- Email: [email protected]
Sounding resource management for QoS support in massive MIMO systems
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Computer Networks xxx (2014) xxx–xxx 1
Contents lists available at ScienceDirect
Computer Networks journal homepage: www.elsevier.com/locate/comnet 5 6
Sounding resource management for QoS support in massive MIMO systems
3 4 7
Q1
Taeseop Lee, Sangkyu Park, Hyung-Sin Kim, Saewoong Bahk ⇑
8
Q2
INMC, Department of Electrical and Computer Engineering, Seoul National University, Republic of Korea
9
a r t i c l e
1 2 1 3 12 13 14 15 16
i n f o
Article history: Received 5 January 2014 Received in revised form 29 April 2014 Accepted 2 May 2014 Available online xxxx
17 18 19 20 21 22
Keywords: Massive MIMO Pilot contamination QoS Sounding resources
a b s t r a c t In massive multiple-input multiple-out (M-MIMO) systems, the base station estimates each forward-link channel by using the reverse-link sounding pilot and the channel reciprocity property of time division duplex (TDD) operation. However, conventional noncooperative sounding schemes, like a sounding sequence assignment scheme with reuse factor-1, cause cell edge users to suffer from pilot contamination in multi-cell scenarios. The pilot contamination problem becomes even worse in an environment where mobile users are travelling through cell edge areas frequently. To alleviate this problem, a cooperative sounding resource reuse method that assigns each cell a set of resources different from other neighboring cells by using a reuse factor-3 scheme, can be considered, but it significantly lowers the number of served users. In this paper, we propose a partial sounding resource reuse (PSRR) method which aims to seamlessly support the quality of service (QoS) of each mobile user by mitigating the pilot contamination and to minimize the reduction in the number of served users. To this end, the PSRR divides each cell area into center and edge areas, and applies a reuse factor-1 scheme for center users while a reuse factor-3 scheme for edge users. We use a Markov chain model to analyze the performance of the PSRR, and evaluate the accuracy of our analysis through simulations. Then we confirm that the PSRR considerably improves QoS performance over the conventional competitive schemes. Ó 2014 Published by Elsevier B.V.
24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
45 46
1. Introduction
47
Recently, wireless data traffic has dramatically increased with the advent of smart phones, so the wireless capacity enhancement in cellular systems has attracted a lot of attention from many researchers [1–3]. To this end, the deployment of small cells and the use of wide bandwidth have been considered as ways of increasing the wireless
48 49 50 51 52
⇑ Corresponding author. Address: Seoul National University, Institute of
Q1
New Media and Communications, 1 Gwanak-ro, Gwanak-gu, Seoul 151-742, Republic of Korea. Tel.: +82 2 880 8434; fax: +82 2 880 8198. E-mail addresses: [email protected] (T. Lee), [email protected]. ac.kr (S. Park), [email protected] (H.-S. Kim), [email protected] (S. Bahk).
capacity [4–6]. Another approach to increase wireless system capacity is to use a large-scale antenna system, also known as massive multiple-input multiple-output (MMIMO) system. M-MIMO systems enable many users with good link quality to be multiplexed in spatial domain, exploiting the cancellation of inter-user interference and noise on desired signals [7–10]. In M-MIMO systems, the base station uses the downlink channel information toward each terminal to generate a pre-coding matrix. A quantization-based channel information feedback method needs huge amounts of downlink and uplink resources when applied to M-MIMO systems, due to the downlink per-antenna pilot signal broadcasting and the uplink feedback overhead, respectively. Thus, uplink sounding techniques are expected to be adopted,
http://dx.doi.org/10.1016/j.comnet.2014.05.001 1389-1286/Ó 2014 Published by Elsevier B.V.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120
2
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
which use the channel reciprocity of time-division duplex (TDD) systems [11–13]. The performance of an M-MIMO system greatly depends on how to use limited sounding resources1 and the accuracy of channel estimation [14,15]. Especially in multi-cell environments, the performance of a cell edge user is significantly degraded due to pilot contamination when each cell exploits the whole sounding resources in a noncooperative manner [11,17]. This means that the pilot contamination severely degrades system performance of an M-MIMO system in terms of QoS for each cell edge user. To provide seamless handover and QoS for each cell edge user, sounding resources need to be reused with reuse factor-3 in a cooperative manner. It provides high signal to interference ratio (SIR) for each edge user at the cost of having a lowered number of served users.2 It has been shown that non-cooperative methods provide higher system capacity while cooperative reuse methods achieve better edge user QoS [11]. Thus, a new sounding technique needs to be designed by considering the trade-off relation between the pilot contamination and the reduced number of served users. In related works, the trade-off relation between the QoS degradation of users and the total system capacity. In this paper, we propose a partial sounding resource reuse (PSRR) scheme that aims to provide desired QoS to each mobile user seamlessly as many as possible. To this end, the PSRR first divides each cell area into center area and edge area; center area users achieve high SIRs, while edge area users suffer from the pilot contamination problem. Then it splits the whole sounding sequences into two sets for center and edge area users, respectively. The center area sounding resources are reused with reuse factor-1, and the edge area ones with reuse factor-3 to mitigate pilot contamination. Whenever a mobile terminal initiates a handover (i.e. center to edge area or edge to edge area), it will be reassigned a new sounding resource to achieve appropriate QoS. We mathematically analyze QoS provisioning performance of the PSRR according to the two ratio parameters; one is the ratio of edge area to the total cell area in size, and the other is the ratio between the sounding resources in number for edge area and total cell area. In the analysis, we simplify the operation of PSRR with the above simple two parameters that are enough to control the QoS provisioning performance of the PSRR. Through simulations, we evaluate the accuracy of our analysis, and confirm that the PSRR improves QoS provisioning performance. We summarize main contributions of this paper as follows.
We propose a partial sounding resource reuse (PSRR) method that allocates sounding resources for mobile users by considering the trade-off relation between pilot contamination and reduced served users in number. For PSRR operation, we analytically obtain the QoS provisioning performance of the PSRR by using the Markov chain model with simple parameters.
123 124 125 126 127 128 129 130
2. Proposed PSRR scheme
140
In this section, we consider the model for an M-MIMO system, explain the pilot contamination problem, and propose the PSRR scheme to maximize the number of users to be served while supporting each user’s QoS.
141
2.1. System model
145
As depicted in Fig. 1, we consider a TDD-based M-MIMO system where a time slot comprises T symbols. Here T is a given parameter, determined by the coherence time during which the channel information does not change. The base station estimates the downlink channel condition for each terminal from uplink sounding pilots received for the preceding sr symbols, and then transmits downlink data for the next ðT sr Þ symbols according to the estimated channel condition. Letting N smooth be the number of consecutive subcarriers in frequency domain which have the same channel gain (i.e. coherence bandwidth), a block of N smooth T subcarriers have the same channel condition. Using the sounding subcarrier block which is denoted by a bold rectangular in Fig. 1, the M-MIMO system generates
146
Fig. 1. TDD-based M-MIMO system.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
122
The remainder of this paper is organized as follows. In Section 2, we discuss the system model and presents the proposed PSRR method. In Section 3, we mathematically analyze the QoS provisioning performance of PSRR, determine the proper values for the two system parameters and validate the correctness of our analysis. In Section 4, through simulations, we evaluate the performance of PSRR and its performance improvement over the competitive schemes. Finally, we summarize the contributions of this paper and conclude our work.
To the best of our knowledge, this is the first work which considers partial sounding resource reuse for guaranteeing the QoS of mobile users in M-MIMO systems.
1 This is because the number of spatially multiplexed users is directly related to the number of sounding users. 2 In this paper, the number of served users indicates the number of spatially multiplexed users for the same downlink resources, which is limited by the number of orthogonal sounding sequences.
121
131 132 133 134 135 136 137 138 139
142 143 144
147 148 149 150 151 152 153 154 155 156 157 158 159
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
3
Fig. 2. Pilot contamination problem in a non-cooperative M-MIMO system.
160 161 162
163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
N smooth sr orthogonal sounding resources.3 Thus, in a TDDbased M-MIMO system, each cell can support Nsmooth sr users at a maximum for the given downlink resources. 2.2. Effect of pilot contamination in a non-cooperative M-MIMO system If error free channel information is available, the M-MIMO system can serve multiple users using the same downlink resources without creating inter cell interference [18,19]. However, in reality, a conventional non-cooperative M-MIMO system cannot provide accurate channel information due to the pilot contamination. For example, let’s consider a case of three neighboring cells, each of which serves its own user k with the same sounding sequence k as depicted in Fig. 2. In this case, the base station of cell l or cell h receives not only the desired sounding pilot signal but also the sounding interference from user k in cell j (i.e. pilot contamination), which results in a polluted pre-coding matrix. As a result, the base stations of cells l and h generate inaccurate beamforming, and the downlink data transmission of cell j suffers from the increased interference from those of cells l and h. In this paper, we adopt the SIR model in [11] where an infinite number of antennas are assumed. Defining blkj as the static channel gain between base station l and user k (with sounding sequence k) in cell j, we have the downlink SIR of user k in cell j as
186
188 189 190 191 192
SIRDkj
b2jkj ¼P 2 j–l blkj
where blkj
zlkj ¼ c rlkj
ð1Þ
where r lkj is the distance between base station l and user k in cell j; c is the path loss exponent, and zlkj is a log-normal random variable [11]. It can be seen that SIRDkj decreases with the interference signal due to the pilot contamination
(i.e. blkj where l – j). Thus, a mobile terminal has difficulty in receiving seamless QoS support due to the pilot contamination when it approaches its cell boundary. To alleviate the pilot contamination problem, sounding sequence assignment with reuse factor-3 that assigns each neighboring cell a different set of sounding sequences can be considered. However, it has the drawback of reduction in number of served users, resulting in critical system capacity degradation. This motivates us to propose the PSRR that aims to provide appropriate QoS for mobile users while minimizing the system capacity degradation.
193
2.3. Proposed scheme
204
The PSRR aims to mitigate the pilot contamination problem and support as many users as possible by applying a different sounding sequence assignment method for each user according to its location. As illustrated in Fig. 3, the proposed PSRR divides each cell area into center area and edge area. Let’s denote the ratio of edge area to cell area in size as F. Then
205
AreaE F¼ AreaCell
ð2Þ
and AreaC is the center area and AreaE is the edge area. F is directly related to the number of users, each of which needs a dedicated sounding sequence (i.e. not used by neighboring cells). Let’s denote the set of total orthogonal sounding sequences as S. The PSRR divides S into subsets SC and SE . Then, it assigns SC to center area users with reuse factor1, and SE to edge area users with reuse factor-3. Considering the hexagonal cell configuration in Fig. 3, the PSRR divides SE into three equal sized subsets SEi for i ¼ 1; 2; 3, and assigns each neighboring cell a different SEi . Defining the ratio of the number of sounding sequences for edge area users to the total number of sequences as H, we have
3
There are two methods for sounding resource generation [16]. In a cyclic shift separation (CSS) system, (Nsmooth T) sounding sequences, each of which has the length of (N smooth T), are generated. In a decimation separation (DS) system, each of (N smooth T) subcarriers is a sounding resource. Although we consider a CSS system in the following sections, the proposed PSRR can also be applied to a DS system.
where AreaCell ¼ AreaC þ AreaE ;
H¼
jSE j ; jSj
ð3Þ
where jSj is the cardinality of set S. The performance of PSRR mainly depends on the interaction between F and
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
194 195 196 197 198 199 200 201 202 203
206 207 208 209 210 211
212 214 215 216 217 218 219 220 221 222 223 224 225 226 227
228 230 231 232
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
4
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
277
F ¼1
Fig. 3. PSRR overview.
238
H. A large number of cell edge users may suffer from pilot contamination when H is too small compared to F. On the contrary, the number of served users may be significantly lowered when H is too large compared to F. In the next section, we mathematically show the effect of these parameters on the QoS provisioning performance of PSRR.
