REAL-TIME SCHEDULING IN LTE FOR SMART GRIDS. Yuzhe Xu, Carlo Fischione

Size: px

Start display at page:

Download "REAL-TIME SCHEDULING IN LTE FOR SMART GRIDS. Yuzhe Xu, Carlo Fischione"

Jennifer Bridges
5 years ago
Views:

1 REAL-TIME SCHEDULING IN LTE FOR SMART GRIDS Yuzhe Xu, Carlo Fschone Automatc Control Lab KTH, Royal Insttue of Technology 1-44, Stockholm, Sweden Emal: ABSTRACT The latest wreless network, 3GPP Long Term Evoluton (LTE), s consdered to be a promsng soluton for smart grds because t provdes both low latency and large bandwdth. However, LTE was not orgnally ntended for smart grds applcatons, where data generated by the grd have specfc delay requrements that are dfferent from tradtonal data or voce communcatons. In ths paper, the specfc requrements mposed by a smart grds on the LTE communcaton nfrastructure s frst determned. The latency offered by the LTE network to smart grds components s nvestgated and an emprcal mathematcal model of the dstrbuton of the latency s establshed. It s shown by expermental results that wth the current LTE up-lnk scheduler, smart grd latency requrements are not always satsfed and that only a lmted number of components can be accommodated. To overcome such a defcency, a new scheduler of the LTE medum access control s proposed for smart grds. The scheduler s based on a mathematcal lnear optmzaton problem that consders smultaneously both the smart grd components and common user equpments. An algorthm for the soluton to such a problem s derved based on a theoretcal analyss. Smulaton results based on ths new scheduler llustrate the analyss. It s concluded that LTE can be effectvely used n smart grds f new schedulers are employed for mprovng latency. 1. INTRODUCTION Smart grd has been proposed as an alternatve to the tradtonal electrcty grd thanks to ts advantages of real tme control on consumpton demands. The term smart grd refers to two-way communcatonal electrcty grd [1]. A smart grd s expected to be capable of remotely detectng statuses of electrcty generatons, transmsson lnes and substatons; of montorng consumpton of user electrcty usage; of adjustng the power consumpton of household applcatons to conserve energy, reduce energy losses and ncrease electrcty We acknowledge the support of the European Insttute of Technology, Smart Energy Systems, the NoE Hycon2 and EU STREP Hydrobonets. grd relablty. It s consdered to be one of relevant solutons for energy conservaton. The dstncton comes from ts unque ablty of real tme control on both generaton and demand sdes. To acheve the real tme control, hgh effcent and relable communcaton networks are needed. 3GPP Long Term Evoluton (LTE), the latest moble communcaton network, s a promsng opton for a smart grd [2, 3]. LTE was developed to fulfll moble users demands for hgher data rates and stabler servce performance. 3GPP whte paper and LTE servce operators announced that the man advantages wth LTE are to provde hgh throughput (up to 3 Mbps n downlnk, and 75 Mbps n uplnk), low latency (less than 5 ms for control plane latency, and less than 5 ms for user plane latency), plug and play, Frequency (FDD) and Tme Dvson Duplex (TDD) n the same platform, an mproved end-user experence and a smple archtecture resultng n low operatng costs [2]. However, LTE has not been desgned for smart grd applcatons. The measurement requrements for smart grd are defned or ntroduced n techncal lterature. Standards [4, 5] and [6] ntroduce the concepts such as message types, reportng rates, for electrcty substatons, Phasor Measurement Unts and Automatc Meter Readngs. The topologes and communcaton nfrastructures of a smart grd are dscussed n [7 11] and [12]. Lteratures [13, 14] and [15] ntroduce several LTE scheduler desgns for common user equpments as moble phones, whch have qute dfferent requrements compared to the servces asked by smart grds components. However, to the best of our knowledge, there s no analyss of LTE latency for smart grds so far. The avalablty of a latency dstrbuton s essental to desgn controllers that are able to take real-tme actons on the smart grds. It s essental that such a dstrbuton meets the typcal requrements of smart grds components. In ths paper, the latency requrements are determned for communcaton n substaton, collectng and dssemnaton of phasors and consumptons data. Experments are conducted to collect the measurements of latency va LTE network. Internet Control Message Protocol (ICMP) echo request packets are sent to the target host va LTE modems, and wat

