EMF: Embedding Multiple Flows of Information in Existing Traffic for Concurrent Communication among Heterogeneous IoT Devices

Transcription

1 EMF: Embedding Muliple Flows of Informaion in Exising Traffic for Concurren Communicaion among Heerogeneous IoT Devices Zicheng Chi, Zhichuan Huang, Yao Yao, Tianian Xie, Hongyu Sun, and Ting Zhu Deparmen of Compuer Science and Elecrical Engineering Universiy of Maryland, Balimore Couny {zicheng, zhihu, yaoyaoumbc, xian, hongyu, Absrac The exponenially increasing number of IoT devices makes he unlicensed indusrial, scienific, and medical (ISM) radio bands (e.g., 2.4 GHz) exremely crowded. Currenly, here is no efficien soluion o coordinae he large amoun heerogeneous IoT devices ha have differen communicaion echnologies (e.g., WiFi and ). To fill his gap, in his paper, we inroduce embedded muliple flows (EMF) communicaion mehod, which (i) embeds differen pieces of informaion in exising raffic and (ii) concurrenly sends ou hese informaion from one IoT sender o muliple IoT receivers ha have a differen communicaion echnology from he sender. By doing his, our EMF mehod (i) enables cross-echnology communicaion among heerogeneous IoT devices, (ii) does no inroduce any exra conrol raffic, and (iii) is ransparen o he higher layer applicaions. Our approach is implemened on USRPs and commercial off-he-shelf (COTS) devices. We also conduced exensive experimens o evaluae our approach in real-world seings. The evaluaion resuls show ha EMF s hroughpu is more han 4 imes higher han he laes cross-echnology communicaion echnique (i.e. FreeBee[]). I. INTRODUCTION Based on Garner, 6.4 billion Inerne-of-Thing (IoT) devices will be in use in 26, and he number of IoT devices will exponenially increase and reach 2.8 billion in 22 [2]. On hese IoT devices, WiFi or is widely used. WiFi and boh operae in he same frequency band (i.e., 2.4 GHz). Therefore, he exponenially increasing number of IoT devices inroduce severe communicaion inerference among hese devices. In order o reduce he inerference and increase specrum uilizaion, more efficien communicaion coordinaion among hese devices is needed. However, WiFi and canno direcly communicae wih each oher because hey have oally differen physical layers. Tradiionally, communicaion beween wireless echnologies is achieved indirecly via gaeways equipped wih muliple radio inerfaces. This approach suffers from several issues including : (i) he cos o purchase he gaeway hardware, (ii) raffic overhead flowing ino and from gaeways, and (iii) deploymen complexiy relaed o posiioning he gaeway in a way ha saisfies user requiremens. To address hese issues, only a handful cross-echnology communicaion (CTC) echniques have been proposed, including Esense [3], GSense [4], HoWiES [5], and FreeBee[]. These approaches have heir own limiaions. Boh Esense and HoWiES need o inser dummy packes which inroduce exra raffic o already crowded 2.4 GHz specrum and reduce he specrum uilizaion. GSense need cusomized hardware plaform which preven i o be widely adoped. FreeBee only uilize beacon for WiFi Sender WiFi Traffic WiFi Receiver Informaion Informaion N Receiver Receiver N Sender Traffic Receiver Informaion Informaion N Receiver Receiver N (a) WiFi-o- (b) -o-wifi Fig. : Example of bi-direcional EMF communicaion (i.e., WiFi o and o WiFi). Figure (a) shows ha he WiFi sender can embed N pieces of informaion ino is regular WiFi raffic and concurrenly send ou hese informaion o N receivers when he WiFi sender is communicaing wih he WiFi receiver. Similarly, Figure (b) shows he reverse direcion. cross-echnology communicaion, herefore i has exremely low hroughpu and high laency. In his paper, we ake a dramaically differen approach by leveraging exising raffic for concurren cross-echnology communicaion. Specifically, we inroduce embedded muliple flows (EMF) communicaion mehod (shown in Figure ), which (i) embeds differen pieces of informaion in exising raffic and (ii) concurrenly sends ou hese informaion from one IoT sender o one or muliple IoT receivers ha have a differen communicaion echnology from he sender. By doing his, our EMF mehod (i) enables cross-echnology communicaion among heerogeneous IoT devices, (ii) does no inroduce any exra conrol raffic, and (iii) is ransparen o he higher layer applicaions. In summary, he main conribuions of his work are as follows: We propose a novel modulaion mechanism ha embeds informaion in exising raffic by slighly shifing he packes and/or flipping he packe-order o form a unique paern ha can represen arbirary srings of daa. By doing his, we enable wo devices wih oally differen radios (i.e., WiFi and ) communicae wih each oher and do no inroduce dummy raffic or modify he hardware. To he bes of our knowledge, his is he firs work ha leverages daa raffic for cross-echnology communicaion. To furher improve he specrum uilizaion, our advanced design leverages he independency among differen window sizes for embedding differen pieces of informaion in a sring of exising packes. In his way, our approach enables one IoT sender simulaneously ransmis muliple pieces of informaion o muliple IoT receivers ha have differen radios han he sender. For example, one WiFi device ransmis hree pieces WiFi

