12 views

Uploaded by Nusrat Hussain Fatema

massive tech

save

- Research Journal Article Critique 2 Romel Espinosa V2
- 25DownlinkMultiUserMIMOcapacity
- VLSI Implementation of OFDM Transceiver for 802.11n systems
- RNL2 LTE
- Advanced 3G and 4G Wireless Mobile
- 31 ijecs
- Performance Analysis of Rayleigh Fading Channel Over Mimo Wireless Communication System
- MIMO in LTE Operation and Measurement
- 1707.00291
- 05 Keysight Agilent Hewlett Packard MIMO in LTE Operation and Measurement - Excerpts 5990-4760EN
- NSN Smart Scheduler for LTE
- lte
- Advanced telecommunication systems
- roze2016.pdf
- Wp Beamforming Revb 11 2010 Lr
- Dm Qm Background Mathematics Lecture 9
- Events
- Thesis-Realistic Multi-Cell Interference Coordination In
- mwoh-samos07
- 212817433-3G-Events.pdf
- RRU3965&RRU3965d Description 01(20151130)
- LTE -1- Overview
- TECHNICAL.pdf
- 4G and Beyond
- LN4_diagonalization
- Sci Lab Quickies
- Synopsis
- 3G_Events
- Thuraya XT Lite Brochure
- Www Teletopix Org Gsm How 26 and 51 Multiframes in Gsm
- 06884080
- vlan
- 068840800000
- Nat Dhcp Chap 7
- Home M2M Survey
- CCNA4E 40 SG Ch 7 (1)

You are on page 1of 11

6, JUNE 2015

3183

**Downlink User Capacity of Massive MIMO Under
**

Pilot Contamination

Juei-Chin Shen, Member, IEEE, Jun Zhang, Member, IEEE, and Khaled B. Letaief, Fellow, IEEE

**Abstract—Pilot contamination has been regarded as a main
**

limiting factor of time division duplexing (TDD) massive multipleinput-multiple-output (Massive MIMO) systems, as it will make

the signal-to-interference-plus-noise ratio (SINR) saturated. However, how pilot contamination will limit the user capacity of downlink Massive MIMO, i.e., the maximum number of users whose

SINR targets can be achieved, has not been addressed. This paper

provides an explicit expression of the Massive MIMO user capacity in the pilot-contaminated regime where the number of users

is larger than the pilot sequence length. This capacity expression

characterizes a region within which a set of SINR requirements

can be jointly satisfied. The size of this region is fundamentally

limited by the pilot sequence length. Furthermore, the scheme for

achieving the user capacity, i.e., the uplink pilot training sequences

and downlink power allocation, has been identified. Specifically,

the generalized Welch bound equality sequences are exploited

and it is shown that the power allocated to each user should

be proportional to its SINR target. With this capacity-achieving

scheme, the SINR requirement of each user can be satisfied and

energy-efficient transmission is achieved in the large-antenna-size

(LAS) regime. The comparison with two non-capacity-achieving

schemes highlights the superiority of our proposed scheme in

terms of achieving higher user capacity. Furthermore, for the

practical scenario with a finite number of antennas, the actual

antenna size required to achieve a significant percentage of the

asymptotic performance has been analytically quantified.

Index Terms—Massive multiple-input-multiple-output (Massive

MIMO), user capacity, pilot contamination, pilot-aided channel

estimation, power allocation.

I. I NTRODUCTION

M

**ASSIVE MIMO is regarded as an efficient and scalable
**

approach for multicell multiuser MIMO implementation, and it can significantly improve both the spectrum efficiency and energy efficiency [1]–[3]. By equipping each base

station (BS) with an antenna array whose size is greatly larger

than the number of active user equipments (UEs), the asymptotic orthogonality among MIMO channels to different UEs is

achieved and in turn it makes intra- and inter-cell interference

more manageable. Hence, using simple linear precoder and

detector can approach the optimal dirty-paper coding capacity

**Manuscript received June 27, 2014; revised November 12, 2014; accepted
**

January 13, 2015. Date of publication February 13, 2015; date of current

version June 6, 2015. This work is supported by the Hong Kong Research Grant

Council under Grant No. 610212. The associate editor coordinating the review

of this paper and approving it for publication was E. A. Jorswieck.

The authors are with the Department of Electronic and Computer Engineering, the Hong Kong University of Science and Technology, Kowloon,

Hong Kong (e-mail: eejcshen@ust.hk; eejzhang@ust.hk; eekhaled@ust.hk).

Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TWC.2015.2403317

**[4]. Channel side information (CSI) at BSs plays an important
**

role in the exploitation of channel orthogonality. In practice,

TDD operation is assumed for CSI acquisition through uplink

training. The advantage of uplink training is that the required

length of pilot training sequences is proportional to the number

of active UEs rather than that of BS antennas. The training

length is fundamentally limited by the channel coherence time,

which can be short due to UEs of high mobility. It has been

shown, nevertheless, in [5] that the effect of using short training

sequences diminishes in the LAS regime. Specifically, among

the poorly estimated channels, the asymptotic orthogonality can

still hold.

