Adaptive IEEE802.15.4 TSCH in Sub-GHz Industrial Wireless Networks

(1)

Wireless Networks

Adaptive IEEE802.15.4 TSCH in Sub-GHz Industrial

Academic year 2019-2020 Technology

Master of Science in Electrical Engineering - main subject Communication and Information Master's dissertation submitted in order to obtain the academic degree of

Counsellors: Jan Bauwens, Robbe Elsas

Supervisors: Prof. dr. ir. Eli De Poorter, Prof. dr. ir. Jeroen Hoebeke

Student number: 01305850

(2)

Preface

First of all I would like to say that I completed this master’s dissertation with pleasure and that it was very interesting to work on a topic like this as wireless communication has always been of great interest to me. My previous master’s dissertation mainly focused on the physical layer, so it was a fun challenge to use this knowledge to try and implement a robust MAC protocol for a challenging environment. The fact that this topic is of industrial interest makes it even more attractive. I would therefore like to thank prof. Eli De Poorter for giving me the chance to write my master’s dissertation related to this topic next to the advise and feedback throughout the year. I would also like to thank Jan Bauwens for the close follow up, the insights and advise related to the implementation, the feedback regarding the thesis and the meetings to align my progress. Prof. Jeroen Hoebeke and Robbe Elsas also provided me with implementation feedback during the interim presentation, for which I am also grateful. Finally, a special thanks to my parents and girlfriend, who helped me with the line of sight measurements and provided me with support all through the process of writing my master’s dissertation.

(3)

Copyright

The author gives permission to make this master dissertation available for consultation and to copy parts of this master dissertation for personal use. In all cases of other use, the copyright terms have to be respected, in particular with regard to the obligation to state explicitly the source when quoting results from this master dissertation.

De auteur geeft de toelating deze masterproef voor consultatie beschikbaar te stellen en delen van de master-proef te kopi¨eren voor persoonlijk gebruik. Elk ander gebruik valt onder de bepalingen van het auteursrecht, in het bijzonder met betrekking tot de verplichting de bron uitdrukkelijk te vermelden bij het aanhalen van resultaten uit deze masterproef.

(4)

Adaptive IEEE802.15.4 TSCH in Sub-GHz

Industrial Wireless Networks

by

Dries Van Leemput

Master’s dissertation submitted in order to obtain the academic degree of

Master of Science in Electrical Engineering - main subject Communication and Information Technology

Academic year 2019–2020

Supervisors: Prof. dr. ir. E. De Poorter, Prof. dr. ir J. Hoebeke Counsellors: ing. J. Bauwens, ing. R. Elsas

Faculty of Engineering and Architecture Ghent University

Abstract

This master’s dissertation describes a robust, adaptive Medium Access Control (MAC) protocol to establish wireless voice communication in the context of the Internet of Ships (IoS). The implementation enhances the current IEEE802.15.4 Time Slotted Channel Hopping (TSCH) MAC mode with sub-GHz frequencies to yield better propagation characteristics and the ability to use multiple Physical layer (PHY)s depending on the link quality. This provides a trade-off between high datarate PHY configurations and low datarate but reliable configurations. The link quality is assessed using Link Quality Estimation (LQE) techniques able to distinguish between good or bad links resulting in an at runtime re-configuration. The implementation is tailored to the used development platform: the Zolertia RE-Mote with the Contiki-NG Operating System (OS). The implementation is verified with three different test setups, to evaluate the Line of Sight (LoS) characteristics of the used PHY configurations, to evaluate the re-configuration and LQE working principle and to compare the implementation to a standard sub-GHz TSCH network in terms of slot occupancy. Results show that an adaptive sub-GHz TSCH network provides a lower slot occupancy and higher throughput while still allowing robust communication.

Keywords

Multi-PHY, IEEE802.15.4 Time Slotted Channel Hopping, Wireless Sensor Networks, Sub-GHz, Internet of Ships

(5)

Adaptive IEEE802.15.4 TSCH in Sub-GHz

Industrial Wireless Networks

Dries Van Leemput

Supervisors: Prof. dr. ir. Eli De Poorter, ing. Jan Bauwens,

Prof. dr. ir. Jeroen Hoebeke, ing. Robbe Elsas

Abstract— This article describes enhancements made to the current

IEEE802.15.4 Time-Slotted Channel Hopping (TSCH) Medium Access Control (MAC) mode to establish wireless voice communication in the con-text of the Internet of Ships (IoS). The current standard is adapted to use sub-GHz frequencies in the Industrial, Scientific and Medical (ISM) 868 MHzband and is made adaptive by providing multiple Physical Layer (PHY)s. This provides a dynamic trade-off between throughput and latency and is dependent on the link quality, estimated by a Link Quality Estima-tion (LQE) process. The implementaEstima-tion is tailored to the used develop-ment platform: the Zolertia RE-Mote with the Contiki-NG Operating Sys-tem (OS). The implementation is verified with test setups and an adaptive sub-GHz TSCH network is compared to a non-adaptive TSCH network, showing a reduced slot occupancy, higher throughput while still allowing robust communication.

Keywords— multi-PHY, IEEE802.15.4 Time Slotted Channel Hopping, Wireless Sensor Networks, Sub-GHz, Internet of Ships

I. INTRODUCTION

I

N order to ensure safety of personnel and correct operation of machinery, it is crucial to provide communication in in-dustrial environments. One of those environments is a container ship at high seas, making it ever more crucial to communicate emergency messages. The use of cables is expensive and im-practical and requiring personnel to be stationary at the bridge to react in case of an emergency leads to an increased personnel cost and is not sufficient for the most outlying locations on the ship. Therefore, the use of wireless communication provides a solution although this brings problems of its own. A high density of equipment and metal structures acting as a Faraday shield limit the propagation range of wireless signals and give rise to multi-path propagation, reflections and fading. This re-quires the use of sub-GHz frequencies, yielding better propaga-tion behaviour [1] and less interference with existing technolo-gies [2] than 2.4 GHz and 5 GHz. Another requirement is to use a mesh Wireless Sensor Network (WSN) to cover the most remote locations on the ship. As some nodes are deployed on personnel, nodes will have to be mobile and the network needs to cope with those dynamic requirements. This leads to the need for a robust and adaptive MAC protocol. Moreover, the mobility of nodes results in a constraint in terms of energy consumption, as mobile nodes are battery powered.

This paper proposes an adaptive, robust, sub-GHz MAC pro-tocol that tries to cope with the aforementioned difficulties to enable voice communication between personnel. The imple-mentation is based on the already existing IEEE802.15.4 TSCH protocol designed for Low-Rate Wireless Personal Access Net-works (LR-WPANs) at 2.4 GHz. Two enhancements are made D. Van Leemput is a master student in the Department of Informa-tion Technology, Ghent University (UGent), Ghent, Belgium. E-mail: dries.vanleemput@ugent.be.

to the current protocol: IEEE802.15.4 TSCH is extended to use sub-GHz frequencies in the ISM 868 MHz frequency band and the ability to use multiple PHYs depending on the link character-istics is added. The latter provides a high throughput link in case of high link quality and is able to fall back to a more reliable link at runtime. As this provides links able to adapt to the current cir-cumstances, the implementation is calledadaptive TSCH. To estimate the link quality, a Received Signal Strength Indicator (RSSI)-based LQE technique is used to distinguish a link as ei-ther good or bad, leading to two possible PHY configurations.

The remainder of this paper is structured as follows: first, the protocol stack used in the implementation is examined in sec-tion II while highlighting the most affected protocol stack lay-ers. Next, section III describes the used development platform and OS before outlining the adaptive TSCH implementation in section IV. Three test setups are used in section V to evaluate the working principle of adaptive TSCH and to compare a simple sub-GHz adaptive TSCH network to a non-adaptive sub-GHz TSCH network in terms of slot occupancy. Section VI mentions some related research and the paper is concluded in section VII.

