"Enhancing Wireless Networks Performance through Learning-based Dynamic Channel Bonding and Dynamic Spectrum Access" by Sergio Barrachina
The number of hungry-bandwidth devices accessing the Internet through Wireless Local Area Networks (WLANs) Access Points (APs) such as laptops, smartphones, and tablets, is increasing drastically at the same time that users' bandwidth requirements do. Nonetheless, as WLANs operate in the industrial, scientific and medical (ISM) radio bands, even coexisting in dense scenarios (e.g. home departments), they are usually managed autonomously by different operators, which prevents applying centralized interference management techniques such as spectrum allocation.
By means of channel bonding (CB), a technique whereby nodes (i.e., devices and APs) are allowed to use contiguous sets of available channels for transmitting, higher throughputs are potentially achieved. However, due to the use of wider channels increases the contention among nodes, undesirable lower performances may be experienced in overlapping networks.
To mitigate such a negative effect, dynamic channel bonding (DCB) techniques are used to select the channels based on the instantaneous spectrum occupancy. Nonetheless, their dynamics and impact on networks performance is still not well known. Preliminary studies modeling WLAN scenarios as Continuous Time Markov Chains (CTMNs) reveal interesting particularities when implementing DCB, such as the fact that selecting always the maximum number of available channels may cause throughput losses in the long-term.
In this first stage of the PhD we aim to deeply characterize channel selection effects, and to propose and validate learning-based policies for enhancing WLANs performance through efficient DCB and dynamic spectrum allocation in order to make networks operate close to their optimal throughput even in highly variable scenarios.