2 CM Deecing CM 2 WiFi # #3 #5 #4 #6 WiFi #2 WiFi # Deecing #3 #5 #4 #6 WiFi #2 Deecing WiFi # #3 #5 #4 #6 WiFi #2 (a) Time slo (b) Time slo 2 (c) Time slo 3 Fig. 2: Human aciviy or gesure recogniion hrough coordinaed IoT devices: In ime slo (see Figure 2(a)), WiFi device # sends packes o WiFi device #2. These packes also carry embedded conrol messages CM and CM2 o ell devices #3 and #5 saring heir ransmissions in ime slos 2 and 3, respecively. Afer receiving hese embedded conrol messages, devices sar o ransmi packes a heir own ime slos (i.e., slos 2 and 3 in Figures 2(b) and 2(c), respecively). All he raffic from WiFi and devices can be leveraged for sensing as well. In his way, we can also minimize he inerference among WiFi and devices due o heir overlapped channels. of message o hree differen devices. Our approach is symmeric and generic. I enables he bidirecional communicaion beween WiFi and devices (i.e., WiFi o and o WiFi). Moreover, our approach does no require any modificaions on he hardware and is compaible wih he commercial off-he-shelf (COTS) devices. We implemened our approach on WiFi and plaforms and conduced exensive evaluaion under real-world seings. The evaluaion resuls show ha EMF s hroughpu is more han 4 imes higher han he laes cross-echnology communicaion echnique (i.e. FreeBee[]). Moreover, EMF is robus agains he environmenal noise. The hroughpu is very sable across differen communicaion ranges (from.5 meer o 4 meers). II. MOTIVATION In his secion, we firs inroduce he poenial applicaions ha moivae our design, hen describe he main observaion ha provides he foundaion for our design. A. Moivaing Applicaions In his secion, we inroduce wo differen applicaions which demonsrae a wide range of benefis he EMF echnology has o offer. Human aciviy or gesure recogniion hrough coordinaed IoTs: Human aciviy or gesure recogniion hrough radio frequency (RF) sensing has been invesigaed by muliple researchers [6], [7], [8]. In order o achieve coninuous monioring over ime, hese recogniion sysems need o coninuously send ou he radio signals, which includes he arificial raffic ha will inerfere wih he surrounding IoT devices. To address his problem, we need o coordinae he radio signals across muliple IoT devices for he RF sensing over ime. Figure 2 illusraes he example of coordinaing WiFi and devices raffic for human aciviy recogniion by using our EMF communicaion mehod, which enables muliple flows from WiFi o devices. Augmened Realiy (AR): As shown in Figure 3, he compuer needs o ransmi huge volume of high qualiy Compuer Command WiFi-o- Graphic Daa WiFi-o-WiFi Command WiFi-o- -based Lamp AR Headse -based Thermosa Fig. 3: Augmened realiy applicaion Applicaion Laency Tolerance (ms) Regular Websies ( , news) -8 Heavy Websies (javascrip, images) 5-4 Web Based Remoe Sysems 3-3 Casual Games (Facebook, flash) 2- Acion Games (firs person shooers) -5 Remoe Adminisraion 5-5 TABLE I: Laency olerance range of nework applicaions [9] graphic daa o he AR headse. These graphic daa will occupy he channel all he ime and prohibi he communicaions of nearby devices ha have overlapped frequency band as he WiFi devices. To address his problem, we can use our EMF communicaion mehod ha embeds he commands from he WiFi sender (i.e., compuer in Figure 3) o devices in he WiFi raffic (i.e., graphic daa packes). In his way, we can conrol he -based smar lamp and smar hermosa by using regular WiFi raffic and do no need explici conrol raffic or cusomized hardware (e.g., gaeway for WiFi o communicaion). B. Moivaing Observaion In his secion, we inroduce he main observaion ha provides foundaion for our design. Observaion: Mos of he nework applicaions have wide range of laency olerance (from ens of milliseconds o hundreds of milliseconds, see Table I). Therefore, slighly shifing he packe ransmission will no affec he performance of hese applicaions. Based on he above observaion, we can slighly shif he packe ransmission or flip he packe ransmission order. Our goal is o embed he cross-echnology messages in exising raffic and wih negligible impac on he original raffic. For example, we wan o enable he concurren communicaions from WiFi o muliple devices using exising WiFi raffic and inroduce minimum delay o he exising WiFi raffic. By doing his, our approach (EMF) will no affec he higher layer applicaions. Similarly, we can do he o WiFi communicaion. Since mos of he applicaions have less resricion on he delay, i is more flexible o send ou packes. III. DESIGN CHALLENGES AND SYSTEM OVERVIEW In his secion, we firs discuss he design challenges and hen provide overview of EMF o address hese challenges. A. Challenges In order o achieve he muliplex cross-echnology communicaion, we have o address he following challenges: C. How o efficienly ransmi informaion beween WiFi and devices? Since WiFi and use oally differen physical layer, hey canno direcly communicae wih each oher. The mos laes echnique (i.e., FreeBee [])

3 enables WiFi o communicaion by using beacons insead of daa packes (due o he complexiy of daa packes). However, he major par of wireless raffic is daa packes. Therefore, FreeBee has exremely low hroughpu (i.e., as low as 7 bps). To enable efficien communicaion beween WiFi and, we need o uilize all he raffic. Unlike beacons, he challenge is ha he daa packes size and ransmission ime are dynamically changing over ime. To address his challenge, we propose a novel packe reordering scheme ha conains wo pars: packe shifing and packe order flipping (deailed in Secion IV-B). C2. How o minimize he bi error rae under unpredicable raffic? Oher han hroughpu, bi error rae (BER) is anoher imporan meric ha is used in communicaion sysems. BER usually depends on Signal-o-Noise Raio (SNR) and modulaion schemes. I reflecs he discriminaion abiliy