In this section, we present several researches related to long-thin wireless sensor networks and CDMA.
As mentioned in Chapter 1, our objective is to design a sensor network system with dual modes. The traditional TDMA-based energy-saving mode is used during the regular operation, and a CDMA-based high-speed mode is used in emergency situations.
In CDMA high-speed mode, each node aggregates its own sensor data to the packet it receives from its down-stream neighbor and transmits to its up-stream neighbor in its transmitting time slots.
With this one-by-one transmission procedure, one of our goals is to find a pipelined aggregated convergecast [6] scheduling scheme to schedule all transmission with less time slots in CDMA WSNs with long-thin topologies. Several researches [6, 11, 12, 13]
have discussed aggregated convergecast in TDMA networks; it can be modeled as a graph coloring problem since each receiver in TDMA networks can only successfully receive the packet from one transmitter at a time. Incel et al. have been shown that the lower bound of the length of TDMA pipelined aggregated schedule is Ω(△T), where △T is the maximum node degree in the network [6]. Conceptually, this lower bound gives an upper limit to the effective throughput which can be achieved by a TDMA network.
8
The authors of [6] also designed a heuristic algorithm which constructs a degree-constrained routing tree to decrease the number of aggregated time slots.
However, it is not guaranteed the algorithm can find a solution. In the extreme case, if there are many chains connecting to the gateway, this algorithm would not find a solution since the number of nodes within the receiving range of the gateway is too large thus the number of interfering links could not be eliminated to a given value of node degree.
Radunovic and Boudec [3] performed a theoretical analysis to derive the optimal scheduling and power control in networks with ring topologies and infinite line topologies in terms of transmission rate with a physical interference model. They found that a node transmitting with maximum power in active mode with the optimal scheduling can be represented as a combination of periodic and symmetric (and rotational) links in networks with infinite line (and ring) topologies. The result matches with researches based on protocol (graph-based) interference model. Similarly, Baccelli et al. [7] give a theoretical analysis to the scheduling scheme of the infinite line topologies. They discuss some performance metric maximizations such as the optimum duty cycle of scheduling under a given SINR threshold, i.e., the ratio of times of each link transmitting to the number of total time slots. As both [3] and [7] use the simplified assumption that the number of nodes in the line network is infinite, all links, as a result,
9
can use the same transmission power using the symmetric property of the network. As the assumption is not realistic in these researched, the results are not applicable in a real-world WSN. In this thesis, we evaluate WSNs with single-chain and multiple-chain topologies with a limited number of nodes. We assume that the gateway can receive data from multiple transmitters at each time slot. Our goal is to find scheduling principles that take advantage of the simplified topologies and can achieve higher throughput than that of a TDMA-based system.
Power control is an important aspect for CDMA-related researches. With the spreading code technique, the lack of spatial-temporal separation in CDMA networks could still result in considerable interference to receivers. The use of a power control scheme to deal with the interference and ensure the signal quality is a common technique used in CDMA networks. There are many different power control schemes proposed by researchers. An objective function common-used [4, 8, 9, 14] for the power control scheme is to minimize the total power of all transmitting nodes under the constraint that the SINR of every link is no less than a pre-specified threshold 𝛽 in a time slot. This is appropriate for WSNs since sensor nodes are usually energy-constrained. However, the scheme only ensures the one-hop signal quality.
When applying the scheme to the chain topology, the number of hops to reach the end node is so large that the route packet error rate (PER) to the gateway is too poor. In
10
addition, it is essential to obtain the knowledge of accurate pairwise link gains in order to use this approach. However, in practice to obtain the information it takes a great amount of overhead. In an environment with rapid small-scale fading, the link gains change so frequently and rapidly that tracking them becomes impractical. In this thesis, we aim to design a power allocation scheme that is suitable for CDMA WSNs with single-chain topologies based on the aforementioned objective function. With our scheme, each link can achieve the maximum SINR instead of just over a SINR threshold, and, in turn, lower the route PER and increase the overall system throughput without the need of having the knowledge of pairwise link gains.
Last, we introduce some researches about CDMA-based WSN in brief. LEACH [16]
is a clustering scheme MAC protocol which combines TDMA and CDMA techniques.
Cluster heads schedule the nodes in their corresponding cluster. Nodes in the same cluster use the same PN code to spreading the original data to the cluster head, and nodes in different clusters use different PN code to reduce inter-cluster interference.
Since all the transmission links in the same cluster still cannot overlap, the scheme would not match the high throughput demand in urgent situations. [17] and [18] use CDMA-based MAC protocols which are combined with frequency division techniques to further reduce MAI. PN code assignment and frequency division can both be modeled as two-hop vertex coloring problems. In [18], receiver-based frequency
11
division scheme, i.e., nodes with a common neighbor cannot use the same frequency, and transmitter-receiver pair based code assignment scheme, i.e., adjacent links in the logical topology cannot use the same PN code, are proposed. In [17], the authors proposed a broadcast scheme where each node uses a different PN code but a common frequency to transmit broadcast packets in order to reduce expensive frequency division cost while broadcasting. Thus they employed one receiver for unicasting and the other one for broadcasting in a sensor node. In the general case networks, the frequency division approaches could reduce the MAI significantly, but with the natural feature where the number of neighbors of nodes in the chain networks is usually not large, applying both power control and frequency division in the sensor nodes seems unnecessary.
12