LoRaWAN-Based Adaptive MACs for Event Response Applications

Received 21 August 2022, accepted 26 August 2022, date of publication 5 September 2022, date of current version 20 September 2022. Digital Object Identifier 10.1109/ACCESS.2022.3204654 LoRaWAN-Based Adaptive MACs for Event Response Applications VASILEIOS ASTERIOU, ANASTASIOS VALKANIS , GEORGIA BELETSIOTI , KONSTANTINOS KANTELIS, GEORGIOS PAPADIMITRIOU , (Senior Member, IEEE), AND PETROS NICOPOLITIDIS , (Senior Member, IEEE) Informatics Department, Aristotle University of Thessaloniki, 54124 Thessaloniki, Greece Corresponding author: Petros Nicopolitidis ([email protected]) This work was supported by the European Union and Greek National Funds through the Operational Program Competitiveness, Entrepreneurship and Innovation, under the call RESEARCH CREATE-INNOVATE 2, under Project T2EDK-02617. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 ABSTRACT Low Power Wide Area Networks have emerged as a leading communications technology in the field of Internet of Things sensor and monitoring networks. In such networks, uplink traffic is characterized as a combination of periodic data reports and event-triggered alarm reports. When an many devices detect an event in a short timespan, a burst of concurrent transmissions can occur, leading to a surge of collisions, and thus severe data delivery performance degradation. In this paper, a hybrid random/scheduled access strategy is proposed for mitigating the impact of traffic-triggering events on network performance. Under periodic report traffic the LoRaWAN standard Class A protocol is in effect, but after an event a TDMA scheme is applied. Three implementations of this strategy are described. The first is a pair of novel MACs for LoRaWAN, allowing (a) synchronization of end devices with the network using the event detection as a crude synchronization point, and (b) the dynamic scheduling of groups of devices. The other two implementations build upon a single-hop and a two-hop previously proposed LoRaWAN-based wake-up architectures, respectively. The above approaches are validated and studied through extensive simulation. The results show improved packet delivery ratio over the Class A MAC. The effect is more prominent as the event propagation velocity increases. The proposed approach also surpasses LoRaWAN in energy per delivered bit for high event propagation velocities. Finally, the novel protocol has a lower hardware and deployment complexity than the wake up radio based alternatives, at the cost of higher energy consumption. INDEX TERMS Internet of Things, LPWAN, LoRaWAN, MAC protocol, synchronization, TDMA, event triggered traffic, event response. I. INTRODUCTION In Internet of Things monitoring sensor networks, some cases may arise in which events, with causes external to the network, can be detected almost simultaneously by the devices in the network. In networks with random access upstream protocols, such as LoRaWAN, this can cause a burst of simultaneous transmissions, which can lead to serious congestion, increased number of collisions, and performance degradation more generally. Thus, there is a need for an appropriate The associate editor coordinating the review of this manuscript and approving it for publication was Hosam El-Ocla VOLUME 10, 2022 . response to such conditions at the medium access control (MAC) layer. Depending on the nature of the event and the application, the collection of event reports from the sensors may be of critical operational importance. The above statement is particularly true in applications such as forest monitoring for management and fire prevention and detection. Another relevant application is smart metering, including power grid and water supply grid monitoring for failures. In such applications, the ability of the network to facilitate swift event report data transfer is directly linked to monitoring performance, and can ultimately have an impact on operational functions. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ 97465 28 29 30 31 32 33 34 35 36 37 38 39 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 Ensuring adequate LoRaWAN network performance in an event response scenario is particularly challenging. The reason is the random access uplink protocol, in which there is no coordination between sensor devices, in combination with the downlink scalability issues of LoRaWAN, which prevent efficient signaling. Thus, while LoRaWAN is highly energy efficient and can support low power wide area network (LPWAN) requirements in normal use case monitoring scenarios with periodic upstream reports [1], additional work is needed to improve performance in event response scenarios without sacrificing energy efficiency and battery lifetime. Being an open network standard, LoRaWAN has attracted a significant number of researchers who have proposed modifications and improvements to the LoRaWAN MAC to optimize network performance. However, the focus of these works is LoRaWAN operation under normal traffic conditions, i.e. periodic uplink and sparse downlink messages. In this paper, a hybrid random/scheduled access approach is proposed for designing MAC layers capable of providing event response service to IoT applications at a reasonable energy consumption overhead for devices in the field. Based on this approach a novel MAC layer is proposed, which provides a basis for adaptive, application-controlled scheduling, to facilitate the efficient transfer of event report data to the application server. Additionally, two wake-up receiver-based (WuRx) alternative implementations of the same service, based on previously proposed architectures and MAC layers, are discussed. The performance of these architectures under external traffic-triggering events is studied extensively via simulation, and compared with Class A LoRaWAN. Thus, the contributions of this paper are: • A MAC layer, Post-Event Synchronization (PES), which enables devices affected by an event to synchronize with the network server shortly after the event. • A MAC layer, Group-Announcement TDMA (GAT), which provides scheduled access to end devices in order to increase aggregate bandwidth and reduce the time it takes for all event report data to pass through the network while using device grouping to reduce signaling overhead. • A proposed application of the wake-up radio-based architectures introduced in [2] – a two-hop architecture – and [3] – a single hop architecture – for event response scenarios. • A simulation-based performance comparison between the above solutions and LoRaWAN to the event response problem, focusing on packet delivery performance and energy consumption. The rest of this paper is organized as follows. In section II the background for this research is discussed. In section III, previous works in the design of LoRaWAN MAC extensions are discussed, and their applicability to the event response problem is assessed. In section V, a detailed introduction of the PES and GAT protocols is made, while in section VI, the WuRx-based alternative event-response solutions are 97466 discussed in detail. In section VII, the results of the simulation study are presented. In section VIII, the protocols, their performance trade-offs, and architectural considerations are discussed, while in section IX the shortcomings of this work are discussed and some future research directions are outlined. Finally, section X concludes the paper. II. BACKGROUND 96 97 98 99 100 101 102 LoRaWAN is an LPWAN network standard, published and maintained by the LoRa Alliance. In order to fulfill the long range, low power consumption requirements of LPWANs, LoRaWAN takes advantage of LoRa, a proprietary and patented physical layer, based on Frequency Shift Chirp modulation [4], a form of Chirp Spread Spectrum modulation system. This modulation system provides robust transmissions which can withstand significant levels of interference. The main parameters of the physical transmission are the transmission bandwidth and the spreading factor (SF), which determine the duration of the transmission and its data rate. Specifically, the most common bandwidth setting in LoRaWAN is 125 kHz, and SFs range from 7 to 12, with each increment roughly corresponding to a doubling in the transmission’s time on air and a halving in its data rate. An important aspect of the LoRa physical layer is the quasi-orthogonality of transmissions on different SFs, which enables, to some extent, simultaneous decoding of frames on the same frequency channel if they are sent over different SFs [5]. Regarding the network layer, LoRaWAN is a star-of-stars network operating in the sub-GHz ISM bands, and in the 868 MHz in the EU specifically. All IoT end devices (ED) establish a link with the network server (NS) via one or more gateways, which are radio devices controlled directly by the network server. On top of this link, traffic between the end device application layer and the application server (AS) can be exchanged. The LoRaWAN standard defines three MAC modes: Class A, which is mandatory, and Classes B and C, which are optional and mutually exclusive. For EDs, the Class A MAC is a frequency-hopping random access-based protocol. In the EU regulatory domain, which is the focus of this work, since no carrier sense mechanism is involved, a duty cycle limitation also applies. Limiting device duty cycle also helps with power conservation, as the device stays in deep sleep for prolonged periods and only wakes up to transmit or receive data. In Class A, after a random access uplink transmission, an ED is required to turn its receiver on during two predefined reception windows (RX1 and RX2), in order to provide an opportunity to the network server to send a downlink frame. A Class B end device is required to additionally open extra reception windows periodically at specific time intervals. The timing of those reception windows, termed ping slots, is synchronized at predefined offsets from beacon frames transmitted simultaneously by all gateways periodically, and which must be received by the end devices to continue using Class B. End devices operating in Class C keep their receiver on at all times. The three classes VOLUME 10, 2022 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications 164 represent three different options on the downlink latencypower consumption trade-off, with Class A being the most energy efficient and Class C offering the lowest downlink latency of the three. LoRaWAN is designed to function as an LPWAN tuned for low power consumption in the periodic data collection scenario. The combination of a robust physical layer and dutycycled MAC ensures the LPWAN requirements under that scenario. However, the standard protocol is not as well suited for the case where end device transmissions are triggered almost simultaneously due to some external occurrence, such as a power outage. In this scenario, as a result of the random access MAC, a high number of collisions occur resulting in severe performance degradation [6]. 