2 for an ICMP response. The round trp tme (RTT) values for transmsson and recepton are then measured and analyzed. Based on these measurement, a mxture of Gussan dstrbuton model s proposed to ft the latency dstrbuton. To mprove the LTE performance for communcaton n smart grd, a new scheduler s suggested. Ths scheduler s able to fnd out those user equpments that serve the smart grds, and gve hgher prorty to them. We show that the new scheduler gves a substantal mprovements n terms of latency and ts dstrbuton. Compared wth the lterature for LTE schedulers [13 15], our scheduler s new n terms of theoretcal analyss and for the ablty to dstngush components of smart grd from other publc users. The smulaton results show that the proposed scheduler can decrease the latency, thus t offers much better performance for the smart grd. The rest of the paper s structured as follows. Secton 2 ntroduces the communcaton requrement n a smart grd wth focus on latency. Secton 3 presents the experment for LTE latency performance. Secton 4 ntroduces a schedule. Smulaton results are gven n Secton 5. Fnally, Secton 6 concludes ths paper. 2. LATENCY REQUIREMENT FOR COMMUNICATION IN SMART GRID For the smart grd communcaton network, latency s one crtcal techncal requrement. Smart grds measurements must be avalable wthn some delay for the grds supervsor to take real-tme acton on, e.g., the dstrbuton of the load. In ths secton, we nvestgate the latency requrements mposed by a smart grd. In most cases, a smart grd focuses on three man areas: household devces for communcaton automatc meter readng; remote sensng devces for electrcal network montorng and control and dstrbuted power energy source, such as wnd and solar management. The key components n a smart grd are: the Advanced Meter Infrastructures (AMI) at houses or buldngs, Phasor Measure Unts (PMU) for transmsson lnes and power generatons such as dstrbuted generatons and substatons. In the followng, we descrbe these components n detal wth focus on ther latency requrements. PMU provdes phasor measurements of voltages and currents n an electrcal grd. The phasor measurements are calculated va Dscrete Fourer Transform (DFT) and delvered to devces called Phasor Data Concentrators (PDC). In PDC, the measurements are tme-synchronzed, stored for future reference and forwarded to applcaton and Super PDCs. Generally speakng, PMU measurements are 1 2 bytes and reported at a rate of 2 6 tmes a second, namely 2 6 Hz. They are expected to meet real-tme control system requrements wth tme delay less than 1 ms [16]. AMI s an upgrade of Automatc Meter Readng (AMR) provdng two-way communcaton and specfc actuators. AMI collects nformaton of consumpton records, alarms Fg. 1. Measurement Setup and status from customers and mpose consumptons of customers. Based on ts two-way communcaton and consumpton meterng, AMI enables real tme prcng and peak shavng n a smart grd. Referrng to Wde-Area Measurement System (WAMS), latency less than 1 second (typcally 1 2 ms) s requred to acheve real tme prcng requrements [17] [18]. The communcaton requrements for functons and devce models n a substaton are defned by Standard IEC [4]. The sze of message vares from 1 to 124 bts. Those messages delvery latences vary from 3 ms to 1 s. Latency less than 3 ms s requred for fast and raw data messages. To sum up, the latency s requred to be less than 1 ms for nterconnectng PMUs and AMIs. In the followng secton, we nvestgate f the LTE latency s low enough for communcaton among PMUs and AMIs n smart grd. 3. EXPERIMENTAL LATENCY ANALYSIS OF LTE In ths secton, we provde an emprcal analyss of the latency offered by LTE. Experments have been performed usng LTE modems. The expermental results and estmated latency dstrbuton are provded n ths secton. We consdered two LTE network servces n Stockholm, Sweden. One s provded by operator TELE2 (Operator 1) and the other one by TELIA (Operator 2). The round trp tme (RTT) and data loss rate of messages travelng between clent and servce are measured and recorded. Fg. 1 llustrates the measurement setup. Wthout consderng the bytes used for header,the length of messages vares from bytes to 124 bytes n each packet. Every message corresponds to one or a group of readngs generated by the components of the smart grd Expermental Results Fg. 2 shows the mean values and standard devatons of RTT tme measured by png command usng two LTE modems. Fg. 2(a) llustrates the values for small data packets, less or equal to 1 bytes, whle Fg. 2(b) shows those for larger packets. These fgures ndcate that when the length of data packets s smaller than 1 bytes, the RTT s shorter than 2 ms under the servce provded by Operator 1, whle t s around 2 ms for Operator 2. The RTT values ncrease wth the length of

