An Energy-Efficient Relaying Scheme for Internet of Things Communications

Transcription

1 An Energy-Efficient Relaying Scheme for Internet of Things Commnications Ahmad Alsharoa, Xiaoyn Zhang, Daji Qiao, and Ahmed Kamal Electrical and Compter Engineering Iowa State University, Ames, Iowa Abstract In this paper, we investigate the problem of optimal planning and deployment of mltiple relays to spport energyefficient plink transmissions of Internet of Things (IoT) devices. A novel approach is proposed to optimize the relay locations with the objective of minimizing the total energy consmption of the network. In addition, the plink transmit power of IoT devices and the device-relay-channel association are jointly optimized to meet the QoS reqirement of IoT devices. A mixed-integer linear programming (MILP) problem is formlated to obtain the optimal soltion. We also design a low-complexity genetic algorithm to provide a sb-optimal soltion to the problem. Index Terms IoT (Internet of Things) Devices, IoT Relays, Relay Planning, Genetic Algorithm. I. INTRODUCTION The emerging Internet of Things (IoT) systems consist of a large amont of small-scale, battery-powered devices with limited compting and commnication capabilities. While long-range commnication technologies are available for the IoT [], many practical IoT systems are still composed of lowcost, short-range IoT devices that are nable to transmit over a long distance [], []. In sch IoT systems, IoT relays can be deployed to collect data from IoT devices and relay to a remote server, which is ot of direct commnication range of IoT devices. IoT relays sally have more resorces than ordinary IoT devices, bt cost higher and consme more energy. Deployment of a large nmber of relays will provide increased coverage and network capacity nder peak traffic conditions. However, they may not be needed when the traffic is light; in sch scenarios, they may become nder-tilized or completely redndant leading to inefficient se of energy and commnication resorces. One of the known techniqes in the literatre is dynamic on/off switching, which is also known as sleeping strategy, where a large nmber of relays are deployed in a certain area, bt they can be trned off dring periods of low traffic and their associated devices can be offloaded to nearby relays. As a reslt, the power consmption of lightly-loaded relays can be redced or completely eliminated depending on the configration of their sleep state [], []. However, one drawback of this strategy is that a large nmber of relays need to be deployed in advance, which cold be a costly investment. In practice, the bdget limitation often dictates the maximm nmber of relays to be deployed. In this paper, we stdy the relay planning problem for IoT systems. Or goal is to identify an optimal relay planning strategy to minimize the overall system energy consmption, while satisfying the bdget constraint (which decides the maximm nmber of relays) and the Qality of Services (QoS) constraint of IoT devices (which is indicated by the minimm Signal-to-Interference-and-Noise-Ratio, or SINR). Previosly, many stdies have been condcted to solve the relay connectivity problem [] or the relay placement problem [], []. However, how to address both problems at the same time needs more investigation. For instance, the work in [9] proposed approximation schemes to place minimm nmber of relays to achieve different level of falt tolerance in heterogeneos networks. The relay placement problem nder with a delay constraint was investigated in []. A local search algorithm has been introdced in [] to solve the relay connectivity problem where the sensor nodes are divided into grops and, after that, a local set cover is fond for each grop sing a local search algorithm; however, the athors did not consider the connectivity (interference) problem and the placement problem jointly together. In this paper, relay planning is considered jointly with device-relay-channel association in an IoT network. Specifically, we consider plink commnications in an IoT system where each IoT device reports to the remote server via a relay. Or contribtions can be smmarized as follows: Proposing an energy-efficient relay planning framework for plink commnications in an IoT network. Formlating a mixed-integer linear programming problem to minimize the network-wide energy consmption. Optimizing the relay locations, the device-relay-channel association, and the transmit power of IoT devices, nder the bdget constraint, the devices maximm transmit power constraint, the device/relay association constraint, and the devices QoS constraint. Proposing a low-complexity green optimization approach based on Genetic Algorithm (GA) to find a sb-optimal low-complexity soltion []. The remainder of the paper is organized as follows. Section II presents the system model. The problem formlation is given in Section III. The proposed low-complexity genetic algorithm is described in Section IV. Section V discsses the evalation reslts. Finally, the paper concldes in Section VI. II. SYSTEM MODEL In this paper, we stdy plink transmissions in an IoT system that consists a total of U IoT devices. As shown in Fig., each IoT device is connected to an IoT relay which forwards the data from IoT devices to a remote server. We assme that the channel gain between an IoT device and