165 III. RELATED WORK 151 152 153 154 155 156 157 158 159 160 161 162 163 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 The LoRaWAN standard, although simple and energy efficiency-oriented, has come under scrutiny from researchers. A lot of research into LoRaWAN performance focuses on scalability, a major performance characteristic for IoT networks, which are required to support device densities of tens of thousands of devices. In [7], it is pointed out that the random access, ALOHA-like protocol of Class A LoRaWAN poses a challenge as the number of devices that join the network increases due to channel saturation which leads to collisions. This observation is also made in [8]. Furthermore, in [9] it is found that duty cycle-related downlink scalability issues of LoRaWAN networks can become an obstacle to the scalability of confirmed transmissions. The impact of event-triggered traffic on LoRaWAN networks is assessed in [6]. There, it is found that in dense networks, traffic triggered by external events can lead to a burst of collisions which severely degrade the packet delivery performance of the network. Scalability and performance issues, both under regular traffic and under event-triggered traffic, also impact energy consumption as collisions result in wasted transmissions. For the above reasons, several researchers have proposed extensions or modifications to the LoRaWAN MAC layer protocol in order to improve network performance, reduce power consumption, or both. In this section, a series of LoRaWAN protocol extensions are summarized and their suitability for event response applications is assessed. These MAC extensions are grouped into (a) random access-based, (b) scheduled access based, (c) wake-up radio-based and (d) hybrid random/scheduled access-based. These related works are summarized in Table 1. This paper is differentiated from the referred protocols in Table 1 by addressing event-triggered traffic in addition to regular traffic, and thus overcomes the limitations of previous works in ways that are discussed in detail below. A few LoRaWAN-based MAC proposals focus on other random access-based protocols. In [10] devices are synchronized to a real-time clock based on the finish time of a transmission event, which is common between a gateway and an end device. Based on that clock, the end devices VOLUME 10, 2022 are programmed to only transmit at the beginning of a slot, i.e. a Slotted ALOHA-style protocol. A synchronization procedure based on the Class B beacon frames is used for establishing slotted random access in [11]. In [12], apart from the slotted random access uplink protocol, aggregated acknowledgments are piggybacked in periodic downlinks. In [13], a periodic beacon structure is used for synchronization, and also a lightweight scheduling scheme is proposed. In that scheme, the gateways constrain what SFs and power levels can be used on each channel, and announce those constraints with the beacon frame. Then, the end devices access the channel, which is slotted, randomly but within the announced constraints. Finally, a CSMA-based protocol is evaluated in [14]. Although slotted protocols and CSMA mitigate to some extent the scalability issues of pure LoRaWAN, these protocols still suffer from collisions that limit performance and energy efficiency due to their random access nature. Regarding the event response scenario, these proposals suffer from the same issue as plain LoRaWAN. Specifically, under eventtriggered traffic, if the devices try to transmit event reports as soon as possible, a high contention situation arises on the channel. As no coordination between the stations exists, this strategy leads to transmission collisions. The implication is that not only data is lost, and possibly has to be re-transmitted at a later stage, but also that the energy expended for the collided transmissions is wasted, reducing the per bit transferred energy efficiency of the network. A different strategy is to use time division multiple access (TDMA)-based protocols, in which contention is resolved in a collision-free way. In [15], two offline scheduling algorithms are presented, that increase overall network capacity and are well suited for data collection applications. The design is iterated in [16], where a two-stage protocol is presented. In the first stage, the end devices join the network. During the join handshake, the end devices also synchronize with the network and receive their slot schedule. In the second stage, the TDMA parameters, such as the frame structure and timing information get transferred to participating end devices. Notably in this work, multiple transmission parameters, channel characteristics, and hardware limitations are taken into account in the system model. Another line of work is the one first presented in [17] and then better evaluated in [18]. In this work, the end device initiates the synchronization handshake with a sync request command. The server responds with precise timing information along with a slot schedule encoded in a bloom filter. Scheduled access, as proposed in the above works is a significant network optimization for IoT networks under regular traffic. Eliminating contention and coordinating the transmissions of IoT end devices enables better channel utilization which can improve overall network performance for dense networks, even if the protocols used are heavier. Although the above approaches eventually achieve the goal of scheduled access over LoRaWAN, a few characteristics make 97467 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications TABLE 1. Research papers related to this work, their main contribution, and their limitations in event response scenarios. 97468 VOLUME 10, 2022 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 their as-is use ineffective for quick response under eventtriggered traffic. First, the synchronization scheme depends on the initiative of the end devices in order to keep the protocols lightweight. Thus if, right after an event, ad-hoc synchronization was required – which in the above works is established using a random access stage – the same contention, collision and data loss problems found in random access protocols would arise. Second, the schedule is decided upon the end devices joining the network, which provides enough flexibility for data collection applications, but cannot easily be adjusted if dynamic scheduling is required. Even if dynamic scheduling were addressed, the spatiotemporal auto-correlation characteristics of event-triggered traffic need to be taken into consideration, as they provide a basis for optimization. A promising direction in the extension of LoRaWAN networks is the use of WuRx radios. This is specialized hardware designed to deliver ultra-low power selective remote asynchronous interruption service. Devices are equipped with a secondary radio, the WuRx, which is always listening, has address decoding capabilities, and consumes considerably less energy than the main radio, the LoRa transceiver in this case. The downside is a significantly shorter communication range than typical LPWAN range requirements of several kilometers. A hardware survey can be found in [19] and a high-level characterization of wake-up radio capabilities about IoT networks is presented in [20]. Recent works focusing on the integration of WuRx with LoRaWAN networks include [2], [21] and [3]. In [2], the On Demand TDMA (ODT) protocol for LoRaWAN using WuRx is introduced, which enables ad-hoc scheduling of neighborhoods of end devices. Always listening Cluster Head nodes are employed to relay asynchronous signaling by the network server to the end devices. The network server starts a TDMA cycle by broadcasting a frame for the cluster heads, which then transmit a wake-up beacon containing a specific address. The end devices that are activated, transmit according to a predefined schedule. An improvement of this design is presented in [21], where wake-up beacon signaling is additionally used to relay device status information to the Cluster Head. To improve latency, the Cluster Head then performs a secondary level scheduling, and includes schedule information in the wake-up beacon addressed to the end devices. In [3], a different approach is taken, with a less energy efficient WuRx design, which is however long range. At the protocol level, the side radio