However, a major problem with TDD Massive MIMO is that

inevitably the same pilot sequences will be reused in multiple

cells. The channels to UEs in different cells who share the

same pilot sequence will be collectively learned by BSs. In

other words, the desired channel learned by a BS is contaminated by undesired channels. Once this contaminated CSI is

utilized for transmitting or receiving signals, intercell interference occurs immediately which limits the achievable SINR.

This phenomenon, known as pilot contamination, can not be

circumvented simply by increasing the BS antenna size [6], [7].

Several attempts have been made to tackle the problem of

pilot contamination. In [8], a sophisticated precoding method

is proposed to minimize intercell interference due to pilot contamination. A more direct approach for pilot decontamination,

harnessing second-order channel statistics, can be found in [9].

A recent study [10] claims that pilot contamination is due to inappropriate linear channel estimation. Hence, by using the blind

subspace-based estimation, unpolluted CSI is obtainable. In the

context of Massive-MIMO-aided OFDM, pilot contamination

can be eliminated by a novel combination of downlink training

and scheduled uplink training [11]. However, this method, suffering from similar drawbacks as the blind estimation, requires

a substantial training period.

The effect of pilot contamination is usually quantified as

SINR saturation due to intercell interference. A number of

studies have examined this saturation phenomenon and the

corresponding uplink or downlink throughput [6], [12], [13].

The former two provide analysis in the LAS regime with a fixed

number of active UEs, while in [13], it analyzes asymptotic

SINRs with a fixed ratio of the BS antenna size to the activeUE number. All these studies lead to a similar conclusion that

the SINR will saturate with an increasing antenna size, making

system throughput limited. In [14], several possible definitions

of SINR are provided and discussed.

So far, however, there has been little discussion about the

user capacity of TDD Massive MIMO, which is the maximum

**1536-1276 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
**

See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

while the multicell scenario is stressed for pilot-contaminated Massive MIMO. B. Section III addresses the issue of admissible users in the pilot-contaminated system. i. Prior Works The idea of using user capacity as a metric to characterize the system performance has been explored in [15]. i.. NO. i. the capacity depicted in the same work assumes an infinite number of UEs.” originally coined for analyzing CDMA systems [15]. [16]. JUNE 2015 number of UEs whose SINR requirements can be met for a given pilot sequence length. VOL. in the LAS regime. the discussion of uplink/downlink user capacity is confined to CDMA systems. When the signature/pilot sequence length is less than the number of served UEs. In [15]. 3) Not only is the user capacity characterized as an admissible region of SINR requirements. Simulation results will verify the accuracy of this analytical investigation. we characterize the average SINR that is actually achieved for a given number of BS antennas. Specifically. The results highlight the advantage of using the proposed scheme. In contrast. we confine our discussion to the user capacity of single-cell downlink TDD Massive MIMO. using our proposed joint pilot sequence and power allocation can achieve these targets concurrently as the BS antenna size grows without limit. it sheds light on the possible existence of a similar region in the general multicell scenario. is employed here for investigating Massive MIMO from a new angle. 6. in which the role of the signature sequences is analogous to that of pilot sequences in TDD Massive MIMO. This advantage becomes increasingly apparent when UEs have more diverse SINR requirements. With perfect CSI at the BS. If crosscorrelations of pilot sequences are not well designed. A. On the other hand. the prediction of the required antenna size. Meanwhile. This upper bound is further translated into an admissible region in which the SINR requirements can be satisfied by using the proposed joint scheme of pilot sequence design and transmit power allocation.e. this paper makes the first attempt to study the impact on a group of UEs and to articulate conditions under which their SINR targets can be met simultaneously in the LAS regime. However. the conventional pilot sequence design is taken into account in one of the non-capacityachieving schemes. The analysis in the non-LAS regime is . the number of the required antennas depends on the cross-correlations of pilot sequences and the SINR requirements. Basically. It means that this region exists without relying on advanced estimation or precoding methods. Briefly. This terminology “user capacity. Meanwhile. the optimal number of UEs for maximizing the ergodic sum throughput can be identified. Though this region is identified under the single-cell assumption. we specify the system model including the uplink training method and the downlink transmit beamforming scheme.e. it is described as overloading in CDMA while as pilot contamination in Massive MIMO [17]–[19]. characterizing the user capacity. In this paper. This result is significant in the sense that within the specified region we have no worries about pilot contamination by using the proposed allocation scheme. The main purpose of this study is to develop an understanding of the mentioned effects on downlink user capacity of Massive MIMO. Moreover. the previous investigation of downlink user capacity of multiuser MIMO in [16] is conducted without considering the effects of pilot contamination. Our discussion will show that the user capacity can be characterized by a specific region within which there exists a capacity-achieving pilot sequence and power allocation scheme such that the SINR requirements are satisfied and the user data can be energy-efficiently transmitted. the study of how many BS antennas are needed to approach a significant percentage of SINR performance limits is presented. which provides a useful guide to practical BS antenna deployment. we consider a more general set of pilot sequences whose cross-correlations can range from −1 to 1. the foregoing studies of the user capacity of multiuser MIMO have not examined the effects of pilot contamination with general SINR constraints. the maximum number of admissible UEs increases with the BS antenna size in the downlink..e. In contrast. but we also propose a joint pilot sequence design and transmit power allocation scheme to achieve this capacity. the derivation of this result only involves a simple least-squares (LS) channel estimator and maximum ratio transmission (MRT) precoding. Particularly. Contributions The major contributions of this paper are summarized as follows: 1) Previous research on Massive MIMO has examined the effects of pilot contamination on the received SINRs of individual users. more emphasis is placed on the single-cell scenario for over-loaded CDMA. We show that if the SINR requirements are located in a prescribed admissible region. is fundamentally limited by the length of uplink pilot sequences. In addition. we provide a fresh insight into how this UE will cause the demand for a larger antenna size.. Generally. We find that the size of the prescribed region. the size of this region can be further reduced. Organization and Notations The rest of the paper is as follows. 14. In Section II.3184 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. The detailed comparisons between our proposed capacity-achieving scheme and other non-capacity-achieving schemes have been made. The recent study of user capacity in multicell CDMA has shifted its focus to the effects of hard-handoffs instead of over-loaded channels [20]. the potential of admitting more UEs into the specified Massive MIMO system. in most existing studies of Massive MIMO. the cross-correlations are restricted to be 1 or 0. while it is governed by the pilot sequence length in the uplink. C. where pilot contamination will occur once the number of UEs is greater than the pilot sequence length. 2) We advance this research by evaluating the performance in the non-LAS regime. uplink/downlink user capacity of multiuser MIMO with identical SINR constraints is presented in [21]. Based on it. When there exists a UE with a relatively high SINR requirement.

