Cross-layer framework and optimization for efficient use of the energy budget of IoT Nodes

Cross-Layer Frame work and Optimiz ation for Ef ficient Use of the Ener gy Bu d get of IoT Nodes Gilles Callebaut, Geoffre y Ott oy , Liesbet V an der Perre KU Leuven, ESA T -DRAMCO, Ghent T echn o logy Campus Ghent, Belgium gilles.callebaut@ku le u ven.be, geoffrey .ottoy@kuleuven.be, liesbet.vanderperre@ku leuven.be Abstract —Both physical and MA C - layer parame ters impact the autonomy of IoT devices. W e present an open-source cross- layer assessment framewor k fo r Low Power Wide Area net- works (LPW ANs) in this paper . It extends the state-of-the-art with energy models, downlink messages, an d adaptive datarate features. Hence, hypotheses and transmission schemes can be tested and ev aluated. As a representati ve case, the LoRaW AN protocol is assessed. The fin dings demonstrate th at a cross- layer is imperative to effectively rea lize LPW ANs in terms of energy efficien cy and throughput. For instance, up to a factor of three reduction i n energy consumption can be achiev ed by transmitting longer packet on quasi-static channels. Y et, under adverse dynamic conditions, an energy penalty will occur . Index T erms —IoT , Energy Efficiency , LPW AN, cross-lay er I . I N T R O D U C T I O N There is a rapid ly increasing deman d to interconn ect de- vices. Sensors and actuators cooper a te, often through a cloud infrastructu re, e n abling new a pplications in smart homes, cities, and sustainable environmen ts. Many of these In ternet- of-Thin gs (IoT) application s face strict e n ergy co nstraints as they rely on batter y powered devices. T ypically , a wirele ss IoT sensor will wake up perio dically to collect measurem e nts, e.g., temperatu re or chemical substances. The data can be sent immediately to the cloud infrastructu re or accumu lated in the node to reduce com munication overhead . The device is in a low-po wer sleep state f or the rest of th e time. Several dedi- cated comm unication systems have been d ev eloped f or long range low power IoT con nectivity , including LoRaW AN [1], SigFox an d NB-IoT [2]. Characterizing the energy of commu nication is a daun ting task. It d epends on nu merous inter related effects and aspects. T o start with, the applicatio n deter m ines how many b ytes are in a packet an d at what rate th ey are being sen t. T h e latter de pends, among o thers, on th e prop a gation co nditions. This paper is a preprint (IEE E ”acce pted” status). IEEE copyright notice. ©2019 IEEE. Persona l use of this materia l is permit ted. Permission from IEEE must be obtaine d for all other uses, in any current or future media, including reprinti ng/repub lishing this material for advertisin g or promotiona l purposes, creati ng new collect i ve works, for resale or redistribut ion to servers or lists, or reuse of any copyright ed This paper is accepted and published in the Confere nce Proceedings of 2019 IEEE W ireless Communicatio ns and Networking Conference (W CNC) wit doi: 10.1109/WCNC.2019.8885739. Secondly , the nu mber of devices in the network effects the energy con sumption, i.e., increasing traffic e ventually will cause more collisions. A third aspect is the position of the devices and the state o f the comm unication channel, which determines th e p ath lo ss a n d n o ise. Finally , the commun ication protoco l and radio hard ware deter mine h ow long the radio is activ e an d at what power . All these effects influen ce each o th er . For example, a no isy ch annel with a high path loss, m ay cause more retransmissions, wh ic h in turn increases the cha nce f or collisions. Simulators have been established in order to assess and optimize the L o RaW AN pro tocol specifically [3]–[5]. Here, we presen t a cross-layer simulation framework to r e alistically analyze and op timize the ene rgy consump tion of Io T devices. T o this end , in extension to [6], [7], b oth Adaptive Data Rate (ADR) an d downlink messages have been inc lu ded. W e show that this has a significant impact on the num- ber of collisions an d the data extraction rate. Secon d ly , the payload size and pac ket rate can be change d to co rrespon d with real- lif e applications, wh ereas this was fixed in prior work [8]. Hence, the framework allows application