239
3. Mathematical analysis for PSRR operation
240
251
For QoS provisioning, the PSRR aims to provide the minimum required SIR, denoted as SIRth , for each mobile user located in edge area, while reusing sounding sequences as much as possible for center area users. We first analyze the relationship between F and SIRth to show SIR provisioning performance of the PSRR. Let P sf represent the probability of sounding sequence assignment failure as a performance metric for the PSRR. Then, we analyze the relationship between H and Psf . Finally, we determine the proper values of two system parameters F and H according to the mathematical analysis, and confirm the accuracy of our model through simulations.
252
3.1. Relation between F and SIRth
253
Let’s denote the distance between a user and the base station as d. For ease of analysis, we consider a multi-cell environment where all cells are uniformly deployed, a sufficiently large number of mobiles are served, and all base stations transmit downlink signals with an equal power. Then, the downlink SIR can be approximated as a function of d, denoted as SIRD ðdÞ [20]. The dark dotted line in Fig. 4(a) shows the expected downlink SIR according to d for the sounding assignment scheme with reuse factor-1. It can be seen that the expected SIR exponentially decreases with d due to the pilot contamination, resulting in failure of SIRth guarantee when a mobile user approaches the cell boundary. To meet the SIRth requirement of each cell edge user, the PSRR divides each cell area into center and edge areas according to whether the assignment scheme with reuse factor-1 can meet the condition of SIRth for a mobile user or not. The center area has a circular shape as shown in Fig. 4(b) since SIR is given as a function of d. Then the PSRR determines the center area radius, r, and accordingly F is given as follows.
233 234 235 236 237
241 242 243 244 245 246 247 248 249 250
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273
274 276
r ¼ max fdjSIRD ðdÞ P SIRth g
ð4Þ
pr2 AreaCell
ð5Þ
It can be easily seen that r decreases as SIRth increases, and F increases as r decreases. Therefore, the increase of SIRth results in the increase of F.4
280
3.2. Relation between H and P sf
283
A service failure happens when there is no available sounding sequence for a mobile user. In the PSRR system, we consider two types of service failure for a mobile terminal depending on whether it is newly joining or in the middle of handover. We define service blocking and handover (HO) blocking as sequence assignment failures for a new service user and a handover user, respectively. Therefore Psf represents the probability that a mobile terminal experiences service failure due to service blocking or handover blocking. In this subsection, we mathematically analyze the relation between H and P sf , and calculate Psf by using a Markov chain model.
284
3.2.1. Mobility model and queuing of handover requests Let’s assume that cells are hexagonal and regular. We denote the radii of circumscribed circle, inscribed circle, and center circle for each hexagonal cell as R0 ; R and r, respectively. The mobility model in the PSRR system uses the following assumptions.
296
(a) Each user moves at a constant velocity of V[km/h]. (b) Each user moves across a cell along its maximum diameter.
302
281 282
285 286 287 288 289 290 291 292 293 294 295
297 298 299 300 301
303 304 305
When a user crosses multiple cell areas with the above mobility assumptions, it is assigned a new sounding sequence at each area to achieve seamless QoS provisioning. We define this new sounding sequence assignment as a handover procedure and consider applying a handover prioritization strategy according to the queuing for handover. In Fig. 5, when a mobile terminal crosses multiple cells, it experiences three types of handover that are denoted by HOEE ; HOEC , and HOCE (i.e. handover between edge to edge, edge to center, and center to edge). During each handover process, it transmits a handover request to the base station to trigger the handover. Then the PSRR reassigns it an available sounding sequence in the associated area. However, if all the sequences in that area are already in use, the handover request gets queued until a sounding sequence is available. If the terminal fails to get a new sounding sequence until it gets out of the handover area, a handover blocking occurs. By defining a maximum handover queuing time as the time duration for which the user starts handover and gets out of the handover area, we can set the maximum handover queuing time t max to a fixed value according to w 4 In LTE systems, an Evolved Serving Mobile Location Center (E-SMLC) can estimate each user’s position by using Observed Time Difference Of Arrival (OTDOA) positioning scheme [21]. OTDOA uses the time difference (observed by the user) between the reference signals of neighboring cells and the serving cell. By using the estimated position of each user, a base station can trigger the HO procedures.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
279
306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
5
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
Fig. 4. Determination of F according to SIRth .
328 329
the considered mobility model. For the HOEE case in Fig. 5, we have
330 0
332 333 334 335
336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351
t max ¼ w
R ð1 V
pffiffi 3 Þ 2
at center area and edge area are denoted by P hC and P hE , respectively, they are given by a
’
0:134R0 : V
ð6Þ
t max w
Note that has a different value according to the mobility model and handover case. For simple analysis, we use the same tmax for HOEC and HOCE cases. w 3.2.2. Memoryless property of handover arrivals To design a markov chain model for the sounding resource assignment problem, we need to know the handover arrival rate at each area. For doing this, we first calculate the probability that an ongoing user in service is continuously active until handover happens. Let tdC and tdE represent the intervals elapsed from the arrival instant of a new user at center area and edge area, respectively, to the instant of its entering the handover area. According to the mobility assumptions, t dC and t dE are random variables uniformly distributed in ½0; 2r=V and ½0; ðR rÞ=V, respectively. A random service duration t s is exponentially distributed with mean T m , and an ongoing served user in the center area (or edge area) generates a handover request whenever t s > tdC (or t dE ). When the handover probabilities
Fig. 5. Mobility model and queuing of handover requests.
PhC ðaÞ ¼ Prfts > t dC g ¼
1e
a
2r where a ¼ ; VT m
ð7aÞ
353
354 356 357
1 eb PhE ðbÞ ¼ Prfts > t dE g ¼ b
Rr where b ¼ : VT m
ð7bÞ
An ongoing served user who has already experienced a successful handover travels a deterministic distance of equal to 2r and R r in center area and edge area, respectively, before starting a new handover process. Let’s denote the deterministic inter arrival times for handover at center area and edge area after the initial arrival as t0dC and t 0dE , respectively. Then they are given as 2r=V and ðR rÞ=V, respectively. The residual service time of an ongoing served user after a successful handover is also exponentially distributed with mean T m due to the memoryless property. Let P0hC and P 0hE represent the probabilities that a user having experienced a handover begins another handover in center area and edge area, respectively. Then we obtain
P0hC ð
aÞ ¼ Pr ts > t0dC ¼ ea
2r where a ¼ ; VT m
ð8aÞ
Rr : VT m
ð8bÞ
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373
374 376 377
P0hE ðbÞ ¼ Pr t s > t0dE ¼ eb
where b ¼
379
The analysis above confirms that the sounding resource assignment problem can be modeled by a Markov chain.
380
3.2.3. Markov model for sounding sequence assignment Fig. 6 depicts the two Markov chain models for center and edge areas, respectively. For simple analysis, we assume the following [22].
382
(a) The numbers of permanently allocated sounding sequences are jSC j for center area users and jSEi j for edge area users. (b) New arrivals and handover arrivals are independent Poisson processes with mean rates k and kh , respectively.
386
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
352
381
383 384 385
387 388 389 390 391
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
6
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
Fig. 6. Markov chain models for sounding sequence assignment in PSRR.
(c) The sequence holding times are approximated as exponentially distributed random variables with means l1 and l1 for center area and edge area, respeC E ctively. (d) The maximum handover queuing time5 t max is w assumed as the mean of an exponentially distributed random variable. (e) The queue size is infinite.
392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411
412 414
Using the above assumptions, we model the sounding resource assignment problem for the PSRR as an M/M/jSA j queuing system for area A which is C or E depending on the considered area. To obtain the service blocking probability and the handover blocking probability in each area, we calculate the steady state probabilities first. For a given number of available sounding sequences for area A; jSA j, we can obtain the average new arrival rate kA , the average handover arrival rate khA , and the average sounding sequence holding time 1=lA , when F and H are given. Simply we have
jSC j ¼ ð1 HÞjSj;
ð9aÞ
415 417 418 419 420 421
422 424
jSEi j ¼
jSE j HjSj ¼ 3 3
for i ¼ 1; 2; 3:
ð9bÞ
When the average new service arrival rate at each cell is given as k, we obtain the average new service arrival rates kC and kE for center area and edge area, respectively, as follows.
kC ¼ ð1 FÞk;
5
ð10aÞ
This value is approximated as a constant in Eq. (6) according to the mobility model and handover case.
kE ¼ Fk:
ð10bÞ
Then we obtain the average handover arrival rates khC and khE for center area and edge area, respectively, as follows.
khC ¼ khC ð1
PhbC ÞP0hC
þ kC ð1 PsbC ÞPhC ;
ð11aÞ
428 429 430
431 433 434
khEE þ khCE ¼ khEE ð1 PhbE ÞP0hE þ khCE ð1 PhbE ÞP 0hE þ kE ð1 PsbE ÞP hE ; khE ¼ khE ð1 P hbE ÞP0hE þ kE ð1 P sbE ÞPhE ;
ð11bÞ
436
ð11cÞ
437 439
where P hbC and P hbE denote handover blocking probabilities for center area and edge area, respectively, and PhC ; PhE ; P 0hC ; P 0hE are handover probabilities derived from (7) and (8). To help understanding, Fig. 7 depicts the statistical equilibrium state in the PSRR system. For the center area, the average handover arrival rate is khC , and the average handover departure rate is khC ð1 PhbC ÞP 0hC þ kC ð1 PsbC ÞPhC , as shown in Eq. (11a). On the other hand, for the edge area, we should consider the handover arrival rate from the center area khCE and that from the neighboring cells, khEE . Then the average of total handover arrival rate at the edge area, khE , becomes the sum of khCE and khEE . Then we can calculate the handover arrival rates khC and khE , when the service blocking probabilities, P sbC and P sbE , and the handover blocking probabilities, P hbC and P hbE are given. To obtain P hbC ; P hbE ; P sbC and P sbE from the Markov chain model, we need the handover arrival rates khC and khE again. To compute these, we use a recursive approach. Let’s denote the sounding sequence holding times for a new service user in center area and edge area as t HC and tHE , respectively. Then they can be expressed, respectively, as
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
425 427
440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
7
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
Fig. 7. Statistical equilibrium between mobile terminals’ arrivals and departures at each area.