operates on a duty cycle and relies on the transmission of prolonged wake-up signal preambles to ensure successful wake-up. The architectures described in the above works highlight the new possibilities for IoT network MAC design that WuRx based signaling brings. However, the above works do not focus on network performance under event-triggered traffic. In section VI, two WuRx-based event response approaches based on the above proposals are discussed. VOLUME 10, 2022 A hybrid random access and TDMA protocol is presented in [22]. In this protocol, devices use confirmed transmissions to treat the absence of an acknowledgment as a proxy for collision detection. Consequently, the devices briefly switch to Class C (always listening) mode, which enables the network server to synchronize the devices with a downlink transmission. Subsequently, the collided end devices access the channel with time-multiplexed transmissions which enables collision-free transfer of the data. However, applicability to event-triggered traffic is not discussed. In this paper, a synchronization primitive similar to the above is adapted for the event response use case and augmented with a dynamic scheduler to offer effective network service under traffic generated by external events. Another hybrid protocol is that in [23], which is based on slotted ALOHA and ODT (TDMA). In this work, the choice of protocol at any given moment is controlled by a learning automaton executed on the network server. The process is driven by network feedback information, which is used to detect events and gauge traffic intensity. A protocol is then chosen to minimize delay and maintain high network performance. This work, notably, takes the event response scenario into account, by using the metadata carried by event packets both in the feedback process Finally, regarding actual scheduling that can improve network service performance in an event response scenario, a spatial scheduler has been presented in [24], based on event epicenter estimation and scheduling of end devices or groups close to the epicenter estimate. However, the underlying MAC mechanism that supports the scheduling has not been sufficiently specified. In this paper, ad-hoc dynamic scheduling approaches for LoRaWAN networks, both with and without the use of WuRx, are defined that can support this type of service. IV. EVENT-TRIGGERED TRAFFIC IN IoT 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 In [6], a traffic model is proposed for IoT applications, in which the aggregate network traffic can be caused by two distinct processes, the periodic monitoring function, and external events. In other words, a distinction is made between periodic and event-triggered traffic. On that basis, a traffic model is developed for event-triggered traffic, and the performance of LoRaWAN under the event scenario is modeled. It is shown that external, traffic-triggering events can cause a packet delivery performance drop of more than 70 percentage points, and thus highlight the need for mitigation strategies. The purpose of this paper is to develop such strategies in the form of LoRaWAN extensions. In this section the traffic model assumed in the rest of this paper is outlined. The model is largely based on that in [6], but it is extended to accommodate general event shapes. The basis of the model is a LoRaWAN network with end devices that generate traffic in order to transfer data to their respective application servers. The model consists of two traffic types, (a) regular traffic, which includes periodic sensor 97469 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 measurement data transmissions, as well as sparse downlink transmissions, and (b) event-triggered traffic that carries event alarm reports upstream. The model is suitable for cases where external events can be detected by the end devices and can trigger an immediate uplink transmission. In this scenario, the event is considered to be local, i.e. the devices are not triggered randomly but rather in a strongly spatially autocorrelated way. Regular traffic is modeled using a Poisson arrival process with a fixed rate, which is the same for all end devices. The rest of this section describes how event-triggered traffic is modeled. An event is defined using the following parameters, (a) an origin time te , (b) an epicenter e, (c) a propagation velocity vp and (d) a spatial correlation factor (SCF) 9(x). The event initially starts at the origin time at the epicenter and it propagates through the environment with a fixed propagation velocity. If x is the location of an end device, then the event will reach that end device after time tp = kx − ek/vp from the origin time. At that point, the end device will detect the event with probability equal to the SCF 9(x). Upon event detection, the end device immediately transmits one uplink event report frame to notify the application server of the event occurrence. A simple model for the SCF is the raised cosine radial function, defined by the formula:  1 d ≤a        (1) 9RC (x) = 1 1 + cos π d − a a<d ≤b  2 b−a    0 d >b where d = kx − ck is the distance of the end device from an arbitrarily chosen center for the radial function, and a, b are parameters controlling the radius of the SCF. In particular, the SCF has the value 1 up to a distance a from the point c. As the distance increases to b, the SCF gradually decreases to 0 in the shape of a cosine. Finally, the SCF value is 0 at distances larger than b. Regarding the modeling of arbitrary shaped SCFs, individual raised cosine SCFs can be combined to model general event shapes. Let 9i (x) where i = 1, 2, . . . , N be SCFs that focus on individual areas an event might affect. Assuming an end device detects the event if it is affected by at least one individual SCF, the combined SCF will be 9(x) = 1 − 408 N Y [1 − 9i (x)] (2) i=1 410 In this way, arbitrary shaped SCFs may be defined as a combination of raised cosine SCFs. 411 V. A NEW LoRa-BASED MAC FOR EVENT RESPONSE 409 412 413 414 415 The main goal of this work is to establish on-demand adaptive TDMA service over LoRaWAN networks for event response applications. Under this type of service, the network normally operates using the low power consumption Class A MAC, 97470 but the application may request higher throughput service for particular end devices in an event response scenario. For those end devices, extra bandwidth should be allocated on a synchronous collision-free frame. This type of hybrid random/scheduled access MAC-based service can afford short-term higher throughput performance at the expense of increased power consumption. On top of this service, spatial scheduling algorithms can be used to efficiently map how a physical event has interacted with the network, and allocate bandwidth preferentially to affected end devices. The main advantage of the hybrid protocol strategy is that in the absence of external events, no power consumption overhead is ever incurred on the end devices. For this hybrid MAC to be implemented, two distinct features are required of the network: (a) synchronization, and (b) scheduling. In the rest of this article, the scheduling will be assumed to be executed centrally by the application server, rather than in a distributed manner by the end devices. Scheduling involves a signaling aspect, whereby the computed schedule has to be communicated to the end devices. Techniques for spatial scheduling will be considered out of scope for this paper. In this paper, a novel approach to event response over LoRaWAN networks is proposed, based on the rough synchronization that is caused by an external detectable event e.g. a power outage event. This mechanism does not require wake up receiver hardware and can be implemented on the existing LoRaWAN infrastructure, thus offering the service on homogeneous hardware architecture. The proposed mechanism consists of two protocols: the Post Event Synchronization protocol, which enables the end devices to participate in a synchronous protocol, and the Group Announcement TDMA protocol, which enables the application server to allocate channel airtime by scheduling end device group transmissions, a strategy analogous to that of [2]. These protocols can be implemented using existing LoRa hardware, with the resulting network relying on a homogeneous physical layer. A. POST EVENT SYNCHRONIZATION 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 The Post Event Synchronization (PES) protocol uses the external event to establish synchronization of the end devices with the network. To achieve that, the approximate synchronization of the end devices that is caused by the external event is leveraged. Specifically, in this mechanism the end devices that have detected the event initially transmit the generated event report under Class A. Then, after a predefined delay has expired, these devices turn on their receivers and listen on a predefined channel. During this reception window, the network server can transmit a multicast downlink to the listening devices. If such a downlink is transmitted, since it is a multicast, the reception will stop for all listening end devices at the same time. This results in the synchronization of those end devices. If no frame is transmitted, the listening end devices return to sleep mode. This method of synchronization is similar to that of [22], as well as standard LoRaWAN procedures, VOLUME 10, 2022 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications FIGURE 1. Illustrative example of the PES and GAT protocols. Five end devices, triggered by an external event, attempt to transmit event reports, resulting in collisions. The end devices then attempt to synchronize with the network server using PES. After a multicast downlink transmission by the network server, the end devices enter TDMA mode and the network server schedules transmissions in groups for two cycles. Each end device transmitting within a TDMA cycle reverts to Class A mode. At the end of the second cycle, no TDMA beacon is transmitted and the last end device also reverts to Class A. 