II. PROTOCOL STACK

Figure 1 depicts the protocol stack. The transport and appli-cation layer are left out as no adaptations are required at those layers. The PHY and MAC layer are both comprised in the IEEE802.15.4 standard [3]. At the PHY, three possible fre-quency bands are possible: the European 868 MHz band, the North American 915 MHz band and the 2.4 GHz band. Later, these bands were extended by the IEEE802.15.4c task group to 315 MHz, 432 MHz, 783 MHz and 953 MHz. Initial modula-tion schemes were based on Direct Sequence Spread Spectrum (DSSS), being Binary Phase-Shift Keying (BPSK) and Orthog-onal Quadrature Phase-Shift Keying (O-QPSK). Later, other modulation schemes such as Gaussian FSK (GFSK) were added to the standard. A Maximum Transmission Unit (MTU) of 127 bytes is defined to reduce network overhead and restrict packet error rates. The combination of PHY and MAC headers and footers comprises 15 to 31 bytes depending on the addressing mode. This results in a maximum payload length of 96 to 112 bytes. IEEE802.15.4 packets are acknowledged by Enhanced Acknowledgement (EACK)s and Information Element (IE)s can be appended to each packet or EACK to encapsulate necessary information such as drift compensation.

TSCH is one of the specific MAC behaviors of IEEE802.15.4, enrolled in the IEEE802.15.4-2015 standard [4] and referred to as IEEE802.15.4 TSCH mode. Frequency division and time di-vision multiplexing are combined to form a reliable, predictable

(6)

Fig. 1. Protocol stack up until network layer used in the implementation of adaptive TSCH. Transport and application layers are left out as not relevant for this research.

Fig. 2. Standard TSCH timeslot timing. Each transmission is preceded by an offset minus a guard time to account for clock drift.

and low duty cycle protocol with an increased network capac-ity compared to the original IEEE802.15.4 MAC protocol. To advertise the network, nodes send Enhance Beacon (EB)s com-prising slotframe, timeslot, synchronization and channel hop-ping information. After reception of an EB, nodes synchronize to the network and follow a periodically repeated slotframe con-sisting of a number of timeslots. Nodes communicate using a link, defined as the pairwise assignement of a directed commu-nication between nodes in a given timeslot on a given channel offset [4]. This offers an extra degree of freedom as in each timeslot different channel offsets can be used.

Figure 2 illustrates the standard timing within a TSCH times-lot. Before each transmission the transmitter waits for a time offset; TsTxOffset before the packet and TxACKOffset prior to the EACK. The receiver listens Guardtime before the transmis-sion to account for clock drift. At the end of the timeslot, both transmitter and receiver wait for Stoptime before the next times-lot starts. To calculate the channel, (1) is used, where ASN indicates the Absolute Slot Number, a shared variable by all nodes indicating the number of slots since the network creation, ChannelOf f seta predefined constant and N the number of channels.

f = G((ASN + channelOf f set)%N ) (1) As TSCH does not specify how to schedule links to nodes, an external scheduler must be used. In this project, Orches-tra, focusing on dynamic, low-power Internet Protocol version 6 (IPv6) and IPv6 Routing Protocol for Low-Power and Lossy Networks (RPL) networks is used where nodes autonomously

compute their own, local schedules based on their RPL parents and children [5]. Orchestra defines two example schedules, one of which is used in this project: Sender-Based Unicast Schedule (SBS). In SBS, three slotframes are defined for different pur-poses: the EB slotframe consists of 397 dedicated TSCH slots and is used to send EBs. A RPL broadcast slotframe is made up of 31 TSCH slots and can be used by all nodes in the net-work contending for channel access. The final unicast slotframe defines unicast slots between nodes. For each node, one shared TX link and a RX link for each parent and child is issued. As all slotframes cycle independently, the length of the unicast slot-frame must be mutually prime with the EB slotslot-frame length of 397 and the RPL broadcast slotframe length of 31. In case slots of different slotframes collide, the slot associated with the slot-frame of lowest priority is skipped in favour of the other slot. A SBS with a unicast slotframe of X slots is referred to as SBS-X. In order to accommodate IPv6 on top of TSCH, some adap-tation layers are needed. To ensure constrained devices can be used with IPv6, the IPv6 over Low-Power Wireless Personal Area Network (6LoWPAN) layer is employed and 6TiSCH Op-eration Sublayer (6top) enables distributed scheduling. The im-plementation of the adaptation layers falls out of the scope of this paper. When IPv6 is used on top of TSCH, it is referred to as 6TiSCH [6].

RPL is defined by the Internet Engineering Task Force (IETF) Routing Over Low-power and Lossy networks (ROLL) working group as a distance-vector routing protocol for low-power IPv6 networks [7]. Is uses a Destination Oriented Directed Acyclic Graph (DODAG) structure giving a rank to each node using a predefined cost function. Messages include DODAG Informa-tion Object (DIO) messages to distribute metrics, DODAG So-licitation (DIS) messages to request DODAG information and Destination Advertisement Object (DAO) messages to provide routes. For this project, RPL storing mode is used meaning that each node keeps a routing table to its children. This is in con-trast to non-storing mode, where all RPL traffic has to pass the root node.

III. DEVELOPMENT PLATFORM

To implement adaptive TSCH, the Zolertia RE-Mote de-velopment platform is used, combining the Texas Instruments CC2538 and CC1200. The CC2538 System on Chip (SoC) con-tains a built in 2.4 GHz IEEE802.15.4 compliant transceiver, an ARM Cortex M3 with 32 MHz clock speed, 512 kB of flash memory and 32 kB RAM. The CC1200 is a sub-GHz transceiver able to transmit in the ISM 868 MHz band with a maximum TX power of 14 dBm and supporting multiple modulation schemes and datarates [8]. The possible modulation schemes are On-Off Keying (OOK), 2/4-Frequency-Shift Keying (FSK) and 2/4-GFSK, but for the adaptive TSCH implementation, only 2/4-GFSK is used because of superior spectral efficiency. It should be noted that Orthogonal Frequency Division Multiplex-ing (OFDM) cannot be accomplished although specified in the IEEE802.15.4-2015 standard [4].

The OS used on the Zolertia RE-Mote is the open source Contiki-NG OS focused on resource constrained devices and Internet of Things (IoT). It has a ROM usage in the order of 100 kB, a RAM usage as low as 10 kB and runs in C.

(7)

Fig. 3. Adaptive TSCH steps: receiving a packet, performing LQE based on the received packet, feeding back the LQE decision and potential reconfigura-tion.

Throughput Modulation scheme RX filter BW

1.2 kbps 2-GFSK 13 kHz 19.2 kbps 2-GFSK 52 kHz 50 kbps 2-GFSK 104 kHz 100 kbps 2-GFSK 208 kHz 200 kbps 2-GFSK 416 kHz 500 kbps 2-GMSK 832 kHz 1 Mbps 4-GFSK 1667 kHz TABLE I

CC1200 PHYCONFIGURATIONS WITH THROUGHPUT,MODULATION SCHEME ANDRXFILTER BANDWIDTH.

IV. ADAPTIVETSCHIMPLEMENTATION

Adaptive TSCH is based on the standard TSCH implemen-tation with the extension of using multiple PHYs. In order to facilitate this, to each TSCH link and neighbor a default PHY is assigned and depending on the quality of the link, this PHY can be changed at runtime. This principle is illustrated in fig-ure 3. In the first step, node A sends a packet to node B using link 1 associated with certain PHY. Node B performs LQE in step 2 based on the received packet and decides if the link qual-ity is good enough (or not sufficient) to switch to another PHY. In case the decision was taken to switch PHY, node B feeds its decision back to node A. Next time link 1 is used, both nodes use the new PHY to communicate. An Orchestra schedule is used in combination with TSCH and depending on the network characteristics (latency, throughput, contention, energy etc.), the unicast slotframe length can be adjusted. The current adaptive TSCH implementation includes two PHYs: a default PHY for reliable communication and a high-datarate PHY for increased throughput. To ensure a robust network, the EB and RPL broad-cast slotframes always use the default PHY. Only links inside the unicast slotframe are allowed to re-configure and are able to use the high-datarate PHY.

A. PHY configurations

Table I lists seven possible CC1200 PHY configurations with corresponding datarate, modulation scheme and RX filter band-width. Datarates range from 1.2 kbps to 1 Mbps and all config-urations use 2-GFSK as modulation scheme except for 500 kbps and 1 Mbps. The RX filter bandwidth gives an indication of the bandwidth needed to send this configuration on a single chan-nel. The current adaptive TSCH implementation employs two of the available configurations: 50 kbps is defined as the default PHY and and 1 Mbps as the high-datarate PHY.

Figure 4 shows the allocation of these PHY configurations in

the ISM 868 MHz frequency band. The available 7 MHz are di-vided in sixteen 50 kbps 200 kHz channels and two 1 MHzps 1800 kHzchannels. This provides a guard band of 48 kHz at each side of the 50 kbps channels and 66.5 kHz for the the 1 Mbpschannels.

