2282 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 8, AUGUST 2016

Transcription

1 2282 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 8, AUGUST 2016 A Low-Latency Collaborative HARQ Scheme for Control/Uer-Plane Decoupled Railway Wirele Network Li Yan, Student Member, IEEE, Xuming Fang, Member, IEEE, Geyong Min, Member, IEEE, and Yuguang Fang, Fellow, IEEE Abtract The control/uer C/U) plane decoupled railway wirele network i an innovative architecture recently poed to meet the communication demand of both train control ytem and onboard paenger. The core idea i to completely eparate the C-plane and the U-plane into different network node operating at different frequency band. Although the ytem capacity of thi network architecture can be highly increaed, the forwarding latency of X3 interface to link the C-plane and the U-plane become a eriou blem, particularly for hybrid automatic repeat requet HARQ) tocol that demand frequent interaction between the C-plane and the U-plane. To addre thi challenging blem, we poe a low-latency collaborative HARQ cheme in thi paper. Specifically, we develop a new collaborative tranmiion framework where the poible pare reource on lower frequency band of macrocell excluding thoe ued by C-plane tranmiion can be utilized to help mall cell relay erroneouly received data. Compared with the conventional HARQ cheme, the poed cheme require fewer retranmiion to reach the ame tranmiion reliability, thereby mitigating the latency blem caued by HARQ retranmiion. Furthermore, channel mapping i alo redeigned to conform to the poed collaborative tranmiion framework. Through theoretical analyi, we derive the expreion of the average number of retranmiion related to the um of independent Gamma variable. Finally, the reult of imulation experiment how that the poed cheme can largely decreae the retranmiion latency for railway wirele network. Index Term C-plane and U-plane decoupling, collaborative tranmiion, low latency, HARQ, railway communication. I. INTRODUCTION RECENT rapid development of railway technologie and mobile Internet have timulated increaingly preing demand on high-peed wirele acce. However, no mobile Manucript received September 30, 2015; revied November 13, 2015; accepted January 11, Date of publication February 15, 2016; date of current verion July 29, Thi work wa upported in part by the 973 Program of China under Grant 2012CB316100, by the National Natural Science Foundation of China under Grant , and by the EU FP7 QUICK ject under Grant PIRSES-GA The Aociate Editor for thi paper wa F.-Y. Wang. L. Yan and X. Fang are with the Key Lab of Information Coding and Tranmiion, Southwet Jiaotong Univerity, Chengdu , China liyan @my.wjtu.edu.cn; xmfang@wjtu.edu.cn). G. Min i with the College of Engineering, Mathematic and Phyical Science, Univerity of Exeter, Exeter EX4 4QF, U.K. g.min@exeter. ac.uk). Y. Fang i with the Department of Electrical and Computer Engineering, Univerity of Florida, Gaineville, FL USA fang@ece.ufl.edu). Color verion of one or more of the figure in thi paper are available online at Digital Object Identifier /TITS network operator would vide full and dependable wirele coverage for parely-populated railway cenario with low revenue return. Furthermore, due to the limited ignal ceing capability of mobile equipment, it i difficult to overcome the evere challenge in high-peed railway cenario, uch a large penetration lo, high mobility and fat group handover [1]. A a reult, a novel unified railway wirele network i urgently needed to meet the communication demand of both train control ytem and onboard paenger. To thi end, indutrial participant have reached the conenu that the current narrowband Global Sytem for Mobile Communication for Railway GSM-R) will evolve to Long Term Evolution LTE) for railway communication ytem in the near future [2], [3]. In addition, a C/U-plane decoupled railway wirele network wa poed [4], [5]. To addre the blem caued by the direct connection between onboard paenger and wayide bae tation, an acce point AP) i deployed inide the train to collect paenger ervice in thi network. Thee ervice are then relayed to wayide bae tation by a mobile relay MR) deployed on the roof outide the train [1], [6]. More importantly, a concept of decoupling C-plane and U-plane i applied to thi network. To guarantee the tranmiion reliability and coverage, the more important C-plane i kept in the macro cell with relatively high-quality lower frequency band, i.e., conventional 800 MHz 2 GHz frequency band. In contrat, a the main data carrier, the U-plane need more tranmiion capacity, thereby being moved to mall cell operating at broadband higher frequency band, including frequency band higher than 5 GHz, up to 300 GHz. In [4], to vide a better undertanding of thi network architecture, we tudied the bandwidth matching between macro cell and mall cell, where 96 paenger C-plane can be upported in a ubframe with a bandwidth of 5 MHz in macro cell, and the correponding required bandwidth in mall cell i 21.2 MHz with a date rate of 500 Kbp per paenger. With more available bandwidth in macro cell, more paenger can be erved, and therefore more bandwidth i required in mall cell. Beide, a the data rate of paenger increae, the required bandwidth in mall cell will alo be enlarged. Conidering more tringent requirement for tranmiion reliability, train control information i entirely kept at macro cell without decoupling. Through a newly introduced interface, namely, X3, macro cell and mall cell can interact and ynchronize with each other. From a purely technical point of view, X3 interface employ the ame tocol of traditional X2 interface in LTE network [7] IEEE. Peronal ue i permitted, but republication/reditribution require IEEE permiion. See for more information.