of differen saes. Since we uilize he exising raffic, which is unpredicable, o convey embedded messages, he wo saes (represening and ) hemselves are no sable. Therefore, i will inroduce high BER wihou a careful design of he modulaion scheme. To address his challenge, we propose a novel packes scheduling algorihm in our modulaion scheme o minimize he BER and have minimum impac o exising raffic (deailed in Secion IV-B). C3. How o enable muliple flows of informaion from WiFi o devices and vice versa wih one sream of exising raffic? In order o more efficienly uilize he specrum, i will be beer if we can embed muliple pieces of informaion in exising raffic and send hem o differen receivers. However, he challenge is ha we only have one sream of exising raffic. To address his challenge, we propose an advanced design ha uilizes he independency of muliplied window sizes o convey differen messages on a single ougoing packes series o ransmi o eiher muliple receivers or one single receiver (deailed in Secion V). B. Sysem Overview In his secion, we inroduce he sysem archiecure of EMF (shown in Figure 4). We firs describe he basic design for one-o-one communicaion (i.e., one WiFi device communicae wih one device and vice versa), hen inroduce he advanced design for one-o-many communicaion (i.e., one WiFi device sends ou muliple messages o differen devices and vice versa). In he res of he paper, unless explicily saed oherwise, we will use WiFi as he sender and as he receiver o describe he WiFi o EMF communicaion jus for clariy purpose. By exchanging he erms WiFi and, we can ge he o WiFi EMF communicaion. One-o-one communicaion: As shown in Figure 4, he WiFi o EMF communicaion conains wo pars: modulaion a he WiFi sender side and demodulaion a he receiver side. On he WiFi sender side, we add an EMF modulaor (deailed in Secion IV-B) which rearranges he packes sending orders wihin he buffer based on he informaion needs o be sen from he WiFi device o he device. As described in EMF Daa Sender Nework Layer EMF Modulaor Normal WiFi Daa WiFi Buffer Link & PHY Layers Sending Queue Receiver N Receiver 2 Receiver EMF Daa EMF Demodulaor RSSI Sampler Link & PHY Layers Fig. 4: Overview of WiFi o EMF sysem archiecure. By exchanging he erms WiFi and in he above figure, we can also ge he o WiFi EMF sysem archiecure. Secion II-B, if we can conrol he packes laency (deailed in Secion VI) wihin he applicaions wide olerance range, EMF makes negligible impac on he ransmission of exising WiFi raffic. Therefore, EMF is ransparen o he upper layers (i.e., nework and applicaion layers). On he receiver side, since he noise floor measuremen or received signal srengh (RSS) measuremen module (for channel deecion purpose) is popular on off-he-shelf radios, we uilize he RSS reading sequence from he sampler (shown on he righ hand side of Figure 4) o demodulae he embedded informaion from WiFi sender (deailed in Secion IV-C). One-o-many communicaion: The mos aracive feaure of EMF one-o-many communicaion is ha i can embed muliple pieces of informaion in exising WiFi raffic and send hem o differen devices (shown in Figure 4) wih minimum impac o original WiFi raffic. To enable his feaure, he WiFi sender needs o adjus he daa packes o embed differen messages based on differen window sizes and he receivers will measure he RSS readings wih corresponding window sizes (deailed in Secion V). IV. BASIC DESIGN: ONE-TO-ONE COMMUNICATION In his secion, we inroduce he deailed design of one-oone communicaion. The definiions of noaions used in he res of his paper are lised in Table II. A. Communicaion Esablishmen The main purpose of communicaion esablishmen is o synchronize he receiver wih he sender. We use a simple folding approach which is originally proposed o search for weak pulsars in he radio noise picked up by elescopes []. In wireless communicaion sysem, beacons are periodical packes o announce he presence of a wireless device for ohers o discover i. We uilize his periodic feaure of beacons o synchronize he receiver wih he sender, so he firs sep is o idenify he beacons. Wihou loss of generaliy, he measured received signal srengh (RSS) values over ime can be represened as RSS(). When hese RSS readings are chopped wih ime period P, hey can form a marix RSS (i, j), where i [, P ], j = /P. Then, we build he hisogram h(i) of RSS (i, j) by using he following equaion: j h(i) = RSS (i, n) () n=

4 Normalized RSS (-) Beacons locaed Tp(m,n) Hisogram Fig. 5: An example of folding Tg(m,n) Ts symbol() Tb Beacon Daa Packe Fig. 6: Modulaion of EMF symbol(m) 2.7 MAX Based on he maximum value of h(i), we can idenify he locaion of beacon i = arg max(h(k)), because only he k periodical beacons RSS are aggregaed ogeher afer folding. Therefore, we can synchronize he receiver wih he sender. Figure 5 shows an example of he folding process o locae he beacons. By looking a he original ime series, i is hard for he receiver o idenify he beacons. Afer dividing he whole series by he period P = 5 and calculae he hisogram, i is easy o idenify he maximum value where he index is he locaion of beacons. B. Modulaion In his secion, we inroduce he modulaion scheme of EMF by leveraging he exising WiFi packes. Due o various nework applicaions and proocols, he lengh of packes and packes ransmission inervals are dynamically changing, which inroduce challenges and also opporuniies for our design. As described in Secion IV-A, he receiver can idenify when he sender s periodical beacons come. Therefore, he sender