specific monitorin g of the LoRaW AN Networks. Th irdly , the en ergy consump tion correspo nding to the IoT nodes are b ased on measuremen ts from a power-optimized n ode [9]. Many nod es can b e po sitioned relatively to the g atew ay and subjected to different signa l lo ss and n oise. Th e source cod e (Python ) of the simulator is publicly a vailable [10]. W e welcome researchers to tailor the simulator to th eir own needs. The next section will briefly summ arize th e features of LoRa and L oRaW AN with an emp h asis on parameters affecting the power consum ption. Section III hig hlights th e key comp onents and op eration of the simulato r . In Sec t. IV we show the simulations and discu ss the results of a c a se study . Conclusions and future work are presented in Sect. V. I I . L P W A N A S S E S S M E N T - L O R A A N D L O R A W A N In our in vestigation, we consider the LoRa PHY an d Lo- RaW AN MAC schemes as a first representative case. W e introdu c e both layers with a f o cus o n th e governab le p aram- eters that affect energy consump tio n. No tably , a nu mber of parameters can be adju sted in the application lay er as well. 1) LoRa P HY scheme: LoRa, shor t f or L o ng Range , is a propr ietary mod ulation technique developed by Cycleo; later acquired by Semtech. The m odulation techniq u e is based on 978-1-5386-5 541-2/18/$ 31.00 ©2018 European Union Fig. 1: LPW AN Simulation Framework based on a m odular design for cro ss-layer assessment an d optimization . Chirp Spread Spectru m (CSS), which is similar to Direct- Sequence Spread Spectr um ( DSSS). LoRa en codes inform a - tion b y means of chirp s, in co n trast to modulation with pseu- dorand om binar y sequen ces in DSSS. A chirp is a sinu soidal signal wh ose f r equency mon o tonically increases ( upchirp ) or decreases ( do wnchirp ). T h e symb ol duration is b ased o n the spreading factor ( S F ) and the b andwidth ( B W ). E ach L oRa symbol is com posed of 2 S F chirps each covering th e entire bandwidth . The symbol d uration o f a LoRa symbol is defined as: T sym = 2 SF BW (1) A LoRa m e ssage co n sists of a pr e amble and data. The pr e am- ble contains on ly u p chirps wh ile the data part c o mprises up- chirps with discontinu ities. The position of the d iscontinuities –in freq uency– is what enco d es the tr ansmitted informa tio n. T o ensure that multiple packets can be demodu lated concur rently , LoRa packets ca n be encoded with different or thogon al spread- ing factors. This yields a r obust and lon g-rang e communicatio n link for IoT devices. 2) LoRaW AN MA C scheme: O n to p of the physical LoRa layer, the Lo RaW AN d efines the multiple access contro l (MA C) layer and the n etwork arc h itecture. Opp osed to the propr ietary m odulation technique LoRa, Lo RaW AN is an open standard sp e c ified by the LoRa Allian c e . LoRaW AN d efines three device classes each targeting d ifferent use cases. In general, Lo Ra devices initiate comm unication b y means o f transmitting a message to the gateway . By mea n s of co nfirmed messages , the node s can reque st acknowledgments to ensur e that the packets ar e successfully r eceiv ed by th e gateway . After an u plink message, the node o pens two slots to receive down- link traf fic fr om the gateway . This communicatio n schem e is optimized fo r low power because of its up link-centr ic design . LoRaW AN mandates that each LoRa device implemen ts this scheme. The compliant devices are called class A devices. Class B and C devices extend the co mmunicatio n capabilities of class A devices by defining ad ditional re c ei ve slots. Class B devices h av e per iodic receive slots while class C devices continuo usly listen for incoming m essages. These add itio nal downlink receive slots reduce the d ownlink latency y et yield a higher power con su mption. 