3 Mean Value of RTT [ms] Standard Devaton for Operator 1 Standard Devaton for Operator 2 Mean Value RTT for Operator 1 Mean Value RTT for Operator Length of Data Packet [bytes] (a) Small data packets 2 1 Standard Devaton of RTT [ms] Table 1. Summary of fts of RTT values for both Operators n Fg. 3 Operator 1 Operator 1 MLE LSE MLE LSE SSE r Mean Value [ms] µ Standard Devaton σ HARQ Tme [ms] T up T down HARQ Probablty p up p down Max. Resend Tmes H Mean Value of RTT [ms] Mean Value RTT for Operator 1 Mean Value RTT for Operator 2 Standard Devaton for Operator 1 Standard Devaton for Operator Length of Data Packet [bytes] (b) Large data packets Fg. 2. Mean values and standard devatons of RTT data packets when the length s larger than 1 bytes. But the standard devatons of RTT are approxmately the same whatever szes for the packets for each servce operator. The mean values of RTT under Operator 1 are lower than those under Operator 2. However, the standard devaton of RTT under Operator 1 s around 4 5 tmes larger than that under Operator 2. Ths mght be due to that the measurement equpment s closer to enodeb of Operator 1 than that of Operator 2. In addton, the mnmum latences for RTT transmsson are 1 ms, 13 ms va Operaton 1 and 2 respectvely. If we calculate the latency by dvdng RTT by half, the mnmum values of the latency for small sze packet agrees wth the theoretcal latency gven by 3GPP whte paper, whch s 5 ms. However, note from Fg. 2(a) that ths has a low probablty of occurrence Latency Dstrbuton Model An accurate latency dstrbuton model s necessary to predct the communcaton latency va LTE network. We propose to model the probablty densty functon (pdf) for RTT as f(x) = H = j= H (1 p up )(1 p down )p 1 G(µ T up jt down, up Standard Devaton of RTT [ms] p j 1 down σ 2 up + σ 2 down ), (1) where p up and p down are the probabltes of repeatng request, T up and T down are the tme spent on resendng n up- and down-lnk respectvely, and G(, ) s a Gausan dstrbuton. Usng maxmum lkelhood (MLE) and least squares estmaton (LSE), the parameters that gve the best fttng dstrbuton functons can be obtaned. The estmated values for those parameters n the model are lsted n Tab. 1, and the fttng results are shown n Fg. 3. Accordng to the results, one of the estmated tme for HARQ s approxmately 8 ms. It ndcates that FDD s deployed n LTE servce, snce the tme for HARQ n uplnk s fxed to 8 ms usng FDD. Fg. 3 ndcates that the probablty of resendng n uplnk va Operator 2 (around 2%) s much lower than that va Operator 1 (around 2%). From the expermental results t follows that the mnmum latences for less than 3 bytes data packets are close to the theoretcal latency announced by 3GPP whte paper (less than 5 ms). However, the most of the packets are transmtted wth a latency that s around 15 ms for one operator and 25 ms for another. Ths s n contrast wth the latency requrements of a smart grd. To overcome ths problem, n the next secton we propose the desgn of a new scheduler for LTE transmsson. 4. SCHEDULER DESIGN In ths secton, we propose a new LTE scheduler that frstly allocates resources for smart grd, thus attemptng to guarantee the latency requrements of smart grds components. Recall that the man purpose of a scheduler s to allocate sutable physcal resources for the set of users for communcaton. To make good schedulng decsons for hgh performance, a scheduler requres knowledge of both channel condtons and users devce condtons. Due to lmted sgnallng channel resources, a user equpment (UE) seldom reports all subcarrres channel condtons, but average condton n the best subcarrers. The smallest resource unt allocated by a scheduler to a UE s two consecutve resource blocks (RB), spannng transmt tme nterval (TTI) of 1 ms and a bandwdth of 18 khz.

