Final Report Summary - COMONO (Cooperative Mobile Network Optimization)
Mobile broadband assess is already a common technology today, and we expect an even faster increase in future. Wireless communication is becoming more and more important across the world. Increasingly higher data rate is needed to satisfy many demanding mobile applications such as mobile television (TV), real-time Internet gaming, large file transfer etc. The development of the high speed packet access (HSPA) and the multimedia broadcast and multicast service (MBMS) have brought significant enhancements to the existing 3GPP standards, allowing many services which have not been possible before. However, in terms of delivery of bandwidth demanding applications such as mobile TV, HSPA still lags behind broadcasting technologies such as digital video broadcasting - Handheld (DVB-H) or broadband wireless access such as worldwide interoperability for microwave access (WiMAX).
In response, a technology known as the long term evolution (LTE) has been proposed in order to significantly improve the network throughput as well as user bit rate and latency.
By using an orthogonal frequency division multiple access (OFDMA), LTE can potentially achieve a performance improvement between two-five times in the downlink, and two-three times in the uplink over the HSPA. Also, LTE is based on a clean slate of a global IP (internet protocol) network, ensuring a good compatibility with the Internet environment, making mobile accesses to the Internet more pervasive.
Although one of the driving philosophies behind LTE is to reduce complexity and improve system scalability, LTE networks are nevertheless complex, involving a large number of parameters which can potentially affect the system performance. Therefore optimal setting of these parameters is crucial for the effective operation of an LTE network. Furthermore, due to the fast changing traffic and even environmental conditions, this must be done in a dynamic fashion.
To reflect to these challenges, the goal of the project was investigating the following three fundamental concepts.
- Dynamic radio resource allocations
Spectrum allocation serves two purposes: to mitigate inter-cell interference effectively, and to direct the usage of the spectrum to where it is needed. To meet these goals, the OFDMA bases LTE technology provides three means of controlling the radio resources: (a) Resource block (RB) allocation, (b) modulation-and-coding scheme (MCS) selection for each RB and (c) transmission power control. Their interdependent effect on the overall network performance calls for a dynamic joint optimization of these parameter.
- Self-optimising networks (SONs)
To be able to efficiently operate a network, a good set of parameter values must be selected. Unfortunately, such a selection is often very difficult due to the intricate interactions among these parameters. To make the matters worse, a good set of parameter values varies in time due to the mobility, the changing traffic loads, and channel conditions among users in the network. Thus, the concept of self-configuring, self-optimising and self-healing networks is becoming increasingly important.
- Centralised vs distributed network management and control
There are two alternative fundamental approaches to network management, the centralised and the distributed approach.
In the first case, a central network management equipment (CNM) manages the network and makes decisions about the system processes. For doing that the CNM must gather all the information necessary for the decision making, and must also refresh it periodically. Once the necessary information is collected, then the decision is made and sent back to the appropriate network component. The problem with this approach is in one hand the bad scalability due to the high amount of control information to be sent and due to the unavoidable high latency, and in the other hand, the reliability issues due to the fact that the central managing node becomes a single point of failure.
In case of the distributed approach the network elements make the decisions on the processes individually, based on a very limited knowledge about the overall network state. As the decisions are made in place, it based on accurate and up-to-date information and the latency is very low. On the other hand, the global state of the network is not taken into account and only locally optimal decisions can be achieved in this way.
The project was terminated early for personal reasons. The results achieved in the first eight months can be summarised as follows.
1. Simulator framework development
As one of the predefined goal of the project, a fast but still flexible LTE simulation framework has been developed.
It is written in C++, keeping the portability in mind. As a result, the simulator is able to run on various platforms (Linux, Windows, OSX etc.). The basis of the simulator is an event-driven simulation engine built around a simple but efficient dynamic event scheduler, which allows handling the sequence of the events generated dynamically during the simulated virtual time-line. Events are used for example to make a user to appear, disappear or move in the scenario. Also events are used to implement e.g. the self-organisation procedures of femtocells. Based upon this engine, both frame level and packet level simulators have been developed.