3) Governable parameters: L oRaW AN facilitates contro l- ling the airtime (E q.1), d ata rate and en e rgy co n sumption of LoRa nodes in orde r to optimize the overall en ergy consum p - tion of the network. This is d o ne by adap ting the data rate an d transmission power to the pro pagation char acteristics o f the LoRa link. Inc reasing th e spreading factor results in a higher airtime, which allows the receiver to better demo d ulate the message. Despite the better range, a node will consu m e mo r e power when transmitting with a highe r spreading factor . In addition to mo difying the spread in g factor, the transmission power can be altered to furth er inc r ease the ran ge o r dec r ease the energy co nsumption . LoRaW AN devices need to co mply with the regu latio ns imposed in the industrial, scientific and medical (ISM) radio bands in which they o perate. Th ese r egulations includ e a limitation in the du ty cycle of transmission s and excited transmit power . Concretely , LoRaW AN enforce s a per b and duty-cycle lim itatio n. Af ter transmitting a m essage, the node needs to wait T off seconds before transmitting again in that band as per Eq . 2. Con sidering the ca se 1 of send in g a me ssag e with a payload size of 5 1 bytes and a spreading factor of 12 and respecting a duty cycle limit of 1%, the time off is 4 minutes. T off = T air T dc − T air [s] (2) I I I . C RO S S - L AY E R A S S E S S M E N T F R A M E W O R K The presented cross-layer simulator (Fig. 1) provid e s a generic f r amew ork to ev aluate and c o-optimize PHY , MAC and network pa rameters un d er realistic con ditions. T o ac- complish a use-case agnostic mod ular platfor m, th e simulator 1 This is a worst-case scenario, where the airtime of one packet is maxi- mized. Fig. 2: Measured power states of a LoRa node [9]. These m e asurements are a lso summarized in T able I. This pr ofile clearly shows the ene rgy impa ct of transmitting a message. In this case a confirmed message was sen t with SF9 and a pay load of 32 bytes. is structured on th e basis of individual com ponen ts. I n the framework, each class A Node sends Lo Ra packets to th e Air I nterface where collision, prop agation, and SNR m odels operate on the messages in progr ess. Finally , the Gateway receives and p r ocesses the p ackets. In the case of co nfirmed messages, th e gateway will transmit a message in th e downlink to the cor r espondin g node to ac knowledge the received uplin k message. A. Nodes Each node is character ized by an energy profile, a set of LoRa param e ters and a location. The default energy p r ofile used in the simulator is b ased on th e energy consu m ption of [ 9]. W e op te d for th is p rofile d ue to its power efficiency . Still, the simulato r is no t co nstrained to on e energy profile. Different d istinct pro files c an be allocated to nod es, mimicking various nod e s. T h e different energy states of th e simulated node are summar ized in T able I a n d illustrated in Figu r e 2. A similar experiment has been conduc ted in [ 11]. No tab ly , th ere are significa n t differences between the en ergy p rofile as mea- sured on our LoRaW AN-enabled node and those reported on in [1 1]. For instance, th e no de –ev aluated in [ 11]– con sumes twice the power of the n ode in our experiments [9] in transmit mode. As can be observed from Figure 2 and T able I, o ther power states are taken into acco unt besides transmit, receive and sleep. First, simp le p rocessing ( state 2 ) is simu lated. Secondly , the states prior to transmitting and receiving ( state 6 ) are related to waking-up and setting u p th e r adio. Finally , after re c ei ving a downlink message, the d ownlink m essage is processed and the MAC -related functio nality is executed ( state 9 ). The beh avior of the no de is d esigned as specified by Semtech [ 12]–[14] and their LoRaW AN no de imp lementa- tion. 2 In spite o f the open ness of the LoRaW AN MA C protoco l, not all L oRa-specific d o cuments are pu b licly acces- sible. In addition, the network operato r can, to some extent, freely define the network’ s behavior . I n our a ssessment, this 2 http:/ /stackfor ce.github .io/LoRaMac- d oc/ function ality is based on the o pen-sou rce implemen tation of The Things Network. 