2 a relay at location l over commnication channel c can be modeled as: q h c,l = d $ h c,l,l, () where d,l is the Eclidean distance between device and relay l, $ is the path loss exponent, and h c,l is the fading coefficient of channel c. Notation h c,l L U Ū l P P P T c,l l th R th Table I: List of Notations Description Channel gain between device and relay l over channel c Maximm nmber of relays (decided by bdget limitation) Total nmber of devices Maximm nmber of devices served by relay l Relay transmit power Device transmit power Maximm transmit power of an IoT device Time slot length Binary variable representing association between, l, and c Binary variable representing if a relay is placed at location l Amplification power factor Offset site power when a relay is active Target SINR between device and relay Target data rate from device to relay Device energy consmption Relay energy consmption Candidate locations with an active relay deployed Fig. : System model. Candidate locations withot a relay Up to L IoT relays may be deployed to collect data from IoT devices, where L is determined by the bdget limitation. Each relay can be placed at any one of L candidate locations specified by the network admin. Each candidate location can accommodate at most one relay. In the following, we se the phrase relay l and a relay placed at candidate location l interchangeably. The energy consmption of an active relay can be represented as follows: [] U l P +, () where () corresponds to the power consmption that scales with the radiated power de to amplifier and feeder losses, () models an offset of site power that is consmed independently of the average transmit power, () U l is the set of devices connected to relay l, and () P corresponds to the relay radiated powers over all sers in U l. In this work, we assme that P is constant and eqal to P/Ūl, where P is the maximm allowable transmit power of an IoT relay (to commnicate with the remote server), and Ūl is the maximm nmber of devices each relay can accommodate de to the backhaling constraint. The achievable plink data rate from device to relay l over channel c is given by R c = log + P! h c,l, () I c + N where N is the noise power, I c = P ûn c,û= P û h c û,l is the interference from other devices that se the same channel. We se N c to denote the set of devices sing channel c. To simplify the analysis, we assme that Pû is fixed and eqal to P, where P is the maximm transmit power of an IoT device. Table I smmarizes the notations sed in the paper. III. PROBLEM FORMULATION AND SOLUTION In this section, we formlate and solve a relay planning and QoS problems aiming to minimize the network energy consmption. We jointly optimize planning and channel assignment of IoT relays, device-relay-channel association, and plink transmit power of IoT devices. A. Relay Planning Problem We se to denote a binary matrix of size L C U. Its entry c,l is given as follows: ><, if device commnicates with relay l c,l = over channel c, () >:, otherwise. A binary vector of size L is sed to indicate relay placement at candidate locations. Its entry l is given as:, if a relay is placed at candidate location l, l =, otherwise. () To ensre that devices cannot connect to a candidate location where no relay is placed, the following condition mst be satisfied: c,l apple l, l,, c. () The total energy consmption of IoT devices is: = T P, () = while the total energy consmption of IoT relays is:! = T c,lp +. () l= l = c=

3 Using (), we can re-write () as follows! = T c,lp + l l= = T P = c= l= = c= = T PU + T c,l + T l. Therefore, the relay planning problem that aims to minimize the total consmed energy while satisfying the QoS of IoT devices (defined by th - the minimm SINR) can be formlated as follows: sbject to: l= l= l (9) minimize E total = +( ), () l, c,l,p P apple P,, () c,l P h c,l I c + N th,, () = c= c,l apple Ūl, l, () c,l =,, () c,l apple l, l,, c, () where is a weight constant. Constraint () represents the maximm allowable transmit power of device, and constraint () represents the minimm QoS reqired by an IoT device. Constraints () and () aim to satisfy the backhaling condition and ensre that each ser is served by at most one relay, respectively. The formlated optimization problem is a mixed-integer non-linear programming problem de to the SINR expression