463 465
t HC ¼ min½t s ; tdC ;
ð12aÞ
466 468
t HE ¼ min½t s ; tdE :
ð12bÞ
469 470 471 472
473 475 476 478
479 480 481
482
Denoting the sounding sequence holing time of an ongoing served user in center area (or edge area) which has already experienced a first handover by t0HC (or t 0HE ), we have
t 0HC ¼ min½t s ; t0dC ;
ð13aÞ
t 0HE ¼ min½t s ; t0dE :
ð13bÞ
where the system idle probability is given as P 0A ¼
( jS j A X ðkA þ khA Þn n!lnA
n¼1
Let’s denote the average sounding sequence holding times for each area as 1=lC and 1=lE . Then they are given, respectively, as
lC 484
lE
490 491 492 493 494 495 496 497 498 499 500 501 502 503 504
505
kC khC ¼ E½t HC þ E½t0 kC þ khC kC þ khC HC kC khC ¼ T m ð1 PhC Þ þ T m ð1 P0hC Þ; kC þ khC kC þ khC kE khE ¼ E½tHE þ E½t 0 kE þ khE kE þ khE HE kE khE T m ð1 PhE Þ þ T m ð1 P0hE Þ: ¼ kE þ khE kE þ khE
ð14aÞ
As a service blocking occurs if there is no available sounding sequence, the service blocking probability in area A can be expressed as
512
PsbA ¼
1 X
PnA
513 514
ð17Þ 517
n¼jSA j
In the case of handover request queuing, a handover blocking occurs if the queuing time exceeds t max w . Then we obtain the conditional handover blocking probability at state n as
PhbAjn ¼ 1
njS YA j
" 1
8 ðk þk Þn A hA > < n!lnA P0A ;
ð14bÞ
> :
1 6 n 6 jSA j;
ðkA þkhA ÞjSA j kh AðnjSA jÞ jSA j!l
jSA j A
QnjSA j j¼1
ðjSA jþjlw Þ
P0A
jSA j þ 1 6 n;
lw ðjSA jlA þ lw Þ2j
ð15Þ
PhbA ¼
1 X
:
519 520 521
ð18Þ 524 525 526
527
PhbAjn PnA :
ð19Þ 529
n¼jSA j
3.2.4. Calculation of P sf We compute the service failure probability Psf , using the above results. Let’s denote the forced service termination probabilities due to the handover blocking by PdropC and P dropE , respectively, according to the user’s service originating location. Then we can express P sf as
530
Psf ¼ ð1 FÞ PsbC þ ð1 P sbC ÞPdropC þ F PsbE þ ð1 PsbE ÞPdropE :
536
531 532 533 534 535
ð20Þ
538
Here the first term indicates the forced service termination probability due to the service blocking or handover blocking for a center area user and the second term represents the same for an edge area user. Before computing PdropC and P dropE , let’s denote the probabilities that a service attempt is served and it has i (¼ 1; 2; 3; . . .) successful handovers originating from center area and edge area as HiC and HiE , respectively. Then we obtain
539
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
518
522
#
Then the handover blocking probability in area A is given by
As depicted in Fig. 6, we define the system state as the number of users who need sounding sequences in each area. This equals the sum of the number of users in service and the number of queued handover requests. If the current state is lower than the number of available sounding sequences in area Að¼ C or Ei Þ, denoted by jSA j, the PSRR system can accept all the users and the total arrival rate equals kA þ khA . If the current state is higher than jSA j, the total arrival rate becomes khA because we the PSRR blocks newly joining users to give handover users higher priority. When the queuing system state is (jSA j þ j) for j ¼ 1; 2; . . ., we need to consider the case in which a handover blocking occurs because the queuing delay exceeds the maximum handover queuing time. Thus we represent the death rate due to the handover blocking as jlw which equals j=tmax w . Therefore, the steady state probabilities in area A can be expressed as
PnA ¼ 507
: 511
j¼0
1
489
509
!)1
ð16Þ
485
488
n¼jSA jþ1
ðkA þ khA ÞjSA j kh AðnjSA jÞ QnjSA j jSA j!ljSA A j j¼1 ðjSA j þ jlw Þ
515
1
487
þ
1 X
508
540 541 542 543 544 545 546 547
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
8
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
548
HiC ¼ 550 551
HiE ¼ 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568
569
8 bði1Þ=3c > > P ðP0 P 02 Þ > < hC hC hE
bði1Þ=3c PhC P 0hE ðP 0hC P02 hE Þ > > > : 0 02 bði1Þ=3c PhC P 02 hE ðP hC P hE Þ
for i%3 ¼ 1 ð21aÞ
for i%3 ¼ 2 for i%3 ¼ 0;
8 bði1Þ=3c > > P ðP0 P02 Þ > < hE hC hE
bði1Þ=3c PhE P0hE ðP0hC P02 hE Þ > > > bði1Þ=3c : PhE P0hE P0hC ðP0hC P02 hE Þ
for i%3 ¼ 1 ð21bÞ
for i%3 ¼ 2 for i%3 ¼ 0;
where % represents a modulo operation, and bxc chooses a largest integer that is smaller than or equal to x. As shown in Fig. 8, a user starting to receive a service from a center area moves according to the considered mobility model and experiences multiple handovers. The probability that the user experiences the first handover is given as PhC (see Eq. (7a)), and the probability that the user experiences the second handover is given as PhC P 0hE by applying the memoryless property of the service time. Then we can compute the i-th handover probability of a user originating from the center area through the same procedures. Let’s denote the probabilities that a user in service is dropped at the i-th handover (i ¼ 1; 2; 3; . . .) and it has started from center area and edge area as BiC and BiE , respectively. Then we obtain 8 n obði1Þ=3c > PhbE ð1 P hbE Þ2 ð1 P hbC Þ > > > > < n obði1Þ=3c BiC ¼ ð1 P hbE ÞP hbE ð1 P hbE Þ2 ð1 P hbC Þ > > > n obði1Þ=3c > > 2 : ð1 P Þ2 P hbE hbC ð1 P hbE Þ ð1 P hbC Þ
for i%3 ¼ 1; for i%3 ¼ 2; for i%3 ¼ 0;
571 572
ð22aÞ 8 n obði1Þ=3c 2 > > for i%3¼1; > P hbE ð1P hbE Þ ð1P hbC Þ > > < n obði1Þ=3c BiE ¼ ð1P hbE ÞP hbC ð1PhbE Þ2 ð1P hbC Þ for i%3¼2; > > > n obði1Þ=3c > > 2 : ð1P Þð1P ÞP for i%3¼0: hbE hbC hbE ð1P hbE Þ ð1P hbC Þ
574 575
ð22bÞ
From (21) and (22), we have
576 578
PdropC
1 X ¼ HiC BiC ;
582 583
584 585 586 587 588 589 590
3.3.1. Relation between F and SIRth Fig. 9 shows the average downlink SIR according to the distance between the base station and a user, when the sounding assignment scheme with reuse factor-1 (i.e. non-cooperative scheme) and our proposed PSRR with F = 0.3 and 0.6 are considered. Here we omit the results for the case of distance below 150 m because all the three schemes show the same SIR performance. In the non-cooperative scheme, the downlink SIR dramatically decreases with the distance due to the pilot contamination. On the contrary, it can be seen that the PSRR tries to provide the downlink SIR greater than SIRth to each edge user. The PSRR is not able to solve this problem completely since it only mitigates the pilot contamination problem. For example, for a required SIRth of 21.8 dB, the PSRR operates with F ¼ 0:6 to guarantee the SIR value larger than 21.8 dB. Nevertheless, the PSRR with F ¼ 0:6 provides the SIR of 17.6 dB (i.e. lower than 21.8 dB) at distance 300 m, which corresponds to the cell boundary. In worst case, the PSRR may not be able to meet SIRth for cell boundary users eventually due to the path loss. Note that overcoming the path loss problem is beyond the scope of this paper. In PSRR systems, we can calculate the expected SIR as a function of the distance between a user and a base station, when all base stations are uniformly deployed and transmit with an equal power [20]. Therefore, the proper value for F can be determined by a given SIRth value. Table 1 presents the required center area radius r and ratio F accord-
i¼1
579 581
ð23aÞ
Fig. 8. Example of handover operation for a user originating from the center area.
PdropE ¼
1 X HiE BiE :
ð23bÞ
i¼1
Finally, we obtain the service failure probability from (20). 3.3. Verification of analysis and values of F and H for QoS provisioning In this section, we show the correctness of our mathematical model with simulation results and how values of F and H can be determined by the above analysis. The simulation results of this section are driven from a simulation scenario in Section 4.1.
Fig. 9. Average SIR vs. distance between the base station and a user.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
9
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx Table 1 SIRth vs. required edge area ratio.
620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644
rðmÞ
F
262 241 223 206 190 175 160 147 134 122 110 100 90
0.08 0.22 0.33 0.43 0.52 0.59 0.66 0.71 0.76 0.8 0.84 0.87 0.89
ing to different SIRth values. The required values of r and F to meet the SIRth condition are determined by Eqs. (4) and (5) in Section 3.1. It can be seen that F increases with SIRth . Again the PSRR eventually has difficulty in guaranteeing SIRth when SIRth is larger than 17.56 dB due to the path loss problem. Since the limitation comes from radio signal attenuation, the cell size and transmit power need to be determined according to the given SIRth . If all the users achieve SIRs greater than SIRth even at the cell boundary and obtain error free channel information, we can say, the PSRR guarantees SIRth successfully. 3.3.2. Relation between H and P sf Fig. 10 shows Psf according to H for F ¼ 0:2 and 0.4, and compares the results of analysis and simulation. In simulations, we consider two types of mobility model; One is the random waypoint model in which a user chooses its destination randomly and the other is the Manhattan mobility model where a user moves along the road indicated by a dotted line in Fig. 11 [24,25]. It can be observed that the analysis results have slightly lower Psf than the simulation results, but they show about the same tendency. It can be inferred that the gap between analysis and simulation results comes from the mobility assumption since users change their moving directions and experience handover more frequently in simulations with two mobility models than in the analysis model. Furthermore, we can see that
Fig. 11. Cell configuration for simulations.
the results with the Manhattan model show higher Psf than those with the random waypoint model because users in the Manhattan model change moving direction more frequently than users in the random waypoint model. In Table 2, we show optimum values of H which minimize Psf according to F, obtained from analysis and simulations with two mobility models. The optimum H is obtained by minimizing Psf , where the PSRR supports a maximum number of users. Minimizing P sf directly indicates maximizing the number of supported users because more users can be supported without experiencing service and handover blockings with lower Psf . The optimum values of H obtained from analysis and simulation are very close, although there exists some gap between them in terms of P sf . This leads us to focus on the minimum Psf instead of the absolute value of Psf , when we choose an optimum H. Furthermore, Table 2 shows that the required resource ratio H increases with F.
645
4. Performance evaluation
663
In this section, we evaluate the performance of PSRR through simulations and show its performance improvement over the competitive schemes.
664
0.8
Service failure probability
619
SIRth (dB) 3.5 7 10.5 14 17.5 21 24.5 28 31.5 35 38.5 42 45.5
analysis with F=0.2 analysis with F=0.4 random waypoint with F=0.2 random waypoint with F=0.4 manhattan with F=0.2 manhattan with F=0.4
0.7 0.6 0.5
Table 2 Optimum H vs. F.
0.4 0.3 0.2 0.1 0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
H
0.9
1
F
H (analysis)
H (simulation – random waypoint)
H (simulation – manhattan)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0.3393 0.4464 0.5357 0.6071 0.6785 0.7321 0.8035 0.8571 0.9286
0.3214 0.4286 0.5357 0.6071 0.6785 0.75 0.7857 0.8571 0.9286
0.3214 0.4286 0.5357 0.5714 0.6428 0.75 0.7857 0.8571 0.9286
Fig. 10. P sf vs. H.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662
665 666
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 10
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx 800
reuse−1 (60km/h) PSRR (60km/h) AREA (60km/h) reuse−3 (60km/h)
number of QoS users
700 600 500 400 300
43.3% improvement
200 100 0
800
reuse−1 (100km/h) PSRR (100km/h) AREA (100km/h) reuse−3 (100km/h)
700
number of QoS users
Q1
600 500 400 300
65.4% improvement
200 100
5
10
15
20
25
30
35
40
45
0
5
10
SIR threshold (dB)
15
20
25
30
35
40
45
SIR threshold (dB)
Fig. 12. Number of QoS users vs. SIRth .
667
4.1. Scenario for simulations
668
686
We consider a TDD M-MIMO system based on the LTE standard where the system bandwidth is 20 MHz, sr ¼ 12 symbol time, and N smooth ¼ 14 symbol bandwidth [11]. We consider the user mobility of up to 100 km/h (i.e. coherence time of 1.6 ms), which provides large T enough to support 12 sounding symbols (e.g., T 6 24 symbol time when the symbol time is 66.7 ls as in LTE systems) [23]. Assume that the path loss exponent is 3.8 and the standard deviation of shadow-fading is 8 dB [11]. As depicted in Fig. 11, we assume that each cell has the radius of 300 m and evaluate the performance considering the dotted square area of 19 cells. Also assume that the number of service requests from users in each cell follows a Poisson process with the average of 6 req/s, and the service time is exponentially distributed with mean 15 s. For user mobility, we use the Manhattan mobility model instead of the random waypoint model because the Manhattan mobility model is more appropriate for a real high mobility scenario than the waypoint model.
687
4.2. Performance of the PSRR
688
Now we evaluate the QoS provisioning performance for mobile users in terms of the number of QoS users, system capacity, and Jain’s fairness. The sounding sequence assignment schemes with reuse-1, reuse-3 and AREA6 are considered for performance comparison.
