PAPER Effect of Joint Detection on System Throughput in Distributed Antenna Network

Transcription

1 DOI: /transcom.2018EBP3077 Publczed:2018/08/29 Ths advance publcaton artcle wll after proofreadng.

2 IEICE TRANS.??, VOL.Exx??, NO.xx XXXX 200x 1 PAPER Effect of Jont Detecton on System Throughput n Dstrbuted Antenna Network Haruya ISHIKAWA a), Student Member and Yuktosh SANADA b), Fellow SUMMARY Ths paper evaluates the throughput of a dstrbuted antenna network (DAN) wth multple moble termnal schedulng and the usage of jont maxmum-lkelhood detecton (MLD). Moble termnals are closer to the desred antennas n the DAN whch leads to hgher throughput and better frequency utlzaton effcency. However, when multple moble termnal schedulng s appled to the DAN, nterference can occur between transmtted sgnals from antennas. Therefore, n ths research, moble termnal schedulng along wth jont MLD s appled to reduce the effects of nterference. A system level smulaton shows that the usage of jont MLD n a densely packed DAN provdes better system throughput regardless of the numbers of moble termnals and fadng channels. key words: Dstrbuted Antenna Network, Jont Detecton, User Schedulng, Small Cell 1. Introducton Wth the on-gong popularty of moble devces such as smartphones and tablet PCs, there are ncreasng demands for hgh-speed and large capacty wreless communcatons [?,?]. To accommodate ths explosve traffc growth, the ffth generaton moble communcaton system (5G) s currently beng developed. One soluton to mprove spectrum effcency s the dstrbuted antenna network (DAN) [?]. DANs were orgnally ntroduced for ndoor rado communcatons smply as a mean of extendng coverage to dead spots [?]. However, the DAN s advantages n reducng power consumpton and ncreasng system capacty were quckly exploted for use n broadband mult-cell envronments [?]. In the DAN, several antennas, known as dstrbuted antennas (DAs), are dspersed n a cell. The DAs are physcally connected to a centralzed baseband unt (BBU) va hghspeed optcal fronthaul lnks [?]. Moble user termnals (MT) are connected to DAs dependng on BBU s confguraton such as schedulng crtera. Snce low-power DAs are geographcally placed throughout the cell to reduce access dstances, the DAN mtgates channel mpacts such as shadowng and fadng, whch deterorates transmsson capacty and coverage [?]. It has been shown that a conventonal cell model, such as a centralzed antenna network (CAN), does not allow as much, f not any, frequency reuse, resultng n the The author s wth the Dept. of Electroncs and Electrcal Engneerng, Keo Unversty a) E-mal: haruyashkawa@snd.elec.keo.ac.jp b) E-mal: sanada@elec.keo.ac.jp Fg. 1 Dstrbuted antenna network (DAN). DAN provdng better spectrum effcency [?]. Wth cell models that deploy multple antennas near each other, such as the DAN, frequency and channel reuse s a common technque for mprovng spectral effcency. However, reusng the same frequency n moble wreless systems often leads to a degraded lnk and servce qualty due to co-channel nterference (CCI) at cell coverage boundares. In order to suppress nterference from adjacent antennas, nterference rejecton combnng has been specfed n the 4G Long- Term-Evoluton (LTE) standard as a part of coordnated mult-pont transmsson, whch consequently ncreases fronthaul traffc [?]. The amount of fronthaul traffc may be a sgnfcant problem f the same technologes are appled to the DAN. In ths research, the DAN wth jont maxmum lkelhood detecton (MLD) s proposed for suppressng nterference. Jont MLD treats the receved sgnals as the superposton of the desred and the nterference sgnals and regards them as a sgnal wth a larger number of constellaton ponts [?,?,?]