2 YAN et al.: HARQ SCHEME FOR CONTROL/USER-PLANE DECOUPLED RAILWAY WIRELESS NETWORKS 2283 Through retranmiion to enure tranmiion reliability, the hybrid automatic repeat requet HARQ) tocol play an important role in LTE network [8] [11]. However, in C/U-plane decoupled railway wirele network, the C-plane ignaling, including HARQ acknowledgment, i tranmitted on relatively high-quality lower frequency band, while the U-plane data are carried by higher frequency band. In other word, the ACK/NACK meage receiving and data tranmitting of an HARQ ce are in a macro cell and a mall cell, repectively, not in the ame network node. Therefore, macro cell need to forward the received HARQ acknowledgment to mall cell via X3 to intruct mall cell what to do next, and handle retranmiion if the acknowledgement i an NACK or end new data if it i an ACK. Obviouly, the decoupling C-plane and U-plane caue more overhead in X3. Since in thi paper we focu on the HARQ retranmiion latency, the overhead i taken into account in the form of forwarding latency of X3 interface. According to the field tet reult in [12], the forwarding latency of X2 interface i about 1 m on average and 6 m at maximum. Becaue X3 interface adopt the ame tocol a X2 interface, the above reult alo apply to X3 interface. Conequently, the time interval between two adjacent re)tranmiion of an HARQ ce in C/U-plane decoupled railway wirele network i enlarged. How to reduce the aggravated latency of HARQ retranmiion in thi network become a ignificant challenge. In the conventional network architecture, C-plane and U-plane hare the ame frequency reource in macro cell. The available reource for the C-plane in LTE network are very limited becaue only the firt three OFDM ymbol in a ubframe are ued for control channel [11]. In contrat, in the C/U-plane decoupled railway wirele network, the reource of lower frequency band in macro cell are all ued for the C-plane. A a conequence, ome relatively high-quality and pare pectrum reource may become available in macro cell, which could be further exploited to ait mall cell to handle retranmiion, aiming at reducing the retranmiion latency. To um up, the main contribution and noveltie of thi paper include: 1) The timing of the conventional HARQ cheme under C/U-plane decoupled railway wirele network i analyzed to invetigate how decoupling C-plane and U-plane can affect the retranmiion latency of HARQ, demontrating the critical iue of aggravating the latency blem of HARQ in C/U-plane decoupled railway wirele network. 2) To reduce the HARQ retranmiion latency in C/U-plane decoupled railway wirele network, we poe a collaborative HARQ cheme in which the potential pare lower frequency band reource in macro cell excluding thoe ued for C-plane tranmiion are exploited to help mall cell handle retranmiion. Furthermore, double copie of data are concurrently tranmitted in macro cell and mall cell in the ce of retranmiion in order to reduce the total number of required retranmiion and the retranmiion latency. Accordingly, the timing of the poed collaborative HARQ cheme i deigned. Conidering the ignificant difference of ignal pagation between the lower and higher frequency band, the ce of data combining i tudied from the hardware point of view. 3) To determine how to realize collaborative retranmiion between different network node in macro cell and mall cell, a collaborative tranmiion framework between macro cell and mall cell i poed to reduce the aggravated latency and enhance the flexibility of bandwidth extenion for C/U-plane decoupled railway wirele network. Thi framework can alo be generalized to common U-plane data collaborative tranmiion beyond railway wirele network. To thi end, the channel mapping i redeigned to conform to thi framework. 4) For both the conventional and poed cheme, we conduct theoretical analyi of the average number of retranmiion related to the um of independent Gamma variable. Baed on the analytical reult, the average number of retranmiion, the average retranmiion latency and the average tranmiion rate of the conventional and poed cheme are compared through numerical experiment. The remainder of thi paper i organized a follow. In Section II, we decribe the timing of the conventional HARQ cheme applied to C/U-plane decoupled railway wirele network and preent the aggravated retranmiion latency blem. In Section III, the collaborative HARQ cheme to mitigate thi blem i poed. In Section IV, to enhance the flexibility of bandwidth extenion of the C/U-plane decoupled railway wirele network, a general collaborative tranmiion framework i poed. In Section V, analytical reult for the average number of retranmiion related to the um of independent Gamma variable are derived. In Section VI, imulation reult are illutrated and analyzed in detail. Finally, concluion are drawn in Section VII. II. CONVENTIONAL HARQ SCHEME IN C/U-PLANE DECOUPLED RAILWAY WIRELESS NETWORKS Thi ection preent the C/U-plane decoupled railway wirele network architecture and then dicue the blem of timing and aggravated retranmiion latency of the conventional HARQ cheme working in thi network architecture. A. C/U-Plane Decoupled Railway Wirele Network In order to increae the ytem capacity to meet the wirele acce demand of train paenger, a C/U-plane decoupled railway wirele network wa poed in [4], [5]. A hown in Fig. 1, an AP and an MR are deployed on the roof inide and outide train, repectively, o a to vide a dependable connection for train paenger. The inide AP firtly collect the ervice of train paenger, and then thee ervice are forwarded to the wayide bae tation via the MR. In thi way, the blem caued by the direct connection between onboard paenger and wayide bae tation, uch a large penetration lo, high mobility and group handover, can be avoided. According to [1], backhaul link between MR and wayide

3 2284 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 8, AUGUST 2016 Fig. 1. C/U-plane decoupled railway wirele network. bae tation are the key capacity bottleneck. Therefore, the C/U-plane decoupled architecture i applied to thee link. To enable efficient mobility upport, the critical C-plane i kept in macro cell operating at relatively high-quality lower frequency band. The correponding U-plane, which i the main data carrier, i moved to broadband but relatively poor-quality higher frequency band to expand the ytem capacity. Specifically, ome low-rate ervice with abolute demand on tranmiion reliability, e.g., train control information, can be entirely ditributed to macro cell without decoupling. Since mall cell are only reponible for the U-plane without handling any control function, they are olely connected to the SGW, but not to the MME a hown in Fig. 1. Via X3 interface, macro cell and mall cell can exchange control ignaling and data a well a ynchronize with each other. In order to vide ome lowrate ervice that have tringent requirement for tranmiion reliability, in addition to the MME, macro cell can alo be connected to the SGW. B. Timing of the Conventional HARQ in C/U-Plane Decoupled Railway Wirele Network Stop and Wait SaW) i a typical HARQ retranmiion mechanim where the next tranmiion i performed only when the acknowledgment to the previou tranmiion i received [11]. To fully utilize the reource while waiting for acknowledgment, multiple HARQ cee are carried out. In the conventional network architecture, the number of ynchronou HARQ cee in frequency diviion duplex FDD) ytem i eight, i.e., the maximum latency caued by a retranmiion i 8 m. While in the C/U-plane decoupled railway wirele network, the maximum latency of HARQ i 8 + T d,max,where T d,max denote the maximum forwarding latency of X3 interface. According to the field tet reult in [12], the forwarding latency of X2 interface i about 6 m at maximum. Then, the time interval between two adjacent re)tranmiion of an HARQ ce in thi network i 14 m, which i much larger than that of HARQ in the conventional network. For the aynchronou HARQ and time diviion duplex TDD) ytem, the time interval i even larger. For clarity, in thi paper we take the downlink ynchronou HARQ of FDD ytem a a cae tudy. Neverthele, the ame analyi can be generalized to other cae a well. Moreover, to achieve a commie between the ytem complexity and performance, Fig. 2. Timing of the conventional HARQ cheme in C/U-plane decoupled railway wirele network. chae combining in which retranmiion contain the ame et of coded bit a the initial tranmiion [13] i adopted to combine the retranmitted data and the initially tranmitted data [14] [16]. After each retranmiion, the receiver ue maximum ratio combination MRC) to combine each received ignal and finally feed them to the decoder. Fig. 2 illutrate the timing of the conventional HARQ cheme applied to C/U-plane decoupled railway wirele network. For clarity, we focu on the timing of one HARQ ce. In downlink, the mall cell tranmit the initial data 1) denoted by D1)) on ubframe n. After decoding the received data, the MR generate the correponding acknowledgment. To guarantee the tranmiion reliability, the HARQ acknowledgment, a a kind of C-plane ignaling, are kept at relatively high-quality lower frequency band in macro cell. In the ituation hown in Fig. 2, the data in the initial tranmiion are not correctly received. Thu, the MR generate an NACK and end it to the macro cell on ubframe n + 4). Then, the macro cell forward thi NACK to the mall cell via X3. In conideration of the forwarding latency of X3 interface, the ubequent retranmiion for thee erroneouly received data i performed on ubframe n T d,max ). A hown in Fig. 2, two retranmiion are performed before the data are uccefully decoded. For D1), the total latency caued by retranmiion i T d,max )= T d,max ) m. Compared to the conventional network in which the latency caued by two retranmiion i 16 m, 2 T d,max m extra latency i brought in, aggravating the latency blem in HARQ epecially for latency-enitive ervice. For clarity, the wirele tranmiion delay i neglected in Fig. 2. Therefore, the ubframe, on which the ame data are tranmitted and received, are aligned in the time domain. III. PROPOSED COLLABORATIVE HARQ SCHEME In the above ection, the aggravated retranmiion latency blem of the conventional HARQ cheme working in C/U-plane decoupled railway wirele network i dicued. To mitigate the blem, the principle and timing of the poed collaborative HARQ cheme are preented in thi ection. A. Principle of the Collaborative HARQ Scheme Due to the change in C/U-plane decoupled railway wirele network where the whole lower frequency band of macro