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 such as firmware upgrades over-the-air (FUOTA) [25], for multicast downlinks. The main difference is that the reception window is triggered by an external event rather than a general case collision or a previous arrangement. Fig. 1 shows an example execution of the PES protocol. Five end devices transmit uplink frames during an event, which results in some data loss. After transmitting the event reports, and after a small delay, the end devices turn their receivers on. The network server transmits a multicast downlink which is received by all end devices. At that point, they are synchronized. A suitable choice for the multicast downlink channel and transmission parameters is a SF 12 channel, such as the one used for the RX2 reception window. The timeout of the PES reception window should be long enough to compensate for the uncertainty in the synchronization caused by the event, but otherwise as short as possible to minimize energy consumption. The method for detection of an event affecting the end devices by the network server is considered out of scope in this paper. One suggestion is channel activity detection (CAD) schemes, which have been previously considered for CSMA-based MAC layers [14]. Such modules could be used on the network server side for collision monitoring and, in combination with LoRa-based localization techniques [26], [27], [28], could provide some crude information VOLUME 10, 2022 to the network about the origin time and location of the event. Another possible source of information might be a successfully delivered event report fed back to the network server by the application. The assumption made here is that the network server is able to combine multiple indications to determine the presence and approximate location of an event. B. GROUP ANNOUNCEMENT TDMA 497 498 499 500 501 502 503 The Group Announcement TDMA (GAT) protocol is a MAC designed to provide dynamic end device scheduling service in a resource constrained environment such as a LoRaWAN network. As the protocol is entirely-based on the low data rate LoRa physical layer, in order to reduce the communication overhead, the bandwidth allocation strategy is to assign channel time to entire groups of end devices rather than to individual end devices. The protocol consists of three phases, (a) group definition, (b) end device initialization, and (c) operation. The group definition is carried out by the network server. The application server is allowed to define new groups and assign devices to groups. Defining a new group initializes an empty TDMA cycle structure, which consists of multiple frequency channels, each of which is further divided into logical channels corresponding to the different SFs. The multiplexing of LoRa channels and SFs is used to decrease 97471 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 protocol delay and increase throughput. Radio equipment capable of simultaneous decoding of all SFs on at least three channels on the gateway side is assumed, and the SFs are assumed to be orthogonal. The slot duration for each SF is the time on air of a frame with the maximum allowed payload Nmax for GAT uplink frames, plus a guard time to avoid collisions due to synchronization error. For each device added to a group, a slot, uniquely identified by its channel, SF and offset, is allocated to that end device, but only if the resulting cycle length is below the maximum allowed value TCYCLE . In this setup, arbitrary groups may be defined but, in this work, the end devices are partitioned in groups based on physical proximity. In order for end devices to be initialized, they must know configuration values common for all devices, and also, for each group to which the end device belongs, the corresponding slot parameters. Common configuration may be transferred out of band before network initialization. However, slot allocation is based on the end device data rate, which is determined only upon joining the network as it depends on channel conditions. As such, slot information must be transferred after (or during) the join procedure. The operation phase starts with the transmission of a group announcement beacon frame by the network server. This is a multicast downlink frame that contains the identifier of the group to which the channel has been assigned. Devices that receive this frame, schedule a reception window to be open after a duration of TCYCLE (Fig. 1), in order to receive the next cycle group announcement frame. Since that duration is also the maximum cycle duration, all TDMA transmissions will have concluded by the time the next reception window is opened. The first group announcement frame conveniently coincides with the synchronization downlink of PES. Since the end devices re-synchronize with every group announcement frame, a relatively short guard time of TGUARD = 5 ms was chosen – see section VII for more details. Devices that belong to the announced group can schedule a transmission in their pre-allocated slot. The payload of this transmitted frame depends on the application client that is executed on the end device. For the event response scenario, it is a copy of the latest event report that has been generated and stored away for this purpose. Devices that do not belong to the announced group enter deep sleep for the duration of the cycle, waking up again prior to the next reception. At the end of the cycle, if the network server transmits another group announcement frame, the end devices stay in TDMA mode and another TDMA cycle begins. If no multicast frame is sent at the end of the cycle, the protocol is concluded and all end devices return to Class A mode. At any moment, an end device may withdraw from the protocol and revert to Class A, for example, if no more data are readily available to transmit, to conserve energy. An important note to be made here concerns duty cycling. In the EU, the radio frequency regulation agency, ENISA, subjects the 868 MHz ISM band to various duty cycling limitations for different sub-channels [29]. However, these 97472 limitations concern the aggregate duty cycle over an observation period of 1 h. Specifically, for the 869.525 MHz channel used for GAT downlinks, a 10% duty cycle limit applies, which amounts to an airtime ratio of 360 s/h. Thus the repeated downlink transmissions remain within EU regulatory standards. The LoRaWAN standard also defines duty cycle limitations but only as a back-off strategy for situations in which traffic re-transmissions can cause severe persistent network overloading. In Fig. 1, an example of a possible execution of the protocol is shown. After synchronizing with PES, the end devices receive the first group announcement frame. In this example, devices 2 and 3 belong to the announced group, so they repeat the event report transmission, and afterward revert to Class A immediately. The other devices proceed to listen for the next group frame at the end of the cycle, in which devices 4 and 5 are instructed to transmit. After their transmission, they also revert to Class A, leaving device 1 as the only participant. At this point, the scheduler decides that the TDMA should end and does not send any more group announcement frames. Device 1 interprets the absence of a group frame transmission as the end of the protocol and reverts to Class A. VI. WAKE-UP RADIO-BASED EVENT RESPONSE ARCHITECTURES 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 Wake-up receivers (WuRx) are radio systems that enable remote device interruption at low power consumption overhead. This is achieved through a low power or ultra-low power radio system, which can detect and perform limited address decoding on wake-up beacon transmissions. If the address in a particular beacon matches a WuRx device configuration, the device drives an interruption signal upon which the host device can exit deep sleep mode. Therefore, a WuRx can achieve low latency, low power selective asynchronous remote interruption. Currently, the trend in WuRx research and development is the reduction in power consumption all the while receivers become more sensitive [19], [20]. WuRx radios provide an energy-efficient means for asynchronous downstream communication with powerconstrained devices such as LoRaWAN end devices. In the context of event response with on-demand adaptive TDMA, a WuRx can be a power efficient primitive for end device signaling and synchronization. However, one of the challenges that arise in the attempt to integrate WuRx with LPWANs is the difference in the range supported by the physical layers. In this section, two alternative ways using WuRx to provide a similar service to the combination of PES and GAT described above, are discussed. These alternatives are based on previously proposed hardware architectures and MACs that extend the LoRaWAN network using WuRx and tackle the range problem of WuRx in distinct ways. Since the basic strategy remains the same, both the LoRaWAN only-based PES/GAT and the WuRx-based protocols described below achieve comparable performance, and a clear improvement over plain LoRaWAN, as will be shown in the next section. VOLUME 10, 2022 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications FIGURE 2. Comparison of the PES/GAT and WuRx-based architectures considered in this paper. 