669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685
689 690 691 692
693 694 695 696 697 698
4.2.1. Number of QoS users For evaluation, we define a QoS user as a user who has achieved an SIR value greater than SIRth for 95% or more of its service duration. Fig. 12 shows the numbers of QoS users in the system with respect to SIRth for the reuse factor-1 scheme, the reuse factor-3 scheme, the AREA scheme, 6
In the AREA scheme, the number of sounding sequences assigned to each area is proportional to its size. For example, when the ratio of center area to edge area is 2:1, the ratio of the numbers of sounding sequences in each area also becomes 2:1.
and our proposed PSRR scheme. We consider the user velocities of 60 km/h and 100 km/h. In both cases, the proposed PSRR system outperforms the three compared schemes, and the performance gap increases with the increase of SIRth in the practical range of lower than 17.6 dB. On the other hand, the performance gap starts to decrease when SIRth is higher than 17.6 dB because the PSRR cannot guarantee the SIR in that range. As the SIRth range of higher than 17.6 dB is not practical, we can say that the PSRR outperforms the other three schemes, which is shown on the left side of the dotted line in Fig. 12. The PSRR shows the performance gain of up to 43.3%, 30.8%, and 10.9% compared to the reuse factor-1, reuse factor-3, and AREA schemes, respectively. Fig. 12(b) plots the number of QoS users in each scheme for a velocity of 100 km/h. The PSRR shows the performance gain of up to 65.4%, 27.1%, and 8.4% compared to the reuse factor-1, reuse factor-3, and AREA schemes, respectively. By comparing the results for the user velocities of 60 km/h and 100 km/h, we observe that the performance gap between the PSRR and the reuse factor-1 scheme increases from 43.4% to 65.4%. This result reveals the strength of the PSRR in fast moving environments. On the contrary, the performance gap between the PSRR and the reuse factor-3 schemes decreases from 30.8% to 27.1%, because the PSRR scheme increases the proportion of sounding sequences for edge area and gradually operates like the reuse factor-3 scheme with the mobility speed of each user. Although, the performance gap is also reduced in the AREA scheme for the same reason, we can see that the PSRR still outperforms both the reuse factor-3 and AREA schemes.
699
4.2.2. System capacity Fig. 13 illustrates the average system capacity, which is the sum of served users’ capacity, for each scheme according to the SIRth when the user velocity is 60 km/h. In the reuse factor-1 and reuse factor-3 schemes, the system capacity remains the same regardless of the SIRth because they do not change their operations according to
732
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731
733 734 735 736 737 738
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx 4
12
x 10
11
system capacity
10 9 8 7 reuse−1 (60km/h) PSRR (60km/h) AREA (60km/h) reuse−3 (60km/h)
6 5 4
5
10
15
20
25
30
35
40
Fig. 13. System capacity vs. SIRth for a velocity of 60 km/h.
741 742 743 744 745 746 747 748 749 750 751 752 753 754 755
756 757 758 759 760
SIRth . The PSRR performs like the reuse factor-1 scheme for low SIRth and gradually like the reuse factor-3 scheme with the increase of SIRth . The AREA scheme shows the same tendency with the PSRR scheme, because it increases the proportion of sounding sequences for edge area users, as its area size increases with SIRth . In Fig. 13, it is also shown that the system capacity of the PSRR gradually decreases with the increase of SIRth . For the practical range of SIRth , the PSRR scheme shows the system capacity close to that of the reuse factor-1 scheme and the maximum performance gain of 37% and 13.3% over the reuse factor-3 and AREA schemes, respectively. Considering the results in Figs. 12 and 13, we can confirm that the PSRR accommodates a largest number of QoS users, while minimizing the degradation in system capacity, owing to the partial reuse of sounding resources.
4.2.3. Jain’s fairness Fig. 14 plots Jain’s fairness index for each scheme according to SIRth [26]. The fairness performance of the PSRR gradually approaches that of the reuse factor-3 scheme as SIRth approaches 17.6 dB, and the PSRR always
1 0.9
Jain’s fairness index
740
0.8 0.7 0.6 0.5
reuse−1 (60km/h) PSRR (60km/h) AREA (60km/h) reuse−3 (60km/h)
0.4 0.3 0.2 0.1 0
5
10
15
20
25
30
35
40
SIR threshold (dB) Fig. 14. Jain’s fairness vs. SIRth for a velocity of 60 km/h.
performs better than the reuse factor-1 scheme, and similarly to the reuse factor-3 scheme. Interestingly, the fairness index of the PSRR and AREA schemes are slightly higher than that of the reuse factor3 scheme for the SIRth range of greater than 28 dB. This is because the partial reuse of sounding sequences leads to improved performance for edge area users and mitigates the throughput gap between center area users and edge area users.
761
5. Conclusion
770
In this paper, we proposed the partial sounding resource reuse (PSRR) method to provide seamless QoS provisioning for mobile users in M-MIMO systems. The QoS provisioning is considered in the two aspects for each user; achievable SIR and seamless sounding sequence assignment. We mathematically analyze the QoS provisioning performance of the PSRR according to the system design parameters that are the ratio of edge area to cell area, and the ratio of the number of sequences for edge area users to the total number of sequences. Then, the effectiveness of our analysis model on performance is evaluated through simulations. The results confirm that the PSRR system performs best among the competitive schemes in terms of the number of QoS users and fairness, while minimizing the degradation in system capacity.
771
References
786
[1] Analysis of the impact of traffic growth on the evolution of Internet access, White Paper delivered to the US FCC by the Corporate CTO Organization, Bell Labs, Alcatel-Lucent, 2010. [2] Y. Kishiyama et al., Evolution concept and candidate technologies for future steps of LTE-A, in: Proc. of IEEE ICCS, 2012, PP. 473–477. [3] J.-G. Choi, S. Bahk, Cell-throughput analysis of the proportional fair scheduler in the single-cell environment, IEEE Trans. Veh. Technol. 56 (2) (2007) 766–778. [4] T. Nakamura et al., Trends in small cell enhancements in LTE advanced, IEEE Commun. Mag. 51 (2013) 98–105. [5] M. Iwamura et al., Carrier aggregation framework in 3GPP LTEadvanced, IEEE Commun. Mag. 48 (2010) 60–67. [6] Y.-J. Choi, S. Park, S. Bahk, Multichannel random access in OFDMA wireless networks, IEEE J. Sel. Areas Commun. 24 (3) (2006) 603– 613. [7] H.Q. Ngo, E. G Larsson, T.L. Marzetta, Energy and spectral efficiency of very large multiuser MIMO systems, IEEE Trans. Commun. 61 (2013) 1436–1449. [8] H.Q. Ngo, E.G Larsson, T.L. Marzetta, Uplink power efficiency of multiuser MIMO with very large antenna arrays, in: Proc. of IEEE Communication 49th Annual Allerton Conference on Control, and Computing (Allerton), 2011, pp. 1272–1279. [9] F. Rusek et al., Scaling up MIMO: opportunities and challenges with very large arrays, IEEE Signal Process. Mag. 30 (2013) 40–60. [10] C. Shepard et al., Argos: practical many-antenna base stations, in: Proc. of ACM 18th Annual International Conference on Mobile Computing and Networking, 2012, pp. 53–64. [11] T.L. Marzetta, Noncooperative cellular wireless with unlimited numbers of base station antennas, IEEE Trans. Wireless Commun. 9 (2010) 3590–3600. [12] H. Yang, T.L. Marzetta, Performance of conjugate and zero-forcing beamforming in large-scale antenna systems, IEEE J. Sel. Areas Commun. 31 (2013) 172–179. [13] R. Kudo, S. Armour, J.P. McGeehan, M. Mizoguchi, A channel state information feedback method for massive MIMO-OFDM, J. Commun. Netw. 15 (4) (2013) 352–361.
787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822
45
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
762 763 764 765 766 767 768 769
45
SIR threshold (dB)
739
11
772 773 774 775 776 777 778 779 780 781 782 783 784 785
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861
12
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
[14] G. Caire, N. Jindal, M. Kobayashi, N. Ravindran, Multiuser MIMO achievable rates with downlink training and channel state feedback, IEEE Trans. Inform. Theory 56 (2010) 2845–2866. [15] S. Park, C.B. Chae, S. Bahk, Before/after precoding massive MIMO systems for cloud radio access networks, J. Commun. Netw. 15 (4) (2013) 398–406. [16] H.S. Kim, S.H. Lee, Y.H. Lee, Uplink channel sounding in TDD-OFDM cellular systems, J. Wireless Pers. Commun. (2013). [17] J. Jose et al., Pilot contamination problem in multi-cell TDD systems, IEEE Int. Symp. Inform. Theory (2009) 2184–2188. [18] D. Gesbert et al., Shifting the MIMO paradigm, IEEE Signal Process. Mag. 24 (2007) 36–46. [19] J. Hoydis, S. Brink, M. Debbah, Massive MIMO in the UL/DL of cellular networks: how many antennas do we need?, IEEE J Sel. Areas Commun. 31 (2013) 160–171. [20] T.D. Novlan, R.K. Ganti, A. Ghosh, J.G. Andrews, Analytical evaluation of fractional frequency reuse for OFDMA cellular networks, IEEE Trans. Wireless Commun. 10 (2011) 4294–4305. [21] 3GPP Technical Specification 36.305, ‘Evloved Universal Terrestrial Radio Access Network (E-UTRAN); Stage 2 functional specification of User Equipment (UE) positioning in E-UTRAN (Release 9)’. http:// www.3gpp.org. [22] E. Del Re, R. Fantacci, G. Giambene, Efficient dynamic channel allocation techniques with handover queuing for mobile satellite networks, IEEE J. Sel. Areas Commun. 13 (1995) 397–405. [23] B. Sklar, Rayleigh fading channels in mobile digital communication systems. I. Characterization, IEEE Commun. Mag. 35 (1997) 90–100. [24] C. Bettstetter, H. Hartenstein, X. Perez-Costa, Stochastic properties of the random waypoint mobility model: epoch length, direction distribution, and cell change rate, in: Proc. of ACM International Workshop on Modeling Analysis and Simulation of Wireless and Mobile Systems, 2002, pp. 7–14. [25] F. Bai, N.Sadagopan, A. Helmy, Important: a framework to systematically analyze the impact of mobility on performance of RouTing protocols for adhoc networks, in: Proc. of IEEE INFOCOM, March 2003, pp. 825–835. [26] R. Jain, D. Chiu, W. Hawe, A Quantitative Measure of Fairness and Discrimination for Resource Allocation in Shared Computer Systems, DEC, Res. Rep. TR-301, 1984.
862 865 866 867 868 869 870 871 872 873
Taeseop Lee is currently a Ph.D. candidate in the School of Electrical Engineering and Computer Science, Seoul National University. He received his B.S. degree from Seoul National University, in 2011. His research interests include fourth-generation wireless networks, Massive MIMO, and multi-path TCP.
Sangkyu Park received B.S degree in Electrical Engineering from Yonsei University, Seoul, Korea in 2009. He is currently working towards the Unified Master and Doctor course with Institute of New Media and Communications, Seoul National University. His current research interests include cloud radio access network, massive MIMO system, and smallcell network.
875 Hyung-Sin Kim received the B.S. degree from Seoul National University (SNU), Seoul, Korea in 2009 and the M.S. degree from SNU in 2011, all in electrical engineering. He is currently working towards the Ph.D. degree in electrical engineering at SNU. His research interests include asynchronous power wireless system, network configuration and network-joining techniques in wireless sensor network, and uplink channel sounding in OFDM systems.
888 889 890 891 892 893 894 895 896 897 898 899
887 Saewoong Bahk received B.S. and M.S. degrees in Electrical Engineering from Seoul National University in 1984 and 1986, respectively, and the Ph.D. degree from the University of Pennsylvania in 1991. From 1991 through 1994, he was with AT&T Bell Laboratories as a Member of Technical Staff where he worked for AT&T network management. In 1994, he joined the school of Electrical Engineering at Seoul National University and currently serves as a Professor. He has been serving as TPC members for various conferences including ICC, GLOBECOM, INFOCOM, PIMRC, WCNC, etc. He is on the editorial boards of IEEE Transaction on Wireless Communications (TWireless), Computer Networks Journal (COMNET), and Journal of Communications and Networks (JCN). His areas of interests include performance analysis of communication networks and network security. He is an IEEE Senior Member and a Member of Whos Who Professional in Science and Engineering.