. Snce the total number of constellaton ponts ncreases to the multplcaton of modulaton orders of sgnal streams, the performance beneft obtaned by jont MLD leads to ncreased complexty. Schedulng the frequency resources to provde farness and mantan better network throughput becomes a problem when there are more MTs n a network. In [?], the authors dscuss three MT schedulng schemes, proportonal-far (PF), round-robn (RR), and maxcarrer-to-nterference (C/I), and ther effects on downlnk (DL) transmsson n the DAN. It was concluded that RR schedulng acheves hgh farness and throughput wth a smple algorthm. MT schedulng n the DAN s also dscussed n [?], where combnng multple schedulng algorthms to create a schedulng crtera acheved a far and hgh throughput system. The DAN wth DAs that transmt on the DL to multple MTs s assumed n ths research. In the as- Copyrght c 200x The Insttute of Electroncs, Informaton and Communcaton Engneers

3 2 IEICE TRANS.??, VOL.Exx??, NO.xx XXXX 200x sumed DAN, t s proposed that each MT demodulates sgnals for ts own and the other va jont MLD and mult-mt schedulng s appled. For the mult-mt schedulng, RR schedulng s used wth max-(c/n) as a selecton algorthm for DAs. The proposed scheme s compared aganst the conventonal scheme as well as a scheme wthout the use of jont MLD, where the throughput evaluaton s based on the numercal analyss of mult-cell smulaton usng constellaton constraned capacty (CCC). Ths paper s organzed as follows. Secton 2 descrbes the system model and the conventonal and proposed schemes. Secton 3 presents numercal results obtaned through computer smulaton. Secton 4 concludes ths research. 2. System Model 2.1 Cell Model Fg. 2 DAN cell model (N DA = 7). A hexagonal-cell concept s assumed for the DAN cell model [?]. Each macro cell has unformly dstrbuted N DA DAs. In ths research, N DA s set to seven. Each DA s placed n a smaller hexagonal-cell wth a radus of R, as shown n Fg.??. Surroundng the center cell, there are sx other outer macro cells wth the same confguraton. Also, the center cell s assumed to be the cell of nterest. In the assumed DAN, the same channel s shared by the DAs n the cell. Also, the same frequency channel can be assgned to two DAs n each macro cell. In the recever of a MT, jont MLD s appled to mtgate the CCI n the macro cell. The MT receves the sgnals transmtted from two DAs n the same macro cell. Wth the jont MLD, the recever extracts ts desred sgnal even though the sgnal for the other MT s transmtted from one of the other DAs on the same channel. 2.2 Schedulng For the multple MT schedulng, the RR schedulng of the MTs and the max-c/n schedulng (or selecton) of the DAs are appled at the same tme. For comparson, a sngle MT schedulng scheme s also consdered n ths research where each MT s chosen one at a tme and the DA wth the best C/I s chosen va the max- C/N schedulng. In the DAN, usng all of the DAs n a dense cell can lead to CCI and, as reserached n [?], turnng off some of the DAs n a mult-mt envronment could ncrease transmsson throughput. Ths s also the reason to reduce the number of recevng MTs to two per resource block (RB). Lmtng the number of MTs lead to low CCI and mproves system throughput wth jont MLD. Algorthm 1: Mult-MT schedulng algorthm. Data: P, N perm, N frames, N RB, H f α, H f β begn = 1 for f = 1 N frames do H r f α = { r} H f α H r f β = { r} H f β for r = 1 N RB do 1 l max α k α, k β P(), lβ max = arg max l α,l β (l α l β ) ( H r f α (k α, l α ) + H r f β (k β, l β )) = + 1 f > N perm then = 1 end end end end In detal, the proposed mult-mt schedulng algorthm s shown n Algorthm??. The k α -th and k β -th MTs are chosen, and each MT receves ts desred sgnal from the l α -th and l β -th DAs, respectvely. N MT represents the number of the MTs. N frame and N RB are the two resource factors, where N frame represents the total number of frames and N RB represents the total number of RBs. P defnes all of the permutatons of selectng two MTs out of the total N MT, and the total number of permutatons s gven by N perm = NMT P 2 where n P r = n! (n r)!. For example, the -th ndex P() contans the scheduled MT par, MT k α and MT k β, whch means (k α, k β ) P. Two MTs are selected per RB at each tme ndex. H f α and