657 The approaches differ in their energy consumption characteristics as well as their hardware and deployment complexity. In particular, WuRx-enabled LoRaWAN networks leverage two heterogeneous physical layers integrated in a unified service offering that can be challenging due to inherent range incompatibilities between the main and side radios. One approach is to introduce extra intermediary nodes and another is to use a longer range but more energy-consuming WuRx. The benefit to the LoRaWAN-only based approach is lower power consumption overhead in the case of an event. Fig. 2 shows an architectural comparison between the PES/GAT combination and the WuRx-based approaches described below. On the left, the original LoRaWAN star network is shown. During the GAT protocol, the network server broadcasts group announcement frames to all devices, but only the ones corresponding to the announced group respond. In the middle, the two-hop WuRx architecture is shown. In order to schedule a cluster, the network server broadcasts commands to Class C cluster heads, which in turn send wake-up beacons to neighboring devices. Only wokenup devices respond. Finally, on the right, in the single-hop WuRx architecture, the gateway sends wake-up beacons directly to end devices. All devices receive the beacon but only those devices with a matching address are interrupted. 658 A. ON-DEMAND TDMA 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 659 660 661 662 663 In order to overcome the short range of WuRx one proposed solution is the introduction of intermediate nodes that can relay signaling from the gateways to the end devices [2]. In this two-hop architecture, end devices are partitioned into clusters, each of which contains one cluster head (CH) node, VOLUME 10, 2022 such that all the WuRx of the end devices are in the cluster head’s range. The cluster heads are assumed not to be powerconstrained and, as such, they can be in always listening mode (e.g. with the Class C protocol), to relay signaling information with little delay. In [2] this architecture is used to implement the On-Demand TDMA (ODT) protocol. In this protocol, a slot duration per SF is fixed, and for each end device in a given cluster a slot is pre-allocated e.g. while or immediately after the end device joins the network. The network server can initiate a TDMA frame by sending the appropriate command to the corresponding cluster head, which immediately transmits a wake-up beacon. The end devices woken up from this beacon (a) wait for their pre-allocated slot, (b) transmit an uplink frame, and finally (c) go back to sleep mode. Sufficient synchronization of the end devices is attained through the simultaneous reception of the wake-up beacon and driving of the interrupt signal. To respond to a traffic triggering event, ODT can be used to schedule uplink transmissions from clusters that contain end devices affected by the event. Specifically, the end devices transmit the event reports by random access as soon as the event is detected, as in PES. The network server then starts scheduling end device clusters with the ODT MAC. While in [2] only one spreading factor is considered at a time per cluster for the TDMA schedule, in this paper different devices in the same cluster are allowed to occupy different spreading factors, as in GAT. To further reduce power consumption overhead, the WuRx can be only enabled right after the event, instead of being on all the time, similar to PES. Regarding end device and network server duty cycling, the same argument as for GAT applies to the case of ODT as well. 97473 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 B. SINGLE-HOP WAKE-UP RECEIVER ARCHITECTURE Another approach for integrating WuRx with LPWANs is the use of long-range WuRx hardware. One such approach is proposed in [3], where a long-range WuRx design is proposed. This single-hop WuRx over LoRaWAN (SH-WuRx) architecture has the advantage of eliminating the need for cluster heads for wake-up signaling. The drawback of this approach is that the WuRx consume almost an order of magnitude more power to operate than more commonly proposed ultra-low power alternatives. In order to manage this overhead, the authors propose the DC-MAC, in which the WuRx operates on a duty cycle to conserve battery life, while the transmitter side uses extended length preambles to increase the chance the wake-up beacon will be received. In the context of event response, the single-hop architecture described above can be used to support a PES/GAT- and ODT-inspired protocol to achieve the same type of service as the previously described architectures. At network initialization, the end devices are partitioned into groups similar to GAT, with each end device receiving one slot in a TDMA cycle structure according to its SF. The end devices, as in the ODT-based alternative, initially transmit event reports as they are generated and then proceed to turn on their WuRx. In this case, the network server schedules the groups by transmitting their respective wake-up beacons directly through the gateways, rather than over relay nodes. Each end device that wakes up, responds by transmitting the stored event report at its pre-specified slot. VII. SIMULATION RESULTS To evaluate and compare the performance of the aboveproposed event response over LoRaWAN architectures, a custom discrete event LoRaWAN simulator was developed. The simulator design is divided into independent modules implementing the kernel, the channel model, the components of the LoRaWAN standard, the extensions, and the driver. Each module is individually tested. In this section, the simulation model is described and the simulation results are presented. A. SYSTEM MODEL The model system on which the simulator was developed is largely based on that of [6]. The simulated network consists of a single gateway, around which IoT end devices are distributed uniformly in a 2.5 km radius, with a density of 500 km−2 . In line with [6], it is assumed that the network is located in an urban environment, with end devices placed indoors at a variety of heights and depths in buildings. Two contributions to signal loss are considered in the channel model: (a) path loss, which is computed according to the Hata model, and (b) building penetration loss which is computed according to the relevant 3GPP model, specifically scenario 1, in [30]. 97474 After initial placement and computation of loss, for each end device, the maximum LoRaWAN SF is computed, as the minimum such that the signal reaching the gateway and vice versa is above the receiver sensitivity threshold [9]. The reader is referred to [6] for the receiver sensitivity values for each data rate. In particular, some end devices can never establish a link with the gateway and are thus pruned altogether from the simulation [6]. Subsequently, and just before the simulation run, the end devices and network server are initialized by simulating the joining procedure and the exchange of protocol parameters. The network operates on the default channels defined in the EU868 regional parameters specification of LoRaWAN, and specifically the 868.100 MHz, 868.300 MHz and 868.500 MHz channels for uplinks and RX1 downlinks, as well as the 869.525 MHz channel for RX2 and RXC downlinks [29], as well as for the group announcement frames. All nodes transmit with a power of 14 dBm. Each simulation consists of one event episode, where the event can be of a general shape and in general positioning, as determined by the simulation parameters. The event occurs at time te = 10 s. Detecting end devices attempt to transmit event reports that are 10 B long. In episodes in which one of the proposed event response architectures is used, the network responds to the event at a later time tr = te + 2vdp + 5 s, where d/2 is a measure of the maximum distance of an event triggered node from the epicenter. In other words, the network response commences after all event report transmissions have concluded, and a several second long arbitrarily chosen margin has passed. To fairly compare the architectures, the groups are defined in the same way for each protocol with the following procedure. First, a hexagonal lattice with a distance between adjacent vertices of 250 m is generated. Each vertex corresponds to a group. Then, each end device is assigned to the group whose vertex is the closest. In the case of ODT, the cluster heads are placed exactly on those vertices and are exempt from the aforementioned pruning. The distance between vertices was chosen by assuming a WuRx range of up to 200 m for ODT, and in order to keep the comparison direct, the same lattice parameter was used under PES/GAT and SH-WuRx as well. Energy consumption modeling is done using a simplified five state model. The five states are (a) deep sleep, (b) processing, (c) listening, (d) listening for wake-up beacons, and (e) transmitting. Each state is associated with a power level. By tracking the time an end device spends in each state during a simulation run, a total consumed energy figure can be calculated. Table 2 summarizes the power values used in each state. The following assumptions are also made: • It is assumed that end devices can retain application data in memory during deep sleep. In this way, the event report frame can be stored when it is first generated, and retrieved later in the duration of the event response VOLUME 10, 2022 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications TABLE 2. Power consumption values for the various energy model states. 