akes he beacon as a ime synchronizaion flag o modulae daa. As shown in Figure 6, he duraion beween wo adjacen beacon is denoed as T b, which can be divided ino M pieces and each piece is called symbol duraion window size T s, where T s = T b /M and symbol is he minimum ransmission uni (i.e., each symbol represens - bi informaion from he sender o he receiver). We furher divide T s ino wo pars, where he duraion of upper half and lower half is denoed as T u and T l, respecively. For clariy purpose, we define he raffic occupancy raio R o help us describe he modulaion scheme. Noaions T b T s T u T l R u R l T p T g λ τ Definiions Time duraion beween wo adjacen beacons Symbol duraion window size Time duraion of upper half wihin one symbol Time duraion of lower half wihin one symbol Traffic occupancy raio of upper half Traffic occupancy raio of lower half Time duraion of ransmiing a packe Time inerval beween wo packes Symbol value Lower bound of T g TABLE II: Definiions of Noaions Ru=2/5 Rl= Ru= Rl=2/5 (a) λ= (b) Shifed, λ= Fig. 7: An example of he shifing operaion Ru=3/5 Rl=2/5 2 3 (a) λ= (b) Flipped, λ= Fig. 8: An example of he flipping operaion Definiion: The raffic occupancy raio R denoes he raio of oal packes duraion o he oal ime duraion which can be calculaed by using he following equaion: N T p (m, n) n= R(m) = (2) N T g (m, n) + N T p (m, n) n= n= Where T p (m, n) is he ime duraion of ransmiing he nh packe and T g (m, n) is he packe ransmission inerval beween he nh and (n + )h packes (n [, N], N is he oal number of packes ransmied during ime inerval T ). m [, M], M is he oal number of symbols need o be ransmied by he sender. Wih he above definiion, we can modulae one symbol by comparing he raffic occupancy raio wihin he upper half ime duraion (T u ) and lower half ime duraion (T l ) o decide wheher his symbol is or using he following equaion: { if R u (m) > R l (m) λ(m) = (3) if R u (m) R l (m) Ru=/5 Rl=4/5 Where λ(m) is he value of he mh symbol, R u (m) and R l (m) is he raffic occupancy raio of he upper and lower half of he mh symbol, respecively. Figures 7(a) and 8(a) illusrae simple examples of one symbol. In Figure 7(a), since R u = 2/5 and R l =, R u > R l. Therefore, he symbol value λ =. Similarly, in Figure 8(a), R u = 3/5, R l = 2/5, and R u > R l, herefore λ =. Since he packe lengh and inervals beween packes are deermined by nework and upper layers, in order o ransmi an arbirary bi, we need o change he raffic occupancy raio. We propose a novel modulaion scheme which combines he following wo operaions: Shifing: As shown in Figure 7(a), if he original packeorder yields λ =, bu he sender wans o ransmi, he sender needs o shif he packe from he upper half o he lower half (as shown in Figure 7(b)) as shown in Figure 4. Similarly, if he original packe-order yields λ =, bu he sender wans o ransmi, he sender also can shif he packe from he lower half o he upper half. Shifing operaion works when here exiss enough packes inerval (i.e., whie space). If he whie space is limied, we need o use he flipping operaion. Flipping: As shown in Figure 8(a), if he original packeorder yields λ =, bu he sender wans o ransmi, he sender needs o flip he packe(s) from he upper half o he 3

5 Algorihm One-o-one communicaion scheduling Inpu: D = {T p(),, T p(n)}, T s, τ, and λ. Oupu: Y = {T b (),, T b (n)}. n : if (T p(i) + (i )τ) T s/2 hen i= 2: S = ; 3: else 4: L =, S = ; 5: while L < T s/2 do 6: i = arg max k 7: L = L + T p(i) + τ, S(i) =, D = D {T p(i)}; 8: end while 9: end if : T u =, T l = T s/2; : for i = o n do 2: if λ = S(i) hen 3: T b (i) = T u, T u = T u + T p(i) + τ; 4: else 5: T b (i) = T l, T l = T l + T p(i) + τ; 6: end if 7: end for Symbol Symbol 2 Symbol 3 Symbol 4 λ()= λ(2)= λ(3)= λ(4)= Ru < Rl Ru > Rl Ru > Rl Ru > Rl Fig. 9: An example of demodulaion wih empirical daa lower half (shown in Figure 8(b)). Similarly, if he original packe-order yields λ =, bu he sender wans o ransmi, he sender also can flip he packe(s) from he lower half o he upper half. To generalize our modulaion scheme and minimize is overhead, we propose a lighweigh algorihm (see Algorihm ) o decide how o schedule he packes in he ougoing buffer for modulaion purpose. Where τ is he minimum ime inerval beween wo packes ha is deermined by he physical layer. In order o minimize he bi error rae, we need o maximize he difference beween R u and R l. We firs decide wheher flipping operaion is needed (Line ). If he sum of n packes lenghs and n number of minimum packe inervals is equal or less hen half of a window size (i.e., T s /2), flipping is no required. This is because all he packes can be ransmied wihin T s /2 which resuls in he highes difference beween R u and R l. Oherwise, we calculae he possible combinaions of packes o reach he poenial maximum raffic occupancy raio for half a window size T s /2 and leave he res packes o he oher half of window size (Lines 4 o 8). Finally, we perform he combinaion of shifing and flipping operaions based on he symbol value (λ) (Lines o 7). The ime complexiy of Algorihm is O(n) which can easily be implemened on common IoT devices. Where, n is he number of packe during window size T s. The shifing and flipping operaions will inroduce delay. Their impacs on he performance will be analyzed in Secion VI. C. Demodulaion In his secion, we inroduce how o demodulae he message by leveraging he widely available received signal srengh S(m) S2(m/2) S(m+) Ts T2s Fig. : Modulaion of one-o-many EMF (RSS). By using he communicaion