864
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
876 877 878 879 880 881 882 883 884 885
902 903 904 905 906 907 908 909 910 911 912 913 914 915 901 916 917 918 919 920 921
No. of Pages 12, Model 3G
14 May 2014 Computer Networks xxx (2014) xxx–xxx 1
Contents lists available at ScienceDirect
Computer Networks journal homepage: www.elsevier.com/locate/comnet 5 6
Sounding resource management for QoS support in massive MIMO systems
3 4 7
Q1
Taeseop Lee, Sangkyu Park, Hyung-Sin Kim, Saewoong Bahk ⇑
8
Q2
INMC, Department of Electrical and Computer Engineering, Seoul National University, Republic of Korea
9
a r t i c l e
1 2 1 3 12 13 14 15 16
i n f o
Article history: Received 5 January 2014 Received in revised form 29 April 2014 Accepted 2 May 2014 Available online xxxx
17 18 19 20 21 22
Keywords: Massive MIMO Pilot contamination QoS Sounding resources
a b s t r a c t In massive multiple-input multiple-out (M-MIMO) systems, the base station estimates each forward-link channel by using the reverse-link sounding pilot and the channel reciprocity property of time division duplex (TDD) operation. However, conventional noncooperative sounding schemes, like a sounding sequence assignment scheme with reuse factor-1, cause cell edge users to suffer from pilot contamination in multi-cell scenarios. The pilot contamination problem becomes even worse in an environment where mobile users are travelling through cell edge areas frequently. To alleviate this problem, a cooperative sounding resource reuse method that assigns each cell a set of resources different from other neighboring cells by using a reuse factor-3 scheme, can be considered, but it significantly lowers the number of served users. In this paper, we propose a partial sounding resource reuse (PSRR) method which aims to seamlessly support the quality of service (QoS) of each mobile user by mitigating the pilot contamination and to minimize the reduction in the number of served users. To this end, the PSRR divides each cell area into center and edge areas, and applies a reuse factor-1 scheme for center users while a reuse factor-3 scheme for edge users. We use a Markov chain model to analyze the performance of the PSRR, and evaluate the accuracy of our analysis through simulations. Then we confirm that the PSRR considerably improves QoS performance over the conventional competitive schemes. Ó 2014 Published by Elsevier B.V.
24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
45 46
1. Introduction
47
Recently, wireless data traffic has dramatically increased with the advent of smart phones, so the wireless capacity enhancement in cellular systems has attracted a lot of attention from many researchers [1–3]. To this end, the deployment of small cells and the use of wide bandwidth have been considered as ways of increasing the wireless
48 49 50 51 52
⇑ Corresponding author. Address: Seoul National University, Institute of
Q1
New Media and Communications, 1 Gwanak-ro, Gwanak-gu, Seoul 151-742, Republic of Korea. Tel.: +82 2 880 8434; fax: +82 2 880 8198. E-mail addresses: [email protected] (T. Lee), [email protected]. ac.kr (S. Park), [email protected] (H.-S. Kim), [email protected] (S. Bahk).
capacity [4–6]. Another approach to increase wireless system capacity is to use a large-scale antenna system, also known as massive multiple-input multiple-output (MMIMO) system. M-MIMO systems enable many users with good link quality to be multiplexed in spatial domain, exploiting the cancellation of inter-user interference and noise on desired signals [7–10]. In M-MIMO systems, the base station uses the downlink channel information toward each terminal to generate a pre-coding matrix. A quantization-based channel information feedback method needs huge amounts of downlink and uplink resources when applied to M-MIMO systems, due to the downlink per-antenna pilot signal broadcasting and the uplink feedback overhead, respectively. Thus, uplink sounding techniques are expected to be adopted,
http://dx.doi.org/10.1016/j.comnet.2014.05.001 1389-1286/Ó 2014 Published by Elsevier B.V.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120
2
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
which use the channel reciprocity of time-division duplex (TDD) systems [11–13]. The performance of an M-MIMO system greatly depends on how to use limited sounding resources1 and the accuracy of channel estimation [14,15]. Especially in multi-cell environments, the performance of a cell edge user is significantly degraded due to pilot contamination when each cell exploits the whole sounding resources in a noncooperative manner [11,17]. This means that the pilot contamination severely degrades system performance of an M-MIMO system in terms of QoS for each cell edge user. To provide seamless handover and QoS for each cell edge user, sounding resources need to be reused with reuse factor-3 in a cooperative manner. It provides high signal to interference ratio (SIR) for each edge user at the cost of having a lowered number of served users.2 It has been shown that non-cooperative methods provide higher system capacity while cooperative reuse methods achieve better edge user QoS [11]. Thus, a new sounding technique needs to be designed by considering the trade-off relation between the pilot contamination and the reduced number of served users. In related works, the trade-off relation between the QoS degradation of users and the total system capacity. In this paper, we propose a partial sounding resource reuse (PSRR) scheme that aims to provide desired QoS to each mobile user seamlessly as many as possible. To this end, the PSRR first divides each cell area into center area and edge area; center area users achieve high SIRs, while edge area users suffer from the pilot contamination problem. Then it splits the whole sounding sequences into two sets for center and edge area users, respectively. The center area sounding resources are reused with reuse factor-1, and the edge area ones with reuse factor-3 to mitigate pilot contamination. Whenever a mobile terminal initiates a handover (i.e. center to edge area or edge to edge area), it will be reassigned a new sounding resource to achieve appropriate QoS. We mathematically analyze QoS provisioning performance of the PSRR according to the two ratio parameters; one is the ratio of edge area to the total cell area in size, and the other is the ratio between the sounding resources in number for edge area and total cell area. In the analysis, we simplify the operation of PSRR with the above simple two parameters that are enough to control the QoS provisioning performance of the PSRR. Through simulations, we evaluate the accuracy of our analysis, and confirm that the PSRR improves QoS provisioning performance. We summarize main contributions of this paper as follows.
We propose a partial sounding resource reuse (PSRR) method that allocates sounding resources for mobile users by considering the trade-off relation between pilot contamination and reduced served users in number. For PSRR operation, we analytically obtain the QoS provisioning performance of the PSRR by using the Markov chain model with simple parameters.
123 124 125 126 127 128 129 130
2. Proposed PSRR scheme
140
In this section, we consider the model for an M-MIMO system, explain the pilot contamination problem, and propose the PSRR scheme to maximize the number of users to be served while supporting each user’s QoS.
141
2.1. System model
145
As depicted in Fig. 1, we consider a TDD-based M-MIMO system where a time slot comprises T symbols. Here T is a given parameter, determined by the coherence time during which the channel information does not change. The base station estimates the downlink channel condition for each terminal from uplink sounding pilots received for the preceding sr symbols, and then transmits downlink data for the next ðT sr Þ symbols according to the estimated channel condition. Letting N smooth be the number of consecutive subcarriers in frequency domain which have the same channel gain (i.e. coherence bandwidth), a block of N smooth T subcarriers have the same channel condition. Using the sounding subcarrier block which is denoted by a bold rectangular in Fig. 1, the M-MIMO system generates
146
Fig. 1. TDD-based M-MIMO system.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
122
The remainder of this paper is organized as follows. In Section 2, we discuss the system model and presents the proposed PSRR method. In Section 3, we mathematically analyze the QoS provisioning performance of PSRR, determine the proper values for the two system parameters and validate the correctness of our analysis. In Section 4, through simulations, we evaluate the performance of PSRR and its performance improvement over the competitive schemes. Finally, we summarize the contributions of this paper and conclude our work.
To the best of our knowledge, this is the first work which considers partial sounding resource reuse for guaranteeing the QoS of mobile users in M-MIMO systems.
1 This is because the number of spatially multiplexed users is directly related to the number of sounding users. 2 In this paper, the number of served users indicates the number of spatially multiplexed users for the same downlink resources, which is limited by the number of orthogonal sounding sequences.
121
131 132 133 134 135 136 137 138 139
142 143 144
147 148 149 150 151 152 153 154 155 156 157 158 159
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
3
Fig. 2. Pilot contamination problem in a non-cooperative M-MIMO system.
160 161 162
163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
N smooth sr orthogonal sounding resources.3 Thus, in a TDDbased M-MIMO system, each cell can support Nsmooth sr users at a maximum for the given downlink resources. 2.2. Effect of pilot contamination in a non-cooperative M-MIMO system If error free channel information is available, the M-MIMO system can serve multiple users using the same downlink resources without creating inter cell interference [18,19]. However, in reality, a conventional non-cooperative M-MIMO system cannot provide accurate channel information due to the pilot contamination. For example, let’s consider a case of three neighboring cells, each of which serves its own user k with the same sounding sequence k as depicted in Fig. 2. In this case, the base station of cell l or cell h receives not only the desired sounding pilot signal but also the sounding interference from user k in cell j (i.e. pilot contamination), which results in a polluted pre-coding matrix. As a result, the base stations of cells l and h generate inaccurate beamforming, and the downlink data transmission of cell j suffers from the increased interference from those of cells l and h. In this paper, we adopt the SIR model in [11] where an infinite number of antennas are assumed. Defining blkj as the static channel gain between base station l and user k (with sounding sequence k) in cell j, we have the downlink SIR of user k in cell j as
186
188 189 190 191 192
SIRDkj
b2jkj ¼P 2 j–l blkj
where blkj
zlkj ¼ c rlkj
ð1Þ
where r lkj is the distance between base station l and user k in cell j; c is the path loss exponent, and zlkj is a log-normal random variable [11]. It can be seen that SIRDkj decreases with the interference signal due to the pilot contamination
(i.e. blkj where l – j). Thus, a mobile terminal has difficulty in receiving seamless QoS support due to the pilot contamination when it approaches its cell boundary. To alleviate the pilot contamination problem, sounding sequence assignment with reuse factor-3 that assigns each neighboring cell a different set of sounding sequences can be considered. However, it has the drawback of reduction in number of served users, resulting in critical system capacity degradation. This motivates us to propose the PSRR that aims to provide appropriate QoS for mobile users while minimizing the system capacity degradation.
193
2.3. Proposed scheme
204
The PSRR aims to mitigate the pilot contamination problem and support as many users as possible by applying a different sounding sequence assignment method for each user according to its location. As illustrated in Fig. 3, the proposed PSRR divides each cell area into center area and edge area. Let’s denote the ratio of edge area to cell area in size as F. Then
205
AreaE F¼ AreaCell
ð2Þ
and AreaC is the center area and AreaE is the edge area. F is directly related to the number of users, each of which needs a dedicated sounding sequence (i.e. not used by neighboring cells). Let’s denote the set of total orthogonal sounding sequences as S. The PSRR divides S into subsets SC and SE . Then, it assigns SC to center area users with reuse factor1, and SE to edge area users with reuse factor-3. Considering the hexagonal cell configuration in Fig. 3, the PSRR divides SE into three equal sized subsets SEi for i ¼ 1; 2; 3, and assigns each neighboring cell a different SEi . Defining the ratio of the number of sounding sequences for edge area users to the total number of sequences as H, we have
3
There are two methods for sounding resource generation [16]. In a cyclic shift separation (CSS) system, (Nsmooth T) sounding sequences, each of which has the length of (N smooth T), are generated. In a decimation separation (DS) system, each of (N smooth T) subcarriers is a sounding resource. Although we consider a CSS system in the following sections, the proposed PSRR can also be applied to a DS system.
where AreaCell ¼ AreaC þ AreaE ;
H¼
jSE j ; jSj
ð3Þ
where jSj is the cardinality of set S. The performance of PSRR mainly depends on the interaction between F and
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
194 195 196 197 198 199 200 201 202 203
206 207 208 209 210 211
212 214 215 216 217 218 219 220 221 222 223 224 225 226 227
228 230 231 232
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
4
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
277
F ¼1
Fig. 3. PSRR overview.
238
H. A large number of cell edge users may suffer from pilot contamination when H is too small compared to F. On the contrary, the number of served users may be significantly lowered when H is too large compared to F. In the next section, we mathematically show the effect of these parameters on the QoS provisioning performance of PSRR.
239
3. Mathematical analysis for PSRR operation
240
251
For QoS provisioning, the PSRR aims to provide the minimum required SIR, denoted as SIRth , for each mobile user located in edge area, while reusing sounding sequences as much as possible for center area users. We first analyze the relationship between F and SIRth to show SIR provisioning performance of the PSRR. Let P sf represent the probability of sounding sequence assignment failure as a performance metric for the PSRR. Then, we analyze the relationship between H and Psf . Finally, we determine the proper values of two system parameters F and H according to the mathematical analysis, and confirm the accuracy of our model through simulations.