given in (). This optimization problem can be linearized by introdcing c,l for each link sch that c,l = c,l P where the following ineqalities have to be respected: P c,l, c,l P c,l P + P, c,l apple P c,l. () The first ineqality ensres that c,l is between and P. The second and third ineqalities garantee that c,l =if c,l =, and c,l = P if c,l =. The third ineqality also garantees that c,l cannot exceed P. To deal with I c, we se the following approximation: UP P h c û,l I c Îc = = û= C UP c û=,l P h c û,l C (). () The approximation in () makes sense since the sers will be niformly assigned to different channels in order to minimize the interference. The prpose of () is to pll I c ot of the denominator in (), and it holds tre becase of (). Then, the linearized optimization problem can be reformlated as follows: sbject to: minimize l, c,l,p,,l c,l E total = +( ), (9) P apple P,, () c,l h c,l th(îc + N ), () = c= c,l apple Ūl, l, () c,l =,, () c,l apple l, l,, c () apple c,l apple P, l,, c () P c,l P + P, l,, c () c,l apple P c,l, l,, c, () where is a weight constant. The constraints () to () correspond to the linearization process in (). The optimal soltion to sch a problem can be obtained by sing Grobi/CVX [], []. IV. LOW-COMPLEXITY GENETIC ALGORITHM The formlated optimization problem given in Section III is considered an NP-hard problem de to the existence of binary variables. Hence, we propose a Genetic Algorithm (GA) as a meta-heristic algorithm []. A. Encoding In or genetic based approach, we generate randomly N soltions (n) (n =,,N) of size (L + L C U + U) to form an initial poplation S. Each soltion is encoded as: (n) =[ (n), (n), P (n) ], () where (n) is a binary vector carrying the relay planning information as defined in (), (n) is a binary vector carrying the device-relay-channel association information obtained by reshaping the matrix defined in (), and P (n) is a float vector representing the device transmit power.

4 B. Fitness Fnction and Selection For each soltion, we define its fitness f as the sm of the objective fnction and the scaled penalties of violated constraints. It is shown as follows: f = E total + X " + X + X l max max max(p th(îc + N ) = c= c,l P, )/ P Ū l,! # c,lp h c,l, /( th N ) /Ūl + X c,l + X,l,c max c,l l,, (9) where E total is the total energy consmption defined in () for the soltion. GA selects strings that provide the best tility and keeps them to the next poplation while the remaining strings are generated by applying crossovers and mtations to the srvived parents. We adopt the tornament selection algorithm to elect soltions with smaller fitness. C. Crossover and Mtation Crossovers consist of ctting two randomly-selected parent strings at a random corresponding point. The obtained fragments are then swapped and recombined to prodce two new strings. More specifically, crossover is applied to the srvived parents with a probability of p c to prodce the next generation. Crossover in or genetic algorithm is performed in three steps: Step : Extract the device, relay, and channel association vectors from the two soltions to be crossed. Let (i) and (j) denote the two vectors. Step : Randomly select a device index K, ct (i) and (j) at position L C K, and swap the ct fragments. Step : Update (i), P (i), (j), and P (j), according to () and (). After that, mtation (i.e., changing the vale of a randomlyselected bit in the string) is applied with a probability p m. Mtation in or genetic algorithm is performed in three steps: Step : Extract the relay planning vector from the soltion to be mted. Step : Randomly select a candidate location which has an active relay, sht down the relay, and re-allocate all the devices associated with the relay to other relays, pdate. Step : Update and P according to () and (). V. EVALUATION In this section, selected nmerical reslts are provided to evalate the performance of the proposed relaying scheme. IoT devices transmit their messages periodically every T = sec. All the fading channel gains adopted in the framework are assmed to be i.i.d. Rayleigh fading gains. The target SINR threshold is th = Rth, where R th =. bits/s/hz is the target data rate. The path loss exponent is $ =. The area of interest is [m ]. The noise power is taken to be N = N W, where N = dbm/hz and W = KHz. In Table II, we list the vales of all the simlation parameters [], []. Table II: Simlation Parameters Parameter Description Vale Amplification power factor [W] Offset site power when a relay is active. P [dbm] Device maximm transmit power Ū l Maximm nmber of sers served by relay l N Nmber of poplations in GA (Genetic Algorithm) p c Crossover rate in GA. p m Mtation rate in GA. Weight constant. A. Optimal Performance ) Energy Consmption vs. Nmber of IoT Devices: Fig. shows the energy consmption performance with different nmbers of IoT devices