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 protocol. As LoRaWAN network session information is also persisted on the end device during deep sleep, this assumption is reasonable. • For simplicity, the time needed to wake-up from deep sleep as well as prepare the radio for reception or transmission is considered to be 200 ms, during which the device is in the processing state. • It is considered that LoRaWAN end devices can transit to sleep mode and wake-up in a period of TCYCLE = 2 s, excluding the wake-up time. To ensure successful protocol operation in the presence of synchronization error for PES/GAT, sufficiently long reception windows and synchronization guard times must be defined. To compute synchronization guard time, the following assumptions are made: • All end devices complete decoding of a group announcement frame up to 1TSYNC after the end of the transmission. • End device clock rates differ by up to a skew rate of rSKEW . As such, if T is the time since the last synchronization, a further clock divergence of up to T · rSKEW may occur. Avoiding collisions due to synchronization error requires an idle interval of duration at least equal to the maximum clock difference between two end devices between uplink transmissions. Also, avoiding lost downlink frames requires that a reception window at least as long as the maximum synchronization error is used. For the above cases, the following synchronization guard time is defined: TGUARD = 1TSYNC + TCYCLE · rSKEW + TMARGIN (3) Assuming a maximum skew rate rSKEW = 100 ppm [18], a maximum initial sync error 1T = 1 ms and allowing for a margin TMARGIN = 2 ms, then a guard time of TGUARD = 5 ms should prevent loss of synchronization for TCYCLE = 30 s. Fig. 3 shows an instance of the event scenario that is used in the simulations. In this figure, each point corresponds to an end device location. Grey points are pruned end devices, green points are those end devices that are affected by the event, and yellow points are the rest of the end devices. The event location is away from the gateway VOLUME 10, 2022 FIGURE 3. Example instance illustrating the events simulated in this section. so that end devices of varying SFs participate in the event response protocol. The event area of effect is of elongated shape to further illustrate the importance of adaptive dynamic scheduling. B. RESULTS 846 847 848 849 850 In this section, simulation results are present ed in an attempt to illustrate the operation and performance characteristics of the proposed protocols. Simulation results focus on both packet delivery performance and energy efficiency, in order to illustrate the different trade-offs, on the one hand between plain LoRaWAN and the proposed event response schemes, and, on the other hand, between the LoRaWAN only-based PES/GAT and the WuRx-based alternatives. 1) PROTOCOL OPERATION 851 852 853 854 855 856 857 858 859 860 In this subsection, sample data from the protocol simulation are presented. In Fig. 4, uplink transmissions and receptions of frames in two simulation instances are shown. The simulations were run with plain LoRaWAN and with PES/GAT as the event response protocol respectively. No background traffic was included, and the event velocity was 315 m/s. The diagram on the left shows the result for LoRaWAN. A spike of frames is generated at te = 10 s. All affected stations transmit, but owing to the collisions, only some frames are received. After the transmissions, the end devices return to sleep mode and, since no background traffic is considered, no further uplink transmissions or receptions occur, and the collided event report data are lost. The result for PES/GAT is substantially different. Even though the collision spike still occurs, the event response mechanism kicks in shortly after, and 16 manually scheduled groups transmit their packets. At this stage the channel is contention free, so all transmissions are received successfully. 97475 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications FIGURE 4. Uplink frame transmissions and receptions in an event scenario with no background traffic over time. Left: LoRaWAN. Right: PES/GAT. FIGURE 5. PDR of the considered protocols as a function of event propagation velocity. 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 Thus, although the frames have been transmitted twice, all data has been delivered. 2) PACKET DELIVERY PERFORMANCE To compare the performance of the event response architectures, a series of simulation runs were executed for a range of event propagation velocity values, without any background traffic. The results are shown in Fig. 5. The packet delivery ratio (PDR) metric is computed as the fraction of all event reports generated that were successfully delivered at least once to the network server. As the event propagation velocity increases, the event report packets are generated in a shorter timespan. This results in higher transient traffic intensity, which for LoRaWAN, being a random access protocol, translates to a higher collision probability and a decreasing packet delivery ratio. On the other hand, the PES/GAT, ODT, and 97476 FIGURE 6. PDR of PES/GAT as a function of regular traffic load. Single-Hop Wake-up Receiver (SH-WuRx) protocols attain almost optimal packet delivery performance. That is because, in the absence of other interference, the TDMA phase is collision-free. Since the end devices unaffected by the event are unaware of the fact that a TDMA protocol is simultaneously executed, transmissions by those devices may result in collisions with the scheduled transmissions. To study this effect, a series of PES/GAT simulations were run, with varying values of aggregate background traffic intensity. Fig. 6 contains event report delivery ratio versus load graphs for the event report frames, the background traffic frames, and all frames respectively. It is clear that as aggregate background traffic load increases, frame delivery performance deteriorates rapidly, which is the expected behavior for random access MACs. What is interesting is the packet delivery performance of the event report frames, which, as shown in Fig. 6, is slightly VOLUME 10, 2022 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications FIGURE 7. Total energy spent by all end devices in the network per event report bit delivered by the considered protocols. 922 worse than the overall network performance. The conclusion is that in the presence of background traffic, the performance of the proposed event response approaches does deteriorate, but only slightly more than overall network performance degradation which would manifest anyway. Although only PES/GAT results are reported in Fig. 6, the same conclusions hold for the wake-up radio-based approaches as well. One mitigation strategy would be the use of dedicated channels for event report uplinks, which would lead to a collision-free protocol. 923 3) ENERGY PERFORMANCE 913 914 915 916 917 918 919 920 921 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 Energy consumption is a major factor in the viability of any protocol and proposal in the LPWAN area since the IoT devices are generally power constrained, as they depend on batteries or energy harvesting methods. It is clear that any event response protocol with the approach used in this paper, i.e. ad hoc intervention for contention resolution, is going to incur an energy consumption overhead for every intervention. To assess this overhead, a series of simulations were run, under plain LoRaWAN and the three event response schemes, PES/GAT, ODT, and SH-WuRx, and for a range of propagation velocity values, while no background traffic was considered. To take into consideration the actual data delivery of each scheme, instead of reporting the absolute consumed energy values, the aggregate consumed energy per bit delivered metric is used. Note that if a frame is delivered twice - i.e. during the random access and the event response phase - its bits are counted only once in the total, as the second transmission is redundant. The consumed energy measurement takes into account transmissions, receptions, listening, processing, and wake-up radio listening, as applicable to each protocol, and is a network total value, i.e. a sum over all end devices. Specifically, the initial random access for the event response VOLUME 10, 2022 schemes is taken into account. For ODT, as the cluster heads are considered exempt from power constraints, they are not included in the measurement. Fig. 7 shows the results. For LoRaWAN, with increasing propagation velocity, energy efficiency drops. This is expected as the number of frames transmitted is the same, but the number delivered diminishes rapidly (Fig. 5). The event response protocols deliver more consistent performance, which is expected since frame delivery does not depend on the event propagation velocity. The wake-up radiobased schemes have similar energy performance, as they both use lower power alternatives for listening and decoding the protocol control signals. SH-WuRx exhibits slightly higher power consumption, as its wake-up radio consumes almost 10 times more power than the one considered in ODT. The LoRa-based PES/GAT consistently consumes the most energy among the three event response schemes. For high velocities, PES/GAT total unnormalized energy consumption is on average 2.94 times that of ODT and 4.73 times that of plain LoRaWAN. Furthermore, in slow propagation scenarios, the extended listening that is required for end device synchronization results in reduced energy efficiency, given that the time spreads between the event affecting different end devices are generally longer. The same effect can be observed with the wake-up radio schemes as well, but to a much smaller degree, as the same operation is carried out using the wake-up radios. Finally, it