For the re-configuration, an extra radio driver function was made in Contiki-NG to re-configure the CC1200 at runtime. The CC1200 needs to be configured through the provisioned Serial Peripheral Interface (SPI), taking up time and energy and thus making it important to limit the re-configuration time since this is done inside a timeslot. The current implementation brings this time back to 0.6 ms.

B. Link Quality Estimation

To make a PHY configuration decision, node B in figure 3 needs to correctly assess the link quality of link 1. Multiple LQE techniques exist, all with a certain accuracy and complexity, de-pending on the desired use case. However, every LQE technique can be brought down to three steps: link monitoring, link mea-surement, metric evaluation and decision [9]. Those steps will be highlighted for the implemented LQE technique for adaptive TSCH.

B.1 Link monitoring

Before performing any form of estimation, the link has to be monitored by the transmitting or receiving node. According to figure 3, the receiver will monitor the link in adaptive TSCH. Link monitoring can be further classified in passive or active link monitoring. In passive link monitoring, links are monitored using existing traffic in contrast to active link monitoring, where dedicated packets are sent. To reduce overhead, passive link monitoring is employed.

B.2 Link measurement

Links must be measured with a certain representative metric, classified as either hardware metrics (RSSI, Link Quality Indica-tor (LQI) and Signal To Noise Ratio (SNR)) or software metrics (Required Number of Packets (RNP) or Packet Reception Rate (PRR)). Since RSSI values can be measured with an accuracy of 1 dB on the CC1200 and because RSSI is an adequate metric to distinguish a good and bad quality link [9], RSSI is chosen as the used hardware metric. Furthermore, the RSSI value is read anyway by the CC1200 to perform Clear Channel Assessment (CCA), limiting the overhead.

B.3 Metric evaluation

After measuring the metric, it is evaluated using a statistical procedure such as mean, standard deviation etc. In this project, the Exponential Weighted Moving Average (EWMA) filter is chosen to smoothen the RSSI input and to achieve a certain de-gree of stability. The operation of the EWMA filter is given in (2), where ˆP represents the EWMA estimate, µ the RSSI mea-surement and α a smoothing factor ranging from 0 to 1.

ˆ P = ( µ, t = 1 ˆ P∗ α − (1 − α) ∗ µ, t > 1 (2) The smoothing factor determines the weight of previous mea-surements on the RSSI estimate, a higher value indicating more

(8)

Fig. 4. Frequency channel allocation when using 50 kbps and 1 Mbps PHY configurations. weight to history. If the current PHY configuration is 50 kbps,

the estimator is smoothed with an α of 0.5, resulting in a slower convergence to the current RSSI measurement compared to 1 Mbps, where α yields 0.125. This is done because a rapid drop in RSSI must be detected quickly to be able to re-configure to the more reliable 50 kbps configuration.

B.4 Decision

Based on the RSSI estimate, a decision can be taken in terms of PHY configuration. This decision is based on the sensitiv-ity limits of the used PHY configurations, provided by Texas Instruments [8]. These limits yield −94 dBm for 50 kbps and −82 dBm for 1 Mbps assuming a link margin of around 15 dB. To provide stability, the thresholds for switching to a new PHY configuration are chosen according to a hysteresisloop. To switch from 1 Mbps to 50 kbps, the threshold is −77 dBm, is-suing an extra 5 dB with the sensitivity limit. To switch back to 1 Mbps, the threshold is chosen to be −72 dBm. As a result, a step of 5 dB is ensured between the thresholds.

C. Timing and throughput

In the standard TSCH timing, a single timeslot takes up 10 ms, but this is for 2.4 GHz frequencies. At sub-GHz, more specifically 868 MHz, datarates can be substantially lower. As shown in table I, the PHY configurations for the CC1200 have a datarate ranging from 1.2 kbps to 1 Mbps. Table II lists the timeslot timing of figure 2 applied to 50 kbps and 1 Mbps. The total timeslot time of 50 kbps yields 32 ms, whereas the 1 Mbps has a 6.88 ms timeslot. TX offset and TX ACK delay are ap-proximately equal for both PHYs, except for a small difference due to operations that rely on the datarate.

To not overcomplicate slot scheduling, all timesots are chosen equal to 32 ms. This allows for the transmission of multiple 1 Mbpspackets. When the re-configuration and slack times of table II are not considered, five 1 Mbps packets can be sent and acknowledged. The re-configuration is only accounted for once at the beginning of a timeslot and the slack times are chosen such that both timeslots take up exactly 32 ms.

The resulting net throughput for the PHY configurations is 28.25 kbpsfor 50 kbps and 141.25 kbps for 1 Mbps. At the moment, this does not suffice to cope with the strict timing con-straints for real-time voice communication (i.e. a maximum la-tency of 200 ms) when multiple hops are needed in the end-to-end path. However, with a smart unicast slotframe scheduling, the current latency can be reduced and adapted to the strict tim-ing constraints.

Datarate 50 kbps 1 Mbps

Re-configuration 0.6 ms 0.6 ms

TX offset 2.4 ms 1.62 ms

Sync and preamble 0.8 ms 0.08 ms

Max TX 20.48 ms 1.024 ms TX ACK delay 2.2 ms 1.84 ms Max ACK 3.36 ms 0.168 ms Slack 1.04 ms 1.172 ms Total 32 ms 6.88 ms TABLE II

TIMESLOT TIMING FOR50 kbpsAND1 Mbps.

D. Feedback

Step three in figure 3 depicts the feedback of the PHY deci-sion made by node B. In this project the feedback is chosen to be encapsulated in an EACK to restrict the delay between the de-cision and the actual re-configuration. Three until now unused bits in the Header IE appended to the EACK can be used to store the PHY configuration for the next time this link is used. The Header IE is comprised of 2 bytes storing clock drift compen-sation (12 bits) and NACK indication (1 bit), leaving 3 bits to store up to eight different PHY configurations.

In case of a packet loss, the decision will not reach node A and both nodes will start in a different PHY during the next times-lot. To resolve this problem, a fallback mechanism is installed. In case node A does not receive an EACK from node B for 7 consecutive tries, it assumes node B is in a different PHY and re-configures for the next timeslot. If this packet is still not ac-knowledged, node A keeps on switching PHY during the next timeslots until node B acknowledges.

V. EVALUATION

The proposed adaptive TSCH implementation is tested in the field using three different test setups. First, the practical range and propagation characteristics of both 50 kbps and 1 Mbps PHY configurations are measured to verify the sensitivity lim-its. Next, the LQE and re-configuration mechanism are evalu-ated between two Zolertia RE-Motes. Finally, a small adaptive TSCH network is established and compared to a non-adaptive TSCH networks in terms of slot occupancy.

A. LoS range and sensitivity limits

For the Line of Sight (LoS) measurements, two Zolertia RE-Motes were used with the RX node at a fixed position and the TX node at increasing distance. This is depicted in figure 5, the blue dot indicating the RX node and the yellow stars the TX

(9)

Fig. 5. LoS measurements setup: the blue dot indicates the RX node and the yellow stars the TX node positions.

node positions. For both 50 kbps and 1 Mbps 500 packets of 24 bytes were sent at 868 MHz, with 14 dBm TX power and at a height of 0.5 m. For every packet, the RSSI and sequence number was logged by the RX node to calculate the PRR.

Figure 6 shows the results fot 50 kbps (top) and 1 Mbps (bot-tom). For both PHY configurations, the sensitivity limit pro-vided by Texas Instruments with the extra link margin can be confirmed. As soon as the RSSI drops below the sensitivity limit, the PRR drops drastically. In the 50 kbps configuration, the PRR rises again to about 80-90 %, but this is not sufficient to achieve a reasonable reliability. It should be pointed out that the achieved LoS ranges (12 m for 1 Mbps and 65 m for 50 kbps) do not agree with the ranges specified by Texas Instruments (221 m for 1 Mbps and 575 m for 50 kbps using the same setup [10]). This was due to a malfunctioning antenna connector, resulting in a Mean Effective Gain (MEG) drop of about 30 dB. However, the main focus of this test setup was the verification of the sen-sitivity limits and those conclusions remain valid irrespective of the achieved LoS range.

B. LQE and re-configuration

To test the LQE and re-configuration mechanisms, two Zol-ertia RE-Motes were placed in close proximity to each other to yield a RSSI of about −25 dBm. Adaptive TSCH is imple-mented on the nodes and the resulting packets only exist out of RPL traffic. The RSSI measurements and EWMA estimates of one unicast link are measured and the accompanying PHY con-figuration is stored. The results are shown in figure 7. Squares indicate an EWMA estimate of a 50 kbps configuration, dots those of a 1 Mbps configuration and triangles raw RSSI mea-surements. The re-configuration thresholds of −72 dBm and −77 dBm are also shown.

The EWMA starts with a default value of −74 dBm, in the middle of the two re-configuration thresholds and since the first RSSI measurement is around −25 dBm, the EWMA output yields −50 dBm, resulting in a re-configuration to 1 Mbps. Af-ter 15 minutes, one antenna is disconnected from a RE-Mote to simulate a RSSI drop. Although the RSSI values swing around the −72 dBm threshold, no re-configuration takes place because of the hysteresisloop. 15 minutes later, the other antenna is dis-connected and the RSSI drops to −85 dBm. Because of the agile α for 1 Mbps, the drop is immediately detected. Another 15 minutes later, one antenna is connected again, resulting in an-other re-configuration. One can notice the less agile character of the 50 kbps EWMA, but because of the high RSSI increase, the

0 0.2 0.4 0.6 0.8 1 PRR PRR 0 50 100 150 −100 −80 −60 Distance [m] RSSI [dBm] PRR RSSI SL 0 0.2 0.4 0.6 0.8 1 PRR PRR 0 20 40 60 80 −120 −100 −80 −60 Distance [m] RSSI [dBm] PRR RSSI SL

Fig. 6. LoS measurements for the 50 kbps configuration (top) and 1 Mbps configuration (bottom) with the CC1200. A −127 dBm RSSI value corre-sponds to no received packets. Ranges are shorter than expected due to a malfunctioning antenna connector.

EWMA estimate finds itself just above the −72 dBm threshold. C. Sub-GHz adaptive TSCH vs non-adaptive TSCH

To compare the slot occupancy for an adaptive sub-GHz TSCH network to a non-adaptive TSCH network, a network of four RE-Motes was formed. In the first measurement, adaptive TSCH was flashed on the nodes and in the second one, non-adaptive TSCH with a 50 kbps PHY configuration. The nodes were placed in close proximity to each other resulting in the 1 MbpsPHY configuration for each link. After a period of 15 minutes, both networks were assumed to be stable and a burst transmission of 30 packets of 90 bytes was issued from one node to another. At the receiving node, all incoming and outgoing packets were logged with the corresponding Absolute Slot Num-ber (ASN). Figure 8 shows the incremental slot use at the receiv-ing node durreceiv-ing the reception of the 30 packets. For the 1 Mbps slot, the top of figure 8 shows slots to receive or transmit packets in blue and yellow squares indicate multiple packets received in-side the same timeslot. Due to multiple packets being received inside the same timeslot, a lower slot occupancy is measured when using the 1 Mbps PHY compared to the 50 kbps (bottom

(10)

0 15 30 45 60 −80 −60 −40 −20 Time [min] RSSI [dBm] RSSI EWMA 50 kbps EWMA 1 Mbps -72 dBm -77 dBm

Fig. 7. Reconfiguration test: RSSI values and the corresponding EWMA output are shown for each received packet. Green dots correspond to a 1 Mbps packet and yellow dots to a 50 kbps packet.

of figure 8). On could argue that a non-adaptive 1 Mbps would yield the same results as the adaptive network. This is true, but the benefit of using adaptive TSCH is that a remote node will still be able to cooperate in the network using the 50 kbps PHY configuration, which would not be the case for a non-adaptive 1 MbpsTSCH network. A purely 50 kbps non-adaptive TSCH network on the other hand lacks the slot occupancy and through-put advantages of adaptive TSCH networks.

VI. RELATED WORK

The idea of using different PHYs in TSCH has been explored previously. J. Munoz et al. [11] describe some design consid-erations that need to be taken into account when using multiple PHYs in 6TiSCH networks. Possible issues related to neigh-bor discovery and PHY capabilites, network formation, TSCH timeslot configuration, channel hopping sequence, resource al-location, packet size and objective function are evaluated with-out proposing any solutions as the internet draft is purely infor-mational. The mentioned difficulties range from the MAC layer up to the RPL routing layer. This paper provides some solu-tions, specifically for the MAC layer, for the highlighted design considerations in [11].

Recently, B. Martine et al. [12] proposed a sub-GHz TSCH implementation using multiple PHYs on the same development platform: using the CC1200 and Contiki-NG. Two implemen-tations for multiple PHYs are evaluated, one using a 1.2 kbps timeslot for every PHY configuration. The second implemen-tation uses a 1 Mbps timeslot as basis and multiple slots are concatenated for lower datarate PHYs. The first implementa-tion was used to test the differences of PHYs in terms of range and link symmetry. It turns out that PHYs with a datarate of 50 kbps and lower have much better range expectation and yield almost perfect symmetrical links. The second implemen-tation was employed to emphasize the advantage of using a low datarate 1.2 kbps PHY for EB advertising and network forma-tion and a high datarate 1 Mbps PHY for data transmission. The

2.95 3 3.05 3.1 ·104 0 5 10 15 ASN Slots 2.9 2.95 3 3.05 3.1 ·104 0 10 20 30 40 50 ASN Slots

Fig. 8. Incremental slot use during the reception of 30 burst packets using adaptive TSCH and 1 Mbps links (top) or non-adaptive 50 kbps TSCH. Blue dots indicate a transmitted or received packet inside a slot and yellow squares indicate extra packets received during the same slot.

1.2 kbpsEB band yields a 20x higher channel occupancy com-pared to the 1 Mbps data band with the same ammount of traffic. Conclusions concerning the PHY configurations are similar to the results shown in this paper, as TSCH timeslots of 29.38 ms and 5.704 ms for respectively 50 kbps and 1 Mbps were found compared to 31.4 ms and 5.88 ms excluding a re-confiuration time of 0.6 ms. While [12] mainly focuses on two (or multi-ple) separate PHYs, this paper explored the possibility of using LQE to adapt the PHY of a link at-runtime to the current link characteristics.

VII. CONCLUSION

The goal of this research was to developed a robust MAC pro-tocol able to provide container ship personnel the required level of safety through wireless voice communication. To achieve this, a robust and adaptive MAC protocol was designed to cope with the dynamic characteristics of the environment. The al-ready existing IEEE802.15.4 TSCH protocol was enhanced by

(11)

using sub-GHz frequencies and allowing multiple PHYs that change according to the link characteristics. 50 kbps is used for advertising and RPL traffic while the 1 Mbps data configu-ration is provided for high quality links. The LQE process uses the RSSI hardware metric to provide a stable and still agile es-timate able to distinguish good or bad links. The current timing does not suffice to send real-time voice communication, but a possible extension to this research could be to find an Orchestra unicast scheduling able to cope with the multi-hop latency.

The adaptive implementation was tested in three different test setups, aiming to measure the sensitivity limits for the PHY configurations, evaluating the re-configuration and LQE process and comparing a small sub-GHz adaptive TSCH network to a non-adaptive TSCH network in terms of slot occupancy. It turns out the adaptive TSCH network uses about 25 % of slots while yielding a higher throughput and is still being able to communi-cate to remote nodes.

REFERENCES

[1] R. Sandoval, A.-J. Garcia-Sanchez, F. Garcia-Sanchez, and J. Garcia-Har, Evaluating the More Suitable ISM Frequency Band for IoT-Based Smart Grids: A Quantitative Study of 915 MHz vs. 2400 MHz Sensors, vol. 17, p. 76, December 2016.

[2] M. Woehrle, M. Bor, and K. Langendoen, 868 MHz: A noiseless environ-ment, but no free lunch forprotocol design pp. 1–8, June 2012.

[3] IEEE Standard for Telecommunications and Information Exchange Be-tween Systems - LAN/MAN Specific Requirements - Part 15: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low Rate Wireless Personal Area Networks (WPAN) IEEE Std 802.15.4-2003, pp. 1–680, Oct 2003.

[4] IEEE Standard for Low-Rate Wireless Networks IEEE Std 802.15.4-2015 (Revision of IEEE Std802.15.4-2011), pp. 1–709, April 2016.

[5] S. Duquennoy, B. Al Nahas, O. Landsiedel, and T. Watteyne, Orchestra: Robust Mesh Networks Through Autonomously Scheduled TSCH November 2015

[6] X. Vilajosana, K. Pister, and T. Watteyne, Minimal IPv6 over the TSCH Mode of IEEE 802.15.4e (6TiSCH) Configuration BCP 210, May 2017. [7] T. Winter, P. Thubert, A. Brandt, J. Hui, R. Kelsey, P. Levis, K. Pister, R.

Struik, J. Vasseur, and R. Alexander, RPL: IPv6 Routing Protocol for Low-Power and Lossy Networks RFC 6550, March 2012.

[8] Texas Instruments, CC1200 Low-Power, High-Performance RF Transceiver July 2013. Revised October 2014.

[9] N. Baccour, A. Koubˆaa, L. Mottola, M. Zuniga, H. Youssef, C. Boano, and M. Alves, Radio Link Quality Estimation in Wireless Sensor Networks: a Survey ACM Transactions on Sensor Networks, vol. 8, no. 4, pp. 1–33, 2012.

[10] Texas Instruments, Achieving Optimum Radio Range March 2015. Re-vised September 2017.

[11] J. Munoz, X. Vilajosana, and T. Chang, Problem Statement for Gener-alizing 6TiSCH to Multiple PHYs Internet-Draft draft-munoz-6tisch-multi-phy-nodes-00, IETF, July 2018. Work in Progress.

[12] B. Martina, S. Duquennoy, N. Tsiftes, and T. Voigt, IEEE 802.15.4 TSCH in Sub-GHz: Design Considerations and Multi-band Support October 2019.

(12)

4.6.1 Timeslot timing . . . 36 4.6.2 Slotframe timing . . . 38 4.6.3 Throughput . . . 39 4.7 Feedback mechanisms . . . 40 4.7.1 Acknowledgement . . . 40 4.7.2 Dedicated packet . . . 41 4.7.3 Next packet . . . 41 4.7.4 Feedback link . . . 42 4.7.5 Fallback mechanism . . . 42 4.7.6 Discussion . . . 42 5 Test setup 44 5.1 LoS measurements . . . 44 5.2 Re-configuration . . . 47

5.3 Sub-GHz adaptive TSCH vs non-adaptive TSCH . . . 48

6 Future work 52 6.1 Extra PHY configurations . . . 52

6.2 Advanced Link Quality Estimation . . . 53

6.3 Timeslot timing . . . 53

6.4 Multiple unicast slotframes . . . 54

(14)

List of Figures

1.1 Example mesh network on a containership. . . 2

2.1 Protocol stack. . . 5

2.2 IEEE802.15.4 superframe. . . 5

2.3 IEEE802.15.4 packet structure. . . 6

2.4 Example TSCH slotframe and channel hopping structure. . . 7

2.5 TSCH timeslot. . . 8

2.6 Orchestra slots. . . 9

2.7 RPL modes. . . 11

2.8 Link Quality Estimation process. . . 13

2.9 Available frequency bands in the 863-870 MHz range [20]. . . 14

2.10 Existing technologies that make use of the 863-870 MHz frequency band [20]. . . 15

2.11 Number of hops reached for each node and for 6 different PHY configurations. . . 18

2.12 Channel utilization and TX count for 1.2 kbps and 1 Mbps [25]. . . 18

3.1 Hardware development platform: Zolertia RE-Mote revision A based on the Zoul module [26]. 20 3.2 CC1200 SPI timing. . . 22

3.3 Allocation of frequency channels in the 868 ISM frequency band. . . 25

3.4 Contiki-NG file structure. . . 25

4.1 Adaptive TSCH steps. . . 28

4.2 Example SBS (left) versus RBS (right) in adaptive TSCH. . . 28

4.3 Example schedule highlighting the problem of estimating PRR using sequence numbers. . . . 31

4.4 Effect of smoothing operator α on a step response and an unstable RSSI measurement. . . 33

4.5 Hysteresisloop with RSSI thresholds for stability. . . 35

4.6 Adaptive TSCH timeslot timing. . . 38

4.7 Orchestra slotframe. . . 39

(15)

LIST OF FIGURES XV

5.1 LoS test setup. . . 45

5.2 LoS measurements. . . 46

5.3 Re-configuration test results. . . 47

5.4 Setup for adaptive sub-GHz TSCH and sub-GHz TSCH comparison. . . 48

5.5 Incremental slot use during the reception of 30 burst packets using TSCH and 50 kbps links. . 49

5.6 Incremental slot use during the reception of 30 burst packets using adaptive TSCH and 1 Mbps links. . . 50

5.7 Incremental slot use during the reception of 30 burst packets using TSCH and 1 Mbps links. . 51

(16)

List of Acronyms

AFA Adaptive Frequency Agility . . . 15

AGC Automatic Gain Control . . . 20

ARM Acorn RISC Machine . . . 19

ASK Amplitude-Shift Keying . . . 23

ASN Absolute Slot Number . . . 8

BE Beacon Enabled . . . 5

BI Beacon Interval . . . 5

BPSK Binary Phase-Shift Keying . . . 6

CAP Contention Access Period . . . 5

CCA Clear Channel Assessment . . . 14

CEPT Conférence Européenne des administrations des Postes et Télécommunications . . . 14

CFP Contention Free Period . . . 5

CPU Central Processing Unit . . . 21

CRC Cyclic Redunancy Check . . . 24

DIO DODAG Information Object . . . 10

CS Common Shared . . . 9

CSMA Carrier Sense Multiple Access . . . 30

CSMA-CA Carrier Sense Multiple Access - Collision Avoidance . . . 5

CSS Chirp Spread Spectrum . . . 15

DAO Destination Advertisement Object . . . 10

DC Duty Cycle . . . 13

DIS DODAG Solicitation . . . 10

DODAG Destination Oriented Directed Acyclic Graph . . . 10

DSS Dual Sync Search . . . 24

(17)

LIST OF FIGURES XVII

EACK Enhanced ACKnowledgement . . . 6

EB Enhanced Beacon . . . 7

EC European Commission . . . 14

ERP Effective Radiated Power . . . 14

ETSI European Telecommunications Standards Institute . . . 14

EU European Union . . . 13

EWMA Exponentially Weighted Moving Average . . . 12

eWOR enhanced Wake-On Radio . . . 20

FastA Fast Association . . . 7

FCS Frame Check Sequence. . . .6

FIFO First-In First-Out . . . 20

FSK Frequency-Shift Keying . . . 22

GFSK Gaussian Frequency-Shift Keying . . . 6

GTS Guaranteed Time Slots . . . 5

GUI Graphical User Interface . . . 23

IE Information Element . . . 6

IETF Internet Engineering Task Force . . . 10

IoS Internet of Ships . . . IV IoT Internet of Things . . . 1

IPv6 Internet Protocol version 6. . . .4

ISI Inter-Symbol Interference . . . 23

ISM Industrial, Scientific and Medical . . . 2

KLE Kalman filter based Link Quality Estimator . . . 12

LBT Listen Before Talk . . . 15

LoS Line of Sight . . . IV LPWAN Low-Power Wide-Area Network . . . 15

LR-WPANs Low-Rate Wireless Personal Area Networks . . . 4

LSB Least Significant Bit . . . 40

LQE Link Quality Estimation. . . .IV LQI Link Quality Indicator . . . 12

MAC Medium Access Control . . . IV MEG Mean Effective Gain . . . 45

MFR MAC Footer . . . 6

(18)

LIST OF FIGURES XVIII

MSB Most Significant Bit . . . 20

MSK Minimum-Shift Keying . . . 22

MTU Maximum Transmission Unit . . . 5

NBE Non-Beacon Enabled . . . 5

OFDM Orthogonal Frequency Division Multiplexing . . . 23

OOK On-Off Keying. . . .22

OS Operating System . . . IV O-QPSK Orthogonal Quadrature Phase-Shift Keying . . . 6