4 YAN et al.: HARQ SCHEME FOR CONTROL/USER-PLANE DECOUPLED RAILWAY WIRELESS NETWORKS 2285 output the two parallel baeband ignal of a retranmiion to the demodulator. Then, through chae combining, the two demodulated bit tream are combined with the initial tranmiion tored in the buffer memory. Finally, the combined ignal are fed to the decoder for a new decoding attempt. Fig. 3. a) Principle of the poed collaborative HARQ cheme for C/U-plane decoupled railway wirele network. b) Timing of the poed cheme. cell are ued for C-plane tranmiion, there may be pare pectrum reource on thee high-quality band. Suppoe that the bandwidth of a macro cell i 10 MHz correponding to 8400 available reource element RE) in 1 m, and 30 percent of 1000 uer in a train are active and cheduled in a frame of 10 m. To make a commie, two control channel element CCE), coniting of 72RE are aigned to the phyical downlink control channel PDCCH) of a uer, and every eight uer hare a phyical HARQ indicator channel PHICH) a in conventional LTE network. Then, the total conumed reource by control channel and reference ignal in 1 m are 3008RE and there are till 5392RE unued. Baed on thi obervation, to mitigate the aforementioned retranmiion latency blem, a low-latency collaborative HARQ cheme that exploit the poible pare reource in macro cell to help mall cell handle retranmiion i poed, a hown in Fig. 3a). Conidering the large ignal pagation difference between dicontinuou higher and lower frequency band, if neceary, two dedicated antenna and receiving circuit are integrated at the MR ide to individually handle the radio frequency RF) ignal received from macro cell and mall cell. Obviouly, compared to the conventional network architecture, although the C/U-plane decoupled network architecture can greatly increae the ytem capacity, it come at the expene of higher hardware complexity. After converting RF ignal to baeband ignal, the receiving circuit B. Timing of the Collaborative HARQ Scheme Correpondingly, the downlink timing of the poed collaborative HARQ cheme i depicted in Fig. 3b). After receiving an NACK on ubframe n + 4), which i the HARQ acknowledgment to the initially tranmitted D1) on ubframe n, the macro cell forward it to the mall cell via X3, reulting in a maximum T d,max latency. If there are pare pectrum reource in the macro cell, the mall cell will forward the retranmitted data to the macro cell via X3. Similarly, a maximum T d,max latency i induced again. A a cae tudy, under the ame ituation a depicted in Fig. 2, in the poed cheme the data are correctly decoded after one retranmiion which actually include double copie of the retranmitted data. The latency of a retranmiion with double copie of data concurrently tranmitted in the poed cheme i T d,max ) m. Compared to the conventional cheme hown in Fig. 2, under the ame ituation, the total latency i reduced by 8 m in the poed cheme. When D1) i correctly decoded, the MR feed back an ACK to the macro cell on the ubequent ubframe after 4 m. Then, with a maximum T d,max latency, the acknowledgment i further forwarded to the mall cell. A it i an ACK, the mall cell directly perform another new data tranmiion on ubframe n T d,max ) without the need to forward data to the macro cell any more. Baed on the above analyi, we can figure out the timing between two adjacent re)tranmiion in the poed collaborative HARQ cheme. With repect to a retranmiion, the time interval between thi retranmiion and the previou re)tranmiion i T d,max ) m. The time interval between a new data tranmiion and the previou re)tranmiion i 8 + T d,max ) m. According to thee timing, the receiver can receive all data in equence accurately. IV. PROPOSED COLLABORATIVE TRANSMISSION FRAMEWORK Different from the conventional network architecture, C-plane and U-plane in the C/U-plane decoupled wirele network architecture are completely eparated into different network node. Therefore, channel mapping need to be redeigned. In thi paper, we take LTE tocol a a deign benchmark. A hown in Fig. 4, in the newly redeigned channel mapping method, all control channel are kept at macro cell, and all traffic channel are moved to mall cell. In macro cell, a new MAC uplink or downlink) control channel i.e., UD)L-CCH) i defined to carry ome of up-layer RLC control channel, which i finally mapped to a converged phyical uplink or downlink) control channel CPUD)CCH). In mall cell, a new MAC traffic channel, uplink or downlink) traffic channel UD)L-TCH), i defined and finally mapped to a phyical uplink or downlink) traffic channel PUD)TCH). In term of phyical reource, all C-plane channel are eventually mapped to lower frequency band, and all U-plane channel

5 2286 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 8, AUGUST 2016 Fig. 4. Collaborative tranmiion framework. are mapped to higher frequency band. Therefore, the phyical control format indicator channel PCFICH) in the conventional coupled LTE network, which i ued to dicriminate the boundary of the control and traffic channel haring the ame frequency band, i aved in the C/U-plane decoupled network architecture. To realize collaborative tranmiion between two different network node, i.e., macro cell and mall cell, new peer entitie, namely, collaborative tranmiion entitie CTE) are poed to be deployed in the MAC layer of macro cell and mall cell. Through an X3 interface, the CTE in different bae tation can communicate with each other. Note that only the potential pare reource in collaborative tranmiion macro cell are exploited to help tranmit U-plane data of mall cell. The real control of thee U-plane data tranmiion, including HARQ deciion, i till under the control of mall cell. Therefore, a hown Fig. 4, the forwarded U-plane MAC PDU from mall cell are directly mapped to phyical reource of macro cell without going through the HARQ module of macro cell. After receiving the data, MR will end the correponding HARQ acknowledgement contained in PHICH on more dependable lower frequency band, and then they are forwarded by macro cell to mall cell. Finally, baed on thi information, mall cell make deciion for the forthcoming tranmiion. For clarity, the whole collaborative tranmiion ce conit of the following five tep: Step 1: In the macro cell, it MAC controller periodically inform it CTE of the reource uage of lower frequency band. Step 2: Via X3, the reource uage of the macro cell i forwarded by it CTE to the CTE of the mall cell. Step 3: If there are pare pectrum reource in the macro cell, the MAC controller of the mall cell will tranfer ome data to it own CTE. Step 4: The CTE of the mall cell forward thee data to the CTE of the macro cell. Then, baed on the current ytem tate, the MAC controller of the macro cell make a deciion on the reource and tranmiion format ued for collaborative tranmiion. Step 5: With the help of the macro cell, the U-plane data can be tranmitted on both lower and higher frequency band, enhancing the pectrum utilization of lower frequency band. After receiving the data, MR will feed back HARQ acknowledgement contained in PHICH on more dependable lower frequency band. Then, macro cell forward thee acknowledgement to mall cell via X3, baed on which mall cell make deciion for the following tranmiion. Moreover, a hown in Fig. 4, in the macro cell, a new MAC channel, namely, collaborative tranmiion channel CTCH) and a new PHY channel, namely, phyical collaborative tranmiion channel PCTCH) are brought in to accommodate collaborative tranmiion. To intruct MR to preciely receive data in the framework, the meage indicating whether a collaborative tranmiion i active or not and the reource a well a the tranmiion format ued for tranmiion in macro cell and mall cell i carried by PDCCH. From the phyical reource point of view, the poed framework equivalently etablihe a virtual bridge between completely dicontinuou lower and higher frequency band, and thu enhancing the flexibility of bandwidth extenion for C/U-plane decoupled railway wirele network. Conidering the general applicability, the framework i decribed from a more general perpective above. Although the ytem capacity can be intuitively increaed if the poible pare reource are ued to tranmit the general U-plane data, for broadband C/U-plane decoupled railway wirele network the capacity i not a key blem temporarily. Neverthele, a what will be hown in Section VI, if they are ued to help retranmiion, the average retranmiion latency can be highly reduced. Therefore, to mitigate the latency blem, in thi paper we ue thi collaborative tranmiion framework to upport collaborative retranmiion between macro cell and mall cell. That i, the forwarded data by mall cell are retranmitted data, and they are mapped on both higher frequency band and lower frequency band in duplication. It i alo notable that if there i no pare pectrum reource in macro cell, the ytem will degenerate to the conventional cheme a hown in Fig. 2. V. A NALYTICAL MODELS In thi ection, we firt decribe the wirele channel of high-peed railway cenario. Then, theoretical analyi of the average number of retranmiion related to the um of independent Gamma variable i conducted for the conventional and poed cheme. A. Wirele Channel Modeling Doppler effect in the high-peed movement cenario i a evere hindrance to high performance tranmiion. Fortunately, in the pecial railway cenario, the characteritic of determined running direction, regular running track and repetitive movement of train along fixed running track lead to a regular, repetitive and predictable Doppler hift curve, thereby making it eaier to trace and compenate the Doppler effect under thi cenario [17]. Moreover, there have been many reearch tudie focuing on the Doppler effect etimation and compenation for high-peed railway cenario, uch a in [18], and [19]. Baed on thi obervation, in thi paper we aume that the Doppler effect can be perfectly compenated and ha almot no influence on the final performance for both the conventional cheme and the poed cheme, which alo enure the fairne when