4 Probablty Probablty Data MLE Ft Curve LSE Ft Curve RTT Value [ms] (a) Va Operator 1 LTE network Data MLE Ft Curve LSE Ft Curve RTT Value [ms] (b) Va Operator 2 LTE network Fg. 3. RTT values dstrbuton. The data set s collected va LTE network wth 1 bytes data packets 4.1. A Scheduler for Smart Grds The scheduler desgn can be posed by an optmzaton problem that allows us to make an optmal usage of the performance of the physcal resources. The scheduler allocates resources for N users, and the modulaton for each user s selected only accordng to user s channel condton. The problem s to allocate resources (, j) n tme- and frequencydoman and maxmzng a measure of utlty. We propose the problem max,c s.t. where R (c) N N TTI N RB c=1 =1 j=1 c j 1 R (c) x(c) (2a) {, 1}, j (2b) N TTI j, c (2c) N RB, c (2d) L(c) c, (2e) j = λ(c) x(c) s the utlty weght functon for the s 1 f (, j) c-th user wth some utlty parameters λ (c), x(c) resource block s assgned to c-th user, and zero otherwse. Eq. (2b) ndcates that each resource block can be allocated to one user at most, and Eq. (2c) and (2d) gve the greatest values N TTI, N RB n tme and frequency doman respectvely. Eq. (2e) ndcates that resource blocks allocated to UE are lmted by each UE transmsson demand L (c). We make the natural assumpton that the weght λ (c) depends on UE s nformaton only. In other words, UEs only report the average channel condton for all avalable channels n every TTI. It s possble to show that problem (2) has multple optmal solutons. In the followng, we propose a procedure for determnng one such soluton Optmal Soluton to the Schedulng Problem To solve optmzaton problem (2), we start to observe that Eq. (2c) can be guaranteed by Eq. (2b), snce j, c 1 = N TTI. c It s smlar for Eq. (2b) to guarantee Eq. (2d). Moreover, the fact that UEs only report the average channel condton for all avalable channels n every TTI suggests that the key to solve ths optmzaton problem s to fnd the optmal solutons for the aggregate l (c) defned by Eq.(4d) below. Moreover, we assume that λ (c) s constant for a gven (c), namely λ (c) = λ(c). Then problem (2) can be rewrtten as max l (c) s.t. N λ (c) l (c) c=1 l (c) mn(n TTI N RB, c c L (c) ) (4a) (4b) l (c) mn(n TTI N RB, L (c) ) c (4c) N TTI N RB l (c) = Z. =1 j=1 (4d) where Z s the set of ntegers. The constrants of problem (4) equvalent to Eq. (2e) because the tme and frequency domans are equvalent n ths case. Ths problem s useful n that we solve problem (2) by the followng steps. Frstly, we fnd the solutons set {l (c) } for Eq.(4). Then based on Eq.(4d), we can recover the optmal solutons { } for Eq.(2) usng {l (c) }. We show ths below. To fnd the optmal soluton for l 1, we have the followng result: Proposton 4.1 Consder optmzaton problem (4). Suppose that users are labelled such that λ (1) λ (2)... λ (N). Then l1 = mn(n TTI N RB, L 1 ) { l2 mn(ntti N = RB l1, L 2 ) f N TTI N RB l1 > otherwse