PAN Personal Area Network . . . 4

PHY Physical layer . . . IV PLL Phase-Locked Loop . . . 21

PRR Packet Reception Rate. . . .11

RBS Receiver-Based Shared . . . 9

RPL IPv6 Routing Protocol for Low-Power and Lossy Networks . . . 4

SNR Signal-to-Noise Ratio . . . 13

RAM Random-Access Memory . . . 19

RERUM REliable, Resilient and secUre IoT for sMart city applications . . . 19

RFC Request For Comments . . . 24

RNP Required Number of Packet transmissions . . . 12

ROLL Routing Over Low-power and Lossy networks . . . 10

ROM Read-Only Memory . . . 24

SPI Serial Peripheral Interface . . . 21

RFID Radio-Frequency IDentification . . . 15

RSSI Received Signal Strength Indicator. . . .12

RTCC Real-Time Clock Calendar. . . .19

RX Reception . . . 9

SBD Sender-Based Dedicated . . . 9

SBS Sender-Based Shared . . . 9

SoC System on Chip . . . 19

SRD Short Range Device. . . .4

TDMA Time Division Multiple Access . . . 5

TSCH Time Slotted Channel Hopping . . . IV TX Transmission . . . 9

(19)

LIST OF FIGURES XIX

USA United States of America . . . 13

WDT WatchDog Timer . . . 19

WMEWMA Window Mean Exponentially Weighted Moving Average . . . 12

WSN Wireless Sensor Network . . . 4

6LoWPAN IPv6 over Low-Power Wireless Personal Area Networks . . . 10

6TiSCH IPv6 over the TSCH mode of IEEE 802.15.4e . . . 10

(20)

Chapter 1

Introduction

Communication in industrial environments is crucial for the operation, management, maintenance, monitor-ing, localization and perhaps most importantly security of the employees, machinery and industrial sites. The lack of acceptable communication quality or delay can result in system failure, inefficient operation, injuries or even fatalities. These consequences can become even more pronounced when the applicable environment is a container ship in high seas. However, communication in such settings is not always straightforward as installing cables would require an immense expense and would be highly impractical. Alternatively, one could use employees to be on stand-by at the bridge and interact in case of an emergency. Nevertheless, this demands high personnel costs and is not sufficient for the most outlying locations on site. Therefore, the preferred solution is to establish wireless communication between the personnel, machinery and the con-trol room. When this is instituted on a ship, this is referred to as the IoS (derived from the Internet of Things (IoT)) and will be the main application domain for this thesis. The ultimate goal will be to provide the ability to send short but essential voice messages from the personnel to the control room in the context of improving safety.

Wireless communication brings its own difficulties, especially when integrated in challenging locations con-taining lots of metal structures, containers and machines. After all, conductive metal structures act as Faraday shields and because of the high density of equipment, multi-path propagation, reflections and fading impose a problem. As a result, the use of sub-GHz frequencies can be beneficial since in contrast to ubiquitous frequencies such as 2.4 GHz or 5 GHz, they yield better propagation behaviour [1], less interference with ex-isting technologies [2] and thereby a higher reliability. To cover all remote locations, meshing networks using multi-hop protocols offer a solution creating redundant routes between nodes without the need for a central-ized infrastructure. A downside to this method is a decreased throughput and added complexity. Another

(21)

2

Figure 1.1: Example of a mesh network on a containership: nodes connect to other nodes to form a path to contact the bridge in case of an emergency.

aspect is mobility: some nodes taking part in the meshing network ought to be mobile since often deployed on personnel. This leads to even more challenges in the meshing network organization as the network must be self-organizing with an extremely dynamic network topology. On top of that, some mobile nodes have to be battery powered imposing constraints on the energy consumption to yield a reasonable lifetime.

On that account, the network needs robust PHY, MAC and routing protocols able to adapt to varying and challenging conditions. This thesis will thereby focus on the implementation of such a MAC protocol in combination with a variable PHY. The solution builds on the already existing IEEE802.15.4 TSCH MAC mode for low-power, dynamic, multi-hop networks and provides two additions to the protocol. Since IEEE802.15.4 TSCH is designed for use in the 2.4 GHz Industrial, Scientific and Medical (ISM) frequency band, the first addition will be to extend this protocol to sub-GHz frequencies. Because the desired use case is to send crucial voice messages and since this will be employed in a dynamic environment, the second addition is to make IEEE802.15.4 TSCH adaptive. This means that underneath the MAC protocol, multiple PHY configurations can be used and changed at runtime depending on the current environment. This produces a trade-off between a high throughput to send voice messages with an acceptable latency and reliability to ensure messages reach their destination. To determine which PHY configuration is to be used, LQE is performed to assess the wireless link quality and adapt the protocol accordingly. These additions together with the current IEEE802.15.4 TSCH implementation will be refered to as adaptive TSCH in the remainder of this thesis.

To start, chapter 2 will discuss the related work and the technologies used or referred to in the subsequent chapters. It details the protocol stack, the LQE process, sub-GHz regulations and some related work tackling the same concepts in TSCH. After that, the development platform to implement adaptive TSCH is considered in chapter 3 together with the OS used on this platform. Chapter 4 describes the concept of adaptive TSCH

(22)

3

and highlights the different steps taken in the process. It examines the available LQE techniques for the used hardware and evaluates the timing, throughput and possibility to send voice messages using adaptive TSCH. Next, some tests are performed on the implementation and described in chapter 5 including LoS measurements of different PHY configurations, validation of the re-configuration mechanism and comparison of an adaptive TSCH network and a stardard TSCH network in terms of throughput and latency. Chapter 6 reflects on possible improvements for adaptive TSCH and future work before concluding in chapter 7.

(23)

Chapter 2

Related work

This chapter summons the different technologies and techniques used in this thesis and provides an overview of related research and work. First, the used protocol stack is considered, with an in-depth look at the PHY and MAC layer, where the actual enhancements take place. The higher layer protocols are briefly reviewed as only some concepts will be used. Next, some useful LQE techniques, more specifically in Wireless Sensor Network (WSN)s, are summarized to gain insight in the numerous methods which are applicable for this research. After that it is important to look into the allowed sub-GHz frequency bands for WSNs and Short Range Device (SRD)s, the corresponding regulations and some already existing technologies that use those frequencies. Finally, related work concerning using multiple PHYs in sub-GHz TSCH networks is evaluated.

2.1 Protocol stack

Figure 2.1 depicts the protocol stack. First, the PHY and the MAC layer will be covered in detail. The adaptation layer is mentioned for completeness as the details of this layer are not useful here. The IPv6 Routing Protocol for Low-Power and Lossy Networks (RPL) protocol at the network layer will be briefly discussed as some concepts and terms are mentioned. The Internet Protocol version 6 (IPv6) protocol and higher layer protocols fall out of the scope of this thesis.

2.1.1 IEEE802.15.4

The original IEEE802.15.4 standard [3] defines a PHY and MAC layer for Low-Rate Wireless Personal Area Networks (LR-WPANs). The Personal Area Network (PAN) is formed by a PAN coordinator who manages the network. Depending on the network topology, regular nodes communicate only with the coordinator

(24)

2.1. PROTOCOL STACK 5

Figure 2.1: Used protocol stack. Transport and application layers are left out since they are not relevant for this research.

Figure 2.2: IEEE802.15.4 superframe: beacons separate successive slotframes, followed by an active and inactive period.

(star topology, single hop) or are able to interact directly with each other when in range (cluster-tree or mesh topology, multi-hop). There are two possible channel access methods defined: Beacon Enabled (BE) mode and Non-Beacon Enabled (NBE) mode. The former is based on a superframe illustrated in figure 2.2. It consists of an active period of 16 slots during which nodes can exchange packets. Those slots can either be Contention Access Period (CAP), meaning a slotted Carrier Sense Multiple Access - Collision Avoidance (CSMA-CA) mechanism is used, or Contention Free Period (CFP), corresponding to a Time Division Multiple Access (TDMA) scheme with Guaranteed Time Slots (GTS) to individual nodes. Succeeding the active period is an inactive period where nodes are switched to a low power state to save energy. Each superframe begins with a beacon transmitted by the coordinator and is periodically sent after each Beacon Interval (BI). In NBE mode there is no superframe and all nodes content with each other conform an unslotted CSMA-CA mechanism.

The packet structure of a IEEE802.15.4 frame is depicted in figure 2.3. As can be seen, the Maximum Transmission Unit (MTU) is limited to 127 bytes to reduce network overhead and restrict packet error rates.

(25)

Figure 2.3: IEEE802.15.4 packet structure: both PHY and MAC headers and footers are shown. A MTU of 127 bytes limits the payload to 92-112 bytes.

