Link dynamics accessment in low-power wireless networks

Introduction

Interference and link dynamics constitute great concerns for the stability and performance of wireless sensor network protocols. These phenomena normally manifest in link burstiness, i.e, prolonged periods of time where packet transmissions from a sender to a receiver are lost (Srinivasan et al., 2008). Such spikes of packet losses cause delays and instability in communication protocols with potentially severe consequences

Multichannel communication has been proposed as alternative to adaptive (single-channel) routing protocols for mitigating the impact of interference and link dynamics in wireless sensor networks. While several studies have advocated features of both techniques (not without running up against contradicting arguments) a comprehensive study that aligns these results is still lacking. This work aims at filling this gap.

Testbed

Testbed 

Our testbed deployment consisting of 32 nodes. Three WiFi access points per floor, microwave ovens, and people moving in and in-between offices create different types of interference.

Experiment Overview

TestbedTable 

Our analysis is based on three sets of experiments, leading to a total of more than 240 hours of experiments and 90 Million transmissions.

TestbedTable 

Multi Channel Experiments: Based on industry standards, we select hopping sequences with large inter-channel gaps.

RF characteristics for point-to-point communications

We compare single channel communication and channel hopping in four settings: (i) packet reception ratio (PRR), (ii) maximum burst loss, (iii) temporal correlation of losses (link burstiness), and (iv) frequency correlation of losses.

1. Packet reception ratio

TestbedPrr 

Average links PRR for links in the network over the channels c15, c20, c25 and c26, and for the hopping sequences S_1 = {c_{15}, c_{20}} and S_2 = {c_{15}, c_{20}, c_{25}, c_{26}}.

Observation: Our conclusion are aligned with previous results in (Ortiz and Culler, 2010, Sexton et al., 2005) (i.e. the link quality is not homogenous across channels). However, the sole links PRR analysis, as used in (Ortiz and Culler, 2010, Sexton et al., 2005), is not sufficient to characterize the statistical behavior of a packet losses of a multi-channel channel hopping MAC. To close this gap, we next analyze three alternative metrics.

2. Maximum burst loss

Packet losses are often correlated in time and occur in bursts.Therefore, analyzing only the (long term average) packet reception ratio is not sufficient, as it hides important performance indicator such as link burstiness. A simple metric to describe the link RF characteristics in terms of burstiness is the maximum burst loss, defined as the maximum number of consecutive packets lost over a communication link.

TestbedLinkBurst 

Cumulative distribution functions (CDFs) of the maximum burst length for single- and multi-channel communication for three relevant scenarios.

3. Link burstiness: temporal correlation

To quantify the correlation of packet losses in time for single-channel communication, Srinivasan et al. defined a “link burstiness” metric referred to as the beta-factor (Srinivasan et al., 2008). The metric is based on the conditional packet delivery function, C(n), which describes the conditional probability of successful packet reception given that the n previous packet were received (for n geq 0) or lost (for n < 0)

beta = frac{KW(I)-KW(E)}{KW(I)},

where KW(cdot) is the distance from the ideal bursty link, while E and I are the conditional packet delivery functions of the empirical and independent link with the same PRR, respectively.

A value of beta = 0 identifies a link with independent packet losses i.e. following a Bernoulli process, while a value of beta = 1 indicates a bimodal link, i.e. a link that exists either in a good or a bad state.

Link Beta 

Study of the packet loss characteristics with single-channel and multichannel point-to-point communication by the analysis of the beta-factor (time correlation).

Observation: we use beta to quantify the temporal correlation of packet losses for both singlechannel and multi-channel communication. To compute beta in the case of channel hopping, we consider the sequence of packets received between each transmitterreceiver pair. Thus, in this case beta describes the link-burstiness of a given transmitterreceiver pair across multiple channels and for a given channel hopping sequence. Both the length of the hopping sequence and the channel used will influence beta.

4. Correlation of losses: frequency correlation

We use the metric kappa-factor (Srinivasan et al., 2010), to describe the inter-link reception correlation in case of single-channel communication. This metric quantifies the spatial correlation of packet reception at different nodes receiving packets from the same source.

We adapt the theoretical framework of (Srinivasan et al., 2010) to characterize the inter-frequency reception correlation of a link in case of multi-channel communication.

TestbedLinkBurst 

Analysis of inter-frequency correlation through the CDF of the kappa-factor for length-2 TSCH sequences computed over all links in the network. We consider three cases: (a) adjacent 802.15.4 channels under different 802.11 channels; (b) distant-2 802.15.4 channels under different 802.11 channels; and (c) well separated 802.15.4 channels.

Adaptive routing and channel-hopping in multi-hop networks

In this section we evaluate the benefit of channel hopping in multi-hop single-path routing, and present two core findings:

  • We show that in dense and medium dense networks, channel hopping achieves similar end-to-end performance, such as average delay and reliability, as singlechannel when the channels have roughly the same quality.

  • In sparse topologies channel hopping can outperform single-channel communication.

1. Routing: channel-hopping in multi-hop networks

TestbedLinkBurst 

Observation: increasing channel robustness does not guarantee a better network performance compared to single channel solution. As long as there still exist many good links in the network, routing makes the use of channel hopping superfluous.

2. Multi-hop analysis in sparse networks

TestbedLinkBurst 

comparison between single-channel and TSCH in a sparse network. Channel hopping outperforms single-channel in both the average end-to-end delay, and reliability.