5 Smlarly, we can fnd the solutons set {l } for other l step by step. Once we have such a soluton, we can recover the optmal soluton for problem (2) by the followng results. Proposton 4.2 Let l (c) c be the optmal soluton of (4). One optmal solutons set { } of problem (2) s obtaned by = 1 c where ( ) k k (, j) = + 1, k N RB (5) n=1 N RB n=1 N RB c 1 c 1 k = l (n) + 1, l (n) + 2,..., 4.3. Utlty Functon Desgn c l (n). (6) n=1 Snce LTE transmts data flows not only n a smart grd but also for publc use n a wde area, we would lke choose a utlty functon n the schedulng problem so to gve the hghest prortes to messages from PMUs and ensure that the latency requrements are met. As a result, we proposed a scheduler that conssts of an estmator and a prorty calculator. The estmator determnes the possblty for UEs of beng n a smart grd based on the buffer status reports from UEs. The features of a UE used n a smart grd can be summarzed as follows: a) constant data updatng rates ṙ (c), b) equvalent data packets lengths l (c) and c) approxmately nvarant channel qualtes q (c). Then the prorty calculator gves the prorty value to each UE. Wth combnng all the components n the scheduler, the whole utlty functon for the c-th UE can be obtaned as λ (c) = W (c) P + P (c) (c) PF. Here W P ndcates the weght for the UE n a smart grd and we propose to obtan t as [ W (c) P = α 1 ṙ (c) + α l(c) 2 + α 3 q (c)] 1, (7) where α [, 1] and α = 1. In ths case, we set α 1, α 2 and α 3 to.3,.5 and.2 respectvely. The choce of these parameters can vary wthout changng the valdty of the scheduler. We set these values because they work well n practce, as we show later. Note that P (c) PF s gven by the tradtonal LTE schedulng algorthm: P (c) PF = (C/I) (c) [R (c) (t)] 1, whch s used n wreless communcaton networks. Here (C/I) (c) represents the channel condtons and R(t) (c) s the throughput wthn the tme nterval t for user c. The whole schedulng algorthm s summarzed by Algorthm SIMULATION In ths secton, we perform LTE uplnk system level smulatons based on 3GPP LTE to llustrate the performance of our scheduler. We could not do experments because ths would requre to change the LTE scheduler at the base staton, whch s only possble to network operators. Tab. 2 summarzes the smulaton parameters and assumptons. The parameters used Algorthm 1 Schedulng 1: Let M be the set of avalable RBs at tme nterval t 2: Let N be the set of schedulable UEs 3: for = to n do 4: calculate λ based on buffer status reports and channel condtons 5: map data packets of th UE to requred RBs quantty c usng AMC 6: end for 7: Index 1; 8: whle M do 9: pck up the user k N wth the largest value λ k 1: f M c k then 11: assgn Index : Index + c k RBs to kth user 12: M M c k 13: Index Index + c k 14: N N k 15: else 16: Assgn Index : m RBs to kth user 17: M 18: end f 19: end whle Table 2. Smulaton parameters Parameter Settng System bandwdth 5MHz Subcarrers per RB 12 OFDM symbols per RB 7 RB bandwdth 18Hz Number of RBs 25 Cell-level user dstrbuton Unform Number of PMUs n cell 5 Rate for PMUs 6Hz Data Packet sze for PMUs 1 bytes Number of other actve users n cell 1 Traffc model Unform Transmsson tme nterval (TTI) 1 ms Modulaton and codng settng QPSK,16QAM,64QAM Probablty of HARQ (P HARQ ) 2%,2% Measurement Nose N(, 2) to generate the latency dstrbuton n uplnk va Operator 1 and 2 LTE networks are obtaned from Tab. 1. The latency dstrbutons and the generated dstrbuton usng estmated parameters are shown n Fg 4. It can be seen that the latences for sendng PMU messages are decreased by usng our new scheduler, whch gves hgher prorty on PMUs. It also ndcates that a mnorty of latences value exceeds the short latency requrement (1 ms) due to resendng process. Based on these results, we can compute how many PMUs can be mounted n a cell fulfllng the 1 mcrosecond latency requrement n smart grds. In partcular, we assume that a PMU data message s 2 bytes and sent at a rate of 6 Hz. Snce the scheduler always gves the hghest prortes to PMUs, PMU data messages are always allocated to physcal resource frst. In a 1-ms nterval, LTE can transmt 75 Mbps 1 mcrosecond= 75 Kbts 75 KB n uplnk. Therefore LTE can handle 375 PMUs n one cell n uplnk. For AMIs whose latency deadlne s 1 second, data message length s 1 bytes and sent every 15 mnutes, ths result s