252
3.1. Relation between F and SIRth
253
Let’s denote the distance between a user and the base station as d. For ease of analysis, we consider a multi-cell environment where all cells are uniformly deployed, a sufficiently large number of mobiles are served, and all base stations transmit downlink signals with an equal power. Then, the downlink SIR can be approximated as a function of d, denoted as SIRD ðdÞ [20]. The dark dotted line in Fig. 4(a) shows the expected downlink SIR according to d for the sounding assignment scheme with reuse factor-1. It can be seen that the expected SIR exponentially decreases with d due to the pilot contamination, resulting in failure of SIRth guarantee when a mobile user approaches the cell boundary. To meet the SIRth requirement of each cell edge user, the PSRR divides each cell area into center and edge areas according to whether the assignment scheme with reuse factor-1 can meet the condition of SIRth for a mobile user or not. The center area has a circular shape as shown in Fig. 4(b) since SIR is given as a function of d. Then the PSRR determines the center area radius, r, and accordingly F is given as follows.
233 234 235 236 237
241 242 243 244 245 246 247 248 249 250
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273
274 276
r ¼ max fdjSIRD ðdÞ P SIRth g
ð4Þ
pr2 AreaCell
ð5Þ
It can be easily seen that r decreases as SIRth increases, and F increases as r decreases. Therefore, the increase of SIRth results in the increase of F.4
280
3.2. Relation between H and P sf
283
A service failure happens when there is no available sounding sequence for a mobile user. In the PSRR system, we consider two types of service failure for a mobile terminal depending on whether it is newly joining or in the middle of handover. We define service blocking and handover (HO) blocking as sequence assignment failures for a new service user and a handover user, respectively. Therefore Psf represents the probability that a mobile terminal experiences service failure due to service blocking or handover blocking. In this subsection, we mathematically analyze the relation between H and P sf , and calculate Psf by using a Markov chain model.
284
3.2.1. Mobility model and queuing of handover requests Let’s assume that cells are hexagonal and regular. We denote the radii of circumscribed circle, inscribed circle, and center circle for each hexagonal cell as R0 ; R and r, respectively. The mobility model in the PSRR system uses the following assumptions.
296
(a) Each user moves at a constant velocity of V[km/h]. (b) Each user moves across a cell along its maximum diameter.
302
281 282
285 286 287 288 289 290 291 292 293 294 295
297 298 299 300 301
303 304 305
When a user crosses multiple cell areas with the above mobility assumptions, it is assigned a new sounding sequence at each area to achieve seamless QoS provisioning. We define this new sounding sequence assignment as a handover procedure and consider applying a handover prioritization strategy according to the queuing for handover. In Fig. 5, when a mobile terminal crosses multiple cells, it experiences three types of handover that are denoted by HOEE ; HOEC , and HOCE (i.e. handover between edge to edge, edge to center, and center to edge). During each handover process, it transmits a handover request to the base station to trigger the handover. Then the PSRR reassigns it an available sounding sequence in the associated area. However, if all the sequences in that area are already in use, the handover request gets queued until a sounding sequence is available. If the terminal fails to get a new sounding sequence until it gets out of the handover area, a handover blocking occurs. By defining a maximum handover queuing time as the time duration for which the user starts handover and gets out of the handover area, we can set the maximum handover queuing time t max to a fixed value according to w 4 In LTE systems, an Evolved Serving Mobile Location Center (E-SMLC) can estimate each user’s position by using Observed Time Difference Of Arrival (OTDOA) positioning scheme [21]. OTDOA uses the time difference (observed by the user) between the reference signals of neighboring cells and the serving cell. By using the estimated position of each user, a base station can trigger the HO procedures.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
279
306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
5
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
Fig. 4. Determination of F according to SIRth .
328 329
the considered mobility model. For the HOEE case in Fig. 5, we have
330 0
332 333 334 335
336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351
t max ¼ w
R ð1 V
pffiffi 3 Þ 2
at center area and edge area are denoted by P hC and P hE , respectively, they are given by a
’
0:134R0 : V
ð6Þ
t max w
Note that has a different value according to the mobility model and handover case. For simple analysis, we use the same tmax for HOEC and HOCE cases. w 3.2.2. Memoryless property of handover arrivals To design a markov chain model for the sounding resource assignment problem, we need to know the handover arrival rate at each area. For doing this, we first calculate the probability that an ongoing user in service is continuously active until handover happens. Let tdC and tdE represent the intervals elapsed from the arrival instant of a new user at center area and edge area, respectively, to the instant of its entering the handover area. According to the mobility assumptions, t dC and t dE are random variables uniformly distributed in ½0; 2r=V and ½0; ðR rÞ=V, respectively. A random service duration t s is exponentially distributed with mean T m , and an ongoing served user in the center area (or edge area) generates a handover request whenever t s > tdC (or t dE ). When the handover probabilities
Fig. 5. Mobility model and queuing of handover requests.
PhC ðaÞ ¼ Prfts > t dC g ¼
1e
a
2r where a ¼ ; VT m
ð7aÞ
353
354 356 357
1 eb PhE ðbÞ ¼ Prfts > t dE g ¼ b
Rr where b ¼ : VT m
ð7bÞ
An ongoing served user who has already experienced a successful handover travels a deterministic distance of equal to 2r and R r in center area and edge area, respectively, before starting a new handover process. Let’s denote the deterministic inter arrival times for handover at center area and edge area after the initial arrival as t0dC and t 0dE , respectively. Then they are given as 2r=V and ðR rÞ=V, respectively. The residual service time of an ongoing served user after a successful handover is also exponentially distributed with mean T m due to the memoryless property. Let P0hC and P 0hE represent the probabilities that a user having experienced a handover begins another handover in center area and edge area, respectively. Then we obtain
P0hC ð
aÞ ¼ Pr ts > t0dC ¼ ea
2r where a ¼ ; VT m
ð8aÞ
Rr : VT m
ð8bÞ
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373
374 376 377
P0hE ðbÞ ¼ Pr t s > t0dE ¼ eb
where b ¼
379
The analysis above confirms that the sounding resource assignment problem can be modeled by a Markov chain.
380
3.2.3. Markov model for sounding sequence assignment Fig. 6 depicts the two Markov chain models for center and edge areas, respectively. For simple analysis, we assume the following [22].
382
(a) The numbers of permanently allocated sounding sequences are jSC j for center area users and jSEi j for edge area users. (b) New arrivals and handover arrivals are independent Poisson processes with mean rates k and kh , respectively.
386
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
352
381
383 384 385
387 388 389 390 391
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
6
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
Fig. 6. Markov chain models for sounding sequence assignment in PSRR.
(c) The sequence holding times are approximated as exponentially distributed random variables with means l1 and l1 for center area and edge area, respeC E ctively. (d) The maximum handover queuing time5 t max is w assumed as the mean of an exponentially distributed random variable. (e) The queue size is infinite.
392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411
412 414
Using the above assumptions, we model the sounding resource assignment problem for the PSRR as an M/M/jSA j queuing system for area A which is C or E depending on the considered area. To obtain the service blocking probability and the handover blocking probability in each area, we calculate the steady state probabilities first. For a given number of available sounding sequences for area A; jSA j, we can obtain the average new arrival rate kA , the average handover arrival rate khA , and the average sounding sequence holding time 1=lA , when F and H are given. Simply we have
jSC j ¼ ð1 HÞjSj;
ð9aÞ
415 417 418 419 420 421
422 424
jSEi j ¼
jSE j HjSj ¼ 3 3
for i ¼ 1; 2; 3:
ð9bÞ
When the average new service arrival rate at each cell is given as k, we obtain the average new service arrival rates kC and kE for center area and edge area, respectively, as follows.
kC ¼ ð1 FÞk;
5
ð10aÞ
This value is approximated as a constant in Eq. (6) according to the mobility model and handover case.
kE ¼ Fk:
ð10bÞ
Then we obtain the average handover arrival rates khC and khE for center area and edge area, respectively, as follows.
khC ¼ khC ð1
PhbC ÞP0hC
þ kC ð1 PsbC ÞPhC ;
ð11aÞ
428 429 430
431 433 434
khEE þ khCE ¼ khEE ð1 PhbE ÞP0hE þ khCE ð1 PhbE ÞP 0hE þ kE ð1 PsbE ÞP hE ; khE ¼ khE ð1 P hbE ÞP0hE þ kE ð1 P sbE ÞPhE ;
ð11bÞ
436
ð11cÞ
437 439
where P hbC and P hbE denote handover blocking probabilities for center area and edge area, respectively, and PhC ; PhE ; P 0hC ; P 0hE are handover probabilities derived from (7) and (8). To help understanding, Fig. 7 depicts the statistical equilibrium state in the PSRR system. For the center area, the average handover arrival rate is khC , and the average handover departure rate is khC ð1 PhbC ÞP 0hC þ kC ð1 PsbC ÞPhC , as shown in Eq. (11a). On the other hand, for the edge area, we should consider the handover arrival rate from the center area khCE and that from the neighboring cells, khEE . Then the average of total handover arrival rate at the edge area, khE , becomes the sum of khCE and khEE . Then we can calculate the handover arrival rates khC and khE , when the service blocking probabilities, P sbC and P sbE , and the handover blocking probabilities, P hbC and P hbE are given. To obtain P hbC ; P hbE ; P sbC and P sbE from the Markov chain model, we need the handover arrival rates khC and khE again. To compute these, we use a recursive approach. Let’s denote the sounding sequence holding times for a new service user in center area and edge area as t HC and tHE , respectively. Then they can be expressed, respectively, as
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
425 427
440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
7
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
Fig. 7. Statistical equilibrium between mobile terminals’ arrivals and departures at each area.
