neighbour nodes
set value of 'C' to power level of the node
Congestion_occurrence_possibility()
// calculation of threshold index
if (
congestion_status =0 // no possibility for congestion
else if
for (i = 1; i
end for
else
congestion_status = 1 //congestion
end if
// propagate feedback about congestion to source
piggyback ACK with congestion_status
forward ACK to upstream nodes
This algorithm P1 is represented in terms of flow chart in
Each node in the network executes algorithm (P1) to determine the possibility of congestion. The
pended to the Acknowledgement (ACK) frame using piggybacking concept and it is propagated as feedback among upstream nodes. Hence impact of congestion is minimized in hop-by-hop manner and according to the
The QOS allows to sort out the entire network traffic into an InElastic Real Time (IERT) and Elastic Non Real Time traffic (ENRT). Further ENRT traffic is divided into High Priority ENRT (HPENRT), Medium Priority ENRT (MPENRT) and Low Priority ENRT (LPENRT) traffic. The relation among these traffic flows in terms of QOS parameters is given by
The sensor node is assigned with two priorities: traffic flow class priority
of traffic flow class “j” in node “i”. This priority value can be manually set to provide different levels of service to different traffic flow classes. Traffic flow class with high priority is assigned with larger value compared to traffic flow class with low priority.
For each node “i”, the value of traffic flow class priority
where “j” represents all categories of traffic flow classes such as IERT, HPENRT, MPENRT, and LPENRT.
Generally in WSN, sensor nodes are set with variety of sensors and are geographically located in different places based on their importance. Hence these sensors need preferential treatment to attain dissimilar throughput. To attain this fairness, each sensor node is assigned with location priority
The local priority
Considered D(i) as the child set of node “i”. Hence the overall priority
From Equation (10) it is known that when a node has no child, then the value of both overall priority
sources only. If the source is inactive, then apart from the type of traffic flow class,
The initialization of outgoing data rate is carried out as follows:
Consider
Average service time,
where
The output rate of sink node,
Substituting the value of Equation (12), the outgoing data rate
where
The steps are repeated to determine the initial outgoing data rate of each node in the network.
Congestion prediction and rate adjustment
For each predetermined time period
Compute the possibility of congestion occurrence
The new outgoing data rate of each child node “i” of sink
For each predetermined time interval
Compute the possibility of congestion occurrence
The above is used by each node to determine its new outgoing data rate.
In this algorithm, for each predetermined time interval, the outgoing data rate for each sensor node is determined by its parent node only and the outgoing data rate for each parent node is in turn determined by its own parent node. Since the sink has no parent node, its allowable outgoing data rate is determined from the inverse of its service time average.
The simulation of the present work has been carried out in NS2 simulator.
The existing work on EWPBRC [
In the present work simulation as per the model mentioned in
The traffic flow classes that each sensor node has in the simulation model are shown in
The performance of the present approach during 30 simulation runs is evaluated for various parameters such as throughput, loss probability, sink-to-base station delay and source-to-sink delay and is tabulated in
The results for the present work are also correlated with already existing two congestion control approaches EWPBRC and FEWPBRC (Figures 5-8) and the comparative result statement is shown in
Network field | 750 m × 750 m |
---|---|
Number of sensor nodes | 12 |
Number of sink nodes | 01 |
Packet size | 250 bytes |
Routing protocol | Static routing |
