Frequency adaptation for interference mitigation in IEEE 802.15.4-based mobile body sensor networks

Abstract

The IEEE 802.15.4 standard is an interesting technology for use in mobile body sensor networks (MBSN), where entire networks of sensors are carried by humans. In many environments the sensor nodes experience external interference – for example, when the BSN is operated in the 2.4 GHz ISM band and the human moves in a densely populated city, it will likely experience WiFi interference, with a quickly changing “interference landscape”. In this paper we consider whether frequency adaptation, to be carried out by the BSN, provides performance gains in such an environment. We investigate a range of adaptation schemes and assess their performance both through simulations and experimentally. We furthermore consider one particular problem caused by frequency adaptation: the problem of orphaned devices. We provide simulation results suggesting that the problem indeed is noticeable, but hard to mitigate.

Introduction

Mobile body sensor networks (MBSNs) have recently been identified as a promising technology enabling a range of applications in health and well-being [1]. The IEEE 802.15.4 standard [2] is a mature and well-established standard for low-power wireless sensor networks and has also attracted a lot of interest in the MBSN arena. We expect that due to the availability of cheap and mature components, the IEEE 802.15.4 standard will remain a serious contender for BSN applications for quite some time to come, despite the recent approval of the IEEE 802.15.6 standard for wireless body area networks [3]. On the physical layer the IEEE 802.15.4 standard specifies three different frequency bands that can be used. One of them (and the one this paper focuses on) being the 2.4 GHz ISM band, which is shared with several other technologies, including Wi-Fi. The IEEE 802.15.4 standard subdivides the available spectrum in the 2.4 GHz band into 16 different channels, and an IEEE 802.15.4 network is supposed to pick one of those channels and to stay there.

A key feature of MBSNs is that they move as a whole, along with the movements of their human carrier. Having to share the spectrum with other technologies implies that data transmission in MBSNs can be harmed by external interference, as documented in several publications addressing co-existence between Wi-Fi and IEEE 802.15.4 [4], [5], [6]. At the same time, data transmission in MBSNs is often concerned with human vital signals, hence reliable and timely data transfer is of paramount importance. However, due to their mobility, MBSNs operated in densely populated urban areas are faced with a continuously changing “landscape” of Wi-Fi interference, as for example shown in the measurements presented in [7]. These measurements reveal that the interference landscape can change substantially at time scales of one minute and less, so that any choice of radio parameters that was good initially will become sub-optimal quickly.

Hence, it makes sense for a MBSN to adapt, i.e. to obtain measurements of the current interference situation and to set some of its operational parameters accordingly. Within the scope of the IEEE 802.15.4 standard there is a range of physical and MAC layer parameters that can be adjusted. The main candidates on the physical layer are the frequency channel and the transmit power, on the MAC layer one can pick between different operation modes (the beaconed and unbeaconed modes) and different access schemes (TDMA- versus CSMA-type schemes), one can vary the number of retransmissions, adjust backoff parameters of the CSMA MAC and so on. In this paper we concentrate on frequency adaptation and explore some of the issues and questions around it.

The relatively small time scale of interference changes in urban environments presents two main difficulties. First, the time available for channel measurements is small, which poses a risk of faulty (or noisy) measurements and subsequent poor decisions. Secondly, these measurements, which may involve switching to various channels and listening on each one to estimate the energy or traffic load present, are an energy-consuming process by itself, and it is not a priori clear whether the energy spent for those measurements will pay off (for example by reduced numbers of retransmissions). A further difficulty with frequency adaptation appears when a change of frequency is made in a situation with severe interference. Wherever the decision point within a BSN, any decision to switch to another channel must be communicated to the other nodes, and all nodes must agree on a suitable time to switch and channel to switch to. In the presence of interference it might happen that not all members of a BSN are informed of a channel change, and these nodes (referred to as “orphans” in the paper) spend substantial time and energy while searching all available channels to connect back.

This paper provides three main contributions: First we consider a range of different frequency adaptation schemes (including a non-adapting scheme and an idealized scheme called the genie scheme as baseline schemes) in a scenario which mimics the movement of pedestrians through densely populated urban areas with lots of WiFi interferers. We assess their performance in terms of packet loss rate and energy consumption through simulations. We keep the transmit power fixed to the smallest possible value offered by a popular IEEE 802.15.4-compliant transceiver (the ChipCon CC2420 [8]) in order to highlight the effects of frequency adaptation in isolation. Our results allow to answer the fundamental question of whether frequency adaptation can really give substantial gains over a non-adapting scheme in the affirmative. One of the considered schemes, referred to as the “lazy” scheme, appears to be particularly attractive, not only because it achieves excellent performance, but also because it lends itself to easy implementation on real devices. The general idea of the lazy scheme is simple: stay on the current channel as long as it is “good enough”, but once things turn bad we scan over all channels and pick the best one.

Secondly, we study the orphaning problem in more detail. Especially in the lazy scheme the times where channel switching occurs and the target channel cannot easily be predicted by nodes. Hence, an orphan does not have much information to begin with and potentially needs to spend a lot of time searching on all 16 available channels. We suggest a scheme to help orphans to shorten the search time by providing them with information that allows to come up with a correct guess of the target channel very often. Surprisingly, our results suggest that our schemes are able to significantly reduce the number of channels that an orphan scans before entering the right channel, but at the same time the average overall time spent in the orphan state is not reduced significantly in scenarios with high interference/interferer density, because the orphan often fails to receive beacons on the right channel as well. Because of this and because of our finding that in lower interference scenarios the orphan problem is not very prominent, we conclude that additional mechanisms to deal with orphans need not be included.

As our third main contribution we consider the implementation of the lazy and a blind frequency-hopping scheme in a real IEEE 802.15.4 stack [9] under the TinyOS operating system [10], and we provide experimental results showing the behavior of both schemes under real WiFi interference. Our results confirm the relative performance trends already observed in our simulations and substantiate our claim that particularly the lazy scheme achieves excellent performance. Furthermore, we briefly discuss our implementations. It turns out that both schemes can be implemented with minimal changes to the given IEEE 802.15.4 MAC protocol implementation, and the actual protocol itself is not modified.

This paper is an extended version of the conference papers [11], [12]. It is structured as follows: In Section 2 we provide the necessary background information on IEEE 802.15.4. Section 3 describes the system model used for our simulation-based performance analysis, and the considered frequency adaptation schemes are described in Section 4. The simulation-based performance evaluation results are presented in Section 5, and our experimental results are provided in Section 7. In Section 6 we investigate the issue of orphan recovery in more detail, again using simulations. Related work is discussed in Section 8 and we conclude the paper in Section 9.