is clear that for very small propagation velocities, depending on the application, plain LoRaWAN may be able to deliver acceptable performance. However, already at a propagation velocity of a few dozen meters per second the event response schemes start demonstrating an advantage, very clearly in terms of delivery ratio but also in terms of consumed energy per bit. VIII. DISCUSSION 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 The main idea in this paper is that, although data regarding event occurrences can be important, and although LoRaWAN can be impacted by a severe increase in collisions in the case of event-triggered traffic, events are too rare to warrant changes in the core protocol that are expensive in terms of energy consumption. For the above reason, a hybrid protocol strategy is adopted, whereby normally the network operates under the energy efficient Class A protocol, but right after events adaptive and dynamic scheduled access is allowed and facilitated. The motivation behind using groups for dynamic scheduling was the following. Firstly, organizing transmissions in groups, although more constraining than scheduling end devices individually, allows for multiplexing transmissions on multiple channels and SFs. Secondly, and more importantly, using predefined groups for dynamic scheduling results in a low overhead solution for both the initialization and the dynamic scheduling phase, as protocol signaling is lightweight, consisting only of a slot identifier, 97477 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications 1043 and a group identifier in GAT or a wake-up address in ODT, and SH-WuRx. Regarding the comparison between GAT, ODT, and SH-WuRx, each approach has its advantages and disadvantages. The ODT-based solution achieves the lowest energy consumption for the energy constrained devices. However, it requires extensive modification of the network architecture with the introduction of the cluster heads to support twohop communication. The SH-WuRx-based solution overcomes the architecture complexity by using longer-range wake-up receivers at the expense of a slight increase in energy consumption. Although the wake-up receiver-based solutions feature fast and efficient signaling, they require the integration of distinct physical layer technologies into a heterogeneous network, which increases hardware and deployment complexity, especially in the case of different ranges between main and side radios. On the contrary, PES/GAT is exclusively LoRaWAN-based and constitutes an extension over a homogeneous network architecture, at the cost of increased energy consumption. As such it is probably better suited for less power-constrained applications, perhaps aided by energy harvesting schemes. Overall, since the event response strategy is to deviate from the energy-efficient Class A protocol only when necessary, the energy consumption overhead of the event response protocol is only expended on relatively rare occasions anyway. Another consideration is that in order for an on-demand, adaptive TDMA access service to be available to an IoT application, it is imperative that the TDMA schedule, which is computed by the network server, be driven by applicationgenerated feedback. In the conventional public LoRaWAN network architecture, depending on the deployment, this may require that packets travel to a physically distant server through the Internet. This may introduce delay, impairing service adaptivity and limiting overall service value. Having some application logic executed close to the network server, i.e. moving it closer to the edge, is a modification that can overcome this problem. By allowing the application to examine packets on the network server machine, immediate feedback can be yielded to the TDMA service scheduler, eliminating the performance impact of an unpredictable Internet service delay. 1044 IX. LIMITATIONS AND FUTURE WORK 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 An important limitation of this work is the applicability of the hybrid approach in events with small propagation velocity. Such events cause a measurable but considerably smaller impact on the packet delivery performance of the network. Furthermore, for PES/GAT, as the transmission start times have a larger variance, the devices have to wait for longer periods before the synchronization can happen. Consequently, the proposed protocol could be considered too heavyweight even for the case when the event-triggered traffic only moderately impacts data delivery. 97478 In such scenarios more lightweight mitigation strategies could provide a better performance-energy consumption trade-off. A second limitation of this work concerns the simulation model simplifications. First, the LoRaWAN end devices are assumed to be able to transit to sleep mode and wake-up in the duration of TCYCLE of a couple of seconds. Depending on the hardware, wake-up times may vary as different clocks may need longer times to stabilize. The result would be higher energy consumption in the duration of the TDMA phase, shifting the performanceenergy consumption trade-off more in the favor of LoRaWAN. Furthermore, some issues have been considered out of scope for this paper. The first issue is initial event detection. Some possible techniques have already been outlined in section V-A. However, more comprehensive solutions on the basis of group scheduling may be achievable by using lowresolution groups for localization of the event. Another issue is the actual scheduler scheme. A previously proposed solution is that in [24], which relies on event epicenter estimation to schedule groups close to the epicenter, but the approach lacks support for general shaped events or multiple events, not leveraging the scheduling capabilities provided by the TDMA MACs proposed in this paper. More sophisticated scheduling algorithms are currently under development to alleviate this issue. X. CONCLUSION 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 In this paper, a MAC layer solution based on LoRaWAN was proposed for event response applications over LoRaWAN, in which devices can operate with the low overhead Class A protocol under normal traffic conditions and temporarily switch to adaptive scheduled access in the case of an event, a more energy consuming but better performing MAC layer. The proposed solution consists of the synchronization protocol PES, which uses the event to synchronize reporting sensors with the network server, and the TDMAbased protocol GAT, which schedules transmissions by sensor groups with explicit announcements. In addition, two WuRx-based solutions following similar strategies are also considered, using architectures proposed in previous works. The proposed protocols, as well as the WuRx-based alternatives, are shown to achieve nearly optimal event report delivery performance. The event response protocols introduce energy consumption overhead to the network, but only when an actual event is detected by the end devices. For high propagation velocities, the proposed strategy achieves lower energy consumption per delivered event report bit than plain LoRaWAN, while for low propagation velocities, LoRaWAN remains a more energy-efficient solution. However, assuming traffic triggering events are relatively rare since the event response mechanisms are only triggered under event circumstances, the impact on end device lifetime is kept to a minimum. VOLUME 10, 2022 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 REFERENCES [1] Q. M. Quadir, T. A. Rashid, N. K. Al-Salihi, B. Ismael, A. A. Kist, and Z. Zhang, ‘‘Low power wide area networks: A survey of enabling technologies, applications and interoperability needs,’’ IEEE Access, vol. 6, pp. 77454–77473, 2018. [2] R. Piyare, A. Murphy, M. Magno, and L. Benini, ‘‘On-demand LoRa: Asynchronous TDMA for energy efficient and low latency communication in IoT,’’ Sensors, vol. 18, no. 11, p. 3718, Nov. 2018. [3] A. Froytlog, M. A. Haglund, L. R. Cenkeramaddi, and B. Beferull-Lozano, ‘‘Design and implementation of a long-range low-power wake-up radio and customized DC-MAC protocol for LoRaWAN,’’ in Proc. IEEE Int. Conf. Adv. Netw. Telecommun. Syst. (ANTS), Dec. 2019, pp. 1–5. [4] L. Vangelista, ‘‘Frequency shift chirp modulation: The LoRa modulation,’’ IEEE Signal Process. Lett., vol. 24, no. 12, pp. 1818–1821 , Dec. 2017. [5] D. Croce, M. Gucciardo, S. Mangione, G. Santaromita, and I. Tinnirello, ‘‘Impact of LoRa imperfect orthogonality: Analysis of link-level performance,’’ IEEE Commun. Lett., vol. 22, no. 4, pp. 796–799, Apr. 2018. [6] V. Gupta, S. K. Devar, N. H. Kumar, and K. P. Bagadi, ‘‘Modelling of IoT traffic and its impact on LoRaWAN,’’ in Proc. GLOBECOM IEEE Global Commun. Conf., Dec. 2017, pp. 1–6. [7] O. Georgiou and U. Raza, ‘‘Low power wide area network analysis: Can LoRa scale?’’ IEEE Wireless Commun. Lett., vol. 6, no. 2, pp. 162–165, Apr. 2017. [8] J. Haxhibeqiri, F. Van den Abeele, I. Moerman, and J. Hoebeke, ‘‘LoRa scalability: A simulation model based on interference measurements,’’ Sensors, vol. 17, no. 6, p. 1193, 2017. [9] F. V. D. Abeele, J. Haxhibeqiri, I. Moerman, and J. Hoebeke, ‘‘Scalability analysis of large-scale LoRaWAN networks in NS-3,’’ IEEE Int. Things J., vol. 4, no. 6, pp. 2186–2198, Oct. 2017. [10] T. Polonelli, D. Brunelli, A. Marzocchi, and L. Benini, ‘‘Slotted ALOHA on LoRaWAN-design, analysis, and deployment,’’ Sensors, vol. 19, no. 4, p. 838, Feb. 2019. [11] L. Chasserat, N. Accettura, and P. Berthou, ‘‘Short: Achieving energy efficiency in dense LoRaWANs through TDMA,’’ in Proc. IEEE 21st Int. Symp. World Wireless, Mobile Multimedia Networks (WoWMoM), Aug. 2020, pp. 26–29. [12] G. Yapar, T. Tugcu, and O. Ermis, ‘‘Time-slotted ALOHA-based LoRaWAN scheduling with aggregated acknowledgement approach,’’ in Proc. 25th Conf. Open Innov. Assoc. (FRUCT), Nov. 2019, pp. 383–390. [13] B. Reynders, Q. Wang, P. Tuset-Peiro, X. Vilajosana, and S. Pollin, ‘‘Improving reliability and scalability of LoRaWANs through lightweight scheduling,’’ IEEE Internet Things J., vol. 5, no. 3, pp. 1830–1842, Jun. 2018. [14] L. Beltramelli, A. Mahmood, P. Osterberg, and M. Gidlund, ‘‘LoRa beyond ALOHA: An investigation of alternative random access protocols,’’ IEEE Trans. Ind. Informat., vol. 17, no. 5, pp. 3544–3554, May 2021. [15] D. Zorbas, K. Q. Abdelfadeel, V. Cionca, D. Pesch, and B. O’Flynn, ‘‘Offline scheduling algorithms for time-slotted LoRa-based bulk data transmission,’’ in Proc. IEEE 5th World Forum Internet Things (WF-IoT), Apr. 2019, pp. 949–954. [16] K. Q. Abdelfadeel, D. Zorbas, V. Cionca, and D. Pesch, ‘‘FREE— Fine-grained scheduling for reliable and energy-efficient data collection in LoRaWAN,’’ IEEE Internet Things J., vol. 7, no. 1, pp. 669–683, Jan. 2020. [17] J. Haxhibeqiri, I. Moerman, and J. Hoebeke, ‘‘Low overhead scheduling of LoRa transmissions for improved scalability,’’ IEEE Internet Things J., vol. 6, no. 2, pp. 3097–3109, Apr. 2019. [18] C. Garrido-Hidalgo, J. Haxhibeqiri, B. Moons, J. Hoebeke, T. Olivares, F. J. Ramirez, and A. Fernandez-Caballero, ‘‘LoRaWAN scheduling: From concept to implementation,’’ IEEE Internet Things J., vol. 8, no. 16, pp. 12919–12933, Aug. 2021. [19] R. Piyare, A. L. Murphy, C. Kiraly, P. Tosato, and D. Brunelli, ‘‘Ultra low power wake-up radios: A hardware and networking survey,’’ IEEE Commun. Surveys Tuts., vol. 19, no. 4, pp. 2117–2157, 4th Quart., 2017. [20] S. Benhamaid, A. Bouabdallah, and H. Lakhlef, ‘‘Recent advances in energy management for green-IoT: An up-to-date and comprehensive survey,’’ J. Netw. Comput. Appl., vol. 198, Feb. 2022, Art. no. 103257. VOLUME 10, 2022 [21] G. A. Beletsioti, K. F. Kantelis, A. Valkanis, P. Nicopolitidis, and G. I. Papadimitriou, ‘‘A multilevel TDMA approach for IoT applications with WuR support,’’ IEEE Internet Things J., early access, Jun. 23, 2022, doi: 10.1109/JIOT.2022.3185750. [22] H. Rajab, T. Cinkler, and T. Bouguera, ‘‘IoT scheduling for higher throughput and lower transmission power,’’ Wireless Netw., vol. 27, no. 3, pp. 1701–1714, Mar. 2020. [23] A. Tsakmakis, A. Valkanis, G. Beletsioti, K. Kantelis, P. Nicopolitidis, and G. Papadimitriou, ‘‘An adaptive LoRaWAN MAC protocol for event detection applications,’’ Sensors, vol. 22, no. 9, p. 3538, May 2022. [24] V. Asteriou, G. Papadimitriou, and P. Nicopolitidis, ‘‘Adaptive MAC protocols for IoT edge computing architectures with event-triggered traffic,’’ in Proc. IEEE Int. Black Sea Conf. Commun. Netw. (BlackSeaCom), May 2021, pp. 1–6. [25] J. Catalano, ‘‘LoRaWAN firmware update over-the-air (FUOTA),’’ J. ICT Standardization, vol. 9, no. 1, pp. 21–34, Apr. 2021. [26] M. Jouhari, E. M. Amhoud, N. Saeed, and M.-S. Alouini, ‘‘A survey on scalable LoRaWAN for massive IoT: Recent advances, potentials, and challenges,’’ 2022, arXiv:2202.11082. [27] K. Mikhaylov, M. Valkama, M. A. Lema, S. Andreev, R. Gupta, O. Galinina, G. Destino, T. Mahmoodi, M. Dohler, and Y. Koucheryavy, ‘‘Multi-radio perspectives for massive MTC localization: Energy consumption and utility,’’ in Proc. 11th Int. Congr. Ultra Modern Telecommun. Control Syst. Workshops (ICUMT), Oct. 2019, pp. 1–6. [28] Z. A. Pandangan and M. C. R. Talampas, ‘‘Hybrid LoRaWAN localization using ensemble learning,’’ in Proc. Global Internet Things Summit (GIoTS), Jun. 2020, pp. 1–6. [29] M. Saelens, J. Hoebeke, A. Shahid, and E. D. Poorter, ‘‘Impact of EU duty cycle and transmission power limitations for sub-GHz LPWAN SRDs: An overview and future challenges,’’ EURASIP J. Wireless Commun. Netw., vol. 2019, no. 1, pp. 1–32, Sep. 2019. [30] Cellular System Support for Ultralow Complexity and Low Throughput Internet of Things (CIOT) (Release 13), document TR 45.820, V. 13.1.0, TSG GERAN, 3GPP, 2015. [31] L. Casals, B. Mir, R. Vidal, and C. Gomez, ‘‘Modeling the energy performance of LoRaWAN,’’ Sensors, vol. 17, no. 10, p. 2364, 2017. 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 VASILEIOS ASTERIOU was born in Thessaloniki, Greece, in 1998. He received the B.S. degree in informatics from the Aristotle University of Thessaloniki, Greece, in 2020, where he is currently pursuing the Ph.D. degree in communications networks with the Department of Informatics. His research interests include the IoT networks, LPWANs, and MAC design. 1221 ANASTASIOS VALKANIS received the graduate degree from Hellenic Air Force Military Technical Academy, the B.S. degree in informatics from the Hellenic Open University, the M.Sc. degree in web intelligence technologies from the Department of Information Technology, International University of Thessaloniki, and the Ph.D. degree in telecommunications networks from the Informatics Department of Aristotle University of Thessaloniki. During his career as a Radar and Telecommunication Engineer at Hellenic Air Force, he was trained and specialized in several telecommunications and radar systems. His current research interests include wireless sensors and optical networks. 1229 97479 1222 1223 1224 1225 1226 1227 1228 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 V. Asteriou et al.: LoRaWAN-Based Adaptive MACs for Event Response Applications 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 GEORGIA BELETSIOTI received the B.Sc. degree in computer science from the Computer Science Department, University of Crete, in 2011, the M.Sc. and Ph.D. degrees in computer science from the Department of Informatics, Aristotle University of Thessaloniki (AUTH), in 2014 and 2020, respectively, and the M.Sc. degree in ICT in education from the Department of Educational and Social Policy, University of Macedonia. She is currently a Postdoctoral Researcher with the Network and Communication Systems Laboratory, Department of Informatics, AUTH. She has participated in three EU-funded research projects. Her current research interests include optical networks, LPWAN, and the IoT networks. KONSTANTINOS KANTELIS received the B.Sc. degree in mathematics, in 2004, the M.Sc. degree in computer systems technology from the National and Kapodistrian University of Athens, in 2007, and the M.Sc. degree in nanotechnology and the Ph.D. degree in communication nanonetworks from the Aristotle University of Thessaloniki (AUTH), in 2012 and 2018, respectively. He currently conducts his postdoctoral research in the area of communication systems and networks with the Network and Communication Systems Laboratory, Department of Informatics, AUTH. His main research interests include but not limited to nanonetworks, optical networks, LPWAN, and ubiquitous computing. 97480 GEORGIOS PAPADIMITRIOU (Senior Member, IEEE) received the Diploma and Ph.D. degrees in computer engineering and informatics from the University of Patras, in 1989 and 1994, respectively. In 1997, he joined as the Faculty Member of the Department of Informatics, Aristotle University, Greece, where he is currently a Full Professor with the School of Informatics. He is the Deputy Head of the School of Informatics and the Director of the Network and Communication Systems Laboratory. He teaches the undergraduate courses: communication networks, digital communications, network security, and network engineering; and the postgraduate courses: architectures and security of optical networks and internet security. He has supervised eight Ph.D. theses and three of the Ph.D. Students. He supervised are currently faculty members (Associate/Assistant Professors). He has published 142 articles in peer-reviewed journals (61 in IEEE Journals) and 158 papers in international conferences. He is author of three books published by Wiley and editor of a book published by Kluwer/Springer. His major research interests include wireless networks, optical networks engineering, network security, biological nanonetworks, the Internet of Things, and AI-based networking. He has served as chair/tpc chair for four international conferences and as a TPC member for 61 international conferences. He also serves as a reviewer for 36 Scientific Journals. He has participated in 24 research projects, some of which as a team leader or coordinator. He serves as evaluator for international and national Research and Development programs. He was an Associate Editor of the IEEE Network, the IEEE TRANSACTIONS ON BROADCASTING, the IEEE TRANSACTIONS ON SYSTEMS, MAN AND CYBERNETICS—PART C, the IEEE SENSORS JOURNAL, and the IEEE Communications Magazine. 1269 PETROS NICOPOLITIDIS (Senior Member, IEEE) received the B.S. and Ph.D. degrees in computer science from the Department of Informatics, Aristotle University of Thessaloniki, Greece, in 1998 and 2002, respectively. From 2004 to 2009, he was a Lecturer at the Department of Informatics, Aristotle University of Thessaloniki, where he is currently serves as an Associate Professor. He has published over 170 papers in international refereed journals and conferences. He has coauthored the book titled Wireless Networks (Wiley, 2003) and co-edited three other books by Wiley and Springer. His research interests include wireless networks, mobile computing, and optical networks. He was the TPC chair for over ten international conferences, mostly sponsored by IEEE. He is an Associate Editor for IEEE ACCESS, the International Journal of Communication Systems by Wiley, and Security Communication Networks and Wireless Communications and Mobile Computing journals by Hindawi. 1299 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 VOLUME 10, 2022