6 YAN et al.: HARQ SCHEME FOR CONTROL/USER-PLANE DECOUPLED RAILWAY WIRELESS NETWORKS 2287 TABLE I DESCRIPTIONS OF MCS MODES Fig. 5. Geometric ketch for the theoretical analyi. comparing the performance of two cheme in both theoretical analyi and imulation experiment. For high-peed railway cenario, the mot typical terrain i viaduct, in which wirele channel can be apximated a line of ight LOS) following a Rician ditribution [20], [21]. To tudy the combining performance of re)tranmiion, we need to derive the joint bability denity function PDF) and cumulative ditribution function CDF) of multiple Rician variable. However, according to [22], it i almot impoible to obtain an exact analytical joint PDF or CDF for more than three Rician variable. Fortunately, it i widely reported that Nakagami tatitic can cloely apximate Rician tatitic with a relationhip between the Rician parameter K and Nakagami parameter m a [23], [24] K + 1)2 m = 2K + 1). 1) Under the Nakagami model, the received ignal to noie ratio SNR) follow a Gamma ditribution, i.e., γ Gammam, m/γ), of which the PDF can be expreed a ) m m γ m 1 f γ γ) = γ Γm) e mγ/γ Sγ) 2) where Sγ) i a unit tep function and i defined a { 1, γ > 0 Sγ) = 0, otherwie and γ denote the average received large-cale SNR, i.e., γ = E[γ], wheree[ ] i the expectation operator. Fig. 5 decribe the geometric ketch to calculate the average large-cale SNR of received ignal from macro cell and mall cell in C/U-plane decoupled railway wirele network. In conideration of the linear topology of wirele communication network in railway cenario, the macro cell and mall cell in Fig. 5 are deployed on a traight line with vertical ditance of d min to the rail. R, R m and a denote the radiu of a macro cell, the radiu of a mall cell and the overlapping ditance of mall cell, repectively. To clarify the analyi, only one macro cell i conidered in thi paper, i.e., the analyi cope of d i from a to 2R m a. Since macro cell and mall cell operate at different frequency band with different characteritic uch a different coverage radiue, for clarity we ue the ubcript m and to repreent the parameter of macro cell and mall cell, repectively. A hown in Fig. 5, uppoe that the train tart from the original point and travel through ditance of d along the abcia-axi direction, then the ignal pagation ditance, x m d), from the macro cell to the train i x m d) = d R m ) 2 + d 2 min. 4) 3) Similarly, the ignal pagation ditance, x d), from the current erving mall cell to the train i d x d) = d Di Di ) 2 + d 2 min 5) Di 2 where Di = 2R a i defined to implify the expreion of Eq. 5). Generally, the average received large-cale SNR can be calculated a γx) = P t PLx) N 0 6) where P t i the tranmit power, PLx) i the large-cale path lo, and N 0 i the noie power. Then, the average large-cale SNR of received ignal from the macro cell and the current erving mall cell can be eparately expreed a γ m x m )= P m,t PL m x m ) N 0 7) γ x )= P,t PL x ) N 0. 8) Baed on [25], for a given modulation and coding cheme MCS) i, the packet error rate PER) related to the intantaneou SNR can be apximated a { 1, if γ<γi th F i γ) a i exp g i γ), if γ>γi th 9) where a i, g i,andγi th are three parameter determined by MCS i a lited in Table I [25], [26]. B. Theoretical Analyi of the Conventional HARQ Scheme To make the analyi more undertandable, the effect of incorrect receiving or loing of HARQ acknowledgment on the overall performance are not conidered. Let Nmax con denote the maximum number of permiible retranmiion for the conventional HARQ cheme in C/U-plane decoupled railway wirele network. Since in the conventional HARQ cheme mall cell perform all initial tranmiion and retranmiion, for MCS i where i {1, 2, 3}, the average number of required retranmiion of the conventional cheme, excluding the initial tranmiion, can be derived a N max con n 1 L i,con = E 10) n=1 l=0 F i γ,j

7 2288 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 8, AUGUST 2016 where γ,0 i the received SNR of the initial tranmiion in the mall cell. The detailed derivation of Eq. 10) i given in Appendix A. Suppoe that the velocity of the train i v = 360 km/h and the center frequency of mall cell i f = 5 GHz. Then, we can get the maximum Doppler Shift in mall cell a f d, = vf /c = 1.67 khz, where c i the light peed. Correpondingly, the coherence time of wirele channel in mall cell i T c, = 0.423/f d, = 0.25 m [27], which i much horter than the time interval between two adjacent re)tranmiion of an HARQ ce. Thu, it i reaonable to aume that the re)tranmiion of an HARQ ce experience independent wirele channel. Additionally, taking LTE network a the analyi baeline, the value of Nmax con i 6 [28], and then the time interval between the initial tranmiion and the ixth retranmiion i T d,max ) m. Baed on the field tet reult in [12], let T d,max = 6m.Andthen48+ 6 T d,max = 84 m. During the time interval of 84 m, the running ditance of the train i 8.4 m, which i relatively much horter than the ditance between bae tation and the rail. Therefore, all re)tranmiion in an HARQ ce can be aumed to have the ame average large-cale SNR. A a reult, γ,j with j = 0, 1, 2...in Eq. 10) follow an independent and identical ditribution. According to [29], we can get Υ,l = l γ,j Gammal + 1)m,m /γ ). Therefore, the average number of retranmiion of the conventional HARQ cheme in C/U-plane decoupled railway wirele network can be rewritten a L i,con N max con = n 1 n=1 l=0 ) Γ m γi th low γ, l+1)m ) γ l+1)m +a i 1+g i Γl+1)m ) m ) ) Γ m low γ +g i γ th i, l+1)m 1 Γl+1)m ) 11) where Γ low y, m) i the lower incomplete Gamma function, defined a Γ low y, m) = y 0 xm 1 e x dx. The detailed derivation of Eq. 11) i given in Appendix A. With the obtained average number of retranmiion, the average latency caued by retranmiion of the conventional HARQ cheme in C/U-plane decoupled railway wirele network can be expreed a D con =8 + T d,max ) L i,con. 12) In the conventional HARQ cheme, only a copy of the data that are unuccefully decoded in the initial tranmiion i carried in a retranmiion. Accordingly, only a copy of reource i conumed in a retranmiion. For MCS i,after L i,con retranmiion, the average ytem tranmiion rate i R i,con = R i L i,con ) C. Theoretical Analyi of the Propoed Collaborative HARQ Scheme In the poed collaborative HARQ cheme, initial tranmiion are carried by mall cell, and the following retranmiion will be collaboratively accomplihed by mall cell and macro cell. Due to the difference of ignal pagation characteritic on lower and higher frequency band, for the train at the ame geographic location a hown in Fig. 5, the average large-cale SNR of received ignal from macro cell and mall cell may be different. Hence, the macro cell and mall cell bably chooe different MCS for the two copie of data in a ingle retranmiion baed on the aumption that the adaptive modulation and coding AMC) technique i ued. In thi cae, we cannot obtain the exact analytical expreion of the average number of retranmiion for the poed cheme. Neverthele, baed on the MCS with a higher order among the two MCS elected by the macro cell and mall cell, the upper limit of the average number of required retranmiion can be obtained for the poed cheme. Similarly, baed on the MCS with a lower order, the lower limit of the average number of required retranmiion can be derived. Let i m denote the elected MCS by the macro cell, where i m {1, 2, 3}. A lited in Table I, the lower the value of i and i m are, the lower the order of MCS i. With the definition of x =maxi m,i ),the upper limit of the average number of required retranmiion in the poed collaborative HARQ cheme i obtained a N max L E F i γ,0 )+ F i γ,0 ) n=2 n 1 F x l=1 γ,j + γ m,j. 14) The detailed derivation of Eq. 14) i given in Appendix B. For clarity, we eparately conduct the derivation for each item in Eq. 14). Baed on the above analyi, we can obtain E [F i γ,0 )] = P γ,0 <γi th ) + ai e g i γ,0 P γ,0 >γi th ) ) m γi Γ th low γ,m ) γ m = + a i 1 + g i Γm ) m 1 Γ )) low m /γ + g i ) γi th,m. Γm ) 15) Similarly, uppoe that the frequency center of macro cell i f m = 2 GHz. Then, the coherence time of wirele channel in macro cell i T c,m = 0.423/vf m /c) =0.63 m, which i much horter than the time interval between two adjacent re)tranmiion in the poed cheme. Therefore, we can aume that the wirele channel experienced by re)tranmiion of an HARQ ce in the macro cell are independent on each other. Let T d,max = 6 m. Conidering that double copie of data are concurrently tranmitted in a retranmiion in the poed cheme, the maximum number of permiible retranmiion of the poed cheme i et