463 465
t HC ¼ min½t s ; tdC ;
ð12aÞ
466 468
t HE ¼ min½t s ; tdE :
ð12bÞ
469 470 471 472
473 475 476 478
479 480 481
482
Denoting the sounding sequence holing time of an ongoing served user in center area (or edge area) which has already experienced a first handover by t0HC (or t 0HE ), we have
t 0HC ¼ min½t s ; t0dC ;
ð13aÞ
t 0HE ¼ min½t s ; t0dE :
ð13bÞ
where the system idle probability is given as P 0A ¼
( jS j A X ðkA þ khA Þn n!lnA
n¼1
Let’s denote the average sounding sequence holding times for each area as 1=lC and 1=lE . Then they are given, respectively, as
lC 484
lE
490 491 492 493 494 495 496 497 498 499 500 501 502 503 504
505
kC khC ¼ E½t HC þ E½t0 kC þ khC kC þ khC HC kC khC ¼ T m ð1 PhC Þ þ T m ð1 P0hC Þ; kC þ khC kC þ khC kE khE ¼ E½tHE þ E½t 0 kE þ khE kE þ khE HE kE khE T m ð1 PhE Þ þ T m ð1 P0hE Þ: ¼ kE þ khE kE þ khE
ð14aÞ
As a service blocking occurs if there is no available sounding sequence, the service blocking probability in area A can be expressed as
512
PsbA ¼
1 X
PnA
513 514
ð17Þ 517
n¼jSA j
In the case of handover request queuing, a handover blocking occurs if the queuing time exceeds t max w . Then we obtain the conditional handover blocking probability at state n as
PhbAjn ¼ 1
njS YA j
" 1
8 ðk þk Þn A hA > < n!lnA P0A ;
ð14bÞ
> :
1 6 n 6 jSA j;
ðkA þkhA ÞjSA j kh AðnjSA jÞ jSA j!l
jSA j A
QnjSA j j¼1
ðjSA jþjlw Þ
P0A
jSA j þ 1 6 n;
lw ðjSA jlA þ lw Þ2j
ð15Þ
PhbA ¼
1 X
:
519 520 521
ð18Þ 524 525 526
527
PhbAjn PnA :
ð19Þ 529
n¼jSA j
3.2.4. Calculation of P sf We compute the service failure probability Psf , using the above results. Let’s denote the forced service termination probabilities due to the handover blocking by PdropC and P dropE , respectively, according to the user’s service originating location. Then we can express P sf as
530
Psf ¼ ð1 FÞ PsbC þ ð1 P sbC ÞPdropC þ F PsbE þ ð1 PsbE ÞPdropE :
536
531 532 533 534 535
ð20Þ
538
Here the first term indicates the forced service termination probability due to the service blocking or handover blocking for a center area user and the second term represents the same for an edge area user. Before computing PdropC and P dropE , let’s denote the probabilities that a service attempt is served and it has i (¼ 1; 2; 3; . . .) successful handovers originating from center area and edge area as HiC and HiE , respectively. Then we obtain
539
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
518
522
#
Then the handover blocking probability in area A is given by
As depicted in Fig. 6, we define the system state as the number of users who need sounding sequences in each area. This equals the sum of the number of users in service and the number of queued handover requests. If the current state is lower than the number of available sounding sequences in area Að¼ C or Ei Þ, denoted by jSA j, the PSRR system can accept all the users and the total arrival rate equals kA þ khA . If the current state is higher than jSA j, the total arrival rate becomes khA because we the PSRR blocks newly joining users to give handover users higher priority. When the queuing system state is (jSA j þ j) for j ¼ 1; 2; . . ., we need to consider the case in which a handover blocking occurs because the queuing delay exceeds the maximum handover queuing time. Thus we represent the death rate due to the handover blocking as jlw which equals j=tmax w . Therefore, the steady state probabilities in area A can be expressed as
PnA ¼ 507
: 511
j¼0
1
489
509
!)1
ð16Þ
485
488
n¼jSA jþ1
ðkA þ khA ÞjSA j kh AðnjSA jÞ QnjSA j jSA j!ljSA A j j¼1 ðjSA j þ jlw Þ
515
1
487
þ
1 X
508
540 541 542 543 544 545 546 547
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
8
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
548
HiC ¼ 550 551
HiE ¼ 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568
569
8 bði1Þ=3c > > P ðP0 P 02 Þ > < hC hC hE
bði1Þ=3c PhC P 0hE ðP 0hC P02 hE Þ > > > : 0 02 bði1Þ=3c PhC P 02 hE ðP hC P hE Þ
for i%3 ¼ 1 ð21aÞ
for i%3 ¼ 2 for i%3 ¼ 0;
8 bði1Þ=3c > > P ðP0 P02 Þ > < hE hC hE
bði1Þ=3c PhE P0hE ðP0hC P02 hE Þ > > > bði1Þ=3c : PhE P0hE P0hC ðP0hC P02 hE Þ
for i%3 ¼ 1 ð21bÞ
for i%3 ¼ 2 for i%3 ¼ 0;
where % represents a modulo operation, and bxc chooses a largest integer that is smaller than or equal to x. As shown in Fig. 8, a user starting to receive a service from a center area moves according to the considered mobility model and experiences multiple handovers. The probability that the user experiences the first handover is given as PhC (see Eq. (7a)), and the probability that the user experiences the second handover is given as PhC P 0hE by applying the memoryless property of the service time. Then we can compute the i-th handover probability of a user originating from the center area through the same procedures. Let’s denote the probabilities that a user in service is dropped at the i-th handover (i ¼ 1; 2; 3; . . .) and it has started from center area and edge area as BiC and BiE , respectively. Then we obtain 8 n obði1Þ=3c > PhbE ð1 P hbE Þ2 ð1 P hbC Þ > > > > < n obði1Þ=3c BiC ¼ ð1 P hbE ÞP hbE ð1 P hbE Þ2 ð1 P hbC Þ > > > n obði1Þ=3c > > 2 : ð1 P Þ2 P hbE hbC ð1 P hbE Þ ð1 P hbC Þ
for i%3 ¼ 1; for i%3 ¼ 2; for i%3 ¼ 0;
571 572
ð22aÞ 8 n obði1Þ=3c 2 > > for i%3¼1; > P hbE ð1P hbE Þ ð1P hbC Þ > > < n obði1Þ=3c BiE ¼ ð1P hbE ÞP hbC ð1PhbE Þ2 ð1P hbC Þ for i%3¼2; > > > n obði1Þ=3c > > 2 : ð1P Þð1P ÞP for i%3¼0: hbE hbC hbE ð1P hbE Þ ð1P hbC Þ
574 575
ð22bÞ
From (21) and (22), we have
576 578
PdropC
1 X ¼ HiC BiC ;
582 583
584 585 586 587 588 589 590
3.3.1. Relation between F and SIRth Fig. 9 shows the average downlink SIR according to the distance between the base station and a user, when the sounding assignment scheme with reuse factor-1 (i.e. non-cooperative scheme) and our proposed PSRR with F = 0.3 and 0.6 are considered. Here we omit the results for the case of distance below 150 m because all the three schemes show the same SIR performance. In the non-cooperative scheme, the downlink SIR dramatically decreases with the distance due to the pilot contamination. On the contrary, it can be seen that the PSRR tries to provide the downlink SIR greater than SIRth to each edge user. The PSRR is not able to solve this problem completely since it only mitigates the pilot contamination problem. For example, for a required SIRth of 21.8 dB, the PSRR operates with F ¼ 0:6 to guarantee the SIR value larger than 21.8 dB. Nevertheless, the PSRR with F ¼ 0:6 provides the SIR of 17.6 dB (i.e. lower than 21.8 dB) at distance 300 m, which corresponds to the cell boundary. In worst case, the PSRR may not be able to meet SIRth for cell boundary users eventually due to the path loss. Note that overcoming the path loss problem is beyond the scope of this paper. In PSRR systems, we can calculate the expected SIR as a function of the distance between a user and a base station, when all base stations are uniformly deployed and transmit with an equal power [20]. Therefore, the proper value for F can be determined by a given SIRth value. Table 1 presents the required center area radius r and ratio F accord-
i¼1
579 581
ð23aÞ
Fig. 8. Example of handover operation for a user originating from the center area.
PdropE ¼
1 X HiE BiE :
ð23bÞ
i¼1
Finally, we obtain the service failure probability from (20). 3.3. Verification of analysis and values of F and H for QoS provisioning In this section, we show the correctness of our mathematical model with simulation results and how values of F and H can be determined by the above analysis. The simulation results of this section are driven from a simulation scenario in Section 4.1.
Fig. 9. Average SIR vs. distance between the base station and a user.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
9
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx Table 1 SIRth vs. required edge area ratio.
620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644
rðmÞ
F
262 241 223 206 190 175 160 147 134 122 110 100 90
0.08 0.22 0.33 0.43 0.52 0.59 0.66 0.71 0.76 0.8 0.84 0.87 0.89
ing to different SIRth values. The required values of r and F to meet the SIRth condition are determined by Eqs. (4) and (5) in Section 3.1. It can be seen that F increases with SIRth . Again the PSRR eventually has difficulty in guaranteeing SIRth when SIRth is larger than 17.56 dB due to the path loss problem. Since the limitation comes from radio signal attenuation, the cell size and transmit power need to be determined according to the given SIRth . If all the users achieve SIRs greater than SIRth even at the cell boundary and obtain error free channel information, we can say, the PSRR guarantees SIRth successfully. 3.3.2. Relation between H and P sf Fig. 10 shows Psf according to H for F ¼ 0:2 and 0.4, and compares the results of analysis and simulation. In simulations, we consider two types of mobility model; One is the random waypoint model in which a user chooses its destination randomly and the other is the Manhattan mobility model where a user moves along the road indicated by a dotted line in Fig. 11 [24,25]. It can be observed that the analysis results have slightly lower Psf than the simulation results, but they show about the same tendency. It can be inferred that the gap between analysis and simulation results comes from the mobility assumption since users change their moving directions and experience handover more frequently in simulations with two mobility models than in the analysis model. Furthermore, we can see that
Fig. 11. Cell configuration for simulations.
the results with the Manhattan model show higher Psf than those with the random waypoint model because users in the Manhattan model change moving direction more frequently than users in the random waypoint model. In Table 2, we show optimum values of H which minimize Psf according to F, obtained from analysis and simulations with two mobility models. The optimum H is obtained by minimizing Psf , where the PSRR supports a maximum number of users. Minimizing P sf directly indicates maximizing the number of supported users because more users can be supported without experiencing service and handover blockings with lower Psf . The optimum values of H obtained from analysis and simulation are very close, although there exists some gap between them in terms of P sf . This leads us to focus on the minimum Psf instead of the absolute value of Psf , when we choose an optimum H. Furthermore, Table 2 shows that the required resource ratio H increases with F.
645
4. Performance evaluation
663
In this section, we evaluate the performance of PSRR through simulations and show its performance improvement over the competitive schemes.
664
0.8
Service failure probability
619
SIRth (dB) 3.5 7 10.5 14 17.5 21 24.5 28 31.5 35 38.5 42 45.5
analysis with F=0.2 analysis with F=0.4 random waypoint with F=0.2 random waypoint with F=0.4 manhattan with F=0.2 manhattan with F=0.4
0.7 0.6 0.5
Table 2 Optimum H vs. F.
0.4 0.3 0.2 0.1 0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
H
0.9
1
F
H (analysis)
H (simulation – random waypoint)
H (simulation – manhattan)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0.3393 0.4464 0.5357 0.6071 0.6785 0.7321 0.8035 0.8571 0.9286
0.3214 0.4286 0.5357 0.6071 0.6785 0.75 0.7857 0.8571 0.9286
0.3214 0.4286 0.5357 0.5714 0.6428 0.75 0.7857 0.8571 0.9286
Fig. 10. P sf vs. H.
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662
665 666
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 10
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx 800
reuse−1 (60km/h) PSRR (60km/h) AREA (60km/h) reuse−3 (60km/h)
number of QoS users
700 600 500 400 300
43.3% improvement
200 100 0
800
reuse−1 (100km/h) PSRR (100km/h) AREA (100km/h) reuse−3 (100km/h)
700
number of QoS users
Q1
600 500 400 300
65.4% improvement
200 100
5
10
15
20
25
30
35
40
45
0
5
10
SIR threshold (dB)
15
20
25
30
35
40
45
SIR threshold (dB)
Fig. 12. Number of QoS users vs. SIRth .
667
4.1. Scenario for simulations
668
686
We consider a TDD M-MIMO system based on the LTE standard where the system bandwidth is 20 MHz, sr ¼ 12 symbol time, and N smooth ¼ 14 symbol bandwidth [11]. We consider the user mobility of up to 100 km/h (i.e. coherence time of 1.6 ms), which provides large T enough to support 12 sounding symbols (e.g., T 6 24 symbol time when the symbol time is 66.7 ls as in LTE systems) [23]. Assume that the path loss exponent is 3.8 and the standard deviation of shadow-fading is 8 dB [11]. As depicted in Fig. 11, we assume that each cell has the radius of 300 m and evaluate the performance considering the dotted square area of 19 cells. Also assume that the number of service requests from users in each cell follows a Poisson process with the average of 6 req/s, and the service time is exponentially distributed with mean 15 s. For user mobility, we use the Manhattan mobility model instead of the random waypoint model because the Manhattan mobility model is more appropriate for a real high mobility scenario than the waypoint model.
687
4.2. Performance of the PSRR
688
Now we evaluate the QoS provisioning performance for mobile users in terms of the number of QoS users, system capacity, and Jain’s fairness. The sounding sequence assignment schemes with reuse-1, reuse-3 and AREA6 are considered for performance comparison.
