1. Introduction
Low-power Lossy Network (LLN), such as Wireless Sensor Network (WSN), comprised of many low-cost and low-power nodes aims to support various applications, such as military surveillance, environment monitoring, and factory automation [
1]. At its early stage, it commonly uses single-channel protocols for simple operation on resource-constrained nodes [
2]. However, as the Industrial Scientific Medical (ISM) frequency band in which LLN generally stays is also occupied by many other wireless technologies (e.g., WiFi and Bluetooth), the performance of single-channel communication will be impaired inevitably. To improve the reliability of LLN, multichannel communication, especially the Time Slotted Channel Hopping (TSCH) technology, has received extensive attention [
3,
4,
5,
6,
7,
8], which has the potential to be a link-layer solution for LLN due to its resilience to wireless interference and multi-path fading.
The multihop and self-organization characteristics of LLN determine that its topology and link quality have great impacts on the network performance. To combat the inherent dynamic changes of low-power wireless links, agile and accurate link quality estimation (LQE) is essential. There have been many works in the literature which analyze the characteristics of low-power wireless links through experiments and propose many effective LQE metrics (e.g., Packet Reception Ratio (PRR), Signal to Noise Ratio (SNR), Received Signal Strength Indicator (RSSI) and Link Quality Indicator (LQI)) and models [
9]. However, almost all these metrics and models are established and verified under the single-channel scenario [
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20], and most of them are under the interference-free channel [
10,
11,
12,
13,
14,
15,
16,
18,
19]. The impacts of radio interference and channel change are not fully considered. Therefore, it is hard to directly judge whether these metrics and models are still valid under a multichannel scenario.
In this paper, the applicability of popular LQE metrics and models in many different channels is analyzed empirically. Results show that the interference level of the channel will affect the LQE capability of traditional metrics. LQI, which is thought to have the highest resolution in traditional LLN, is inferior to SNR and RSSI under interfered channels. In particular, its LQE capability has even been completely lost in channels with strong interference. On the other hand, traditional LQE models including theoretical and empirical models are basically not adaptive to radio interference and channel change. They are only valid for channels with similar interference under which they are modeled. Therefore, traditional LQE metrics and models cannot be directly used in the multichannel scenario. On the one hand, the thresholds of popular link quality metrics established in traditional LLN will lead to incorrect judgment of link quality under a multichannel scenario. On the other hand, the link quality will be overestimated inevitably if traditional link quality models are directly used under a multichannel scenario. It is necessary to study and design channel and interference adaptive LQE metrics and models.
The main contributions are as follows: (1) The applicability of popular LQE metrics and models in multiple different channels are analyzed empirically. (2) The problems of these metrics and models directly used under multichannel scenarios are pointed out. (3) The conclusions could provide new insights for multichannel protocol design. To the best knowledge of the authors, this is the first time to discuss the applicability of traditional link quality metrics and models under a multichannel scenario.
The rest of this paper is structured as follows. In
Section 2, the works related to LQE and multichannel communication are described. Popular link quality metrics and models are summarized and their mathematical descriptions are presented in
Section 3. This is followed by the experimental setup in
Section 4, including the experimental field, channel selection, and test methodology.
Section 5 analyzes the spatial characteristics of low-power wireless links under different channels and evaluates the applicability of main spatial models.
Section 6 analyzes the performance of popular physical layer metrics and models under different channels.
Section 7 evaluates the generality of the conclusions acquired in
Section 6 with a completely different experimental setup. Finally, conclusions are presented and suggestions are made for future works.
3. Popular Link Quality Metrics and Models
Popular link quality metrics were considered in this paper, including PRR and various physical layer metrics such as SNR, RSSI, and LQI [
9]. PRR is the most direct metric of link quality. However, the agility of using PRR for LQE is very poor. To estimate the link quality and determine the node locations before network deployment, the spatial distribution model of PRR with distance could be established based on the lognormal path loss model [
10], shown as follows:
where
Q(·) represents the Q function,
Pt is the transmit power (dBm),
Lc is the signal strength gain (or loss, if its value is negative) in the transceiver (dB),
n is the path loss exponent characterizing the attenuation rate of wireless signals,
d is the distance between transmitter and receiver (m),
d0 is the reference distance (usually 1m),
PL(
d0) is the free-space path loss (dBm) at
d0,
Xσ is a normally distributed random variable with zero mean and a standard deviation of
σ (dB),
Pn is the background noise power (dBm),
BN is the noise bandwidth of the transceiver (kHz),
R is the communication data rate (kb/s), and
Nbit is the number of bits in the packet. For the 2.4GHz physical layer of IEEE 802.15.4 standard typically used in traditional LLN, the values of
BN and
R are 384 kHz and 250 kb/s respectively.
The spatial distribution model of PRR is evaluated in this paper. This model is the basis of another important spatial model of low-power links, namely the model for estimating the bounds of the transitional region. If the former is not applicable under the multichannel scenario, the model for estimating the bounds of the transitional region will also fail inevitably. In the following sections, standard deviation σ of Xσ was calculated from the variances of RSSI at different distances, Pn used the measured background noise power, and Lc and n were obtained by using the least squares fitting method.
Using physical layer metrics could solve the agility problem of PRR, as these metrics could be measured per-packet level and are highly correlated with PRR. In practice, the estimated physical layer metrics need to be mapped to PRR to obtain a more granular characterization of link quality [
9]. Therefore, the popular models of physical layer metrics and PRR are also evaluated in this paper, which is formulated as follows. For the 2.4GHz physical layer of IEEE 802.15.4 standard typically used in traditional LLN, a theoretical model of SNR and PRR could be established [
11], as shown in Equation (2):
which is denoted as TH-SNR Model in the following. In addition, logical regression can be used to establish the fitting model of SNR and PRR [
12], as shown in Equation (3):
where
b1 and
b2 can be fitted using the experimental data. It is denoted as LR-SNR Model in the following.
Also using logistic regression, the mapping model of RSSI and PRR could be established [
13], as shown in Equation (4):
where
c1,
c2, and
c3 can be fitted using the experimental data. It is denoted as LR-RSSI Model in the following. For the 2.4GHz physical layer of IEEE 802.15.4 standard typically used in traditional LLN, a simplified theoretical model of RSSI and PRR could be established [
14], as shown in Equation (5):
which is denoted as TH-RSSI Model in the following. In addition, a polynomial mapping model between the normalized RSSI and PRR could be established [
17]. As the RSSI is normalized to the possible maximum RSSI value, its value ranges are fixed. Therefore, the expression given in [
17] was used directly:
where
Rnor is the normalized RSSI obtained using the median filter, and its value ranges from 0 to 0.5. As the normalized RSSI is used, this model is self-adaptive to some extent. It is denoted as PN-RSSI Model in the following.
In [
15], the mapping model between LQI and PRR was obtained based on the multi-segment linear model, as shown in Equation (7):
where
f1~
f9 can be fitted using the experimental data. It is denoted as ML-LQI Model in the following. In [
16], the mapping model between LQI and PRR was obtained using the cubic model, as shown in Equation (8):
where
e1~
e6 can be fitted using the experimental data. It is denoted as CU-LQI Model in the following. In [
12], the mapping model between LQI and PRR was obtained based on logistic regression, as shown in Equation (9):
where
g1 and
g2 can be fitted using the experimental data. It is denoted as LR-LQI Model in the following. It is worth pointing out that the models above have been derived and analyzed carefully when they were proposed. Therefore, only the final equations of these models are given here and the parameters of these models will be fitted using the measured data we acquired from the experiments.
7. Generality of The Conclusions
To verify the generality of the conclusions above, a completely different setup was designed. In this setup, an open playground with little WiFi interference was chosen. To emulate different interference scenarios, two unmanned aerial vehicles (UAV) operating in the 2.4 GHz and 5.8 GHz ISM bands respectively were used. Undoubtedly, the 5.8 GHz UAV should not interfere with the communication of TelosB nodes. However, as the 2.4 GHz UAV uses frequency hopping technology for communication, it will inevitably interfere with all the channels of TelosB nodes. The same methodology as
Section 4.3 was employed. In the process of measurement, the UAVs were always placed near the receiver. The measured data and corresponding models with the 5.8 GHz UAV are shown in
Figure 13, while those with the 2.4 GHz UAV are shown in
Figure 14. For clarity, only one model was plotted and compared with the measured data of channel 26 for each metric. This will not affect the conclusions.
Obviously, the main conclusions above are still valid. The link quality estimation capability of typical metrics will be affected by the interference level of the channel, which makes the conclusions obtained under the single-channel scenario no longer valid. The thresholds of popular link quality metrics established in traditional LLN will lead to incorrect judgment of link quality under the multichannel scenario. Meanwhile, LQI is inferior to SNR and RSSI under interfered channels, although it is thought to have the highest resolution in traditional LLN. In particular, its LQE capability has even been completely lost in channels with strong interference. On the other hand, traditional LQE models are basically not adaptive to radio interference and channel change. They are only valid for channels with similar interference under which they are modeled. If traditional link quality models are directly used under a multichannel scenario, the link quality will be overestimated inevitably.
8. Conclusions
Due to its resilience to wireless interference and multi-path fading, multichannel communication can effectively improve the reliability of LLN under cross-technology interference, which has received extensive attention from academia and industry. However, existing LQE metrics and models are generally established and verified under the single-channel scenario [
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20], and most of them are under the interference-free channel [
10,
11,
12,
13,
14,
15,
16,
18,
19]. In the multichannel scenario, as the channels used will change from time to time, whether these metrics and models are still applicable is hard to judge directly because of the lack of corresponding studies.
In view of this, this paper analyzes the applicability of popular LQE metrics and models in many different channels empirically. Results show that the interference level of the channel will affect the capability of traditional LQE metrics. In particular, the LQE capability of LQI will be completely lost in strong interference channels, although it is thought to have higher resolution than SNR and RSSI in traditional LLN. Traditional LQE models including theoretical and empirical models are basically not adaptive to radio interference and channel change. They are only effective for channels with similar interference levels. Although some models try to provide adaptability by introducing the environmental noise or using normalized physical layer metrics, they are unable to achieve the expected effect because this cannot accurately model the characteristics of interference.
Therefore, traditional LQE metrics and models cannot be directly used in the multichannel scenario. On the one hand, the thresholds of popular link quality metrics established in traditional LLN will lead to incorrect judgment of link quality under a multichannel scenario. On the other hand, the link quality will be overestimated inevitably if traditional link quality models are directly used under a multichannel scenario. In future works, it is necessary to deeply analyze the statistical characteristics of existing LQE metrics in multiple typical channels and design channel and interference adaptive LQE metrics and models to support effective multichannel communication in LLN.