The former one enables less detailed traffic modelling, but on the other hand it has a lower run-time complexity. In this context, frame-level means that the individual packets are not simulated, but instead the user terminals are assumed to always have data to be transmitted. The relatively lower run-time complexity makes this simulator especially suitable for evaluating scenarios and comparing new procedures where a high number of cells and users need to be simulated.
On the other hand, the packet level performs a more detailed traffic modelling, i.e. the traffic can be defined on the packet level. It does detailed simulation of the LTE downlink data plane (packet data convergence protocol (PDCP), radio link control (RLC), medium access control (MAC) and physical layers), on the expense of a higher run-time complexity compared to the frame level simulator.
2. Self-organising OFDMA femtocell frequency assignment
This work deals with femtocell network using OFDMA technology. Being a femtocell base station a highly autonomous system, while it is planned to be deployed at a high density, its efficient operation calls for self-organising and self-optimisation -therefore distributed- techniques. As a response to these challenges, two novel approaches for the self-organisation of OFDMA femtocells has been developed. They rely on femtocells being able to dynamically sense the air interface and tune its sub-channel allocation in order to reduce inter-cell interference and enhance system capacity. In the _sensing_ phase, these techniques make use of either messages broadcast by the femtocells or measurements reported by the users, while in the _tuning_ phase, they provide a good solution for the frequency assignment problem.
Simulation results confirm that it is worth using information collected at the user position (measurement reports), when devising self-organisation algorithms for tuning the parameters of femtocells.
3. Joint modulation scheme, coding rate, RB and power allocation
This work investigates the problem of the allocation of modulation and coding, subcarriers and power to users in LTE. The proposed model achieves inter-cell interference mitigation through the dynamic and distributed self-organisation of cells. Therefore, there is no need for any a prior frequency planning. Moreover, a two-level decomposition method able to find near optimal solutions is proposed to solve the optimisation problem. The lower level optimisation is able to efficiently find an optimal solution to the combined allocation problem assuming that the modulation schemes to be used is known a priori. Using this method as a subroutine, the higher level optimisation solves the original problem with metaheuristics. Based on the observation that the computational requirement of the lower level algorithm is small, a practical resource management architecture is architecture is proposed, where the low level optimisation is run at a high frequency in order to react on rapid changes in network parameters, while the higher level algorithms continuously improves the overall performance of the network node.
In response, a technology known as the long term evolution (LTE) has been proposed in order to significantly improve the network throughput as well as user bit rate and latency.
By using an orthogonal frequency division multiple access (OFDMA), LTE can potentially achieve a performance improvement between two-five times in the downlink, and two-three times in the uplink over the HSPA. Also, LTE is based on a clean slate of a global IP (internet protocol) network, ensuring a good compatibility with the Internet environment, making mobile accesses to the Internet more pervasive.
Although one of the driving philosophies behind LTE is to reduce complexity and improve system scalability, LTE networks are nevertheless complex, involving a large number of parameters which can potentially affect the system performance. Therefore optimal setting of these parameters is crucial for the effective operation of an LTE network. Furthermore, due to the fast changing traffic and even environmental conditions, this must be done in a dynamic fashion.
To reflect to these challenges, the goal of the project was investigating the following three fundamental concepts.
- Dynamic radio resource allocations
Spectrum allocation serves two purposes: to mitigate inter-cell interference effectively, and to direct the usage of the spectrum to where it is needed. To meet these goals, the OFDMA bases LTE technology provides three means of controlling the radio resources: (a) Resource block (RB) allocation, (b) modulation-and-coding scheme (MCS) selection for each RB and (c) transmission power control. Their interdependent effect on the overall network performance calls for a dynamic joint optimization of these parameter.
- Self-optimising networks (SONs)
To be able to efficiently operate a network, a good set of parameter values must be selected. Unfortunately, such a selection is often very difficult due to the intricate interactions among these parameters. To make the matters worse, a good set of parameter values varies in time due to the mobility, the changing traffic loads, and channel conditions among users in the network. Thus, the concept of self-configuring, self-optimising and self-healing networks is becoming increasingly important.