esablishmen echnique described in Secion IV-A, he receiver can be synchronized wih he sender. Afer measuring he RSS values for a duraion of T b beween wo adjacen beacons, he receiver can divide he measured RSS values ino wo pars (i.e., he upper half and lower half over ime) by using he same mehod we described in he modulaion secion, hen calculae he upper half s and lower half s raffic occupancy raios (i.e., R u (m) and R l (m), respecively) by applying Equaion 2. Finally, he receiver is able o demodulae all he symbols by applying Equaion 3. Figure 9 shows an example of he demodulaion process. The RSS readings are measured by a complian TelosB device and he signal is ransmied by a WiFi device using 82.g complian USRP. As shown in he figure, wihin he ime duraion of symbol, R u is less han R l. Thererfore, he TelosB device ges λ() =. By applying he same mehod, he TelosB device ges λ =. V. ADVANCED DESIGN: ONE-TO-MANY COMMUNICATION Build on op of he one-o-one communicaion described in Secion IV, we inroduce one-o-many communicaion which is one of he unique feaures of our EMF mehod. Due o he space consrain, we only inroduce he modulaion of oneo-many communicaion in his secion. The communicaion esablishmen is he same as one-o-one communicaion in Secion IV-A. When he receivers demodulae he signal, hey choose differen window size T s o obain corresponding message from he sender and apply he same scheme as described in Secion IV-C. As shown in Figure, he symbol sequence S(m) has symbol duraion T s, which can be divided ino upper half duraion T u and lower half duraion T l. We can also embed anoher symbol sequence S2(m/2) which has wice of he symbol duraion as S (i.e., T 2 s = 2 T s and T 2 u = T 2 l = T s ). Because of he independence of symbol duraion T s wih a facor of 2 and he feaure of relaive difference beween raffic occupancy raio of upper half R u and raffic occupancy raio of lower half R l, we can modulae S and S2 on he same original raffic wihou inerfere wih each oher. Figure shows an example of our one-o-wo modulaion scheme. The corresponding deailed descripion of differen cases is shown in Table III. Because S2 s ime duraion is wice of S s ime duraion, S2 includes wo bis of symbols while S includes one bi. We show four differen combinaions o illusrae he independence of S and S2. The number of messages can be furher expanded o hree or more. We propose he general algorihm ha can modulae m pieces of messages in Algorihm 2. The iniial symbols Λ are calculaed wih given D and τ based on Equaion (3) (Line ). Where D is he se of packe ransmission duraion and

6 Ru=3/5 Rl=/5 Ru2=2/5 Rl2=/ Ru3=4/ Rl3=3/ (a) λ =, λ 2 = Ru=/5 Rl=3/5 Ru2=/5 Rl2=2/ Ru3=4/ Rl3=3/ Ru=3/5 Rl2= Ru2=3/5 Rl2=/ Ru3=3/ Rl3=4/ (b) Shifed, λ =, λ 2 = Ru= Rl=3/5 Ru2=/5 Rl2=3/ Ru3=3/ Rl3=4/ (c) Flipped, λ =, λ 2 = (d) Shifed & flipped, λ =, λ 2 = Fig. : An example of one-o-wo modulaion: wo messages S and S2 are ransmied by using windows sizes 5 and, respecively. Wih differen shifing and/or flipping operaions for packes # o #5, arbirary λ and λ 2 can be ransmied. For example, in Figure (a), he original λ 2 =. In order o ransmi λ 2 =, we need o shif packe #3 (shown in Figure (b)). More deailed combinaions are shown in Table III. Case Raio R Raio R Raio R Figure of S() of S(2) of S2() λ λ 2 (a) R u > R l R u2 > R l2 R u3 > R l3 (b) R u > R l R u2 > R l2 R u3 < R l3 (c) R u < R l R u2 < R l2 R u3 > R l3 (d) R u < R l R u2 < R l2 R u3 < R l3 TABLE III: Four differen cases of ransmiing wo pieces of arbirary informaion τ is he minimum ime inerval beween wo packes ha is deermined by he physical layer. For each piece of message, if he iniial symbol values λ i do no equal o λ i and he packes can be shifed, we shif he shores packe unil λ i = λ i (Lines 2-6). Oherwise, we flip he packes beween upper half and lower half (Lines 7-). Finally we apply Algorihm for all minimum windows (Lines -3). The ime complexiy of Algorihm 2 is O(nlogn), which can easily be implemened on common IoT devices. Where n is he number of packe during window size T s. VI. PERFORMANCE ANALYSIS In his secion, we mahemaically analyze differen facors ha will affec he communicaion performance and he impac on exising ransmission. RSS sampling rae: To ensure he capure of he smalles packes and inervals beween wo packes, he RSS sampling rae mus mee he following requiremen: f rss 2 MAX( min(t p (m, n)), min(t g (m, n)) ) (4) Where f rss is he RSS sampling rae, min(t p (m, n)) and min(t g (m, n)) are he ransmission ime duraion of he minimum packe lengh and minimum ransmission inerval beween wo packes, respecively. The COTS IoT radios usually provide relaively high RSS sampling rae. For example, CC242 (a common radio) provides RSS sampling rae up o 64 khz [], which is corresponding o 5.6 µs sampling period ha is far less han a WiFi packe ransmission duraion. For example, he minimum packe duraion in 82.g is 92 µs [2]. Laency: As menioned in Secion IV-B, EMF may inroduce laency due o he packe(s) shifing and/or packe-order flipping. To minimize he impac on exising WiFi raffic, ideally, we need o ensure ha he inroduced laency is less han Algorihm 2 One-o-many communicaion scheduling Inpu: D = {T p(),, T p(n)}, Λ = {λ,, λ m}, T s, and τ. Oupu: Y = {T b (),, T b (n)}. : Calculae Λ wih D and τ based on Equaion (3); 2: for i = m o 2 do 3: while λ i λ i τ i + R u + R l T s i do 4: Shif shores packe beween upper and lower half; 5: Updae λ i based on shifed resuls; 6: end while 7: if λ i λ i hen 8: Flip he upper half