3 In ord er to optimize the energy budget of LoRa class A no des, a d ownlink me ssage can on ly be receiv ed after transmitting an u plink message. He n ce, the L o Ra nodes only need to listen to incoming messages at spe c ific times. As previously men tioned, Class A LoRaW AN d evices [1] u tilize two receive wind ows. Th e data rate an d center f requen cy of the downlink messages dep ends on the used receive wind ow , the data rate an d center frequ e n cy o f the uplink me ssag e . By default, th e downlink message sch eduled for the fir st receiv e window (RX 1) u ses the same f r equency and data rate as the uplink messag e . In the seco n d receive window (RX 2), a fixed p r edefined fr e quency and data rate are being used. In the remainder of th is paper, the rece ive win dows will b e denoted as RX 1 and RX 2 . In the case a d ownlink message was r eceiv ed in RX 1 , th e node will n o t use the secon d receive slot. In RX 2, Sem te c h defin es a spread in g factor of 12 while in our assessment we pref er the SF9 as propo sed by The Thing s Network. As a lower spreadin g factor is fa v ored because the base station can transmitted with higher power . The lower spreading factor r e sults in a faster receptio n, which in turn yields a lower energy co nsumption at the node. A channel frequen cy o f 868 . 5 25 MHz was selected f or RX 2; con form to The Thin g s Network. The c h annel frequency of the uplink p ackets ar e selec ted on basis of the ch annel av ailability . Th e end-d evice chooses a channe l with the lowest T off (Eq. 2). The d evice respects the duty cycle r egulations and waits to transmit a message if the re quired T off is not satisfied. A default transmission rate ( λ ) of 0.02 bits per secon d is ch o sen which is eq uiv alent to transmitting a 9 by te m e ssage every ho u r . B. Air I nterface The air interface includes thr ee m ain compo nents. First, the prop agation chann el introduces a path loss. Second ly , a simple SNR mode l is provided to translate the Received Sig n al Strength (RSS) to an SNR value. Finally , a co llision mo del 3 https:/ /github .com/thethingsnetwork T ABLE I: Energy profile [9] used in the case. State State Po wer Duration No. Descripti on (mW) (ms) 1 Sleep 5.7e-3 - 2 Proc essing 15 5 3 Tx prep. 12.5 40 4 Tx T ab . II Eq. 1 5a W ai t Rx 1 5.7e-3 1000 5b W ait Rx 2 5.7e-3 1000 - len(state 7) 6 Rx prep. 8.25 3.4 7 Rx1 36.96 airtime (DR=DR tx) 8 Rx2 34.65 airtime (DR=3) 9 Rx post proc. 8 .3 10.7 T ABLE II: Measur ed transmit power [9] fo r th e defined finite transmit power states. T ransmit Power (dBm) 2 5 8 11 14 Powe r (mW) 91.8 95.9 101.6 120.8 146.5 determines the collided packets, which occurs particularly in the uplin k in a typ ic a l LPW AN c a se. 1) Pr o pagation Model: Curr ently , the f ramework fe a tures two channe l m odels. First, a log- d istance chann e l mod el with shadowing is provided, wher e the path loss is ch a racterized by: P L ( d ) = P L ( d 0 ) + 10 · n log d d 0 + X σ [dB] (3) By default, the following par ameters [1 5] are used: d 0 = 1 000 m P L ( d 0 ) = 128 . 95 dB X σ = 7 . 8 dB n = 2 . 3 2 (4) An addition al path loss can be include d in the log-d istance model to simu late indoor p ositioned nodes an d g a te ways to accommo date for the addition al path loss [16] due to the penetration o f a building. Secondly , a COST 231 model [17] implementatio n can be used to model specific scen arios. 2) SNR Model: Th e cu rrent version of the simula to r takes into accou nt the n oise floor, as d escribed in [12]. In f uture extensions more complex mo dels can be includ ed and in ter- ference could be adde d . 3) Collision Model: The collision model considers the cen - ter f requency , spr eading factor, timin g and power to d etermine whether packets c o llide. The mod el is based on the finding s reported in [6]. Due to the orthogo n ality of the specified spreading factors, two message s enco d ed with different sp read- ing factor s can be dem o dulated concu rrently withou t collidin g . C. Gateway The gateway mod el is mainly based on the p o pular RF solution iC88 0A [ 18]. Th is LoRa conc e ntrator is able to receive up to eight packets simultan eously sent with d ifferent spreading factors o n different channe ls. Th is restriction is not considered in the assessment in this paper . A message c an be received by the gateway if it has not co llided and the signal strength is high e r than the sensitivity of the gateway [18], [19]. After demodu lating the re ceiv ed message, th e network executes Adaptive Data Rate (ADR) –if enabled– f ollowing a mechanism inspire d by th e implementation of The Thin gs Network. 