Buffer size of sensor nodes | 75 packets |
Buffer size of sink node | 125 packets |
Simulation time/Number of runs | 75 seconds/30 runs |
Sensor node no. | IERT (W = 7) | HPENRT (W = 5) | MPENRT (W = 3) | LPENRT (W = 1) | Traffic flow class priority |
---|---|---|---|---|---|
Node 1 | OFF | ON | OFF | ON | 6 |
Node 2 | ON | ON | ON | OFF | 15 |
Node 3 | ON | ON | ON | ON | 16 |
Node 4 | OFF | ON | ON | OFF | 8 |
Node 5 | OFF | OFF | ON | ON | 4 |
Node 6 | ON | ON | OFF | ON | 13 |
Node 7 | ON | OFF | ON | OFF | 10 |
Node 8 | ON | ON | ON | OFF | 15 |
Node 9 | ON | OFF | ON | OFF | 10 |
Node 10 | OFF | OFF | ON | ON | 4 |
Node 11 | ON | OFF | ON | OFF | 10 |
Node 12 | ON | ON | ON | ON | 16 |
Run# | PQACC | |||
---|---|---|---|---|
Source-to-sink delay | Throughput | Loss probability | Sink-to-base station delay | |
1 | 0.318182 | 318.1521 | 10.39962 | 0.001641 |
2 | 0.301435 | 359.8536 | 11.6875 | 0.001639 |
3 | 0.300866 | 362.9049 | 11.45454 | 0.001757 |
4 | 0.279221 | 344.9772 | 11.6553 | 0.001727 |
5 | 0.274892 | 380.0494 | 11.91288 | 0.001787 |
6 | 0.266576 | 347.6483 | 10.88258 | 0.001747 |
7 | 0.31784 | 315.1008 | 10.75758 | 0.001668 |
8 | 0.32684 | 334.3118 | 11.15151 | 0.001757 |
9 | 0.311688 | 366.9734 | 12.39583 | 0.001737 |
10 | 0.316017 | 360.5209 | 11.81629 | 0.001866 |
11 | 0.311802 | 335.4429 | 11.52651 | 0.001866 |
12 | 0.287081 | 303.2852 | 12.81439 | 0.001844 |
13 | 0.320346 | 348.8555 | 12.10606 | 0.001648 |
14 | 0.313853 | 358.8365 | 12.27273 | 0.001787 |
15 | 0.307587 | 369.6882 | 12.46023 | 0.001817 |
16 | 0.322511 | 328.7776 | 11.23674 | 0.001913 |
17 | 0.32684 | 314.0836 | 11.91288 | 0.001896 |
18 | 0.30303 | 353.751 | 12.39583 | 0.001767 |
19 | 0.298701 | 360.8707 | 12.29924 | 0.001956 |
20 | 0.316017 | 306.2586 | 10.78598 | 0.001826 |
21 | 0.299385 | 315.1008 | 12.52462 | 0.001876 |
22 | 0.307359 | 326.289 | 11.81818 | 0.001737 |
23 | 0.320346 | 348.1369 | 12.13826 | 0.001747 |
24 | 0.324675 | 360.5209 | 11.81629 | 0.001836 |
25 | 0.30303 | 327.27 | 11.20454 | 0.001826 |
26 | 0.305195 | 358.8365 | 12.62121 | 0.001836 |
27 | 0.284689 | 341.5456 | 11.06061 | 0.002075 |
28 | 0.323992 | 318.1521 | 11.42992 | 0.001872 |
29 | 0.318182 | 325.0494 | 12.42803 | 0.001856 |
30 | 0.302727 | 357.057 | 12.62121 | 0.001836 |
Average | 0.30703 | 341.61 | 11.78624 | 0.001805 |
Performance metric | EWPBRC | FEWPBRC | PQACC |
---|---|---|---|
Throughput (MB) | 313.403 | 322.3384 | 341.61 |
Sink-to-base-station delay (s) | 0.002119 | 0.00204 | 0.001805 |
Loss probability (%) | 14.03 | 13.34 | 11.78 |
Source-to-sink delay (s) | 0.34243 | 0.33034 | 0.30703 |
The protocol PQACC, for congestion control is developed by considering the requirements of multimedia applications on WMSNs. To achieve QOS for various types of data, each traffic type is set with a priority. The outgoing data rate of each source is adjusted based on the priority of data, predicted level of congestion and deployed location of the sensor nodes. The proposed approach allocates more network resources to high-priority traffic flow compared to low-priority traffic flow, so that real-time traffic gets preferential treatment over non- real-time traffic which is a requirement for WMSN. The threshold value used for comparing buffer occupancy to predict congestion intensity is dynamic, so that efficient prediction is achieved. The results from the simulation data for PQACC efficiently diminish the congestion for a better QOS. The probabilistic method used in the present study, for predicting the level of congestion gives good result compared with EWPBRC and FEWPBRC approaches. Hence an improvement is achieved in throughput, delay and packet loss of the network. This proposed approach can be employed in tracking systems and monitoring applications. In future the work can be optimized using neuro-fuzzy approach.
Muthuselvi Mayandi,Kavitha Velayudhan Pillai, (2016) Probabilistic QOS Aware Congestion Control in Wireless Multimedia Sensor Networks. Circuits and Systems,07,2081-2094. doi: 10.4236/cs.2016.79181