A Fast Network Configuration Algorithm for TDMA Wireless Sensor Networks

Abstract

The deployment of large-scale wireless sensor networks (WSNs) presents various challenges whose solution requires the design and development of power-and-time efficient protocols. In this context many proposals and various standards have suggested the use of time division multiple access (TDMA) in order to guarantee tight-time scheduling and high overall network throughput under high load conditions. However, in TDMA networks the time and overhead required during the setup phase are major drawbacks that are often overlooked. In this paper we introduce a simple and robust algorithm specially tailored to be used during the setup phase of a TDMA-based WSN. The proposed algorithm makes use of 2C, a conflict resolution protocol with some advantageous properties. As a case study, we consider the setup phase of the synchronous protocol SA-MAC. Our results show that the proposed algorithm is able to configure highly populated networks in significantly shorter times than traditional CSMA/CA. Furthermore, an experimental prototype has been developed allowing us to show the feasibility of deploying the proposal using off-the-shelf components.

1. Introduction

Wireless sensor networks (WSNs) provide a new way of working for traditional applications such as environmental monitoring, security, and robotics [1]. The current interest in these networks is due to the potential number of applications supported by a large number of small wireless sensor nodes with some computing capabilities at reduced cost. However, the battery life of sensor nodes strongly relies on the development of efficient communication protocols. These protocols must be based on strategies to minimize power consumption. In fact, power saving has been the main driving force behind the development of several protocols that have recently been introduced in the literature (see [2] for a recent survey). In this context, the largest energy savings are achieved by protocols whose communications are based on time division multiple access (TDMA). In order to achieve collision-free communications and minimum end-to-end latency, TDMA communications require a network configuration phase where all node transmissions must be scheduled. In this phase all nodes will have to establish a father-and-child relation in order to create the network and there will be contention and its related effects such as collisions and delays. This might be a negligible issue in networks with a few nodes; but with large and dense WSNs this problem becomes more relevant. Therefore, network configuration algorithms must be fast, scalable, and flexible enough to handle networks of various sizes with no human intervention.

It is interesting to note that although network configuration is a common phase of diverse TDMA-based MAC protocols, so far the development of fast and efficient setup algorithms has not been given enough attention.

The work reported in this paper focuses on a proposal for the efficient setup of TDMA WSNs. At the core of the protocol there is a conflict resolution algorithm since, at the network start time, there is no scheduling and channel access conflicts most likely will arise. To remedy this problem we can make use of the usual choice for solving the problem of channel access, that is, the CSMA/CA algorithm. However, as we will discuss in more detail later, we believe that this algorithm is not the best solution to this problem. The conflict resolution approach used in this work is derived from the definition of the two-cell (2C) algorithm introduced by Paterakis and Papantoni-Kazakos in [3]. The resulting protocol can be used as the core of the setup phase in a number of TDMA based protocols. As a case study, we make use of the SA-MAC protocol, a TDMA synchronous communication protocol previously introduced in one of our recent works [4]. Throughout extensive simulation work we evaluate the performance and operation of our proposal and show that the 2C-based approach is able to speed up the network configuration time as compared with solutions based on traditional CSMA/CA. We also show and verify the operation of the proposed algorithm using an experimental setup.

The remainder of this paper is structured as follows. Section 2 reviews related work in the context of TDMA-based protocols for WSNs. Section 3 overviews the principles on which the present proposal is based. Section 4 describes our proposal using the SA-MAC protocol as a case study. In Section 5 we describe the simulation results that we obtained from the performance evaluation. Section 6 describes our experiences from implementing the protocol in a testbed system. Finally, in Section 7 we provide our conclusions.

2. Network Setup in TDMA-Based WSNs

TDMA MAC protocols are an appealing approach for densely populated WSNs. In the context of networks composed of a large number of power-constrained nodes, TDMA protocols avoid some important sources of power wastage, such as idle listening, collisions, and overhearing. In addition, when an efficient synchronization mechanism is available, TDMA protocols are able to provide guarantees for efficient and robust communications [5]. Nevertheless, the creation of the logical network structure along with the specific transmission schedule are two issues that remain as the major challenges during the setup phase of TDMA protocols.