669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685
689 690 691 692
693 694 695 696 697 698
4.2.1. Number of QoS users For evaluation, we define a QoS user as a user who has achieved an SIR value greater than SIRth for 95% or more of its service duration. Fig. 12 shows the numbers of QoS users in the system with respect to SIRth for the reuse factor-1 scheme, the reuse factor-3 scheme, the AREA scheme, 6
In the AREA scheme, the number of sounding sequences assigned to each area is proportional to its size. For example, when the ratio of center area to edge area is 2:1, the ratio of the numbers of sounding sequences in each area also becomes 2:1.
and our proposed PSRR scheme. We consider the user velocities of 60 km/h and 100 km/h. In both cases, the proposed PSRR system outperforms the three compared schemes, and the performance gap increases with the increase of SIRth in the practical range of lower than 17.6 dB. On the other hand, the performance gap starts to decrease when SIRth is higher than 17.6 dB because the PSRR cannot guarantee the SIR in that range. As the SIRth range of higher than 17.6 dB is not practical, we can say that the PSRR outperforms the other three schemes, which is shown on the left side of the dotted line in Fig. 12. The PSRR shows the performance gain of up to 43.3%, 30.8%, and 10.9% compared to the reuse factor-1, reuse factor-3, and AREA schemes, respectively. Fig. 12(b) plots the number of QoS users in each scheme for a velocity of 100 km/h. The PSRR shows the performance gain of up to 65.4%, 27.1%, and 8.4% compared to the reuse factor-1, reuse factor-3, and AREA schemes, respectively. By comparing the results for the user velocities of 60 km/h and 100 km/h, we observe that the performance gap between the PSRR and the reuse factor-1 scheme increases from 43.4% to 65.4%. This result reveals the strength of the PSRR in fast moving environments. On the contrary, the performance gap between the PSRR and the reuse factor-3 schemes decreases from 30.8% to 27.1%, because the PSRR scheme increases the proportion of sounding sequences for edge area and gradually operates like the reuse factor-3 scheme with the mobility speed of each user. Although, the performance gap is also reduced in the AREA scheme for the same reason, we can see that the PSRR still outperforms both the reuse factor-3 and AREA schemes.
699
4.2.2. System capacity Fig. 13 illustrates the average system capacity, which is the sum of served users’ capacity, for each scheme according to the SIRth when the user velocity is 60 km/h. In the reuse factor-1 and reuse factor-3 schemes, the system capacity remains the same regardless of the SIRth because they do not change their operations according to
732
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731
733 734 735 736 737 738
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx 4
12
x 10
11
system capacity
10 9 8 7 reuse−1 (60km/h) PSRR (60km/h) AREA (60km/h) reuse−3 (60km/h)
6 5 4
5
10
15
20
25
30
35
40
Fig. 13. System capacity vs. SIRth for a velocity of 60 km/h.
741 742 743 744 745 746 747 748 749 750 751 752 753 754 755
756 757 758 759 760
SIRth . The PSRR performs like the reuse factor-1 scheme for low SIRth and gradually like the reuse factor-3 scheme with the increase of SIRth . The AREA scheme shows the same tendency with the PSRR scheme, because it increases the proportion of sounding sequences for edge area users, as its area size increases with SIRth . In Fig. 13, it is also shown that the system capacity of the PSRR gradually decreases with the increase of SIRth . For the practical range of SIRth , the PSRR scheme shows the system capacity close to that of the reuse factor-1 scheme and the maximum performance gain of 37% and 13.3% over the reuse factor-3 and AREA schemes, respectively. Considering the results in Figs. 12 and 13, we can confirm that the PSRR accommodates a largest number of QoS users, while minimizing the degradation in system capacity, owing to the partial reuse of sounding resources.
4.2.3. Jain’s fairness Fig. 14 plots Jain’s fairness index for each scheme according to SIRth [26]. The fairness performance of the PSRR gradually approaches that of the reuse factor-3 scheme as SIRth approaches 17.6 dB, and the PSRR always
1 0.9
Jain’s fairness index
740
0.8 0.7 0.6 0.5
reuse−1 (60km/h) PSRR (60km/h) AREA (60km/h) reuse−3 (60km/h)
0.4 0.3 0.2 0.1 0
5
10
15
20
25
30
35
40
SIR threshold (dB) Fig. 14. Jain’s fairness vs. SIRth for a velocity of 60 km/h.
performs better than the reuse factor-1 scheme, and similarly to the reuse factor-3 scheme. Interestingly, the fairness index of the PSRR and AREA schemes are slightly higher than that of the reuse factor3 scheme for the SIRth range of greater than 28 dB. This is because the partial reuse of sounding sequences leads to improved performance for edge area users and mitigates the throughput gap between center area users and edge area users.
761
5. Conclusion
770
In this paper, we proposed the partial sounding resource reuse (PSRR) method to provide seamless QoS provisioning for mobile users in M-MIMO systems. The QoS provisioning is considered in the two aspects for each user; achievable SIR and seamless sounding sequence assignment. We mathematically analyze the QoS provisioning performance of the PSRR according to the system design parameters that are the ratio of edge area to cell area, and the ratio of the number of sequences for edge area users to the total number of sequences. Then, the effectiveness of our analysis model on performance is evaluated through simulations. The results confirm that the PSRR system performs best among the competitive schemes in terms of the number of QoS users and fairness, while minimizing the degradation in system capacity.
771
References
786
[1] Analysis of the impact of traffic growth on the evolution of Internet access, White Paper delivered to the US FCC by the Corporate CTO Organization, Bell Labs, Alcatel-Lucent, 2010. [2] Y. Kishiyama et al., Evolution concept and candidate technologies for future steps of LTE-A, in: Proc. of IEEE ICCS, 2012, PP. 473–477. [3] J.-G. Choi, S. Bahk, Cell-throughput analysis of the proportional fair scheduler in the single-cell environment, IEEE Trans. Veh. Technol. 56 (2) (2007) 766–778. [4] T. Nakamura et al., Trends in small cell enhancements in LTE advanced, IEEE Commun. Mag. 51 (2013) 98–105. [5] M. Iwamura et al., Carrier aggregation framework in 3GPP LTEadvanced, IEEE Commun. Mag. 48 (2010) 60–67. [6] Y.-J. Choi, S. Park, S. Bahk, Multichannel random access in OFDMA wireless networks, IEEE J. Sel. Areas Commun. 24 (3) (2006) 603– 613. [7] H.Q. Ngo, E. G Larsson, T.L. Marzetta, Energy and spectral efficiency of very large multiuser MIMO systems, IEEE Trans. Commun. 61 (2013) 1436–1449. [8] H.Q. Ngo, E.G Larsson, T.L. Marzetta, Uplink power efficiency of multiuser MIMO with very large antenna arrays, in: Proc. of IEEE Communication 49th Annual Allerton Conference on Control, and Computing (Allerton), 2011, pp. 1272–1279. [9] F. Rusek et al., Scaling up MIMO: opportunities and challenges with very large arrays, IEEE Signal Process. Mag. 30 (2013) 40–60. [10] C. Shepard et al., Argos: practical many-antenna base stations, in: Proc. of ACM 18th Annual International Conference on Mobile Computing and Networking, 2012, pp. 53–64. [11] T.L. Marzetta, Noncooperative cellular wireless with unlimited numbers of base station antennas, IEEE Trans. Wireless Commun. 9 (2010) 3590–3600. [12] H. Yang, T.L. Marzetta, Performance of conjugate and zero-forcing beamforming in large-scale antenna systems, IEEE J. Sel. Areas Commun. 31 (2013) 172–179. [13] R. Kudo, S. Armour, J.P. McGeehan, M. Mizoguchi, A channel state information feedback method for massive MIMO-OFDM, J. Commun. Netw. 15 (4) (2013) 352–361.
787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822
45
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
762 763 764 765 766 767 768 769
45
SIR threshold (dB)
739
11
772 773 774 775 776 777 778 779 780 781 782 783 784 785
COMPNW 5302
No. of Pages 12, Model 3G
14 May 2014 Q1 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861
12
T. Lee et al. / Computer Networks xxx (2014) xxx–xxx
[14] G. Caire, N. Jindal, M. Kobayashi, N. Ravindran, Multiuser MIMO achievable rates with downlink training and channel state feedback, IEEE Trans. Inform. Theory 56 (2010) 2845–2866. [15] S. Park, C.B. Chae, S. Bahk, Before/after precoding massive MIMO systems for cloud radio access networks, J. Commun. Netw. 15 (4) (2013) 398–406. [16] H.S. Kim, S.H. Lee, Y.H. Lee, Uplink channel sounding in TDD-OFDM cellular systems, J. Wireless Pers. Commun. (2013). [17] J. Jose et al., Pilot contamination problem in multi-cell TDD systems, IEEE Int. Symp. Inform. Theory (2009) 2184–2188. [18] D. Gesbert et al., Shifting the MIMO paradigm, IEEE Signal Process. Mag. 24 (2007) 36–46. [19] J. Hoydis, S. Brink, M. Debbah, Massive MIMO in the UL/DL of cellular networks: how many antennas do we need?, IEEE J Sel. Areas Commun. 31 (2013) 160–171. [20] T.D. Novlan, R.K. Ganti, A. Ghosh, J.G. Andrews, Analytical evaluation of fractional frequency reuse for OFDMA cellular networks, IEEE Trans. Wireless Commun. 10 (2011) 4294–4305. [21] 3GPP Technical Specification 36.305, ‘Evloved Universal Terrestrial Radio Access Network (E-UTRAN); Stage 2 functional specification of User Equipment (UE) positioning in E-UTRAN (Release 9)’. http:// www.3gpp.org. [22] E. Del Re, R. Fantacci, G. Giambene, Efficient dynamic channel allocation techniques with handover queuing for mobile satellite networks, IEEE J. Sel. Areas Commun. 13 (1995) 397–405. [23] B. Sklar, Rayleigh fading channels in mobile digital communication systems. I. Characterization, IEEE Commun. Mag. 35 (1997) 90–100. [24] C. Bettstetter, H. Hartenstein, X. Perez-Costa, Stochastic properties of the random waypoint mobility model: epoch length, direction distribution, and cell change rate, in: Proc. of ACM International Workshop on Modeling Analysis and Simulation of Wireless and Mobile Systems, 2002, pp. 7–14. [25] F. Bai, N.Sadagopan, A. Helmy, Important: a framework to systematically analyze the impact of mobility on performance of RouTing protocols for adhoc networks, in: Proc. of IEEE INFOCOM, March 2003, pp. 825–835. [26] R. Jain, D. Chiu, W. Hawe, A Quantitative Measure of Fairness and Discrimination for Resource Allocation in Shared Computer Systems, DEC, Res. Rep. TR-301, 1984.
862 865 866 867 868 869 870 871 872 873
Taeseop Lee is currently a Ph.D. candidate in the School of Electrical Engineering and Computer Science, Seoul National University. He received his B.S. degree from Seoul National University, in 2011. His research interests include fourth-generation wireless networks, Massive MIMO, and multi-path TCP.
Sangkyu Park received B.S degree in Electrical Engineering from Yonsei University, Seoul, Korea in 2009. He is currently working towards the Unified Master and Doctor course with Institute of New Media and Communications, Seoul National University. His current research interests include cloud radio access network, massive MIMO system, and smallcell network.
875 Hyung-Sin Kim received the B.S. degree from Seoul National University (SNU), Seoul, Korea in 2009 and the M.S. degree from SNU in 2011, all in electrical engineering. He is currently working towards the Ph.D. degree in electrical engineering at SNU. His research interests include asynchronous power wireless system, network configuration and network-joining techniques in wireless sensor network, and uplink channel sounding in OFDM systems.
888 889 890 891 892 893 894 895 896 897 898 899
887 Saewoong Bahk received B.S. and M.S. degrees in Electrical Engineering from Seoul National University in 1984 and 1986, respectively, and the Ph.D. degree from the University of Pennsylvania in 1991. From 1991 through 1994, he was with AT&T Bell Laboratories as a Member of Technical Staff where he worked for AT&T network management. In 1994, he joined the school of Electrical Engineering at Seoul National University and currently serves as a Professor. He has been serving as TPC members for various conferences including ICC, GLOBECOM, INFOCOM, PIMRC, WCNC, etc. He is on the editorial boards of IEEE Transaction on Wireless Communications (TWireless), Computer Networks Journal (COMNET), and Journal of Communications and Networks (JCN). His areas of interests include performance analysis of communication networks and network security. He is an IEEE Senior Member and a Member of Whos Who Professional in Science and Engineering.
864
Q1 Please cite this article in press as: T. Lee et al., Sounding resource management for QoS support in massive MIMO systems, Comput. Netw. (2014), http://dx.doi.org/10.1016/j.comnet.2014.05.001
876 877 878 879 880 881 882 883 884 885
902 903 904 905 906 907 908 909 910 911 912 913 914 915 901 916 917 918 919 920 921