4 ISHIKAWA and SANADA: EFFECT OF JOINT DETECTION ON SYSTEM THROUGHPUT IN DISTRIBUTED ANTENNA NETWORK 3 H f β are the channel matrx on the -th subcarrer wth a sze of N MT N DA and they are also assumed to be constant over N frames. H f α (k α, l α ) = h k α,l α and H f β (k β, l β ) = h k β,l β are the channel responses from the l α -th and l β -th DAs to the k α -th and k β -th MTs on the -th subcarrer, respectvely. In ths algorthm, the DAs for each MT s chosen for the scheduled MTs, k α and k β, by fndng the maxmum sum of the channel responses, Hr fα (k α, l α ) + H r f β (k β, l β ). As shown n lne 1, the scheduled DAs are represented by lα max and, and the total channel response for these pars s l max β h,max f = H r f α (k α, lα max ) + H r f β (k β, lβ max ). 2.3 Throughput Calculaton x lα Desred sgnal Modulaton lα-th DA(Transmt sde) x lβ Interference sgnal Modulaton lβ-th DA (Transmt sde) Fg. 3 lα-th Antenna lβ-th Antenna h kα, lα h k α, lβ h, h kα, lα kα, lβ n Nose kα n kα + Channel response System model for proposed scheme. Detecton kα-th MT In ths research, each MT s equpped wth a sngle antenna and DL transmsson s assumed. The system model for the proposed scheme s shown n Fg.??. In the assumed DAN, at most two DAs transmt at once on the same channel n the mult-mt schedulng. Each MT receves the desred sgnal as well as the nterference sgnal from the other DA. Each MT also receves nterference from the surroundng macro cells. Suppose that x l β (s l β ) s treated as an nterference sgnal from the other l β -th DA n the same cell and s l β s the ndex for the constellaton ponts of the nterference sgnal on the -th subcarrer. Also assume that E[ x l α (s l α ) 2 ] = [ x l β (s l β ) 2 ] = 1 and the number of constellaton ponts of the nterference sgnal s Nl β (0 s l β Nl β 1). The receved sgnal at the k α -th MT, yk α, on the -th subcarrer s gven as y k α = h k α,l α x l α (s l α ) + h k α,l β x l β (s l β ) (1) + l h k α,l x l (s l ) + n k α and smlarly, the k β -th MT s receved sgnal, y k β, s gven as y k β = h k β,l β x l β (s l β ) + h k β,l α x l α (s l α ) (2) + l h k β,l x l (s l ) + n k β, where h k,l s the channel response from the l-th DA to the concernng k-th MT. For throughput evaluaton, the nter-cell nterference s regarded as a Gaussan nose and s subject to Gaussan dstrbuton. Therefore, Eqs. (??) and (??) can be rewrtten as y k α = h k α,l α x l α (s l α ) + h k α,l β x l β (s l β ) + z k α, (3) y k β = h k β,l β x l β (s l β ) + h k β,l α x l α (s l α ) + z k β, (4) where zk s the sum of the addtve whte Gaussan nose (AWGN) and the nter-cell nterference and ts total varance s gven as σ 2. Two demodulaton schemes for MLD are researched n ths paper. In the frst scheme, system throughput when the MT apples MLD for the desred sgnal only s calculated. In ths case, the sgnal for the other MT n the same cell s treated as CCI. From Eqs. (??) and (??), the CCC wth the MLD s gven as Eqs. (??) and (??) where σ 2 ncludes nterference power that s subject to a dstance [?,?]. The system throughput s then calculated by the sum of the entre RB gven by, T m = T j = [ NRB N RB r T mkα r + T m r kβ r ]. (7) On the other hand, the throughput of the jont MLD scheme s also calculated. In ths case, the recever of the MT jontly demodulates the sgnals for the concernng MT and another one n the same cell. Compared to the MLD, the jont MLD requres the modulaton orders and the channel responses of the nterference sgnal as well as the desred sgnal and the complexty of the jont MLD ncreases snce the receved sgnal pont has to be compared wth N lα N lβ constellaton ponts. The CCC wth the jont MLD s gven as Eqs. (??) and (??). The system throughput s then calculated by [ NRB ] T j r k α r NRB + T j r k β r. (10) As for the sngle-mt schedulng, a sngle MT n each macro cell s scheduled for a partcular frame s RB. Therefore, MLD s appled. The system throughput s calculated by takng only the k r -th MT n Eq. (??), whch s gven by, T = N RB 3. Numercal Analyss 3.1 Smulaton Condtons Tk r r. (11) Mult-cell system-level smulaton s conducted