Nowadays, various approaches are being pursued to enable the setup phase of TDMA networks. In some proposals it is assumed that network creation is solved by using other protocols. For instance, the R-MAC protocol [6] assumes that a separate protocol, operating during the setup period, synchronizes the clocks in the sensor nodes with the required precision. Once the network nodes are synchronized, R-MAC sends a small control frame along the data forwarding path to allow all nodes along the path to learn when to awake in order to receive the data packet from the immediate upstream node and forward it to the immediate downstream node.

Other protocols assume that network creation is solved by means of external hardware. In this category we find RT-Link [7] which considers that global synchronization is based on an add-on hardware consisting of a radio module for synchronization in indoor environments and an atomic clock receiver for outdoor operation. After detection of the periodic synchronization signal, the microcontroller updates its local time. This marks the beginning of a slotted data communication period. This period is defined as a fixed-length cycle and it is composed of multiple frames. Each frame is divided into multiple slots: SS (scheduled slots, transmit and receive slots) and CS (contention slots, transmit slots of random access as in slotted aloha). In the case of a scheduling error, communication is still possible using contention slots, but nodes in CS do not have guarantees of successful transmission. This situation produces loss of information and repetition of the scheduling phase.

In spite of these efforts, dense networks still pose significant challenges to network configuration mechanisms. This is due to the fact that at one time there might be several nodes trying to join the network. Furthermore, several nodes may simultaneously reply to join requests issued by a newly arriving node. As previously mentioned, arising conflicts during the setup phase can be resolved by means of the widely known CSMA/CA protocol. In fact, this mechanism has been included in the specifications of IEEE 802.15.4. However, WSNs require protocols that are fast, easy to implement, and flexible enough to be used without modifications across different scenarios. CSMA/CA, on the other hand, does not meet these requirements mainly due to the fact that its performance has a strong dependence on its configuration parameters. For instance, it can be tuned to save energy by limiting its backoff period, but this policy will also lead to a large number of collisions in dense networks. If the backoff window is allowed to grow, this policy will lead to long idle times and energy waste. Besides, channel access is not guaranteed.

At this point it is worth mentioning that the problems previously mentioned have motivated a large number of clever proposals intended to improve the performance of CSMA/CA. For instance, Sift is a medium access control which was introduced in [8]. It makes use of a fixed-size contention window and nonuniform probabilities for selecting transmission slots. By reducing the probability of choosing the first slots, stations selecting these slots most likely will access the channel without colliding. This is useful for event-driven WSNs where several nodes may sense the same event and it is enough to let just a few notification messages to pass through the network. The performance evaluation reported in [8] shows that Sift has several attractive features when compared to standard CSMA/CA. Other proposals, such as the one described in [9], attempt to reduce the overall number of collisions by adapting the success probability according to the collisions observed in the medium. As a final example we can mention the CARMA protocol introduced by Garces and Garcia-Luna-Aceves [10]. This algorithm makes use of a splitting tree algorithm to resolve collisions and it results in a significant reduction on the number of collisions. In spite of these and other efforts, traditional CSMA/CA is the protocol that is used in real systems such as devices that comply with the IEEE802.15.4 standard. Due to this reason in this work and, for comparison purposes, it is the only one that we will consider.

In the following section we will describe the core of our proposal for the network configuration phase.

3. The Core of the Network Configuration Protocol

In this paper our objective is to introduce a simple, efficient, and robust network configuration algorithm specifically designed to be used during the setup phase of TDMA wireless sensor networks. Such algorithm should provide the means to quickly solve conflicts arising among nodes attempting to simultaneously reach a given node. To this end we propose to develop the collision resolution mechanism based on the 2C protocol introduced in [3].

The 2C algorithm performs collision resolution by means of random access. This algorithm considers that time is slotted and stations are allowed to transmit only at the beginning of a time slot. A time slot basically equals the time it takes to transmit a packet and receive a feedback message from a central station. The feedback message is binary, that is, it is a collision message C when a collision was detected and a no collision message NC otherwise. If only one station transmitted, the corresponding packet will be successfully transmitted. On the other hand, if there were several transmission attempts in the same slot, there will be a collision and its resolution will begin in the following slot. The collision resolution procedure ends when all stations that collided successfully transmit their packets. This time interval is known as a collision resolution interval (CRI). A station that generates a new transmission request, when a CRI is in progress, has to wait until the current CRI ends before attempting channel access. Thus, the 2C algorithm is able to provide guarantees for fair channel access.