and lower half; 9: end if : end for : for i = o 2 m do 2: Apply Algorihm for ime slo [T s (i ), T s i]; 3: end for he lower bound in Table I under differen applicaions. The laency is direcly relaed o he ime window T s. The expeced laency of flipping and shifing operaion is /4T s and /2T s, respecively. Therefore, o minimize he impac on he exising raffic, ideally, we need o ensure ha he maximum laency /2T s is less han he lower bound of differen nework applicaions in Table I. However, we also noe ha mos of he nework applicaions have wide range of laency olerance (shown in Table I). Aggregaed hroughpu: Afer saisfying Equaion 4, we can achieve he expeced T hroughpu = /T s. Because he paricular modulaion we use (inroduced in Secion V), we are able o ransmi N pieces of message simulaneously (one-omany EMF). So he aggregaed hroughpu is as follows: Aggregaed hroughpu = N n= T s /2 n (5) VII. EVALUATION In his secion, we firs inroduce he experimenal seup, hen evaluae one-o-one EMF wih differen window size T s and communicaion range under differen scenarios. We also evaluae he performance of one-o-many EMF. The following wo merics are used o evaluae he wireless communicaion performance: Throughpu: The number of correcly received daa in erms of bi per second (bps). Bi error rae (BER): The number of bis no correcly received divided by he oal number of ransmied bis during he cross-echnology communicaion. A. Experimenal Seup To prove he concep, we implemened he EMF modulaion and demodulaion scheme on 82.g [3] WiFi complian USRP B2 and complian TelosB devices. The using of USRP can help us o beer evaluae he performance of our sysem since we can fully conrol he PHY layer. In he WiFi o communicaion experimens, he USRP coninuously ransmied packes wih varying packe lenghs (from byes o,5 byes) based on he races of real-world WiFi raffic ha we colleced. The devices sample he received signal srengh and esablish he communicaion by

7 WiFi-o- -o-wifi Receiver 4m WiFi Sender 5m 3m Receiver WiFi Sender Concree Wall (a) Line-of-Sigh (b) None-Line-of-Sigh Fig. 2: Experimens Seup EMF FreeBee A-FreeBee Throughpu (bps) Fig. 3: Comparison beween EMF and sae-of-he-ar cross echnology communicaion FreeBee []. applying beacon folding (deailed in Secion IV-A). Afer he communicaion is esablished, he receivers demodulae (deailed in Secion IV-C) he signal and ge he received daa. We compare he received daa wih he originally ransmied daa o calculae he bi error rae. We evaluaed our EMF sysem in an engineering building, which has a lo of oher WiFi access poins ha creae inerference. To evaluae he robusness of our approach, he USRP uses he mos popular WiFi channel and he ransmission power is 25 dbm. The channel is se o be channel 2, which is overlapped wih WiFi channel. devices sample he RSSI wih 32 khz sampling rae. In each experimen, 4, symbols are ransmied. The experimen is repeaed for imes. Two scenarios are evaluaed as follows: Line-of-Sigh (LoS): As shown in Figure 2(a), he LoS scenario is a hallway which is 45 meers long and 4.3 meers wide. We fixed he WiFi sender a one end of he hallway and move he receivers so ha hey are.5m, 3m, 5m, m, 2m, 3m, and 4m away from he sender. Non-Line-of-Sigh (NLoS): As shown in Figure 2(b), in his scenario, he sender and receiver are deployed in wo adjacen rooms which are separaed by a concree wall. B. Evaluaion of one-o-one EMF Communicaion In his secion, we firs compare our approach wih he mos laes echnique (i.e., FreeBee[]), hen evaluae he performance of our approach under diverse seings. ) Comparison wih sae-of-he-ar: Since our work is he firs work ha enables he muliple flows of informaion in cross-echnology communicaion, he sae-of-he-ar research is complemenary, bu provides no appropriae baseline for comparison. Therefore, we can only compare our work wih he mos relaed and laes work (i.e., FreeBee[]) under he same seing wih one flow of informaion. The resuls are shown in Figure 3. Based on hese resuls, EMF is 6.4 and 4.4 imes beer han he FreeBee and A-FreeBee in erms of hroughpu during he WiFi o communicaion. During Throughpu (bps) NLoS LoS Communicaion Range (meer) (a) Throughpu BER (%) NLoS LoS Communicaion Range (meer) (b) BER (he Y-axis is in log scale) Fig. 4: Throughpu and BER of one-o-one EMF: The performance in NLoS scenario is slighly lower han in LoS in mos circumsances, bu hey are very close o each oher when disance increases. Throughpu (bps) BER (%) T s =5ms T s =ms T s =2ms T s =4ms Communicaion Range (meer) (a) Throughpu T s =5ms T s =ms T s =2ms T s =4ms Communicaion Range (meer) (b) BER (he Y-axis is in log scale) Fig. 5: Impac of window size T s: The hroughpu is half when he window size T s doubled (because he modulaion mechanism we used) and i is very sable across all he communicaion range (from.5m o 4m) which shows our EMF modulaion scheme is robus. he o WiFi communicaion, EMF is 7.7 and 5.4 imes beer han he FreeBee and A-FreeBee in erms of hroughpu. This is because we leverage he daa raffic insead of jus using beacons. Moreover, our approach can concurrenly send muliple packes wih differen informaion o differen devices, while FreeBee does no suppor i. 2) Performance in LoS and NLoS Scenarios: As shown in Figure 4(a), he hroughpu under NLoS scenario does no have significan difference comparing o LoS scenario. We hink i is also because of he robusness of he EMF modulaion scheme as we menioned before. An ineresing observaion is he BER (as shown in Figure 4(b)) has 2.5% difference beween NLoS and LoS (i.e., NLoS>LoS). However, when he disance increases, he difference is less han % and he rend is inversed (i.e., LoS>NLoS). We also noice ha he sandard deviaion of boh hroughpu and BER in NLoS scenario is larger han ha in LoS scenario. These migh be because of ha he aenuaion and mulipah fading in NLoS scenario is more complicaed han in LoS scenario. 3) Impac of Window Size T s : We firs invesigaed he impac of window size T s (in LoS scenario) which is he mos imporan parameer relaed o he communicaion per-