8 YAN et al.: HARQ SCHEME FOR CONTROL/USER-PLANE DECOUPLED RAILWAY WIRELESS NETWORKS 2289 to Nmax = 3. Then, the time interval between the initial tranmiion and the third retranmiion i T d,max )= 60 m. During thi period, the running ditance of the train i 6 m, which i relatively much maller than the ditance between bae tation and the rail. Hence, the average large-cale SNR of all re)tranmiion in an HARQ ce i almot equivalent. A a conequence, it can be aumed that wirele channel experienced by all re)tranmiion of an HARQ ce in the macro cell follow an independent and identical ditribution. Then, Υ,l = γ,j Gamma l + 1)m, m ) Υ m,l = For clarity, we define γ γ m,j Gamma lm m, m ) m. 16) γ m α,l =l + 1)m, 1 β = m 1 α m,l = lm m, = m m. 17) β m γ m A macro cell and mall cell operate at completely different frequency band, we can aume that tranmiion in macro cell and mall cell are divere and independent. That i, Υ,l and Υ m,l are two independent Gamma variable with different parameter. According to [30], [31], the PDF of the um of the two variable, denoted by Y =Υ,l +Υ m,l, can be derived a ) δ υ y ρ+υ 1 e y/β q gy) = C υ=0 Γρ + υ)β ρ+υ q γ Sy) 18) where q, C, ρ and δ are interim coefficient to implify the expreion of Eq. 18), and q = min β,β m ) q {,m} C = βq β {,m} q ρ = α,l + α m,l δ υ+1 = 1 δ 0 = 1. ) α{,m} q,l υ+1 υ+1 κ=1 α {,m} q,l 1 β q β {,m} q ) κ δυ+1 κ 19) Then, we can obtain the expectation of the lat item in Eq. 14) a E F x γ,j + γ m,j = C υ=0 Γlow γ th δ x /β q,ρ+ υ ) υ + a x 1 + g x β q ) ρ+υ) Γρ + υ) 1 Γ low 1/βq + g x )γx th,ρ+ υ ) )). Γρ + υ) 20) Conidering the practical application, we take the firt Ω item of the infinite erie in Eq. 20) a an apximation, that i, E F x γ,j + C Ω υ=0 δ υ γ m,j ) Γ γ th x low β q,ρ+ υ + a x 1 + g x β q ) ρ+υ) Γρ + υ) 1 Γ low 1/βq + g x )γx th Γρ + υ) ),ρ+ υ). 21) To ubtitute the above reult into Eq. 14), the upper limit of the average number of retranmiion in the poed collaborative HARQ cheme, denoted by L upper, can be rewritten a Eq. 22), hown at the bottom of the page. Similarly, if we define z = mini m,i ), the lower limit of the average number of required retranmiion in the poed cheme, denoted by L lower, can be obtained a Eq. 23), hown at the bottom of the next page. With the above reult, the cope of the latency caued by retranmiion in the poed cheme i T d,max ) L lower D T d,max ) L upper. 24) In the poed collaborative HARQ cheme, double copie of data are concurrently tranmitted in a retranmiion, leading to the conumption of double copie of reource. Therefore, L L upper = Γ ) low m γi th /γ,m Γm ) + N max n=2 n 1 C l=1 Γlow m γ th i /γ,m ) Ω υ=0 Γm ) ) m γ + a i 1 + g i 1 Γ low m ) m γ + a i 1 + g i m Γlow γ th δ x /β q,ρ+ υ ) υ + a x 1 + g x β q ) ρ+υ) 1 Γ low Γρ + υ) )) m /γ + g i ) γi th,m Γm ) 1 Γ ))) low m /γ + g i ) γi th,m Γm ) 1/βq + g x )γx th,ρ+ υ ) Γρ + υ) ))) 22)

9 2290 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 8, AUGUST 2016 the cope of the average ytem tranmiion rate in the poed cheme i R i 2L upper + 1 R R i 2L lower ) VI. PERFORMANCE ANALYSIS AND COMPARISON Baed on the theoretical analye in Section V, numerical reult are vided in thi ection to compare the performance of the conventional and poed cheme. Simulation experiment are performed under two condition, i.e., under fixed MCS and under AMC. Due to the fact that AMC can enhance the tranmiion reliability in ome degree, the retranmiion latency performance imvement of the poed cheme under AMC i le remarkable than that under fixed MCS. A. Performance Comparion Under Fixed MCS In thi ection, to conduct fair and comprehenive performance comparion, imulation experiment are performed under the condition that the ame MCS i ued in mall cell and macro cell and i not adaptively changed when the train run through the whole wirele coverage. In thi way, we can invetigate the pure performance imvement of the poed cheme without the influence of AMC technique on the tranmiion reliability. Performance comparion of the average number of retranmiion, the average retranmiion latency and the average ytem tranmiion rate are illutrated in Fig. 6 8, repectively. Detailed imulation parameter etting are lited in Table II [4], [5]. For clarity, here two mall cell ymmetrically located in the coverage of a macro cell are conidered. Therefore, the following imulation reult are ymmetrical to the poition of the macro cell. For the purpoe of performance analyi, we focu on the reult of the left-hand ide a cae tudie. A hown in Fig. 6, overall the poed cheme require much fewer retranmiion compared to the conventional cheme. Moreover, the higher the MCS order i, the more remarkable the performance imvement of the poed cheme i. For intance, under the MCS of 64QAM, when the train i at the edge of the macro cell which alo correpond to the edge of mall cell, e.g., at d = 0.2 km, due to the low SNR in thee region both cheme arrive at their maximum permiible retranmiion. A the train move toward the mall cell, with double copie of data concurrently tranmitted Fig. 6. Performance comparion of the average number of retranmiion for different MCS mode. Fig. 7. Performance comparion of the average retranmiion latency for different MCS mode. in a retranmiion, the number of required retranmiion of the poed cheme i greatly reduced. For intance, at d = 0.3 km, the average number of retranmiion of the conventional cheme and the poed cheme are 2 and 1, repectively. At the center area of mall cell in which the L L lower = Γ ) low m γi th /γ,m Γm ) + N max n=2 n 1 C l=1 Γlow m γ th i /γ,m ) Ω υ=0 Γm ) ) m γ + a i 1 + g i 1 Γ low m ) m γ + a i 1 + g i m Γlow γ th δ z /β q,ρ+ υ ) υ + a z 1 + g z β q ) ρ+υ) 1 Γ low Γρ + υ) )) m /γ + g i ) γi th,m Γm ) 1 Γ ))) low m /γ + g i ) γi th,m Γm ) 1/βq + g z )γz th,ρ+ υ ) Γρ + υ) ))) 23)