Each station participating in a CRI maintains a counter that controls its channel access. Let us denote by the value of this counter at the beginning of slot . A station is allowed to attempt channel access in slot only when . Let be the feedback message corresponding to the transmission in slot and received at the end of the same time slot. If the transmission was unsuccessful, , otherwise the feedback message is .

Let us assume that up to the current slot all packets have been transmitted. Stations with new transmission requests will set their counter to 0 and will attempt channel access in the following slot . Depending on the feedback messages each station updates its counter as follows:

• (i) if and , then ,

• (ii) if and , then ,

• (iii) if and , then will increase to 1 with probability 0.5.

Regarding the last policy for updating the counter, it is worth mentioning that in [3] it was proposed to use 0.5 as the probability of increasing the counter to 1 when a collision was detected. However, it has been shown [11] that the optimum value for such probability depends on the actual number of colliding stations. Since the number of contending nodes is most likely unknown at network start time, we will use the value of 0.5 in this work. In following sections we will denote by the probability of increasing the counter after a collision and by the probability of not changing it (with the obvious condition that ).

According to the previous description of the 2C algorithm note that, following an NC feedback message, all stations in the waiting group will attempt to transmit in the following slot. Therefore, two consecutive NC feedback messages can only occur at the end of the CRI.

This algorithm is called 2C because contending stations may be either transmitting or waiting and the two states can be represented using two cells in a stack. The transmission cell (TC) represents the group of transmitting stations (i.e., ) and the waiting cell (WC) the group of stations that have deferred transmission (i.e., ).

The 2C algorithm is not tied to any specific transmission medium so that the original description has to be adapted to the particularities of wireless communications. In the original 2C algorithm it is assumed that there is a central station that is continuously monitoring the channel and providing feedback messages. However in self-configuring wireless ad hoc networks it cannot be assumed that there is such infrastructure in place. In this case the very same network nodes have to assume this role by monitoring the transmission medium and reacting accordingly. This issue leads to a second one. Whereas in wired networks it is rather easy to detect collisions, in wireless networks this is not a trivial matter. We propose that, instead of detecting a collision, the network nodes infer that a collision has happened. A wireless node can infer that its transmission has collided if the reply to its request does not arrive. In this case, and according to the 2C algorithm, a station has to randomly choose whether to retransmit (i.e., to remain in the TC) or to enter into the waiting group (WC). When a successful transmission is sensed, all stations in the WC enter into the TC and contend again for the channel. No new stations are allowed to contend until the initial collision is resolved. Eventually, all stations that collided at the beginning achieve a successful transmission. We will name this proposal 2C-WSN.

4. Network Configuration in SA-MAC

In this section we describe how 2C-WSN solves the conflicts arising during the setup phase of a TDMA protocol. Without loss of generality, we take as a case study the setup phase of SA-MAC, a TDMA protocol specifically designed for wireless sensor networks. It is worth emphasizing that the 2C-WSN mechanism could be easily integrated for solving the conflicts arising during the setup phase of other TDMA protocols.

4.1. SA-MAC Overview

The main aim of the SA-MAC protocol is to schedule transmission opportunities in the network. In the following, the procedure for network configuration will be described by considering one coordinator node which is responsible of gathering all the data having been sensed by all the other nodes. In large networks some of the other nodes may have to act as coordinators thus enabling the forwarding of collected data to the sink station through multihop paths.

The SA-MAC protocol makes use of the superframe structure defined in the 802.15.4 [12] standard for a beacon-enabled network. Network beacons are broadcasted by a coordinator node and they are used to synchronize the network by signaling the boundaries of superframes. In multihop networks the beacons are also used to identify a local coordinator as a possible relay node to the sink station. Superframes are divided into 16 equally sized slots where the first one serves as the beacon. The network can enter into either active or inactive modes. In the inactive mode the coordinator will not interact with its associated nodes and may enter in a low-power mode. In the active mode there are periods for network setup and data transmission. The setup period is where the network configuration takes place. To this end, the network nodes exchange three types of packets, namely, discovery packets (DSC), delay packets (DLY), and acknowledgement packets ( and ) to establish father-and-child relations and to get slot allocations to be used for data transmission. The exchange of these four packets forms an atomic operation, from now on referred to as atomic association operation (AAO). In this work we will only focus on the setup phase of the protocol, other aspects of its operation can be consulted in [4].

