## Final Activity and Management Report Summary - DISTMO4WNET (Distributed multiobjective optimization for ad hoc wireless networks)

Wireless networks have become a well-established and convenient means of communication. This project concentrates on specific wireless networks that do not rely on a fixed infrastructure to carry the data between distant users that are called ad hoc networks. Such networks are rapidly deployable and can provide wireless connectivity to groups of people, as it would be needed in disaster relief scenarios for instance. Wireless devices (or nodes) have to auto-organise themselves to provide the best possible communication service to their users. When direct transmission between an emitter and its receiver isn't possible, some users of the network have to relay the transmission. Picking the appropriate set of relays to forward the messages is known as a routing problem, which is of interest in this project.

The design of wireless networks is strongly constrained by interference. Interference arises when multiple users transmit at the same time and on the same frequency and has for consequence of degrading the throughput of the communication. To avoid such effect, the communication channel has to be shared between the users and hence, proper transmission times or frequencies have to be assigned to them. This problem is commonly referred as the resource allocation problem and is of interest in this project as well. Since the topology of an ad hoc network is changing over time, it is important to find distributed algorithms that solve the routing and resource allocation problem. Distributed means that the nodes take their own routing and resource allocation decisions. The performance of such algorithms is commonly measured with performance criteria such as throughput, reliability, delay or energy expenditure. Depending on the application context, some criteria are more or less important. For instance, wireless networks transmitting video need to provide high throughput communications.

The DISTMO4WNET project aimed at answering the two following questions:

1 What are the optimal performance points of an ad hoc wireless network with respect to capacity, reliability, transmission delay and energy?

2 Which distributed algorithms can reach such optimal performance points?

To answer the first point, we have formulated the joint routing and resource allocation problem as a multi-objective optimisation problem. In multi-objective optimisation problems, several performance criteria are optimised concurrently. With such perspective, we are looking for a set of optimal working points that give the best possible trade-offs with respect to capacity, energy and delay for instance. This set is called the Pareto set and in our case it provides the Pareto bound in performance for the wireless ad hoc system. Knowing such Pareto bound is a great tool to quantify the quality of already existing distributed routing and resource allocation algorithms. The other very interesting point is that it helps us understanding the limits in performance of a wireless network.

The framework developed in the DISTMO4WNET project is complex and unique because it relies on a network model that accounts for both interference and the interactions between routing and resource allocation decisions. The Pareto bounds are derived using a multi-objective Tabu search optimisation algorithm that we have designed to get the most precise description of the Pareto set and bound. This framework has been illustrated on a wireless network of about 300 users when criteria such as reliability, delay and energy are considered. We have shown through simulations that the bounds derived are tight.

Within the DISTMO4WNET project, we have worked as well on the design of distributed algorithms. We have designed routing protocols inspired by the network model of the multi-objective framework. These algorithms are closely related to gradient broadcasting algorithms where nodes decide with a given probability whether they broadcast a received packet or not. Two types of algorithms were designed, one where the routing decision accounts for the interference it may create and the other one where it leverages the instantaneous channel load.

The design of wireless networks is strongly constrained by interference. Interference arises when multiple users transmit at the same time and on the same frequency and has for consequence of degrading the throughput of the communication. To avoid such effect, the communication channel has to be shared between the users and hence, proper transmission times or frequencies have to be assigned to them. This problem is commonly referred as the resource allocation problem and is of interest in this project as well. Since the topology of an ad hoc network is changing over time, it is important to find distributed algorithms that solve the routing and resource allocation problem. Distributed means that the nodes take their own routing and resource allocation decisions. The performance of such algorithms is commonly measured with performance criteria such as throughput, reliability, delay or energy expenditure. Depending on the application context, some criteria are more or less important. For instance, wireless networks transmitting video need to provide high throughput communications.

The DISTMO4WNET project aimed at answering the two following questions:

1 What are the optimal performance points of an ad hoc wireless network with respect to capacity, reliability, transmission delay and energy?

2 Which distributed algorithms can reach such optimal performance points?

To answer the first point, we have formulated the joint routing and resource allocation problem as a multi-objective optimisation problem. In multi-objective optimisation problems, several performance criteria are optimised concurrently. With such perspective, we are looking for a set of optimal working points that give the best possible trade-offs with respect to capacity, energy and delay for instance. This set is called the Pareto set and in our case it provides the Pareto bound in performance for the wireless ad hoc system. Knowing such Pareto bound is a great tool to quantify the quality of already existing distributed routing and resource allocation algorithms. The other very interesting point is that it helps us understanding the limits in performance of a wireless network.

The framework developed in the DISTMO4WNET project is complex and unique because it relies on a network model that accounts for both interference and the interactions between routing and resource allocation decisions. The Pareto bounds are derived using a multi-objective Tabu search optimisation algorithm that we have designed to get the most precise description of the Pareto set and bound. This framework has been illustrated on a wireless network of about 300 users when criteria such as reliability, delay and energy are considered. We have shown through simulations that the bounds derived are tight.

Within the DISTMO4WNET project, we have worked as well on the design of distributed algorithms. We have designed routing protocols inspired by the network model of the multi-objective framework. These algorithms are closely related to gradient broadcasting algorithms where nodes decide with a given probability whether they broadcast a received packet or not. Two types of algorithms were designed, one where the routing decision accounts for the interference it may create and the other one where it leverages the instantaneous channel load.