10 YAN et al.: HARQ SCHEME FOR CONTROL/USER-PLANE DECOUPLED RAILWAY WIRELESS NETWORKS 2291 Fig. 8. Performance comparion of the average ytem tranmiion rate for different MCS mode. TABLE II SIMULATION PARAMETERS Alo double copie of reource are conumed by a retranmiion in the poed cheme. A depicted in Fig. 8, when the train i at the edge of the macro cell, the conventional cheme outperform the poed cheme in term of the average ytem tranmiion rate. Neverthele, with reduced retranmiion in the poed cheme, the performance gap between two cheme i very mall. On the contrary, for 16QAM and 64QAM, at the center area of the macro cell, thank to the high SNR collaborative retranmiion from the macro cell, much fewer retranmiion are needed in the poed cheme. A aforementioned, at d = 1 km in Fig. 6, the poed cheme reduce the average number of required retranmiion from about 5.8 to 1 and from 6 to 1 for 16QAM and 64QAM, repectively. Therefore, in thi region, the poed cheme obtain a higher average ytem tranmiion rate a hown in Fig. 8. Correpondingly, at d = 1 km, the performance imvement of the poed cheme under 64QAM and 16QAM i about 0.9 bit/ymbol and 0.6 bit/ymbol, repectively. However, a dicued above, for QPSK which ha a higher ability in enhancing the tranmiion reliability, the performance imvement of the average number of required retranmiion i le remarkable. At d = 1 km of Fig. 6, the average number of required retranmiion are reduced from about 1.3 to 0.9 by the poed cheme. Therefore, in Fig. 8, even at the center area of the macro cell, the conventional cheme till obtain a higher average ytem tranmiion rate than the poed cheme. SNR of received ignal from mall cell i high, e.g., at d 0.5, 0.7) km, the number of required retranmiion of both cheme are cloe to zero. When the train i at the center area of the macro cell which correpond to the edge of thee two mall cell, thank to the high SNR collaborative retranmiion from the macro cell, the performance imvement of the poed cheme in thi region i the mot remarkable. At d = 1 km, the number of required retranmiion of the poed cheme i about 1, while for the conventional cheme it reache up to 6. For other MCS, the whole trend i the ame a 64QAM. Neverthele, becaue lower order MCS can achieve higher retranmiion reliability, a hown in Fig. 6, the lower the MCS order i, the le remarkable the performance imvement i. Baed on the above reult, a hown in Fig. 7, with double copie of data concurrently tranmitted in a retranmiion in the poed cheme, to achieve the ame tranmiion reliability, the poed cheme require fewer retranmiion, thereby mitigating the latency blem caued by HARQ retranmiion. For example, at d = 1 km, the average retranmiion latency of 64QAM, 16QAM and QPSK i reduced by the poed cheme from about 84 m to 14 m, from about 80 m to 14 m and from about 18 m to 13 m, repectively. B. Performance Comparion Under AMC In practical wirele communication ytem, AMC technique are uually employed to adapt to variou wirele channel. If a wirele channel i of high quality, higher order MCS with higher tranmiion rate will be ued to imve the pectrum efficiency. Otherwie, lower order MCS will be implemented to enhance the tranmiion reliability. That i to ay, AMC can enhance the tranmiion reliability in ome degree. Hence, under AMC, the performance imvement of the poed cheme will not be very remarkable. To make thi more undertandable, the MCS adaption reult of the macro cell and mall cell during the train running through the coverage of the macro cell are illutrated in Fig. 9. The higher the received ignal quality i, the higher the MCS order i elected. For the center area of the macro cell and mall cell, the elected MCS mode i 64QAM. In contrat, QPSK i ued for cell edge. And 16QAM i ued for other area with intermediate ignal quality. For implicity, in thi paper, we only conider the three MCS lited in Table I, in which the SNR threhold of MCS witching are given in the lat row. Neverthele, by ubtituting the correponding MCS related parameter, the ame analyi method alo hold for other MCS. Fig. 10 depict the average number of retranmiion of both cheme under AMC, confirming the fact that the poed cheme till outperform the conventional cheme. Baed on the theoretical analyi, in the region of d 0.2, 0.34) km and d 0.72, 0.74) km where both the macro cell and mall cell adopt the ame MCS a hown in Fig. 9, the obtained number of retranmiion of the poed cheme in Fig. 10

11 2292 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 8, AUGUST 2016 Fig. 9. MCS adaptation reult. Fig. 11. Performance comparion of the average retranmiion latency and average ytem tranmiion rate under AMC. Fig. 10. Performance comparion of the average number of retranmiion under AMC. are exact value, not the upper or lower limit. For other region, a hown in Fig. 9, the macro cell and mall cell alway ue different MCS. And we can only obtain the upper limit and lower limit of the average number of retranmiion in thoe region for the poed cheme. Beide, a hown in Fig. 9, when the train i in thoe region, either the macro cell or the mall cell can vide a high SNR retranmiion. Therefore, in the poed cheme which employ both the macro cell and mall cell to handle retranmiion, for all MCS, the number of required retranmiion are very low and the performance difference among them i mall. A a conequence, a hown in Fig. 10, the two curve of the upper limit and lower limit of the average number of retranmiion in the poed cheme are almot overlapped. Then, in the following analyi, we can ue one of them a an apximate curve of the real average number of retranmiion for the poed cheme. Due to the fact that AMC technique can enhance the tranmiion reliability in ome degree, a hown in Fig. 10 the total number of required retranmiion of both cheme are le than 1.5 and the performance imvement of the poed cheme i not very remarkable. Fig. 11 compare the average retranmiion latency and the average ytem tranmiion rate comparion between two cheme under AMC. In the poed cheme, with double copie of data concurrently tranmitted in a retranmiion, the HARQ retranmiion latency i reduced. Neverthele, a aforementioned, due to the fact that AMC technique can enhance the tranmiion reliability in ome degree, the performance imvement i not very remarkable. A a kind of acrifice, alo with double copie of reource conumed in a retranmiion, the poed cheme ha lower average ytem tranmiion rate comparedto the conventionalcheme. However, the performance gap i very mall. In addition, thank to the concept of C-plane and U-plane decoupling which dramatically expand the bandwidth, the ytem capacity in thi network i not a critical blem. On the contrary, it i very important and neceary to mitigate the latency blem when linking the completely eparated C-plane and U-plane in different network node, which i alo the main goal of thi paper. Alternatively, excluding the pectrum reource ued by C-plane tranmiion, if there are till pare reource in macro cell, they can alo be exploited to tranmit general U-plane data via the poed collaborative tranmiion framework. Baed on thi obervation, in order to preent a more comprehenive performance tudy, in Fig. 12, the effective tranmiion rate of pare reource of macro cell ued in two different cae, i.e., collaborative HARQ retranmiion and general tranmiion, are compared. A hown in Fig. 12, for the cae that pare reource are ued for HARQ retranmiion, in the region of d 0.2, 0.5) km and d 0.7, 1) km thee reource are inveted to repeatedly retranmit the received erroneou packet, leading to a lower effective tranmiion rate in thee region compared to the cae that thee reource are ued for general tranmiion. Neverthele, a demontrated above, with the help of thee reource to handle retranmiion, the retranmiion latency i highly reduced. A a