Let us consider a simple scenario consisting of a coordinator (i.e., the sink node) and a set of nodes within its transmission range. The coordinator announces its presence as a parent node, using a PA packet as beacon, so that all other nodes can start trying to establish a father-and-child relation with it. All nodes that become aware of the presence of the coordinator start to broadcast DSC packets. Upon receiving a DSC packet, the coordinator replies with a DLY packet. The delay packet indicates the time slot that is assigned for transmissions from the sensor node to the coordinator. The node acknowledges the slot allocation with an packet, and finally the parent node finishes the association procedure by sending an packet. Once a sensor node ends its association, it may become a parent node for other nodes.

In order to illustrate the operation of the SA-MAC protocol in a more complex scenario consider a set of nodes as illustrated in Figure 1. This scenario consists of a base station (BS) and nodes , , and . Let us assume that and are located within the coverage area of the BS and that is located within the coverage area but out of the reach of the BS. Once the BS announces its presence, using a PA packet as beacon, nodes and can start sending DSC packets and collisions may occur at this time. Thus, a policy has to be implemented in order to resolve collisions. Let us assume that the collision resolution protocol lets successfully transmit its DSC packet first and in this way it establishes a father-and-child relation with the BS, completing an AAO. Node detects the end of the AAO between and the BS and it sends its DSC, establishing a father-and-child relation too.

Figure 1

Example of a packet exchange in SA-MAC for the network shown in the upper right corner.

Nodes that are already part of the network may serve as coordinators of a new association domain. This process is initiated when these nodes broadcast their beacon (i.e., a PA packet). In our example node starts an association domain and is able to use it as a relay node in a route to the BS. By itself, the beacon scheduling mechanism for multihop topologies is an important problem [13]; for this work we assume this problem solved by the time division approach proposed by the Task Group 15.4b [14].

In order to choose the best parent (i.e., the one with the lowest hop count to the BS), nodes that want to join the network can overhear packet exchanges from associations that take place in their neighbourhood. These packets carry information about the number of hops to the sink node and can help other nodes choose the best parent node. At present, only this parameter has been taken into consideration in the design.

As nodes get an association to the coordinator node, they will be assigned guaranteed slots at the end of the superframe.

4.2. Integrating 2C-WSN into the SA-MAC TDMA Protocol

The overall network setup starts when the coordinator node is powered on. As previously mentioned, the coordinator (i.e., sink node or BS) starts the network configuration by issuing a Parent Available (PA) packet or beacon. The configuration process requires that the nodes that are already part of the network offer themselves as local coordinators by broadcasting PA packets. Other nodes that receive a PA packet decide whether to select the transmitting node as their parent node or not by taking into consideration the reported hop count to the BS. Figure 1 depicts a scenario and a possible packet exchange that may take place during the configuration of this network.

From the previous description of SA-MAC operation, there is one situation when conflicts may arise when the nodes aiming to join the network attempt to issue their DSC packets. There is, therefore, a clear need of a reliable and fast collision resolution protocol to be included into the setup phase. In the following, we specify the operating mode of the 2C-WSN when used to solve the conflicts during this time period.

Having detected the presence of a coordinator, two main outcomes are possible when the nodes attempt to join the network: (1) only one station broadcasts its DSC packet or (2) two or more stations broadcast their DSC packets. In the former case, the coordinator will reply to the requesting node by issuing a DLY packet completing, after the two acknowledgement packets, the AAO. In the second case, that is, several nodes issue their DSC packets during the same time slot which results in a collision at the coordinator involving all participating nodes, the nodes involved in the collision will realize that a collision has resulted since they will not get any reply from the coordinator node during the following slot. They will then invoke the 2C-WSN process, that is to say, each one of them, and independent from each other, will decide to transmit once again with probability or refrain from doing so with probability . The nodes will proceed this way till only one of them succeeds by getting back a DLY packet in response to its DSC packet. The coordinator node having issued the DLY packet becomes in this way its parent, and it has to take into account its superframe structure for slot reservation during the data transmission period. The latter has been added to the specification of the overall procedure, and it has, as main purpose, to let all nodes within the transmission range of the coordinator know that the association has been successfully completed. Once this association is completed, the node or nodes, if any, waiting in the WC cell will attempt to place their request and, if needed, the collision resolution mechanisms will be activated as already described.