The MAC frame is preceded by 6 bytes consisting of a sync header and a PHY header. The MAC frame itself is categorized in three parts starting with the MAC Header (MHR) which includes 2 bytes of frame control, 1 byte sequence number and 4 to 20 bytes of addressing fields, depending on the used addressing mode (network and device identifier for a star topology and source and destination identifier for a mesh topology). The MHR is followed by the MAC payload and the MAC Footer (MFR) consisting of a Frame Check Sequence (FCS) of 2 bytes. One notices that the maximum MAC payload is limited to 112 bytes for a mesh topology and 96 bytes for a star topology.

IEEE802.15.4 packets can be acknowledged by the receiver using Enhanced ACKnowledgement (EACK)s. The packet structure is similar to that of figure 2.3 without the addressing fields, as the receiver already knows the original transmitter. [4] describes the use of Information Element (IE)s, which can be appended to packets or acknowledgements to encapsulate information. Two types of IE are defined: Header IEs, part of the MAC header and Payload IEs encapsulated in the payload and therefore able to be encrypted and authenticated.

Three possible frequency bands are possible: the European 868 MHz band with one channel, the North Ameri-can 915 MHz band with 10 channels and the 2.4 GHz band with 16 channels. Later, these bands were extended by the IEEE802.15.4c task group to 315 MHz, 432 MHz, 783 MHz and 953 MHz. Initial modulation schemes were based on Direct Sequence Spread Spectrum (DSSS), being Binary Phase-Shift Keying (BPSK) and Orthogonal Quadrature Phase-Shift Keying (O-QPSK). Later, other modulation schemes such as Gaussian Frequency-Shift Keying (GFSK) were added to the standard.

(26)

Figure 2.4: Example TSCH slotframe and channel hopping structure: to each link a timeslot and channel offset is assigned. A separation between free, dedicated and shared links is made.

2.1.2 Time Slotted Channel Hopping

In 2012, the IEEE802.15.4e MAC Enhancement Document [5] was approved in which a new multi-hop MAC protocol was developed that can be used in industrial applications. Next to some general enhancements such as Enhanced Beacon (EB), Fast Association (FastA), IE etc, IEEE802.15.4e describes five specific MAC behaviors including TSCH, which was later enrolled in the IEEE802.15.4-2015 standard [6] and is referred to as IEEE802.15.4 TSCH mode after this merge. In TSCH, channel hopping, TDMA and CSMA-CA are combined to form a reliable, predictable and low duty cycle protocol but still with an increased network capacity compared to the previous MAC protocol. To advertise the network, a node sends EBs which contain slotframe, timeslot, synchronization and channel hopping information. Upon reception of such an EB, a node synchronizes to the network and is able to send EBs of his own. All nodes synchronized to the network follow a periodically repeated slotframe. Unlike to the original standard, slotframes are not separated by beacons to facilitate synchronization but instead a shared notion of time is employed. Nodes can communicate with each other in one of the timeslots being part of the slotframe, following a TDMA approach. In addition to this, a specific channel offset is appointed to the link implementing channel hopping. Therefore a link in IEEE802.15.4e is defined as the pairwise assignment of a directed communication between nodes in a given timeslot on a given channel offset [5]. This is illustrated in figure 2.4, where a slotframe is divided into 4 timeslots and 4 possible channel offsets are chosen.

Inside a timeslot, packet transmission begins after a TsTxOffset minus a guard time. However, because desynchronization between nodes has to be accounted for (clock drift), the receiver already starts listening at TsTxOffset - Guardtime after the start of the timeslot. The same principle is being used for the acknowledge-ment, with a TsACKOffset and Guardtime. Because the relative desynchronization between nodes cannot be

(27)

Figure 2.5: TSCH timeslot: an offset of TsTxOffset and TxACKOffset is provided before the transmission of data/EACK and a guardtime is installed before the reception to account for clock drift.

greater than Guardtime, nodes have to be able to resynchronize and have a dedicated time-source neighbor. Two types of synchronization are possible: frame-based synchronization, where the receiver denotes the start of the frame transmission and ACK-based synchronization where the start of the acknowledgement is used. When no acknowledgement is received, transmission of the packet starts again in the next slot dedicated to the link. The timing of a timeslot is depicted in figure 2.5.

By using different channels, the effect of multi-path fading and external interference is reduced. Each time nodes communicate a different frequency is used. By default 16 channels are possible to communicate, so each link has a channel offset in between 0 and 15 assigned to it. To calculate the frequency for a certain link, (2.1) is employed, where N denotes the number of channels and channelOf f set is a predefined constant. The Absolute Slot Number (ASN) counts all slots since the starting point of the network creation and is thus shared by all nodes. The function G assures that a link will use all channels in time to take full advantage of the multiple frequencies.

f = G((ASN + channelOf f set)%N ) (2.1)

Figure 2.4 also shows different types of link. A link can either be free, dedicated or shared. When a link is shared, a CSMA-CA mechanism is employed to reduce the chance of a collision. For a full explanation of the used CSMA-CA algorithm, the reader is referred to [5].

2.1.3 Orchestra

To assign links to nodes without interference, a proper link schedule is needed. However, the IEEE802.15.4e standard does not specify a link scheduler and therefore a large degree of freedom is introduced in scheduling the links. For this thesis, a scheduler named Orchestra is chosen, where nodes autonomously compute their

(28)

(a)

Topol-ogy (b) CS slot (c) RBS slot (d) SB(D)S slot

Figure 2.6: Three kinds of Orchestra slots for an example topology of four nodes: CS slot, RBS slot and SB(D)S slot.

own, local schedules based on their RPL (see section 2.1.5) parents and children [7]. The scheduler focuses on dynamic, low-power IPv6 and RPL networks. Since Orchestra does not rely on any centralized scheduling individual and has no need for multi-hop path reservation or signaling, it is well suited for dynamic networks without a predefined traffic pattern.

Orchestra slots are not always equal to TSCH slots, but can comprise multiple TSCH slots. The slot lengths are mutually prime so they overlap one another evenly and use a hash function based on the MAC address of the node to link TSCH slots to an Orchestra slot. Furthermore, four types of slots are defined: Common Shared (CS) slots, Receiver-Based Shared (RBS) slots, Sender-Based Shared (SBS) slots and Sender-Based Dedicated (SBD) slots. In figure 2.6 these slot types are shown based on a example topology in 2.6a. The CS slot (2.6b) is a Reception (RX) and Transmission (TX) slot and can be used by all nodes in the network. A RBS slot (2.6c) is used for neighbor-to-neighbor communication and the RX slot is calculated based on the receiving nodes MAC address. Therefore, a RBS slot consists of one RX slot and as many TX slots, based on the neighbors MAC address, for every neighbor. A SBS slot (2.6d) is similar to a RBS slot except that the TX slot is calculated based on the transmitting nodes MAC address, resulting in one RX slot per neighbor and one TX slot. Finally, the SBD slot is equal to a SBS but the TSCH slots are all dedicated. This means that the slotframe must be large enough in order to schedule a TX slot to every neighbor. Since the slot is dedicated, no CSMA-CA is needed.

Orchestra also defines two example schedules: Receiver-based Unicast Schedule and Sender-based Unicast Schedule. Both schedules use three types of slotframes: a Broadcoast slotframe with one CS slot for RPL broadcast messages, an EB slotframe used for sending EBs which consists of one SBD slot so therefore longer than the number of nodes in the network and a Sender(/Receiver)-based Unicast Slotframe depening on the type of schedule. This consists of a SBS or RBS slot and is for unicast communication beween the children and preferred parent of the node.

(29)

Each slotframe is repeated with a certain period, determined by the slotframe length. In order for the slotframes to cycle independently, the slotframe lengths must be mutually prime. Choosing the length of a slotframe requires a trade-off between latency, energy, network capacity and contention. When short slot-frames are used, the slots repeat more often resulting in higher traffic capacity and lower network latency. However, because nodes have to wake up to listen or transmit more often, this also increases energy con-sumption. In case slots of different slotframes overlap, the slot with the highest priority (lowest slotframe handle) is executed. The probability of one slot colliding with another slot from a different slotframe is given by (2.2), where Blenindicates the slotframe length and Bslotsthe number of slots in slotframe B.

pcollB =

1 Blen/Bslots

(2.2) The probability of a slot being skipped over a slot with a smaller slotframe handle, is therefore given by (2.3), with SF the set of all slotframes in the system and Ah, Bh the slotframe handles of A and B

respec-tively. pskipA= 1−   Y ∀B∈SF,Bh<Ah 1_{− p}collA,B   (2.3)

2.1.4 Adaptation layers

When IPv6 has to run on top of TSCH, TSCH is referred to as IPv6 over the TSCH mode of IEEE 802.15.4e (6TiSCH) [8]. This means that a proper interface is created between TSCH and higher protocol layers. It implements the IPv6 over Low-Power Wireless Personal Area Networks (6LoWPAN) adaptation layer [9] that ensures that constrained devices can be used with IPv6. The 6TiSCH Operation Sublayer (6top) [10] enables distributed scheduling in 6TiSCH networks.