8 Throughpu (bps) Receiver # Receiver #2 Receiver #3 Throughpu (bps) Receiver # Receiver #2 Receiver # Communicaion Range (meer) (a) Throughpu Receiver # Receiver #2 Receiver # Communicaion Range (meer) (a) Throughpu Receiver # Receiver #2 Receiver #3 BER (%) BER (%) Communicaion Range (meer) (b) BER (he Y-axis is in log scale) Fig. 6: Throughpu and BER of one-o-many EMF in LoS scenario: he relaionship of hroughpu for hree receivers is as expeced ha he hroughpu of he firs receiver is wo and four imes of he second and hird receiver (as menioned in Secion V), respecively, across all he communicaion ranges. The sabiliy shows he robusness of he modulaion scheme. formance (as discussed in Secion VI). As shown in Figure 5(a), he hroughpu is halved when he window size T s is doubled (because he modulaion mechanism we used). The highes hroughpu is 28 bps (T s = 5ms). The hroughpu is very sable from.5m o 4m. The resul shows ha our EMF modulaion and demodulaion scheme is robus. This is mainly because he scheme is based on he relaive raffic occupancy raio and his raio is widely deecable even when he signal srengh is geing lower. Figure 5(b) shows he BER of differen window size. We can observe he rend ha he BER slighly increases when he range increases. The BER is less han % when he range is shorer han meers and less han 4% when he range is shorer han 4 meers. A a same range, basically, he BER decreases when he window size T s increases, his is because he longer he window size is, he more packes can be operaed which yields large difference on R u (raffic occupancy raio of upper half) and R l (raffic occupancy raio of lower half). C. One-o-many EMF Performance The mos aracive feaure of EMF is one-o-many communicaion. In his secion, we show he resuls of one-o-many communicaion. ) Line-of-sigh Scenario: As shown in Figure 6(a), he highes hroughpu is 23 bps, 2 bps and 5 bps for hree receivers a.5m and 9 bps, 98 bps and 48 bps a 4m, respecively. The BER does no exceed % for all he ranges. We noice he hroughpu changing rend of one-omany along communicaion range is as sable as in one-o-one communicaion, bu he BER (as shown in Figure 6(b)) is a lile bi higher han in one-o-one communicaion. Inuiively, i is because managing hree ransmission makes he packe Communicaion Range (meer) (b) BER (he Y-axis is in log scale) Fig. 7: Throughpu and BER of one-o-many EMF in NLoS scenario: we noe ha he hroughpu has he rend of decreasing and he BER has he rend of increasing. We can see he concree wall has big impac on one-o-many communicaion Aggregaed Throughpu (bps) Receiver # Receiver #2 Receiver # Communicaion Range (meer) Fig. 8: Aggregaed hroughpu of hree receivers: he highes aggregaed is 356 bps a.5m communicaion range and he hroughpu is 344 bps a 4m. order conrol less flexible han managing one ransmission so ha he difference of wo raffic occupancy raios (R u and R l ) is smaller. In Algorihm 2, he shifing and flipping operaions for larger window size may slighly reduce he BER for smaller window size. 2) Non-line-of-sigh Scenario: In NLoS scenario, he highes hroughpu in NLoS scenario is 88 bps, 3 bps and 5 bps for hree receivers a.5m and 66 bps, 95 bps and 5 bps a 4m, respecively (as shown in Figure 7(a)). The BER increases along wih he communicaion range which reaches 9% a 4m (as shown in Figure 7(b)). The resuls show he concree wall in he middle of sender and receivers makes relaive high impac on one-o-many communicaion. We hink his is because of ha he mulipah fading reduces he difference beween wo raffic occupancy raios R h and R l in NLoS scenario. For he hree receivers a he same communicaion range, we observe he performance of he firs receiver decreases when he hird receiver is relaive sable due o larger window size. 3) Aggregaed Throughpu: Theoreically, EMF can suppor much more concurren receivers o receive differen messages. In our evaluaion, we evaluaed hree concurren receivers and he experimenal resuls are shown in Figure 8. The figure shows he aggregaed hroughpu of hree receivers can achieve 356 bps when receivers are close o he

9 sender and 344 bps when receivers are even 4m far away from he sender. The green and yellow pars are he benefis by using our one-o-many scheme. VIII. RELATED WORK The mos relaed work are in he following wo caegories: Neworking performance opimizaion wihin one communicaion echnology: In he mos crowded 2.4GHz ISM band, researchers have invesigaed how o improve specrum uilizaion [4], [5], [6], [7], [8]. Anoher bunch of papers invesigaed on he collision avoidance mehods [9], [2], [2]. Insead of avoiding inerference by channel sensing ype of scheme, our work invesigae how o esablish he communicaion link across echnologies by leveraging he RSSI framework which is widely available on COTS devices. We noe ha RSSI and CSI have been uilized in various applicaions such as indoor localizaion [22], [23], [24], [25], [26], human aciviy recogniion [6], [7], [8], and crossechnology communicaion. Cross-echnology