A potential new father must detect three consecutive idle slots before attempting to broadcast a beacon packet. In this way, the node makes sure that no neighbouring nodes are still engaged in a collision resolution process. In other words, this period ensures that even the nodes in the waiting cell should be allowed to proceed first before new nodes are invited to join the network. For the same reasons, new nodes willing to join the network must also sense three consecutive empty slots before issuing a DSC packet. Once again, it is worth to mention that the beacon broadcast should be properly scheduled using a scheduling scheme such as the one proposed by the IEEE802.15.4 Task Group [14].

5. Simulation Experiments and Results

We used discrete event simulations in order to observe the performance of the proposed protocol under different scenarios. For our performance study, we implemented the SA-MAC and 2C-WSN protocols using OMNeT++ and the project Castalia [15]. For comparison purposes we also implemented the CSMA/CA protocol in Castalia. Table 1 lists the parameters used in our simulations. The CSMA/CA parameters follow the specifications of the IEEE 802.15.4 standard, that is, default values. We made use of the simulation model for the radio chip CC2420 as implemented in the Castalia project.

Table 1 Relevant simulation parameters.

5.1. Simulation Scenarios

In order to investigate the effect of node density and spatial distribution of the network nodes we set up Cases A and B described below.

Case A.

Irregular topology and increasing density. Network nodes were placed at random over a circular simulation area of radius R. Parameter R corresponds to the transmission range of the nodes and it was set to 50 m. The sink node (ID 0) was located in the center (see Figure 2), and the number of nodes was varied from 3 to 21.

Figure 2

Examples of the spatial node distribution for Case A (black dots) and Case B (white dots).

Case B.

Grid topology and increasing density. In this case the nodes were placed at random in the different intersections of a grid pattern. Although it is generally assumed that sensor nodes will most likely be deployed at random, we used this scenario in order to compare with Case A and determine the effect of having equidistant nodes on the association procedure. We used the same assumptions and parameter values as in Case A (see also Figure 2).

With scenarios C and D, described below, we studied how the algorithm scales when it is used in networks that span across large geographical areas.

Case C.

Irregular topology and increasing area. The area covered by the network was assumed to be circular with the sink node located in its center. All nodes were assumed to have a circular coverage area with a transmission radius of 50 m (i.e., R). We considered areas with different radius from R to 10R, but we maintained a constant node density. In the largest area we used as many as 1959 nodes, and for each simulation run, the network nodes were repositioned.

Case D.

Grid topology and increasing area. We used the same assumptions and parameter values as in Case C.

For each scenario and a particular combination of parameters, we ran 200 simulations in order to obtain 99% confidence intervals for the mean network creation time. This metric is defined as the time elapsed between the transmission of the initial PA packet issued by the base station until the time instant when the last node association takes place. We also report the number of unsuccessful attempts required by the CSMA/CA and the 2C-WSN to transmit the signalling packets of SA-MAC. Following the specifications of IEEE 802.15.4, in case the number of backoffs reaches the value MaxCSMABackoffs, CSMA/CA declares the network as unreachable.

5.2. Simulation Results

Cases A and B

In these cases all nodes are placed within the transmission range of the base station. Figure 3 shows the resulting mean network configuration time as a function of the number of nodes composing the network. As seen from the figure, 2C-WSN outperforms CSMA/CA. Furthermore, CSMA/CA began having problems completing the network configuration for a system consisting of as few as seven nodes. This is due to the fact that once having reached the value defined in the parameter MaxCSMABackoffs, CSMA/CA gives up and reports a network failure to upper layers. In this case, such layers have to decide which action will be applied. This result clearly shows how sensitive CSMA/CA is with respect to its parameter values. In case of the system configuration making use of 2C-WSN, the figure shows that this protocol is able to perform the network configuration with a reasonable increase in the required time as the number of nodes increases. These results also show that our proposal can, in fact, guarantee the network configuration.