6 Probablty T HARQ = 8 ms p HARQ = 2% T HARQ = 8 ms p HARQ =2% wth estmated Operator 2 parameters wth estmated Operator 1 parameters Latency [ms] Fg. 4. Latency Dstrbuton for PMU messages based on smulaton calculated to be 75 Mbps 1 second= 75 Mbts 7.5 MB, whch means LTE can accommodate up to 75, AMIs communcaton n uplnk even n the worst case. 6. CONCLUSION Ths paper nvestgated the latency performance of LTE network utlzed n a smart grd. Frst, the communcaton requrements n a smart grd were studed based on IEEE standards and related lterature. We showed that current LTE can support only a lmted number of smart grds component. Therefore, the second contrbuton of the paper was the characterzaton of a new scheduler to mprove the latency. The smulaton results showed the latency could be reduced to around 5 ms for PMUs communcaton va LTE. In addton, the results showed that LTE can handle more than 2 PMUs n a sngle cell. The study proven that the latency va LTE network s low enough for communcaton among PMUs and AMIs by usng an LTE scheduler that gves hgher prorty to smart grds components. 7. REFERENCES [1] H. Farhang, The path of the smart grd, IEEE Power and Energy Magazne, vol. 8, 29. [2] S. Sesa, I. Toufk, and M. Baker, LTE: The UMTS Long Term Evoluton from theory to practce. John Wley & Sons Ltd., 29. [3] P. Lescuyer and T. Lucdarme, Evolved Packet System (EPS): The LTE and the SAE Evoluton of 3G UMTS. John Wley & Sons Ltd., January 28. [4] IEC, : Communcaton requrements for functons and devce models. IEC, 22. [5] IEEE, C Standard for Synchrophasors for Power Systems. IEEE, March 26. [6] IEC, : Electrcty meterng Data exchange for meter readng, tarff and load control: Part 21 Drect local data exchange. IEC, 22. [7] V. Sood, D. Fscher, J. Eklund, and T. Brown, Developng a communcaton nfrastructure for the smart grd, n IEEE Electrcal Power and Energy Conference (EPEC), 29, 29. [8] D. Bakken, A. Bose, C. Hauser, D. Whtehead, and G. Zwegle, Smart generaton and transmsson wth coherent, real-tme data, Proceedngs of the IEEE, vol. 99, pp , 211. [9] Y.-J. Km, M. Thottan, V. Kolesnkov, and W. Lee, A secure decentralzed data-centrc nformaton nfrastructure for smart grd, IEEE Communcaton Magazne, vol. 48, pp , 21. [1] K. Sou, J. Wemer, H. Sandberg, and K. Johansson, Schedulng smart home applances usng mxed nteger lnear programmng, n Conference on Decson and Control, 211. [11] D. Zhang, L. Papageorgou, N. Samsatl, and N. Shah, Optmal schedulng of smart homes energy consumpton wth mcrogrd, n ENERGY 211, 211. [12] A. Aggarwal, S. Kunta, and P. Verma, A proposed communcaton nfrastructure for the smart grd, n Innocatve Smart Grd Technologes (ISGT), 21, 21. [13] H. Yang, F. Ren, C. Ln, and J. Zhang, Frequencydoman packet schedulng for 3gpp lte uplnk, n IEEE INFOCOM 21, pp. 1 9, 21. [14] S. Lee, S. Choudhury, and A. khoshnevs, Downlnk mmo wth frequency-doman packet schedulng for 3gpp lte, n IEEE INFOCOM 29, pp , 29. [15] S. Lee, I. Pefkanaks, and A. Meyerson, Proportonal far frequency-doman packet schedulng for 3gpp lte uplnk, n IEEE INFOCOM 29, pp vol.2, 29. [16] C. Hauser and D. Bakken, Securty, trust, and qos n next-generaton control and communcaton for large power system, Internatonal Journal of Crtal Infrastructures, 27. [17] J. Ca, Z. Huang, J. Hauer, and K. Martn, Current status and experence of wams mplementaton n north amerca, n IEEE/PES Transmsson and Dstrbuton Conference and Exhbton (Asa and Pacfc), 25. [18] T. Khalfa, K. Nak, and A. Nayak, A survey of communcaton protocols for automatc meter readng applcatons, Communcaton Survey and Tutorals, Second Quarter 211.

Throughput Maximization by Adaptive Threshold Adjustment for AMC Systems

Throughput Maximization by Adaptive Threshold Adjustment for AMC Systems APSIPA ASC 2011 X an Throughput Maxmzaton by Adaptve Threshold Adjustment for AMC Systems We-Shun Lao and Hsuan-Jung Su Graduate Insttute of Communcaton Engneerng Department of Electrcal Engneerng Natonal