2.1.5 RPL

RPL [11] is defined by the Internet Engineering Task Force (IETF) Routing Over Low-power and Lossy networks (ROLL) working group as a distance-vector routing protocol for low-power IPv6 networks. It has a Destination Oriented Directed Acyclic Graph (DODAG) structure with an increasing rank for nodes starting from the root based on the distance to that root through some cost function. There are three types of messages, both multicast and unicast: DODAG Information Object (DIO) messages to distribute metrics, DODAG Solicitation (DIS) messages to request DODAG information and Destination Advertisement Object (DAO) to provide routes. Two modes of operation are defined: non-storing mode and storing mode. In non-storing mode (figure 2.7a), nodes do not keep a routing table to neighbors with a higher rank (children).

(30)

2.2. LINK QUALITY ESTIMATION 11

(a) RPL

non-storing mode

(b) RPL storing

mode

Figure 2.7: RPL modes: storing mode vs non-storing mode.

Therefore, a packet to another node has to go first uplink to the root and then downlink to the destination. In storing mode (figure 2.7b), nodes do maintain a routing table to its children and the packet first goes to a node with smaller rank (parent) shared by origin and destination and is then routed towards the destination.

2.2 Link Quality Estimation

2.2.1 Characteristics of Wireless Sensor Networks

In order to estimate the quality of a radio link it is important to understand what affects the quality of this link and how the characteristics can be classified. It is widely known that effects such as external and internal interference, multi-path propagation, non-ideal radiation patterns, fading and scattering make radio links unreliable and highly variable. Therefore, the radio transmission pattern is far from circular, but exhibits directionality resulting in better propagation in certain directions. Another aspect that has to be taken into account is link asymmetry. When communication from node A to node B is classified as good, the opposite is not necessarily true. In fact, approximately 5-15% of all links are asymmetric [12]. When the Packet Reception Rate (PRR) is related to the distance between the transmitter and receiver, three regions can be classified: a connected region with good reception (90% average PRR), a gray region with highly variable reception rate (between 10% and 90% average PRR) and a non-connected region with very poor reception. Moreover, the PRR is not only space dependent, but also time dependent. In the gray area, nodes experience not only a high degree in spatial variation, but also a time varying packet loss [13].

(31)

2.2. LINK QUALITY ESTIMATION 12

2.2.2 Link Quality Estimation process

The process of LQE can be reduced to three steps: link monitoring, link measurement and metric evaluation [14] seen in figure 2.8. In the first step, a certain metric is evaluated during a window w or based on w items. One can distinguish between active link monitoring, where links are evaluated using probe packets, passive link monitoring using existing traffic to monitor the link and hybrid monitoring, applying a combination of the two aforementioned techniques. It is intuitive to see that passive link monitoring produces less overhead than hybrid monitoring, followed by active monitoring.

After a link is being monitored, actual measurements can take place to quantify the quality of the link. These measurements can be done in both hardware and software. The two hardware metrics available in IEEE802.15.4 are Received Signal Strength Indicator (RSSI) and Link Quality Indicator (LQI). However, it turns out that these hardware metrics can be deceptive when used for LQE in WSNs: both RSSI and LQI can be very misleading under interference and RSSI only indicates interference if a certain threshold is reached, LQI is based on estimated instead of corrected symbols and furthermore the period over which the RSSI is calculated might not be large enough [15], although the used CC2420 has a different RSSI calculation than the CC1200 used in this thesis (see chapter 3). Nevertheless, RSSI and LQI can be used to indicate high or low link quality, but are poor indicators for the gray area. Another possibility is to use software metrics to estimate the link quality. Depending on where the LQE occurs, software metrics estimate either the PRR (receiver side) or Required Number of Packet transmissions (RNP) (transmitter side).

The third and final step in the LQE process is the metric evaluation. Here, the measured metric is used to compute a link quality estimate. Statistical procedures such as average, standard deviation, Exponentially Weighted Moving Average (EWMA) or a combination are possible in case of hardware metrics. When software metrics are considered, more advanced techniques can be used. Since in this thesis the estimation will take place at the receiver side, only those techniques will be considered. A first approach is to use Window Mean Exponentially Weighted Moving Average (WMEWMA). Here, the receiver expects packets at a certain rate R and each packet has a sequence number. The event T of a packet reception is evaluated and the window mean µ = r

r+f is calculated over a certain window w with r a packet reception and f a packet miss. The

actual estimate ˆP is calculated with an EWMA filter shown in (2.4) with α a tuning parameter.

ˆ P =      µ, t = 1 ˆ P_{∗ α + (1 − α) ∗ µ, t > 1} (2.4)

The WMEWMA outperforms other filter-based estimators in terms of efficiencey, stability, reactivity and accuracy [16]. A faster approach is to use a single measurement for LQE, such as the Kalman filter based Link

(32)

2.3. SUB-GHZ FOR SHORT RANGE DEVICES 13

Figure 2.8: Link Quality Estimation process: link monitoring, link measurements and metric evaluation applied on hardware and software metrics.

Quality Estimator (KLE). Because a single measurement is used, a pre-calculated curve relates the RSSI to the Signal-to-Noise Ratio (SNR) and this SNR is mapped onto PRR. This technique requires however a static relation between the PRR-SNR curve, which is not suitable for dynamic environments [17].

2.2.3 Qualitative measures for Link Quality Estimation

According to [14], the effectiveness of a link quality estimator can be assessed by four qualitative measures. First of all, the estimator needs to be energy efficient as it is to be used in low-power energy constrained devices. Secondly, to get a meaningful result an accurate estimator is very important. The last two measures are reactivity and stability. The estimator has to be reactive in order to produce a quick and up to date estimate that gives a reflection of the current link. On the other hand, a stable estimator is desired meaning that short-term variations of the estimate are not attractive. It is trivial to see that this results in a trade-off between reactivity and stability.

2.3 Sub-GHz for Short Range Devices

2.3.1 Duty Cycle and Transmission Power Regulations

The use of frequency bands is tightly regulated and it is therefore important to look at the regulations imposed on the Sub-GHz frequency bands used in this thesis, more specifically for SRDs. In the European Union (EU) those regulations are based on Duty Cycle (DC)s and transmission power, in contrast to the United States of America (USA), where this is done based on harmonic and electrical field strength. The focus of this

(33)

2.3. SUB-GHZ FOR SHORT RANGE DEVICES 14

Figure 2.9: Available frequency bands in the 863-870 MHz range [20].

section will be on the EU regulations, who are provided by the Conférence Européenne des administrations des Postes et Télécommunications (CEPT), European Commission (EC) and European Telecommunications Standards Institute (ETSI).

The first limit of the EU regulations is the DC limit. The CEPT defines the DC as the ratio, expressed as a percentage, of the on-air time of a single transmitter device to the observation period, measured in an observation frequency band as shown in (2.5). Tobs is by default a continious one hour [18]. For the

transmission power limit, both the CEPT and EU use the Effective Radiated Power (ERP) in as a measure. This indicates the power radiated by a half-wavelength dipole antenna to get the same electrical field strength of the actual antenna at a distance along the main beam of this antenna.

DC = Ton Tobs

(2.5) The EU defined the radio spectrum limits for SRDs in a EC decision [19] for all allowed frequencies. In [20] a visual overview is given for the sub-GHz frequencies of the 868 MHz ISM band (863-870 MHz) and the results are shown in figure 2.9. In all frequency bands except band 49 and 51-53 the DC limit can be relaxed to 2.7 % if polite spectrum access is employed. This means that the device must have a Clear Channel Assessment (CCA) and a random backoff interval should the medium be busy. It should be noted that the current CEPT recommendation [18] adds an extra sub-GHz band to this list, namely 862-863 MHz with 25 mW ERP and a DC limit. However, this is not yet defined in the official EC decision.

2.3.2 Existing technologies

Since interference can make radio links unreliable and highly variable as mentioned in section 2.2.1, it is interesting to see which existing technologies use the same sub-GHz frequency bands. Figure 2.10 provides