Data transmission plan adaptation complementing strategic time-network selection for connected vehicles

Abstract

Connected vehicles can nowadays be equipped with multiple network interfaces to access the Internet via a number of networks. To achieve an efficient transmission within this environment, a strategic time-network selection for connected vehicles has been developed, which plans ahead delay-tolerant transmissions. Under perfect prediction (knowledge) of the environment, the proposed strategic time-network selection approach is shown to outperform significantly leading state-of-the-art approaches which are based either on time selection or network selection only. Under realistic environments, however, the efficiency of planning-based approaches may be severely compromised since network presence and available capacities change rapidly and in an unforeseen manner (because of changing conditions due to the uncertainty in car movement, data transmission needs and network characteristics). To address this problem, a mechanism is proposed in this paper that determines the deviation from the anticipated conditions and modifies the transmission plan accordingly. Simulation results show that the proposed adaptation mechanisms help maintain the benefits of a strategic time-network selection planning under changing conditions.

Introduction

Nowadays, mobile nodes typically integrate different wireless network interfaces. An example environment of wireless networks is shown in Fig. 1, covering one mobile network (yellow) and three WiFi networks (blue, green, red) that are available for limited time spans during the trip. To improve connectivity performance, connected vehicles may use these networks in parallel to distribute their data traffic. Moreover, the connected vehicle use case provides an additional optimization potential, especially considering automated vehicles: Routes are usually known in advance and, thus, movement can be predicted accurately. As a result, a vehicle can predict future network availability and characteristics using the so-called connectivity maps, which use geographically mapped indicators to estimate the local transmission quality of available networks [1], [2]. An exemplified prediction of network availability over time is visualized in Fig. 1 using colored bars. Furthermore, according to Sandvine [3], a major part of a mobile node’s data traffic is delay-tolerant or heavy-tailed. Assuming networks and data traffic to be roughly known for a certain time horizon, we show in prior work [4] that a transmission planning can provide significant benefits. The transmission planning approach combines network selection [5] with a selection of the transmission time [6]. The approach plans ahead data transmission over multiple networks. In this paper, we present additional insights on the performance characteristics of this approach. However, the presented approach assumes perfect prediction of vehicle movement, network characteristics and data to transmit, as visualized in Fig. 2 left. Such accurate prediction might not always be available. In the real world, further mechanisms have to cope with prediction errors. Accordingly, we present three contributions in this paper:

• A strategic time-network selection approach that maximizes transmission efficiency using heterogeneous wireless networks due to transmission planning (Fig. 2 blue arrow “Plan”).

• An investigation of the effects of erroneous prediction on the performance of transmission plan execution (Fig. 2 red arrow “Real World”).

• A transmission plan adaptation that can mitigate a negative impact of erroneous prediction (Fig. 2 green arrow “Adapt & Execute”).

In Section 2, we briefly outline our previous work on an anticipative data transmission planning assuming static conditions and perfect knowledge of the environment and compare it to an Opportunistic Network Selection (ONS) (no planning or prediction). We show its performance characteristics in different scenarios and compare it also to other state-of-the-art-based approaches. Furthermore, we introduce our prediction error models and show that the performance of the strategic time-network selection approach degrades severely in the presence of prediction errors, due to its inability to react to changing conditions. In contrast, the opportunistic approach ONS – although underperforming with respect to the transmission plan approach under static conditions and perfect knowledge – appears to deliver a constant performance as it does not rely on prediction. Its insensitivity to prediction errors provides the motivation for our proposed transmission plan adaptation mechanism, presented in Section 3, which complements the transmission planning. The benefit of each planned transmission is re-evaluated and the planned transmission is modified by invoking a constrained ONS taking into account the type and magnitude of condition changes (i.e., car movement, data flows or network characteristics); mechanisms detecting relevant condition changes are also introduced. In Section 4, we discuss the performance of our novel adaptation approach under various changing conditions, followed by a related work discussion in Section 5. It turns out that under small to moderate changes in the environment, our responsive adaptation approach can largely sustain the gain foreseen from anticipative transmission planning with strategic time-network selection.