communicaion sysems: Many works uilize he coexisence of differen echnologies on a same band (e.g., WiFi and are on he same 2.4GHz band) o help each oher [27], [28], [29]. And several cross-echnology communicaion sysems [3], [5], [], [4] have been inroduced. Esense [3] and HoWiES [5] enables WiFi o communicaion by modulaing on he packe lengh of WiFi packes. GSense [4] uses special preamble o deliver coordinaion messages. FreeBee [] leverages beacons for cross-echnology communicaion. B 2 W 2 [3] enables muliple BLE o WiFi communicaions by using of CSI. Differen from he above approaches, we explore he possibiliy of embedding muliple flows of informaion in exising raffic for concurren communicaion among heerogeneous IoT devices. Our approach has he poenial o concurrenly coordinae muliple IoT devices and enable muliple applicaions simulaneously. IX. CONCLUSION The increasing number of IoT devices and limied available specrum moivae us o leverage exising raffic for concurren cross-echnology communicaion wih muliple flows of informaion. Specifically, we inroduce EMF communicaion mehod, which (i) embeds differen pieces of informaion in exising raffic and (ii) concurrenly sends ou hese informaion from one IoT sender o muliple IoT receivers ha have a differen communicaion echnology from he sender. By doing his, our EMF mehod (i) enables crossechnology communicaion among heerogeneous IoT devices, (ii) does no inroduce any exra conrol raffic, and (iii) is ransparen o he higher layer applicaions. We conduced exensive experimens o evaluae our approach in real-world seings. The evaluaion resuls show ha EMFs hroughpu is more han 4 imes higher han he laes cross-echnology communicaion echnique (i.e. FreeBee []). Moreover, our approach can concurrenly send muliple packes wih differen informaion o differen devices, while FreeBee does no suppor i. ACKNOWLEDGMENT This projec is suppored by NSF grans CNS-5359 and CNS REFERENCES [] S. M. Kim and T. He, Freebee: Cross-echnology communicaion via free side-channel, in MobiCom 25. [2] hp:// [3] K. Chebrolu and A. Dhekne, Esense: Communicaion hrough energy sensing, in MobiCom, 29. [4] X. Zhang and K. G. Shin, Gap sense: Lighweigh coordinaion of heerogeneous wireless devices, in INFOCOM, 23. [5] Y. Zhang and Q. Li, Howies: A holisic approach o zigbee assised wifi energy savings in mobile devices, in INFOCOM, 23. [6] Y. Li and T. Zhu, Gai-Based Wi-Fi Signaures for Privacy-Preserving, in ASIACCS, 26. [7] Y. Li and T. Zhu, Using Wi-Fi Signals o Characerize Human Gai for Idenificaion and Aciviy Monioring, in CHASE, 26. [8] Z. Chi, Y. Yao, T. Xie, Z. Huang, M. Hammond, and T. Zhu, Harmony: Exploiing coarse-grained received signal srengh from io devices for human aciviy recogniion, in ICNP, 26. [9] hps:// [] R. LOVELACE, J. SUTTON, and E. SALPETER, Digial search mehods for pulsars, Naure, vol. 222, pp , Apr 969. [] hp:// [2] C.-J. M. Liang, N. B. Priyanha, J. Liu, and A. Terzis, Surviving wi-fi inerference in low power zigbee neworks, in SenSys, 2. [3] hps://gihub.com/basibl/gr-ieee82-. [4] L. Cheng, Y. Gu, J. Niu, T. Zhu, C. Liu, Q. Zhang, T. He, Taming Collisions for Delay Reducion in Low-Duy-Cycle Wireless Sensor Neworks, in INFOCOM, 26. [5] J. Jun, L. Chen, L. He, Y. Gu, and T. Zhu, Exploiing Sender-based Link Correlaion in Wireless Sensor Neworks, in IEEE ICNP, 24. [6] S. Guo, S. Min Kim, T. Zhu, Y. Gu, and T. He, Correlaed Flooding in Low-Duy-Cycle Wireless Sensor Neworks, in IEEE ICNP, 2. [7] Luo, Hongyao and Huang, Zhichuan and Zhu, Ting, A survey on specrum uilizaion in wireless sensor neworks, Journal of Sensors, vol. 25, 25. [8] S. Kumar, D. Cifuenes, S. Gollakoa, and D. Kaabi, Bringing crosslayer mimo o oday s wireless lans, in SIGCOMM, 23. [9] J. Shi, E. Aryafar, T. Salonidis, and E. W. Knighly, Synchronized csma conenion: Model, implemenaion and evaluaion, in INFOCOM, 29. [2] S. Sen, R. R. Choudhury, and S. Nelakudii, Csma/cn: Carrier sense muliple access wih collision noificaion, IEEE/ACM Trans. New., vol. 2, no. 2, pp , Apr. 22. [2] T. H. T. Zhu, Z. Zhong and Z. Zhang, Exploring Link Correlaion for Efficien Flooding in Wireless Sensor Neworks, in NSDI, 2. [22] Q. Zhang, Z. Zhou, W. Xu, J. Qi, C. Guo, P. Yi, T. Zhu and S. Xiao, Fingerprin-free Tracking wih Dynamic Enhanced Field Division, in IEEE INFOCOM, 25. [23] P. Yi, M. Yu, Z. Zhou, Q. Zhang, W. Xu, and T. Zhu, A Three- Dimensional Wireless Indoor Localizaion Sysem, Journal of Elecrical and Compuer Engineering, 24. [24] D. W. Z. Zhong, T. Zhu and T. He, Tracking wih Unreliable Node Sequence, in IEEE INFOCOM, 29. [25] J. Jun, Y. Gu, L. Cheng, B. Lu, J. Sun, T. Zhu, and J. Niu, SocialLoc: Improving Indoor Localizaion wih Social Sensing, in ACM SenSys, 23. [26] Q. Zhang, W. Xu, Z. Huang, Z. Zhou, T. Zhu, P. Yi, and S. Xiao., Conex-Cenric Targe Localizaion wih Opimal Anchor Deploymens, in IEEE Inernaional Conference on Nework Proocols, 25. [27] T. Hao, R. Zhou, G. Xing, M. W. Muka, and J. Chen, Wizsync: Exploiing wi-fi infrasrucure for clock synchronizaion in wireless sensor neworks, IEEE Transacions on Mobile Compuing, vol. 3, no. 6, pp , June 24. [28] T. Jin, G. Noubir, and B. Sheng, Wizi-cloud: Applicaion-ransparen dual zigbee-wifi radios for low power inerne access, in INFOCOM, 2. [29] R. Zhou, Y. Xiong, G. Xing, L. Sun, and J. Ma, Zifi: Wireless lan discovery via zigbee inerference signaures, in MobiCom, 2. [3] Z. Chi, Y. Li, H. Sun, Y. Yao, Z. Lu, and T. Zhu, B 2 W 2 : N-way concurren communicaion for io devices, in SenSys, 26.