Energy Benefits of Cooperative Docitive over Cognitive Networks Pol Blasco, Lorenza Giupponi, Ana Galindo-Serrano and Mischa Dohler CTTC Barcelona, Spain {name.surname}@cttc.es Abstract—Cognitive radios facilitate an autonomous and dis- tributed deployment for a variety of applications. A truly cognitive device, however, needs to exhibit a certain degree of in- telligence to draw optimum decisions based on prior observations and anticipated actions. Said intelligence comes along with a high complexity and poor convergence, which currently prevents any viable deployment of cognitive networks. However, the recently introduced and largely unexplored concept of docitive networks, where more expert nodes teach less expert nodes, allows both cognitive complexity as well as convergence to be significantly reduced. These achieved performance gains can be translated into energy gains, which will be focus of this paper. The baseline scenario is an ultra-high capacity network topology envisaged in beyond next-generation deployments offering capacity in the order of 1Gbps/km 2 . We show that, in dependency of the re- quired download volumes, a docitive deployment offers significant energy gains over a cognitive one. Typically, shorter download volumes yield larger gains. Index Terms—ignore I. I NTRODUCTION The ICT BuNGee project (www.ict-bungee.eu) aims at breaking the capacity barrier by designing a communication architecture which provides data rates of 1Gbps/km 2 . This is by approximately an order of magnitude larger than envisaged by next generation mobile networks. The reason, however, to provide these large rates is rooted in the applications envisaged to emerge over coming years and thus requiring said rates in urban environments with typical population densities. BuNGee thus relies on some key innovations, spanning from the physical layer to scheduling protocols as well as innovative deployment methodologies. A core architecture element is to have a high-capacity, elevated, multi-beam hub base station (HBS) serve several below-rooftop access base stations (ABSs); these, in turn serve associated mobile stations (MSs) using an omni-directional antenna. A further important element is that all involved wireless links use the very same frequency band. WiMAX naturally lends itself to such an architectural design since WiMAX was originally designed to be a backhaul network and only lately became suitable to serve as an access network too; furthermore, multihop is stipulated by the standard and appropriate frame structures have already been designed. Therefore, WiMAX is able to reflect BuNGee’s ultra-high capacity architecture. Having said this, a variety of technical challenges remain. Notably, it is (NP-)hard to control the scheduling patterns of all ABSs such that interference between them and the backhaul HBS is minimized/avoided. Such an approach would also require a very large overhead and would generally not be scalable. Distributed scheduling and control approaches are hence the only viable way forward, and cognitive protocols are an ideal facilitator at the cost of exhibiting poor convergence time and precision. In more details, cognition (from “cognoscere” = “to know” in Latin) is generally defined as “a process involved in gaining knowledge and comprehension, including thinking, knowing, remembering, judging, and problem solving”. Cognition thus heavily relies on a given level of intelligence, which is impacted by the degree of observation, the ability to learn, and the amount of memory. Learning in cognitive systems however exhibits poor convergence properties which does it not make suitable to operate in dynamic environments. To improve these characteristics, docition (from “docere” = “to teach” in Latin) had been introduced in [1], which encourages more expert nodes to share their knowledge in a cooperative fashion with other nodes in the network. This has been shown to significantly speed up the learning process and increase precision. This in turn minimizes the energy spent to deliver given information. The prime aim of this study is thus to quantify the energy gains of using a docitive system when compared to a purely cognitive one. To this end, the paper is structured as follows. In Section II, we briefly introduce the system model which is heavily based on [2] and only mentioned for reference purposes here. In Section III, we then briefly reflect on the cognitive and doc- itive algorithms which are the best known purely distributed approaches suitable to such an architecture. Thereupon, we quantify the energy gains of a docitive approach in Section IV. We finally conclude in Section V. II. SYSTEM MODEL The BuNGee approach considers an urban area with a fairly low number of HBSs. In the study under consideration, we will work with a single HBS that is providing service N ABSs. Each ABS provides service to its Q associated MSs. HBS and ABSs need to coexist in a meaningful manner as they essentially use the same spectral band(s). We consider orthogonal frequency division multiplexing (OFDM) symbols grouped into R sub-channels. Both HBS and ABS systems have the same amount R of available sub-channels, which allows to increase the spectral efficiency per area through