12 YAN et al.: HARQ SCHEME FOR CONTROL/USER-PLANE DECOUPLED RAILWAY WIRELESS NETWORKS 2293 Fig. 12. Performance comparion of the effective tranmiion rate of pare reource ued in different ituation. matter of fact, thi i very common for the field of wirele communication where ome performance i acrificed for the imvement of other performance. Fortunately, for broadband C/U-plane decoupled railway wirele network, the capacity i not a key blem temporarily. On the contrary, the aggravated retranmiion latency blem in thi network i more nounced. Hence, it i more beneficial here to utilize the poible pare high-quality reource of macro cell to handle retranmiion. VII. CONCLUSION The recently poed C/U-plane decoupled railway wirele network aim to meet dramatically growing wirele acce demand of train paenger a well a the reliable tranmiion requirement of train control ytem. Although the whole ytem capacity i highly increaed in thi network with the U-plane moved to broadband higher frequency band, how to mitigate the latency blem when linking the completely eparated C-plane and U-plane in different phyical node become important, epecially for HARQ tocol which require frequent interaction between the C-plane and U-plane. To addre thi challenge, in thi paper we have poed a lowlatency collaborative HARQ cheme, in which if there are pare pectrum reource in macro cell excluding thoe ued by C-plane tranmiion, they will be exploited to help mall cell handle retranmiion. To realize collaborative tranmiion between two different network node, i.e., macro cell and mall cell, a novel collaborative tranmiion framework i poed. Although the framework in thi paper i ued for HARQ retranmiion, it can alo be developed for general collaborative tranmiion to enhance the flexibility of bandwidth extenion for C/U-plane decoupled railway wirele network. Accordingly, the channel mapping i alo redeigned to conform to thi framework. Both the theoretical analyi and imulation experiment are conducted under two different condition, i.e., with and without AMC. No matter under which condition, the poed cheme alway outperform the conventional cheme in term of the average retranmiion latency. Due to the fact that AMC technique can enhance the tranmiion reliability in ome degree, under the condition with AMC, the average ytem tranmiion rate of the poed cheme i lightly lower than that of the conventional cheme. Neverthele, for C/U-plane decoupled railway wirele network in which broadband higher frequency band are integrated, the ytem capacity i not a key point temporarily. In contrat, in thi network, how to mitigate the aggravated latency during HARQ retranmiion become important, which i exactly the reearch focu of thi paper. A a matter of fact, in the field of wirele communication, acrificing ome pectrum reource to gain the tranmiion reliability i a common mean. For our future work, we will conider the difference of ignal pagation characteritic on higher and lower frequency band, uch a different Doppler hift and frame tructure when combining the data received from macro cell and mall cell, o a to make the collaborative tranmiion cheme more practical and further increae the data combining performance. APPENDIX A DERIVATION OF EQUATIONS 10) AND 11) L i,con =PrSu,1 Fa,0 )+2PrSu,2 Fa,0,Fa,1 )+,..., +Nmax con 1)Pr Su,N con max 1 Fa,0,Fa,1,..., ) Fa,N con max 2 + N con max Pr ) Fa,0,Fa,1,...,Fa,N con max 1 Nmax 1 con n = E n 1 F i γ,j = E = = n=1 n 1 F i N con max 1 γ,j +N max con l=0 l=0 N max con n 1 F i γ,j n=1 l=0 Nmax con n 1 n=1 l=0 Nmax con n 1 n=1 l=0 F i γ,j P Υ,l <γi th ) +ai e g i Υ,l P Υ,l >γi th )) Γ low m γ th i γ, l + 1)m ) Γl + 1)m ) + a i ) l+1)m γ 1 + g i m ) ) Γ m low γ +g i γ th i, l+1)m 1 Γl+1)m ) 26) where Su indicate the event that data in thi tranmiion are uccefully decoded, and Fa denote the otherwie cae. The firt three equation are the derivation tep of Eq. 10), and the lat two equation are the derivation tep of Eq. 11).