in the system. As we can see in Fig. a, as the nmber of devices increases, the total energy consmption of devices increases, and the energy consmption per device first increases and then decreases slightly. The increase of the energy consmption per device is becase when there are more devices in the system, there will be more sers per channel and ths the interference will be higher. On the other hand, when the nmber of devices gets even larger, more relays will be deployed to spport more devices as we can see in Fig. b, which will reslt in a shorter distance from the device to the relay on average. - - (per device) - (per relay) N l Fig. : Energy consmption vs. nmber of IoT devices. The nmber of candidate relay locations is L =, and the nmber of channels is C =. From Fig. b, we can observe that, as the nmber of devices increases, the total energy consmption of relays increases as a reslt of the increased nmber of active relays. Meanwhile, the energy consmption per relay remains almost the same, which means that the load on each relay does not vary mch. ) Energy Consmption vs. Nmber of Candidate Relay Locations: Fig. shows the energy consmption performance with different nmbers of candidate locations for IoT relays. As we can see, as the nmber of candidate locations increases, the total energy consmption of devices, the energy consmption per device, the total energy consmption of relays, and the nmber of active relays all decrease. This is simply becase we cold plan the IoT relays better, with a larger nmber of candidate location available to choose from. ) Energy Consmption vs. Nmber of Channels: Fig. shows the energy consmption performance with different nmbers of channels. As we can see in Fig. a, as the nmber Nmber of Active Relays

5 - - (per device) - (per relay) Fig. : Energy consmption vs. nmber of candidate relay locations. The nmber of IoT devices is U =, and the nmber of channels is C =. of channels increases, the total energy consmption of devices and the energy consmption per device decrease. This is becase when there are more channels, there will be less sers per channel and ths the interference will be smaller. We also observe in Fig. b that, as the nmber of channels increases, the total energy consmption of relays decreases too, as a reslt of the redced interference. - - (per device) - Nmber of Channels Nmber of Channels N l (per relay) Fig. : Energy consmption vs. nmber of channels. The nmber of IoT devices is U =, and the nmber of candidate relay locations is L =. B. Snapshots In this sbsection, we select one simlation rn and show some snapshot reslts. In this rn, we simlate IoT devices, candidate relay locations, and channels. ) Device-Relay Association: Fig. shows the relay locations, together with the device-relay association. In this simlation rn, we only need three relays and they are placed at the candidate locations,, and, which are marked by a solid red triangle, a solid ble sqare, and a solid black circle, respectively. The other nine candidate locations, marked by X s, are not selected. The devices associated with a relay are marked with the same (bt smaller) symbol as the relay. As we can see, a device sally favors a relay close to itself. However, there are some otliers sch as and at the pper-right corner. They connect to relay l instead of the closer relay l, since l already has associated sers and has reached its limit. Another example otlier is, which chooses l instead of l ; this is becase has a higher channel gain and lower interference with l than l, which is an effect of Rayleigh fading. N l Nmber of Active Relays Nmber of Active Relays 9 9 l l l 9 l l l 9 l 9 l l 9 l l 9 l Fig. : Relay placement and device-relay association. Three relays are deployed, which are marked with large solid triangle, sqare, and circle. X s represent nsed candidate locations withot a relay. Devices associated with a relay are marked with the same bt smaller symbol as the relay. ) Transmit Power: Fig. shows the relation between the transmit power of IoT devices and their distances from the associated relays. Generally, devices with a larger distance from the relay tend to choose a higher transmit power. Transmit Power (dbm) Distance between Device and Relay (m) Fig. : Device transmit power vs. distance from the relay. ) Channel Assignment: Fig. shows the channel assignment of IoT devices. As we expect, devices scatter niformly in all channels and each channel is occpied by a similar nmber of devices. Channel Assignment 9 Index of Device Fig. : Channel assignment of IoT devices. ) Approximated vs. Actal Interference: Fig. compares the approximated interference with the actal interference. As we can see, the actal interference of all devices is close to bt less than the approximated interference, which demonstrates the validity of or approximation of I c in ().