4 According to the ADR specification , the network is capable of in creasing the data rate and chang ing th e transmit power of th e node, while the no d es can only dec rease the ir data rate. This can result in a low p ower transmit trap where nodes are no longer capable of commu n icating with the gateway [4]. Dependin g on the MAC LoRaW AN parameter s of the up link message, the gateway respon ds with a downlink message. W e currently assume that every sched uled downlink message will be received b y the en d-device considering gatew ays have a higher p ermitted transmit p ower . Th e gatew ay will first tr y to schedule a m essage in the receive slot which requir es less energy . For instan c e, if a message with SF12 was sen t, the gateway will try to sch e dule a downlink message on th e second receive slo t with S F 9 oppo sed to the fir st receive slot with S F 12 , in o rder to save significant air time, a n d h ence, energy . W e measu red an energy g ain o f four when utilizing this approa c h compared to u sing the first rec ei ve slot. This is one of th e cro ss-layer energy op tim izations already implemen ted in present networks. I V . E N E R G Y A S S E S S M E N T A N D O P T I M I Z A T I O N : R E S U LT S A N D D I S C U S S I O N The imp act of LPW AN param eters on energy con sumption and perform a n ce was assessed by p erformin g experim ents with the fram e work. Consequently , we analy zed the effect of th e packet len gth o n different perf ormance parameter s. Particularly , the impac t of the new capabilities in our f rame- work were a ssessed, most prom inently the o ptions to perfo rm ADR and use con firmed m essages. T o adequate ly evaluate the network, each experiment h as been repeated 1000 times by means of Monto-Carlo simulations. Th e default parameter s – for the condu cted exper iments– are d isp la y ed in T able III. T he perfor mance of the n etwork and individual no des have been ev aluated ba sed on the data extraction r ate (DER), the e n ergy per payload byte and the cha nnel variance. The d a ta extractio n rate d e fines the av erage ratio of the number o f uniq u ely received packets on the base station to the uniquely transmitted packets per no d e. I t in dicates how reliable the inten ded p ayload bytes a re received by th e gatew ay . It differs fro m the pa c ket delivery su c cess ratio becau se it do es not in clude re- transmissions in the calculatio n of the unique transmitted packets. Hen ce, the DER can be improved by utilizing re-tr a nsmissions in or der to ac c ommoda te for p acket loss. DER = Number of uniqu ely received bytes Number of un iquely tran sm itted b ytes (5) 4 https:/ /www .the thingsnet work.org/do cs/lorawan/adr .html W e prefer this parameter over the packet deliv ery success ratio because the main objectiv e is that the inten d ed pay load is receiv ed by the gatew ay . The nu m ber of re-transmission s necessary to achieve this goal is of secon dary importanc e . In the exper iments, th e energy pe r payload by te indicates the impact of the set of d e fin ed parameters on energy efficienc y . The channel is characterized b y its variance σ (in dB ) ac- cording to Eq. 3. This variance with respect to the average path loss is a consequen ce of varying propaga tion ch aracteristics in both space and time. In the fo llowing exper iments, th e defaults of Eq. 4 are used; if no t stated otherwise. T ABLE III: Parameter s fo r th e expe r iments rep orted on . Parame ter Defau lt V alue Channel V ariance σ 7.8 dB Number of Nodes 100 Data Transmission rate λ 0.02 bps Initia l Transmit Power 14 dBm Channel s 868.1, 868.3 and 868.5 MHz RX2 Channel 868.525 MHz RX2 Data Rate DR3 (SF 9) Cell Radius 1000 m A. V alidation o f the Simula tion Model The cross-layer simulation framework has been ev aluated and validated by checking the results with findings fr om re- lated work. As ad vised by Semtec h and The Thin gs Network, the ADR rate should only be en abled when a n o de has a fixed location. Th is is confirme d by o ur simu la tio ns. If the ch annel is dynam ic, the effect of ADR is nullified and even reduces the data extraction ra te (Fig. 4) , i.e. fewer packets ar e successfully received b y the gatew ay . Fur thermor e , the impact of the duty cycle lim it o n the downlink capab ilities of the gatew ay has been assessed as well. The exper iments co nfirm the findin gs reported in [7]. If only th e default channe ls (T able III) are utilized, the g atew ay is incapab le of ackn owledging all c o n- firmed messages. Consequ ently , the n umber o f retransmitted packages in creases, which in its turn yields a lo wer DER, as also ob served in [7 ]. Hence, the scalability of the network is mainly con strained by ( I) employing confirmed messages an d (II) the duty cycle limitation. B. Results - Cr o ss-Layer Appr oach to the R escue W e have assessed the im pact of package length as a first cross-layer optimization op portun ity . As expected, the av erage energy co nsumption p e r payload byte decreases wh en sen ding larger packets (Fig . 3). T o save ene rgy , non-time - critical data can be accum ulated, becau se b y increasing the paylo a d size 1) the overhead related to he ader in f ormation d ecreases, 2) the overhead of startin g and in itializing a transmission lowers, 3) the number of retransmissions in a stable propag a tion en vironm ent red uces, 4) the number o f downlink receive wind ows is also lower . 10 20 30 40 50 10 20 Payl oad size (B) Energy per payloa d byte E B (mJ) ADR CONF ADR NO CONF NO ADR NO CONF (a) Energy consumption per transmitted payload byte for diffe rent configurations. Disabling adapti ve data rate may result in up to an order of magnitude higher av erage energy consumption and considerably increase its spread. 10 20 30 40 50 1 2 Payl oad size (B) E B (mJ) (b) Z oom in showing the further energy reduction by disabling confirmed messages, especially r educing the spread for larger packet sizes. Fig. 3: Energy pe r tr ansmitted payload byte as a f u nction of payload size. The average is depicted by the mar kers while the deviation is illustrated b y the shaded area. The experiment simulated 30 opera tio n days. Substantial energy savings, up to an order of magn itude, can be achieved by enabling ADR as indicated in Fig . 3a. This obviously demonstrates the imp ortance of in c luding ADR in the a ssessment and op timization of tr a nsmission parameters in LPW ANs to ensure long battery lifetime of Io T nodes. Despite th e aforementio n ed beneficial ef fects of increasing the pay load size, sendin g more by tes per pac ket increases the tota l num ber of b ytes which ar e sent sub-optimal. Only after receiving 20 uplin k messages, the network will respon d with the adequ ate ADR p arameters to accomm o date fo r non- optimal propagatio n match ed Lo Ra parameters. For highe r payload sizes this implies that more bytes have been sent before the LoRa par ameters are adjusted to the c h annel. In addition, ADR chang es the parameter s in steps yieldin g an ev en slower adaptio n to th e pro pagation en viron m ent for larger payload sizes. This effect is clear ly notable w h en observ ing the energy con sumption over a short time period or when nodes hav e a slow d ata transmission rate. Th e pheno menon results in a hig her en ergy spread as depicted in Figure 3b. In quasi-static situations the impac t w ill be c ome negligible on the lon ger term. In d ynamic situation s, howe ver , the trade- off on packet length may yield a different r e su lt. 20 40 1 1 . 5 2 Energy per Byte (m J ) 20 40 1 1 . 5 2 2 . 5 20 40 4 6 8 10 12 0 dB 5 dB 7 . 8 dB 15 dB 20 dB 20 40 99 . 96 99 . 98 100 Payl oad size (B) Data E xtract ion Rate (%) (a) ADR on conf on 20 40 70 80 90 100 Payl oad size (B) (b) ADR on conf off 20 40 90 95 100 Payl oad size (B) (c) ADR of f conf off Fig. 4: Impact of the cha n nel variance and pay lo ad size o n the e n ergy and d ata extrac tio n rate. T o faster a d apt to the channel, L o Ra devices could first sent 20 smaller packets. This will result in reduced airtime and energy fo r packets wh ich are sent with non-o ptimal parameters. Th e furthe r in-d epth inv estigation of packet length versus d ynamics in the chann el can be pe r formed con veniently in the presented framework. V . C O N C L U S I O N S A N D F U T U R E W O R K A mod ular cross-layer fram ew ork f or LPW ANs has been presented. It allows assessing energy an d reliability . Furth er- more, di verse scenario s can be analyzed based on detailed energy pro files of IoT nodes. The framework has be en ex- tended w ith downlink messages, Adap ti ve Data Rate an d fine-grain ed monito ring of various parameter s (e.g., energy , collided packets). It is of par ticu lar interest to stud y th e impact of scaling u p to large numbers of no des in a network in both q uasi-static an d dyn amic scenarios. Ou r r esults show that many p arameters imp act the energy on the link , and th ey do influence eac h other . T his has been illustrated fo r example for the p acket leng th and ADR p arameters. T he imp ortance of a cross-layer ap proach is evidenced by a first sp ecific assessment on packet length and payload size. The propo sed cross-lay e r approa c h es will b e validated b y conduc tin g real experiments. W e see many interesting cross-layer opportun ities to further improve en ergy efficiency and reliability for massiv e Mach ine- T ype Commun ication (mMTC). These includ e, amo ng o thers, optimizing packet len gth taking into account channel dy n am- ics. R E F E R E N C E S [1] N. SORNIN and A. YEGIN, LoRaW AN™ Specification , L oRa Alliance T echnical Committee Std., 2017, v1.1. [2] Y . P . E. W ang , X. Lin, A. Adhikary , A. Grovlen, Y . Sui, Y . Blank enship, J. Ber gman, and H. S. Razaghi, “A Primer on 3GPP Narro wband Internet of T hings, ” Comm. Mag. , vo l. 55, no. 3, pp. 117–123, Mar . 2017. [Online]. A v aila ble: https://doi.or g/10.1109/M COM.2017.1600510CM [3] M. Bor and U. Roedig, “LoRa transmission parameter selection, ” in Pr oceedi ngs of the 13th IEEE International Conferen ce on Distribut ed Computing in Sensor Systems (DCOSS), Ottawa, ON, Canada , 2017, pp. 5–7. [4] M. Slabicki, G. Premsankar , and M. Di Francesco, “ Adapti ve Configu- ration of LoRa Networks for Dense IoT Deployments, ” 2018. [5] B. Reynde rs, Q. W ang, P . T uset-Peiro, X. V ilajo sana, and S. Pollin, “Im- provi ng Reliabi lity and S calability of L oRaW ANs T hrough Lightweight Scheduli ng, ” IEE E Internet of Things Journal , vol. PP , no. 99, pp. 1–1, 2018. [6] M. C. Bor , U. Roedi g, T . V oigt, and J. M. Alonso, “Do LoRa lo w- po wer wide-area networks scale?” in Pr ocee dings of the 19th AC M Internati onal Confer ence on Modeling , Analysis and Simulation of W ireless and Mobile Systems . A CM, 2016, pp. 59–67. [7] A.-I. Pop, U. Raza, P . Kulka rni, and M. Sooriyabandara , “Does bidi- rectio nal traf fic do m ore harm than good in LoRaW AN based LPW A netw orks?” arXiv prep rint arX iv:1704.041 74 , 2017. [8] J. Haxhibeqiri , F . V an den Abeele, I. Moerman, and J. Hoebeke, “LoRa scalabi lity: A simulation model based on interference m easurements, ” Sensors , vol. 17, no. 6, p. 1193, 2017. [9] G. Otto y , G. L eenders, and G. Calle baut, “LoRaW AN EFM32, ” doi: 10.5281/ze nodo.1209414 . [Online]. A va ilabl e: https:/ /github .com/DRAMCO/LoRaW AN EFM32 [10] G. Calle baut, “LoRaW AN Network Simulator , ” doi: 10.5281/ze nodo.1217124 . [Online]. A v ail able: https:/ /github .com/GillesC/LoRaEnergySim/tree/v0.1.0 [11] L. Casals, B. Mir , R. V idal , and C. Gomez, “Modeling the Energy Performance of LoRaW AN, ” Sensors , vol. 17, no. 10, p. 2364, 2017. [12] LoRa™ Modulation Basics , Semtech Std., 2015, aN1200.22. [Online]. A vail able: https://www .semtech.com/uplo ads/documents/an1200.22.pdf [13] SX1272/3 /6/7/8: LoRa Desig ner’ s Guide , Semte ch Std., 2013, aN1200.13. [Online]. A vai lable: https:/ /www .semtec h.com/uploa ds/documents/LoraDesignGuide ST D.pdf [14] SX1272/3 /6/7/8: LoRa Modem Low Energy Consumpt ion Design , Semtech Std., 2013, aN1200.17. [Online]. A vaila ble: https:/ /www .semtec h.com/uploa ds/documents/LoraLowEner gyDesign STD.pdf [15] J. Petajaj arvi, K. Mikhaylo v , A. Roiv ain en, T . Hanninen, and M. Pet- tissalo, “On the covera ge of LPW ANs: range e va luatio n and channe l atten uation model for LoRa technology , ” in ITS T elecommunicat ions (ITST), 2015 14th International Confer ence on . IEEE, 2015, pp. 55–59. [16] Compil ation of measurement data relating to buil ding entry loss , Internati onal T elecommunicat ion Union Std., 2015, r eport ITU-R P .2346-0. [Online]. A va ilable : https:/ /www .itu .int/dms pub/itu - r/opb/rep/R- REP- P .2346- 2 015- PDF- E.pdf [17] E. Damosso, Digital mobile radio towar ds futur e gene ration systems: COST action 231 . European Commission, 1999. [18] W iMOD iC880A datasheet , IMST Std., 2015. [Online]. A v ail able: https:/ /wireless- solutions.de/do wnloads/Radio- Modules/iC880A/iC880A Datashee t V0 50.pdf [19] SX1301 Datasheet , Semtech, 2017, v2.3.