Multi-Agent Design for the Physical Layer of a Distributed Base Station Network

The proposed proof-of-concept design described herein demonstrates that minimalist multiagent systems do provide all expected qualities: scalability, dynamic properties, efficiency, simplicity, adaptability and auto-configuration. Moreover, it represents a novel and surprisingly simple solution to resource allocation. It is most obvious with the power allocation scheme which results in drastically lower spatial power distributions when compared with traditional algorithms. The multi agent design approach, based on heuristics appears effective. It requires an understanding of the underlying mechanisms and compromises within the context of the problem at hand, from which insights and intuition can be drawn and used to design the agents. While it is unclear a priori to which end result the system will converge to, it should be noted that it is also unclear a priori to which it should. Indeed, the current context is far different from formal frameworks such as information theory which are often characterized by a single uni-dimension criterion, e.g. the channel capacity. In our multi-dimensional context, information theory remains too limited at the time to model and grasp the many possibilities and compromises facing a multitude of mobiles and DBS with macrodiversity where limited resources lead to interference. And in such a context, the proposed MA approach has the virtue of demonstrating via simulation that some novel allocation solutions (which should be understood as compromises) can lead to much higher efficiency of resource usage (where efficiency is necessarily also a notion of compromise). The design in itself is not so complicated, and one should keep in mind the fuzziness of such an approach. Indeed, the utility functions proposed could have a variety of alternatives. What matters is not their exactness, but that they provide certain properties that will sustain the interactions of agents. These properties remain to be understood and studied to provide insights on the inner workings of the agent system. For example, the shaping function in the utility function of the connection agents uses a logarithm which could be replaced by a first degree approximation (x - 1) and still converge, but a concave function with similar properties (e.g. ( x - 1)2 , null for x = 1 and strictly increasing for x > 1) would, despite providing some degree of convergence, fall short of a more balanced solution. The shaping function is therefore crucial to converge to certain Pareto solutions, and this remains to be studied in detail. Concerning MA design, we considered more specifically the notion of homeostasis which is not explicitly mentioned in Parunak's methodology. The proposed design shows how the search for such an equilibrium helps in designing and tuning the properties of the agents' behavior to obtain the desired global function. Considering future work, the proposed MA design and DBS architecture offers malleability and vast margins for tuning, enhancing, or providing additional functionality. It was studied in (Leroux et al., 2008) that the reuse of channels could be enhanced by pairing mobiles to cooperate in exploiting a single channel while multiplying their diversity gain. Interesting results have been found in this study. Yet, coupled with power-level control, the management of cooperation between mobiles revealed counter synergistic effects. To date, finding a way to have the cooperation and power-control agents interoperate in a synergetic manner remains an open problem. Another additional functionality to be studied is beamforming. Channel allocation agents would need to be improved to account for dynamically-created directive beams and provide network-wide gains by minimizing interference . Channelization also needs to be further studied, including a model to implement the IEEE 802.11 shared random access mechanism (CSMA/CA on conjunction with the so-called dis-

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.