5 4 IEICE TRANS.??, VOL.Exx??, NO.xx XXXX 200x T m kα = log 2 (N lα ) 1 N lα N lβ T m kβ = log 2 (N lβ ) 1 N lβ N lα N lα N l β s lα =0 s z l =0 kα β N l β N lα s l =0 s β lα =0 zk β N lα t E log lα =0 exp( h kα,lα {x lα (s ) lα x lα (t )} + lα h kα,lβ x lβ (s ) + lβ z kα 2 /σ 2 ) 2 exp( h x (s ) + z 2 /σ 2 ) kα,lβ lβ lβ kα E log 2 t l =0 exp( h kβ,lβ {x lβ (s ) lβ x lβ (t )} + lβ h kβ,lα x lα (s ) + lα z kβ 2 /σ 2 ) β exp( h x (s ) + z 2 /σ 2 ) kβ,lα lα lα kβ N l β (5) (6) T j = log kα 2(N ) 1 lα N N lα lβ T j = log kβ 2(N ) 1 lβ N N lβ lα N lα N l β s lα =0 s z l =0 kα β N l β N lα s l =0 s β lα =0 zk β E log 2 E log 2 N lα t lα =0 N l β t l β =0 t l =0 exp( h kα,lα {x lα (s ) lα x lα (t )} + lα h kα,lβ {x lβ (s ) lβ x lβ (t )} + lβ z kα 2 /σ 2 ) β Nl β t l =0 exp( h {x (s ) x (t )} + z 2 /σ 2 ) kα,lα lβ lβ lβ lβ kα β Nl β t lα =0 exp( h kβ,lβ {x lβ (s ) lβ x lβ (t )} + lβ h kβ,lα {x lα (s ) lα x lα (t )} + lα z kβ 2 /σ 2 ) Nlα t lα =0 exp( h {x (s ) x (t )} + z 2 /σ 2 ) kβ,lβ lα lα lα lα kβ N lα (8) (9) to compare the system throughputs of the conventonal schemes, the sngle-mt schedulng, the mult- MT schedulng wth MLD, and the proposed mult-mt schedulng wth jont MLD scheme. Smulaton condtons are presented n Table??. A seven hexagonal macro-cell model wth a seven hexagonal DA cell n each macro-cell s assumed as shown n Fg.??. The nter-antenna dstance between adjacent DAs s changed from 25 m to 250 m. N MT MTs are dropped randomly wth an unform dstrbuton and they are dropped further than 5 m from DAs. Dstance dependent path loss wth a decay factor of 36.7 and lognormal shadowng wth standard devaton of 8 db s used for the propagaton model. The shadowng correlaton between the antennas are set to 0.5. A one-path Rcan fadng channel model wth a K-factor of 10 s assumed for the cell of nterest and a sx-path Raylegh fadng channel model s assumed for the macro-cells surroundng the cell of nterest. For comparson of dfferent channels, a sx-path Raylegh fadng channel model s also used for the cell of nterest. The total power of nterference from the other cells s requred for the calculaton of σ 2 and t s derved based on the dstances between the DAs and the MTs wth the assumed dstance dependent path loss model. The system bandwdth s set to 4.32 MHz. The number of RBs, N RB, s 24 and the number of subcarrers for each RB s 12. The symbols are modulated wth QPSK, 16QAM, 64QAM, or 256QAM on each subcarrer. Each DA has a heght of 10 m and the transmt power s 30 db. Each MT s set at a heght of 1.5 m and the recever nose densty s -174 dbm/hz. The throughputs n Eqs. (??), (??), and (??) of the conventonal and proposed schemes are smulated through Monte Calro smulaton [?]. The number of MT drops s 200 and the number of trals per MT drop s fve. The channel response s renewed for each tral. The number of transmt symbols for each Table 1 Smulaton condtons. Cell layout 7 DA network Inter-antenna dstance 25, 50, 100, 150, 200, 250 m Mnmum dstance 5 m between MT and DA Path loss log 10 (R) db, R: Dstance [km] Heght of antennas 10 m Heght of MTs 1.5 m Shadowng devaton 8 db Shadowng correlaton 0.5 Channel model (nner) One-path Rcan (K = 10) Sx-path Raylegh Channel model (outer) Sx-path Raylegh Number of MTs 5, 7, 10, 15, 20, 25 Transmt power 30 dbm Recever nose densty -174 dbm/hz System bandwdth 4.32 MHz RB bandwdth 180 khz Number of RBs 24 Modulaton scheme QPSK, 16QAM, 64QAM, 256QAM MT drops 200 Trals per MT drop 5 Number of symbols per tral 100 tral s 100. The same RR schedulng s appled over all of the cells and ths mples that only one or two DAs n each surroundng cell cause nterference to the cell on nterest.