- Centralised vs distributed network management and control
There are two alternative fundamental approaches to network management, the centralised and the distributed approach.
In the first case, a central network management equipment (CNM) manages the network and makes decisions about the system processes. For doing that the CNM must gather all the information necessary for the decision making, and must also refresh it periodically. Once the necessary information is collected, then the decision is made and sent back to the appropriate network component. The problem with this approach is in one hand the bad scalability due to the high amount of control information to be sent and due to the unavoidable high latency, and in the other hand, the reliability issues due to the fact that the central managing node becomes a single point of failure.
In case of the distributed approach the network elements make the decisions on the processes individually, based on a very limited knowledge about the overall network state. As the decisions are made in place, it based on accurate and up-to-date information and the latency is very low. On the other hand, the global state of the network is not taken into account and only locally optimal decisions can be achieved in this way.
The project was terminated early for personal reasons. The results achieved in the first eight months can be summarised as follows.
1. Simulator framework development
As one of the predefined goal of the project, a fast but still flexible LTE simulation framework has been developed.
It is written in C++, keeping the portability in mind. As a result, the simulator is able to run on various platforms (Linux, Windows, OSX etc.). The basis of the simulator is an event-driven simulation engine built around a simple but efficient dynamic event scheduler, which allows handling the sequence of the events generated dynamically during the simulated virtual time-line. Events are used for example to make a user to appear, disappear or move in the scenario. Also events are used to implement e.g. the self-organisation procedures of femtocells. Based upon this engine, both frame level and packet level simulators have been developed.
The former one enables less detailed traffic modelling, but on the other hand it has a lower run-time complexity. In this context, frame-level means that the individual packets are not simulated, but instead the user terminals are assumed to always have data to be transmitted. The relatively lower run-time complexity makes this simulator especially suitable for evaluating scenarios and comparing new procedures where a high number of cells and users need to be simulated.
On the other hand, the packet level performs a more detailed traffic modelling, i.e. the traffic can be defined on the packet level. It does detailed simulation of the LTE downlink data plane (packet data convergence protocol (PDCP), radio link control (RLC), medium access control (MAC) and physical layers), on the expense of a higher run-time complexity compared to the frame level simulator.
2. Self-organising OFDMA femtocell frequency assignment
This work deals with femtocell network using OFDMA technology. Being a femtocell base station a highly autonomous system, while it is planned to be deployed at a high density, its efficient operation calls for self-organising and self-optimisation -therefore distributed- techniques. As a response to these challenges, two novel approaches for the self-organisation of OFDMA femtocells has been developed. They rely on femtocells being able to dynamically sense the air interface and tune its sub-channel allocation in order to reduce inter-cell interference and enhance system capacity. In the _sensing_ phase, these techniques make use of either messages broadcast by the femtocells or measurements reported by the users, while in the _tuning_ phase, they provide a good solution for the frequency assignment problem.
Simulation results confirm that it is worth using information collected at the user position (measurement reports), when devising self-organisation algorithms for tuning the parameters of femtocells.
3. Joint modulation scheme, coding rate, RB and power allocation
This work investigates the problem of the allocation of modulation and coding, subcarriers and power to users in LTE. The proposed model achieves inter-cell interference mitigation through the dynamic and distributed self-organisation of cells. Therefore, there is no need for any a prior frequency planning. Moreover, a two-level decomposition method able to find near optimal solutions is proposed to solve the optimisation problem. The lower level optimisation is able to efficiently find an optimal solution to the combined allocation problem assuming that the modulation schemes to be used is known a priori. Using this method as a subroutine, the higher level optimisation solves the original problem with metaheuristics. Based on the observation that the computational requirement of the lower level algorithm is small, a practical resource management architecture is architecture is proposed, where the low level optimisation is run at a high frequency in order to react on rapid changes in network parameters, while the higher level algorithms continuously improves the overall performance of the network node.