In order to observe the time that CSMA/CA would take in order to configure dense networks without being restricted by the value of MaxCSMABackoffs, we proceeded as follows. In order to prevent CSMA/CA from giving up a network configuration when the value of MaxCSMABackoffs was reached, after a failed transmission attempt the corresponding packet was rescheduled for transmission as many times as necessary until its successful transmission was achieved. We used the scenarios described in Case A and Case B, and Table 2 summarizes the corresponding results in terms of the number of collisions and its related effects.

Table 2 summarizes some relevant collision-related measurements, on average, for both algorithms. The column labelled Backoff limit reached indicates the average number of times that a network failure was reported by CSMA/CA to upper layers before a successful association was completed. In these cases the corresponding packets had to be reinserted. The table shows that, with as few as seven nodes, CSMA/CA incurs in network access problems, as previously pointed out. The column Collis. indicates the average number of times that network nodes using CSMA/CA collided before the network configuration was achieved. As seen in the table, it is clear that the number of collisions is significantly higher for CSMA/CA than for 2C-WSN. Furthermore, the number of collisions grew with the number of nodes composing the network, but the grow rate is lower for 2C-WSN.

Table 2 Collision-related results.

Figure 3

Network setup times for Case A and Case B.

Cases C and D

These cases are intended to test the scalability of 2C-WSN with different network sizes. As previously described, we increased the radius of the simulation area from R to 10R in steps of  m and maintained the same node density.

Figure 4 shows the behaviour of the creation time as a function of the network size. Once the network includes the nodes that are far away from the base station, the required time for the network creation increases, but the growth rate is rather slow. For instance, based on the data shown in Figure 4 when the network radius increases from R to 6R, the creation time for a random topology increases from approximately 0.6 s to 1.5 s () when the corresponding area increases from 7,854 m² to 282,743 m² () and the number of nodes increases from 21 to 709 (). This relatively small increase in configuration time is due to the limited transmission range of the network nodes combined with a large network size and the fact that the network configuration functions are not centralized. This situation allows the simultaneous creation of two or more branches of the tree in different parts of the network. This result shows that it is feasible to use 2C-WSN in the configuration phase of large TDMA networks in reasonable time.

We also collected statistics regarding the tree depth in the last hop of the network. Figure 5 shows the corresponding results and depicts the relation between network size and hop count. For instance, for a network radius of 6R, the average hop count was between 7 and 8. However, under the best circumstances in this case a node near the border of the nework should be reached with, at the most, a 6-hop route. A number of factors influence this result such as node density and whether the placement of the nodes is regular or not. As it can be seen in the figure the grid topology achieves a slighly shorter hop count than the irregular one.

Figure 4

Network configuration times for Cases C and D.

Figure 5

Mean tree depth for Case C and Case D.

6. Experimental Platform and Evaluation

In this section, we describe a first prototype of our proposal and provide an experimental assessment using a network composed of four nodes. Our findings, with a small system like this, provide a useful insight on real network performance and help us foresee how such performance would scale to larger networks.

6.1. System Configuration

The experimental platform was developed using MicaZ motes, a commercial product developed by Crossbow [16]. The mote incorporates an Atmel Atmega128L microcontroller [17] operating at 8 MHz and comprising a 128 Kbytes program flash memory and 4 Kbytes of user memory (RAM). The mote also includes a Chipcon CC2420 [18] IEEE 802.15.4 radio system equipped with a 2.4 GHz RF transceiver designed for low-power wireless applications with an effective data rate of 250 kbps. The motes operate under TinyOS [19], a popular operating system for wireless sensor networks. TinyOS features a component-based architecture, that is, the software is structured in modular pieces called components. TinyOS provides a component library including network protocols, services, and sensor drivers. Its network architecture provides a medium access control layer based on the CSMA/CA protocol [20]. We developed our proposed protocol using NesC and evaluated its performance against the CSMA/CA component provided by TinyOS. The packet lengths were fixed as follows: DSC 9 bytes, DLY 12 bytes, 12 bytes, and 12 bytes. In our experiments we monitored the current demanded by the sensor node which is an indication of the instantaneous power consumption and node activity. The curves shown in this section were obtained by using an instrumentation setup that made use of a four-channel digitizing oscilloscope with a 10 MHz sampling rate.