6 ISHIKAWA and SANADA: EFFECT OF JOINT DETECTION ON SYSTEM THROUGHPUT IN DISTRIBUTED ANTENNA NETWORK Average Throughput (bt / RB / sec) Sngle-MT Mult-MT w/o Jont Detecton Mult-MT wth Jont Detecton Average Throughput (bt / RB / sec / m 2 ) Sngle-MT Mult-MT w/o Jont Detecton Mult-MT wth Jont Detecton Number of MTs Dstance (m) Fg. 4 Average throughput versus number of MTs. Fg. 5 Average throughput per area versus nter-da dstance. 3.2 Effect of MT Numbers The average throughput versus the number of MTs are presented n Fg.??. Ths fgure s smulated wth an nter-antenna dstance of 25 m. For the frst characterstc, as shown n Fg.??, the number of MTs (N MT ) does not effect on the over all average throughput of the system. Ths s due to the employment of the RR schedulng. Unlke the PF schedulng, even f the rato of unfavorable condtoned MTs ncreases, the DA assgns the same amount of RBs to each MT. The average throughputs for the mult-mt wth jont MLD, the mult-mt wthout jont MLD, and the sngle-mt schedulng are almost constant throughout the dfferent numbers of MTs. Ths s due to the fact that for each RB at a specfc tme frame, only one or at most two MTs are scheduled for transmsson dependng on the sngle or mult-mt schedulng. Ths means that n the mult-mt schedulng, the best throughput can be acheved f two MTs could transmt 256QAM symbols to ther desred DAs. In Fg.??, the effect of jont MLD s clear snce the average throughput goes up around 15% as compared to that wthout jont MLD and 40% as compared to that of the sngle-mt schedulng. 3.3 Effect of Inter-DA Dstance Fgure?? presents the average throughput per area versus the nter-da dstance of the DAs. The number of MTs are set to seven. The effect of jont MLD can be seen greatly accordng to the nter-antenna dstance of the DAs. As shown n Fg.??, when the dstance between the DAs are longer, such as from 150 to 250 m, the throughput dfferences among those schemes dmnsh. On the other hand, f the dstances between adjacent DAs s shorter, such as from 25 to 50 m, the throughput dfferences are greater. By comparng the sngle and mult-mt schemes, n the shorter nter-da dstance DAN, the mult-mt schemes provde better system throughput. In the mult-mt schemes, when the nter-antenna dstance s 25 m, the throughput wth jont MLD s hgher than the one wthout jont MLD, whereas n the other dstances, the mult-mt schedulng wth and wthout jont MLD acheves the same mprovements. Ths s due to the fact that n the shorter nter-da dstance DAN, the channel s more lkely to be nterfered by the sgnals from the surroundng DAs than that when the cells were more wdely spread out. Jont MLD provdes better throughput snce ths scheme mtgates the effect of the nterference. Throughput curves shown n Fg. 5 also mply that the DAN can acheve better spectrum effcency as compared to the CAN. From Fg. 2, the average dstance between the MT and the antenna n the CAN s about three tmes larger than that of the DAN. Wth a dstance of 75 m, even though the number of transmt MTs s two, the average throughput s less than 1/4 of the sngle MT wth a dstance of 25 m. Ths means that the number of transmt MTs ncreases to seven, whch s equvalent to the CAN wth a dstance of 75 m, the average throughput of the DAN wth a dstance of 25 m s better than that of the CAN wth a dstance of 75 m. In Fg.??, the cumulatve dstrbuton functon (CDF) curves of the system throughput for nter-da dstance of 25 m s presented. In the fgure, sngle-mt scheme only transmts around 8 bts / RB / sec snce only one MT s allowed to transmt at a specfc resource slot. At a partcular resource slot, the assgned MT s not affected by CCI snce nterference wll come only from the outer macro cells, whch s assumed to be farther than the connected antenna. Ths means that these MTs can transmt symbols wth hgher order modulaton, such as 256QAM. For the sngle-mt schedulng, the man source of the nterference s the sgnal from the closest DA n the adjacent outer cell.