13 2294 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 8, AUGUST 2016 APPENDIX B DERIVATION OF EQUATION 14) L =PrSu m,,1 Fa,0 )+2PrSu m,,2 Fa,0,Fa m,,1 ) +,...,+Nmax 1)Pr Su m,,n max 1 Fa,0,..., ) Fa m,,n max 2 + Nmax Pr ) Fa,0,Fa m,,1,...,fa m,,n max 1 Nmax 1 n n E n 1 F x γ,j + n=2 n 1 F i γ,0 ) l=1 F x γ,j + γ m,j γ m,j F x γ,j + γ m,1 F i γ,0 ) +N maxf i γ,0 ) N max 1 l=1 F x N max = E F i γ,0 )+ F i γ,0 ) n 1 l=1 F x n=2 γ,j + γ,j + γ m,j γ m,j. 27) REFERENCES [1] X. Zhu, S. Chen, H. Hu, X. Su, and Y. Shi, TDD-baed mobile communication olution for high-peed railway cenario, IEEE Wirele Commun., vol. 20, no. 6, pp , Dec [2] B. Ai et al., Challenge toward wirele communication for high-peed railway, IEEE Tran. Intell. Tranp. Syt., vol. 15, no. 5, pp , Oct [3] J. Calle-Sanchez, M. Molina-Garcia, J. I. Alono, and A. Fernandez-Duran, Long term evolution in high peed railway environment: Feaibility and challenge, Bell Lab Tech. J., vol. 18, no. 2, pp , Sep [4] L. Yan, X. Fang, and Y. Fang, Control and data ignaling decoupled architecture for railway wirele network, IEEE Wirele Commun., vol. 22, no. 1, pp , Feb [5] L. Yan and X. Fang, Reliability evaluation of 5G C/U-plane decoupled architecture for high-peed railway, EURASIP J. Wirele Commun. Netw., vol. 2014, pp. 1 11, [6] C.-X. Wang et al., Cellular architecture and key technologie for 5G wirele communication network, IEEE Commun. Mag., vol. 52, no. 2, pp , Feb [7] S. Ahmad and D. Datla, Ditributed power allocation in heterogeneou network with dual connectivity uing backhaul tate information, IEEE Tran. Wirele Commun., vol. 14, no. 8, pp , Aug [8] L. Lu, M. Xiao, and L. Ramuen, Relay-aided broadcating with intantaneouly decodable binary network code, in Proc. IEEE ICCCN, Maui, HI, USA, Jul. 2011, pp [9] M. Xiao and M. Skoglund, Multiple-uer cooperative communication baed on linear network coding, IEEE Tran. Commun., vol. 58, no. 12, pp , Dec [10] L. Lu, M. Xiao, L. Ramuen, and M. Skoglund, Efficient cheduling for relay-aided broadcating with random network code, in Proc. IEEE PIMRC, Toronto, ON, Canada, Sep. 2011, pp [11] E. Dahlman, S. Parkvall, and J. Sköld,LTE/LTE-Advanced for Mobile Broadband. Amterdam, The Netherland: Elevier, [12] L. Zhang, Y. Nagai, T. Okamawari, and T. Fujii, Field experiment of network control architecture for CoMP JT in LTE-advanced over aynchronou X2 interface, in Proc. IEEE 77th VTC Spring, Dreden, Germany, Jun. 2013, pp [13] D. Chae, Code combining A maximum-likelihood decoding apach for combining an arbitrary number of noiy packet, IEEE Tran. Commun., vol. 33, no. 5, pp , May [14] P. Frenger, S. Parkvall, and E. Dahlman, Performance comparion of HARQ with chae combining and incremental redundancy for HSDPA, in Proc. IEEE 54th Veh. Technol. Conf., Atlantic City, NJ, USA, 2001, vol. 3, pp [15] Performance comparion of Hybrid-ARQ cheme, Motorola, Stockholm, Sweden, TSG R , 3GPP Input Paper, [16] K. C. Beh, A. Doufexi, and S. Armour, Performance evaluation of hybrid ARQ cheme of 3GPP LTE OFDMA ytem, in Proc. IEEE 18th Int. Symp. PIMRC, Athen, Greece, Sep. 2007, pp [17] Y. Zhao, C. Yin, and J. Li, Learning- and optimization-baed channel etimation for cognitive high-peed rail broadband wirele communication, in Proc. IEEE ISCIT, Gold Coat, Qld., Autralia, Oct. 2012, pp [18] A. Kalakech, M. Berbineau, I. Dayoub, and E. Simon, Time domain LMMSE channel etimator baed on liding window for OFDM ytem in high mobility ituation, IEEE Tran. Veh. Technol., vol. 64, no. 12, pp , Dec [19] Q. Du, G. Wu, Q. Yu, and S. Li, ICI mitigation by Doppler frequency hift etimation and pre-compenation in LTE-R ytem, in Proc. IEEE ICCC, Beijing, China, Aug. 2012, pp [20] L. Liu et al., Poition-baed modeling for wirele channel on highpeed railway under a viaduct at 2.35 GHz, IEEE J. Sel. Area Commun., vol. 30, no. 4, pp , May [21] W. Luo, X. Fang, M. Cheng, and Y. Zhao, Efficient Multiple-Group Multiple-Antenna MGMA) cheme for high-peed railway viaduct, IEEE Tran. Veh. Technol., vol. 62, no. 6, pp , Jul [22] P. Dharmawana, N. Rajatheva, and C. Tellambura, On the trivariate Rician ditribution, IEEE Tran. Commun., vol. 56, no. 12, pp , Dec [23] M. Yacoub, M. Barbin, M. de Catro, and J. B. Varga, Level croing rate of Nakagami-m fading ignal: Field trial and validation, Electron. Lett., vol. 36, no. 4, pp , Feb [24] J. Proaki,Digital Communication, 4th ed. New York, NY, USA: McGraw-Hill, [25] H. Jin, C. Cho, N.-O. Song, and D. K. Sung, Optimal rate election for peritent cheduling with HARQ in time-correlated Nakagami-m fading channel, IEEE Tran. Wirele Commun., vol. 10, no. 2, pp , Feb [26] J. Franci and N. Mehta, EESM-baed link adaptation in OFDM: Modeling and analyi, in Proc. IEEE GLOBECOM, Atlanta, GA, USA, Dec. 2013, pp [27] T. S. Rappaport,Wirele Communication: Principle and Practice. Upper Saddle River, NJ, USA: Prentice-Hall, [28] W. Luo, X. Fang, S. Li, and Y. Xia, Poition aited coordinate HARQ in LTE ytem for high peed railway, in Proc. IEEE 77th VTC Spring, Dreden, Germany, Jun. 2013, pp [29] S. Kotz and J. W. Adam, Ditribution of um of identically ditributed exponentially correlated gamma-variable, Ann. Math. Statit., vol. 35, no. 1, pp , Mar [30] P. G. Mochopoulo, The ditribution of the um of independent gamma random variable, Ann. Int. Statit. Math., A, vol. 37, pp , Mar [31] M.-S. Alouini, A. Abdi, and M. Kaveh, Sum of gamma variate and performance of wirele communication ytem over Nakagami-fading channel, IEEE Tran. Veh. Technol., vol. 50, no. 6, pp , Nov Li Yan S 14) received the B.E. degree in communication engineering from Southwet Jiaotong Univerity, Chengdu, China, in 2012, where he i currently working toward the Ph.D. degree with the Key Laboratory of Information Coding and Tranmiion, School of Information Science and Technology. Her reearch interet include handover, network architecture, and reliable wirele communication for high-peed railway.

14 YAN et al.: HARQ SCHEME FOR CONTROL/USER-PLANE DECOUPLED RAILWAY WIRELESS NETWORKS 2295 Xuming Fang M 00) received the B.E. degree in electrical engineering, the M.E. degree in computer engineering, and the Ph.D. degree in communication engineering from Southwet Jiaotong Univerity, Chengdu, China, in 1984, 1989, and 1999, repectively. In September 1984, he wa a Faculty Member with the Department of Electrical Engineering, Tongji Univerity, Shanghai, China. He then joined the School of Information Science and Technology, Southwet Jiaotong Univerity, where he ha been a Profeor ince 2001 and the Chair of the Department of Communication Engineering ince He held viiting poition with the Intitute of Railway Technology, Technical Univerity at Berlin, Berlin, Germany, in 1998 and 1999 and with the Center for Advanced Telecommunication Sytem and Service, Univerity of Texa at Dalla, Richardon, TX, USA, in 2000 and He ha, to hi credit, around 200 high-quality reearch paper in journal and conference publication. He ha authored or coauthored five book or textbook. Hi reearch interet include wirele broadband acce control, radio reource management, multihop relay network, and broadband wirele acce for high-peed railway. Dr. Fang i the Chair of the IEEE Vehicular Technology Society of the Chengdu Chapter and an Editor of the IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY. Geyong Min M 01) received the B.Sc. degree in computer cience from Huazhong Univerity of Science and Technology, Wuhan, China, in 1995 and the Ph.D. degree in computing cience from the Univerity of Glagow, Glagow, U.K., in He i a Profeor of high-performance computing and networking with the Department of Mathematic and Computer Science, College of Engineering, Mathematic and Phyical Science, Univerity of Exeter, Exeter, U.K. Hi reearch interet include future Internet, computer network, wirele communication, multimedia ytem, high-performance computing, modeling, and performance engineering. Yuguang Michael Fang F 08) received the B.S./M.S. degree from Qufu Normal Univerity, Shandong, China, in 1987; the Ph.D. degree from Cae Wetern Reerve Univerity, Cleveland, OH, USA, in 1994; and the Ph.D. degree from Boton Univerity, Boton, MA, USA, in In 2000, he joined the Department of Electrical and Computer Engineering, Univerity of Florida, Gaineville, FL, USA, where he ha been a Full Profeor ince He held a Univerity of Florida Reearch Foundation Profeorhip from 2006 to 2009; a Changjiang Scholar Chair Profeorhip with Xidian Univerity, Xi an, China, from 2008 to 2011; and a Guet Chair Profeorhip with Tinghua Univerity, Beijing, China, from 2009 to Dr. Fang received the U.S. National Science Foundation Career Award in 2001 and the Office of Naval Reearch Young Invetigator Award in 2002 and wa a recipient of the Bet Paper Award from IEEE ICNP 2006). He ha alo received the UF Doctoral Diertation Advior/Mentoring Award, the 2011 Florida Blue Key/UF Homecoming Ditinguihed Faculty Award, and the 2009 UF College of Engineering Faculty Mentoring Award. He i the Editor-in-Chief of the IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, wa the Editor-in-Chief of the IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS ), and erve/erved on the Editorial Board of everal journal, including the IEEE TRANSACTIONS ON MOBILE COMPUTING , 2011 preent), the IEEE TRANSACTIONS ON COMMUNICATIONS ), and the IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS ). He ha been actively participating in conference organization uch a erving a the Technical Program Cochair for IEEE INOFOCOM 2014 and the Technical Program Vice Chair for IEEE INFOCOM 2005.