6.2. Experimental Evaluation

Our first experiments had as main objective to determine the time required to complete a father-and-child association, AAO. Recall that this operation consists of exchanging four packets: DSC, DLY, , and . Several trials were conducted placing the nodes at a distance ranging from 1 to 20 meters in a line-of-sight situation. The AAO time obtained throughout our trials ranged between 8.09 ms and 8.33 ms. Figure 6 shows a snapshot of an exchange of packets with the nodes placed three meters away from each other. The solid and dotted lines correspond to the base station (coordinator) and the child node, respectively. The total time required to complete the association depicted in the figure was 8.10 ms, measured from the time the node issued the DSC packet till it completely received the acknowledgement from the base station, . The consumption for both nodes is also shown in the figure.

Figure 6

AAO in real implementation using 2C-WSN.

The values obtained experimentally for the AAO that resulted substantially were higher than the ones considered in our simulations. This was mainly due to the fact that the model of the radio chip CC2420 implemented in OMNeT++ does not take into account the switching time from the RX to TX modes nor the data buffer or radio crystal startup delays.

As already mentioned, TinyOS uses by default the CSMA/CA medium access protocol. For comparison purposes, we carried out a second experiment for assessing the AAO time when using CSMA/CA. Figure 7 shows a snapshot of the packet exchanges. The time required to complete the AAO was about 35.5 ms, that is, over four times longer than the time required by our protocol. It is worth mentioning that the CSMA/CA implementation makes use of the Clear Channel Assessment (CCA) mechanism to verify that the channel is free, after a random delay chosen in the interval [0, ].

Figure 7

AAO in real implementation over CSMA/CA.

We configured the network shown in Figure 1, with the difference that Node 2 was moved into the communication range of BS. Figure 8 shows a snapshot of the operation of the network. The crosses over the traces indicate the points at which the DSC packets collided. The first collision involves all three nodes. In a second attempt, Node 1 issues its request and it is able to complete its association (the check mark over the trace indicates this fact). Nodes 2 and 3 having refrained from attempting to issue their request, then attempt once again. A collision results and in a second attempt, Node 2 is able to perform its association, finally Node 3 gets to join the network, achieving the setup time in about 35 ms.

Figure 8

AAO using three nodes and a coordinator over 2C-WSN.

Figure 9 shows the association time obtained for the network used in the previous case using CSMA/CA as underlying MAC protocol. The successful completion of the association event is marked with a check mark. As seen from the figure, the time required for the whole operation takes about 150 ms, which is substantially longer than the time required by our proposal. The results are scalable to multihop networks since the collision resolution algorithm equally works in new areas of the network. That is, no collisions arise among superframes that belong to different network coordinators. Regarding this last statement, superframes are required to use either different time slots or frequencies. However, this superframe schedule is out of the scope of this work.

Figure 9

AAO using three nodes and a coordinator over CSMA/CA.

7. Conclusions and Future Work

In this work we focused our attention on the setup phase of TDMA wireless sensor networks. This phase is often overlooked, but we have pointed out the various conflicts that may arise during it. Based on the particularities of WSNs, we proposed 2C-WSN, a conflict resolution protocol intended to be used during the network configuration. Our proposal is based on the advantageous properties of the 2C conflict resolution algorithm, namely, simplicity and fairness. We took the configuration phase of SA-MAC (a TDMA-based synchronous MAC protocol) as a case study and carried out a performance evaluation by means of computer simulations and measurements in a real system. Our first set of simulation results showed that our proposal is able to set up a highly populated wireless sensor network within reasonable time bounds. From the second simulation campaign we showed that our proposal scales well by keeping within reasonable bounds the time required to configure networks consisting of a large number of nodes spread over a wide geographical area. Based on these results we showed that our proposal is robust and scalable. We also implemented 2C-WSN in real sensor nodes and confirmed the improvement in performance in comparison with the widely used CSMA/CA protocol. As compared with CSMA/CA our proposal is easier to implement, faster and the channel access is guaranteed.

There are a number of directions in which we plan to extend our work. In particular, we plan to conduct a series of experiments in a real-world application such as vineyard monitoring [21].