7 6 IEICE TRANS.??, VOL.Exx??, NO.xx XXXX 200x Sngle-MT Mult-MT w/o Jont Detecton Mult-MT wth Jont Detecton Cumulatve Probablty Sngle-MT Mult-MT w/o Jont Detecton Mult-MT wth Jont Detecton Average Throughput (bt / RB / sec / m 2 ) Throughputs (bt / RB / sec) Dstance (m) Fg. 6 CDF for nter-da dstance of 25 m. Fg. 7 Average throughput per area versus nter-da dstance on Raylegh fadng channel. If only one DA n each of the surroundng outer cell s actvated due to the RR schedulng, sever nterference occurs only when the scheduled MT s located at the cell edge and the closest DA n the adjacent outer cell s transmttng. On the other hand, n the mult-mt schemes, at most two MTs can transmt for each resource slot. If two MTs can transmt at 256QAM, the throughput wll be 16 bts / RB / sec. As clearly shown n the fgure, around half of the resource slots n the mult-mt scheme wthout jont MLD can transmt at 8 bts / RB / sec, whereas the scheme wth jont MLD provdes the MTs that transmt at greater throughput whch means that those resource slots can now transmt two MTs at once. For the mult-mt schedulng, the man source of the nterference s the sgnal from the other MTs wthn the cell of nterest. In ths case, both of the MTs acheve throughput of 4.0 bts / RB / sec n most cases snce the nferences from the adjacent DAs lmt the throughput wthout jont MLD. It also shows how jont MLD can provde better system throughput when the DAs are closer to one another. In Fg.??, around 87% of the MTs are able to transmt at throughput of 9.7 bts / RB /sec wthout jont MLD, whereas 87% of the MTs are able to transmt at throughput of 11.6 bts / RB / sec. Jont MLD enables the smultaneous transmsson of all of the MTs n each resource slot. 3.4 Effect of Raylegh Fadng Channel Fgure?? presents the average throughput per area versus varyng nter-da dstances on the sx-path Raylegh fadng channel. Compared to the curves on the one-path Rcan fadng channel used n Fg.??, the overall throughput curves wth sx-path Raylegh fadng channel are hgher whle the tendences of the curves are smlar. Ths s because more nterference occurs on the Rcan fadng channel wth a K-factor of 10. It s smlar n the manner that as the nter- DA dstance becomes shorter, the throughput becomes larger, and vce versa. Also, the mult-mt schemes are deemed superor than the sngle-mt scheme and, n the mult-mt schemes, the use of jont MLD provdes better throughput performance. 4. Conclusons In ths research, the DAN wth jont MLD has been proposed by takng nto consderaton the schedulng of multple MTs. The system throughout for the proposed scheme was compared wth those of the conventonal sngle-mt schedulng and the mult-mt schedulng wthout jont MLD n a system level smulaton. The results show that usng jont MLD n the mult- MT schedulng scheme ncreases system throughput. In partcular, when the nter-da dstance s 25 m, the system throughput mproves around 15%. The same throughput s mantaned when the number of MTs changes and smlar trends are seen n the Rcan and Raylegh fadng channels. Acknowledgments Ths work s supported n part by a Grant-n-Ad for Scentfc Research (C) under Grant No.16K06366 from the Mnstry of Educaton, Culture, Sport, Scence, and Technology n Japan. References [1] DOCOMO 5G Whte Paper, NTT DOCOMO, INC., July [2] Moble Communcatons System for 2020 and Beyond Whte Paper, ARIB 2020 and Beyond Ad Hoc Group, Oct [3] F. Adach, W. Peng, T. Obara, T. Yamamoto. R. Matsukawa, and M. Nakada, Dstrbuted Antenna Network for Ggabt Wreless Access, 54th IEEE Internatonal Mdwest Symposum on Crcuts and Systems, Aug