6 Interference (dbm) Approximated Interference Actal Interference - Index of Device Fig. : Approximated vs. actal interference. C. Performance of Low-Complexity Genetic Algorithm Fig. 9 compares the proposed low-complexity genetic algorithm with the optimal soltion, in term of total energy consmption, with different nmbers of IoT devices in the system. As we can see, the genetic algorithm yields a soltion close to the optimal one. In particlar, when the problem is in small scale, i.e., U =, the genetic algorithm is able to prodce almost the optimal. Fig. 9: Total energy consmption vs. nmber of IoT devices. The nmber of candidate relay locations is L =, and the nmber of channels is C =. Fig. : Total energy consmption vs. nmber of candidate relay locations. The nmber of IoT devices is U =, and the nmber of channels is C =. Fig. shows the comparison reslts with different nmbers of candidate relay locations. The reslts of the genetic algorithm are still close to those of the optimal algorithm, which verifies the effectiveness of the genetic algorithm. Note that, for the genetic algorithm, the total energy consmption of relays may increase when the nmber of candidate locations increases; this is de to the sb-optimality and randomness of the genetic algorithm. Finally, Fig. shows the comparison reslts with different nmbers of channels. Nmber of Channels Nmber of Channels Fig. : Total energy consmption vs. nmber of channels. The nmber of IoT devices is U =, and the nmber of candidate relay locations is L =. VI. CONCLUSIONS In this paper, we proposed a novel relaying framework for energy-efficient plink transmissions in IoT networks. We formlated a mixed-integer linear optimization problem that aims to place the relays optimally with goal of minimizing the total consmed energy while satisfying a set of constraints, inclding the bdget constraint that decides the maximm nmber of IoT relays, and the QoS constraint of IoT devices. We also proposed a low-complexity algorithm based on genetic algorithm, and compared its performance with the optimal soltion. REFERENCES [] Lorawan. [Online]. Available: [] S. Y. Lien, K. C. Chen, and Y. Lin, Toward biqitos massive accesses in GPP machine-to-machine commnications, IEEE Commnications Magazine, vol. 9, no., pp., Apr.. [] A. Kott, A. Swami, and B. J. West, The internet of battle things, Compter, vol. 9, no., pp., Dec.. [] I. Ashraf, F. Boccardi, and L. Ho, Sleep mode techniqes for small cell deployments, IEEE Commnications Magazine, vol. 9, no., pp. 9, Ag.. [] L. Falconetti, P. Frenger, H. Kallin, and T. Rimhagen, Energy efficiency in heterogeneos networks, in IEEE Online Conference on Green Commnications (GreenCom ), Sept.. [] C. Ma, W. Liang, M. Zheng, and H. Sharif, A connectivity-aware approximation algorithm for relay node placement in wireless sensor networks, IEEE Sensors Jornal, vol., no., pp., Jan. [] S. Misra, S. D. Hong, G. Xe, and J. Tang, Constrained relay node placement in wireless sensor networks: Formlation and approximations, IEEE/ACM Transactions on Networking, vol., no., pp., Apr.. [] D. Djenori and M. Bagaa, Energy-aware constrained relay node deployment for sstainable wireless sensor networks, IEEE Transactions on Sstainable Compting, vol., no., pp., Jan.. [9] X. Han, X. Cao, E. L. Lloyd et al., Falt-tolerant relay node placement in heterogeneos wireless sensor networks, IEEE Transactions on Mobile Compting, vol. 9, no., pp., May. [] C. Ma, W. Liang, and M. Zheng, Set-covering-based algorithm for delay constrained relay node placement in wireless sensor networks, in IEEE International Conference on Commnications (ICC), Kala Lmpr, Malaysia, May, pp.. [] M. Mitchell, An Introdction to Genetic Algorithms. Cambridge, MA, USA: MIT Press, 99. [] G. Aer, O. Blme, V. Giannini et al., Energy efficiency analysis of the reference systems, areas of improvements and target breakdown, INFSOICT- EARTH (Energy Aware Radio and NeTwork TecHnologies), Tech. Rep, Dec.. [] Grobi optimizer reference manal,. Available [online]: [] CVX Research, Inc., CVX: MATLAB Software for disciplined convex programming, version., Ag.. Available [online]: