Final Report Summary - FSA (Fluid Spectrum Acess)
Radio communication networks have become critical in our societies and their usage continues to grow at a dramatic speed. To meet this ever-increasing demand, we envision scenarios where radio devices may access a very large part of the radio spectrum, and aim at designing a set of protocols allowing each device to explore and determine which part of the available spectrum yields the highest throughput, but also allowing devices to dynamically share spectrum in a fair and efficient manner.
We have first investigated the problem of spectrum exploration. More precisely, when a radio device may exploit many radio channels but transmit on a few of them simultaneously, the objective is to determine when packet transmissions should be scheduled, which spectrum part and rate (depending on the chosen modulation and coding scheme) should be used for each packet transmission so as to maximize the throughput, i.e. the number of packets successfully transmitted per second. This task is complicated by fading phenomenon, i.e. by the fact that channel conditions vary over time in an unpredictable manner. We have formulated the problem of optimal spectrum exploration as a bandit optimization problem. Through a collection of novel results in the theory of bandit optimization, we have been able to devise sequential channel and rate selection strategies that optimally learn and track the best channel and rate. The latter have been implemented in a test-bed, thereby illustrating their superiority against existing strategies.
We have further studied the design of distributed protocols, aiming at fairly and efficiently sharing radio resources among various links. We have proposed the first fully distributed spectrum access strategy achieving a Proportionally Fair (PF) allocation of radio channels to links. This allocation achieves a perfect trade-off between fairness among transmitters and efficiency (defined through the total throughput of the network). The design of the proposed protocol leverages game-theoretical concepts. We have finally devised optimal radio sharing protocols that account for the heterogeneity of the devices for example in terms of their maximum transmit power. This design is based on recent theoretical advances in learning Nash Equilibriums in games.
Importantly, the project has promoted the use of stochastic optimization and machine learning techniques to devise radio resource management schemes. These techniques allow us to devise schemes that "optimally" exploit the radio spectrum, and can be adapted to the level of feedback and information available at the transmitters. This is in sharp contrast with existing approaches that mainly use heuristics to design such schemes.
We have first investigated the problem of spectrum exploration. More precisely, when a radio device may exploit many radio channels but transmit on a few of them simultaneously, the objective is to determine when packet transmissions should be scheduled, which spectrum part and rate (depending on the chosen modulation and coding scheme) should be used for each packet transmission so as to maximize the throughput, i.e. the number of packets successfully transmitted per second. This task is complicated by fading phenomenon, i.e. by the fact that channel conditions vary over time in an unpredictable manner. We have formulated the problem of optimal spectrum exploration as a bandit optimization problem. Through a collection of novel results in the theory of bandit optimization, we have been able to devise sequential channel and rate selection strategies that optimally learn and track the best channel and rate. The latter have been implemented in a test-bed, thereby illustrating their superiority against existing strategies.
We have further studied the design of distributed protocols, aiming at fairly and efficiently sharing radio resources among various links. We have proposed the first fully distributed spectrum access strategy achieving a Proportionally Fair (PF) allocation of radio channels to links. This allocation achieves a perfect trade-off between fairness among transmitters and efficiency (defined through the total throughput of the network). The design of the proposed protocol leverages game-theoretical concepts. We have finally devised optimal radio sharing protocols that account for the heterogeneity of the devices for example in terms of their maximum transmit power. This design is based on recent theoretical advances in learning Nash Equilibriums in games.
Importantly, the project has promoted the use of stochastic optimization and machine learning techniques to devise radio resource management schemes. These techniques allow us to devise schemes that "optimally" exploit the radio spectrum, and can be adapted to the level of feedback and information available at the transmitters. This is in sharp contrast with existing approaches that mainly use heuristics to design such schemes.