8 ISHIKAWA and SANADA: EFFECT OF JOINT DETECTION ON SYSTEM THROUGHPUT IN DISTRIBUTED ANTENNA NETWORK 7 [4] A. A. M. Saleh, A. J. Rustako, and R. S. Roman, Dstrbuted Antennas for Indoor Rado Communcatons, IEEE Trans. on Commun., vol. 35, no.12, pp , Dec [5] W. Cho, and J. G. Andrews, Downlnk Performance and Capacty of Dstrbuted Antenna Systems n a Multcell Envronment, IEEE Trans. on Commun., vol. 6, no. 1, pp , Jan [6] H. Kaj, S. Kumaga, K. Temma, and F. Adach, Spectrum-Energy Effcency Tradeoff of Dstrbuted Antenna Network, IEEE 11th Vehcular Technology Socety Asa Pacfc Wreless Communcatons Symposum, Aug [7] S. Kumaga, R. Matsukawa, T. Obara, T. Yamamoto, and F. Adach, Spectral Effcency of Dstrbuted Antenna Network Usng MIMO Spatal Multplexng, IEEE 76th Veh. Technol. Conf., Sept [8] T. Abe, Y. Kshyama, Y. Kakura, and D. Imamura, Rado Interface Technologes for Cooperatve Transmsson n 3GPP LTE-Advanced, IEICE Trans. on Comun., vol. E94- B, no. 12, Dec [9] T. Yazak and Y. Sanada, Effect of Jont Detecton n Far User on Non-orthogonal Multple Access Downlnk, Wreless Pers. Commun., vol. 86, pp , Aug [10] T. Yazak and Y. Sanada, Throughput Performance of Non-orthogonal Multple Access wth Jont Detecton n Far User, Internatonal Symposum on Intellgent Sgnal Processng and Communcaton Systems, Nov [11] H. Myag and Y. Sanada, MMSE Interference Rejecton Followed by Jont Maxmum Lkelhood Detecton for Dstrbuted Antenna Network, 23rd Asa-Pacfc Conf. on Commun., Dec [12] Y. Sek and F. Adach, Downlnk Capacty Comparson of MMSE-SVD and BD-SVD for Cooperatve Dstrbuted Antenna Transmsson usng Mult-user Schedulng, IEEE 86th Veh. Technol. Conf., Sept [13] Y. Xn, R. Zhang, D. Wang, J. L, L. Yang, X. You, Antenna Clusterng for Bdrectonal Dynamc Network Wth Large-Scale Dstrbuted Antenna Systems, IEEE Access, vol. 5, pp , Feb [14] V.H. Mac Donald, Advanced Moble Phone Servce: The Cellular Concept, Bell Syst. Tech., vol. 58, no. 1, Jan [15] H. Tabassum, U. Sddque, E. Hossan, J. Hossan, Downlnk Performance of Cellular Systems Wth Base Staton Sleepng, User Assocaton, and Schedulng, IEEE Trans. on Commun., vol. 13, ssue 10, pp , July [16] G. Ungerboeck, Channel Codng wth Multlevel/Phase Sgnals, IEEE Trans. on Info. Theory, vol. IT-28, no. 1, pp , Jan Haruya Ishkawa was born n Tochg, Japan n He receved hs B.E. degree n electroncs engneerng from Keo Unversty, Japan n Snce Aprl 2018, he has been a graduate student n School of Integrated Desgn Engneerng, Graduate School of Scence and Technology, Keo Unversty. Hs research nterests are Dstrbtued Antenna System. Yuktosh Sanada was born n Tokyo n He receved hs B.E. degree n electrcal engneerng from Keo Unversty, Yokohama Japan, hs M.A.Sc. degree n electrcal engneerng from the Unversty of Vctora, B.C., Canada, and hs Ph.D. degree n electrcal engneerng from Keo Unversty, Yokohama Japan, n 1992, 1995, and 1997, respectvely. In 1997 he joned the Faculty of Engneerng, Tokyo Insttute of Technology as a Research Assocate. In 2000 he joned Advanced Telecommuncaton Laboratory, Sony Computer Scence Laboratores, Inc, as an assocate researcher. In 2001 he joned Faculty of Scence and Engneerng, Keo Unversty, where he s now a professor. He receved the Young Engneer Award from IEICE Japan n Hs current research nterests are n software defned rado, cogntve rado, and OFDM systems.