if i = j.i = STi . The correlation sTi sj = ρij between different training sequences ranges from −1 to 1. in which the BS acquires CSI via uplink training. card(·): cardinality. Consider that channel estimates are available at the BS and only channel statistic E[hH i ti ] is known by the ith UE. if i = j. Following this. Uplink Training During the uplink training phase. The received signal at the ith UE is ⎛ ⎞ K 1/2 H ⎝ ri = β i h i tj x j ⎠ + w i .i y. (1) i=1 Utilizing estimated CSI at the BS. 1. · p : p-norm. 1. the BS acquires downlink CSI through uplink pilot training. where ρij may not be 0 for some j = i. and wi 2 is the zero-mean noise with variance σw . (·)T : transpose. linearly precoded signals are formed and simultaneously transmitted to UEs.: DOWNLINK USER CAPACITY OF MASSIVE MIMO UNDER PILOT CONTAMINATION 3185 can be achieved. can be expressed as y(τ M ×1) = K 1/2 ps. Numerical results are presented in Section IV. which is followed by the concluding remarks in Section V. h = hi + K ρij hj + STi z. A single-cell TDD Massive MIMO network. as shown in Fig. S YSTEM M ODEL Consider a single-cell Massive MIMO system. the received signal (3) is recast as 1/2 . provided in the same section. ·): complex normal distribution. IN : N × N identity matrix. ⊗: Kronecker product. : vector inequalities. The acquired CSI will then be utilized to form MRT precoding vectors for downlink spatial multiplexing. (·)H : Hermitian transpose. the ith UE transmits its own pilot sequence si ∈ Rτ ×1 with si 2 = 1. ◦: Hadamard product. M →∞ M Such asymptotic orthogonality among MIMO channels has been experimentally verified in realistic propagation environments [4]. The fact. synchronously received at the BS. A. With TDD operation. . (2) j =i where PLS. Let us devise the ˆ i / M ( K ρ2 + σ 2 ). Null(·): null space.i Si 1/2 βi hi + z. Without exploiting statistical knowledge of hi . can be easily proved by making use of the following results [12] 1 H 0. Z: integers. B.SHEN et al. CN (·. hi hj = (4) lim 1. where a BS equipped with M antennas serves K singleantenna UEs. tr(·): trace. Downlink Transmission Fig. Notations: R: real number. diag(· · ·): diagonal matrix. The pilot data over the block-fading channel. where the ith precoding vector ti = h j=1 z ij denominator ensures that ti has unit 2-norm as M2 grows 1 ˆH ˆ 2 hi hi = K without limit. the LS method is adopted to provide the channel estimate ˆ i = PLS. 0: zero vector. E[·]: expectation. (3) j=1 where ti is a linear precoding vector. limM →∞ M j=1 ρij + σz . This estimate indicates how the desire channel information is polluted by undesired channels in the pilot-contaminated regime (K > τ ). II. xi denotes uncorrelated zero-mean data symbols with power E[xH i xi ] = Pi .

ri = β i E h H i ti x i part1 1/2 + βi .

. where .d. As shown in [8.i βi = 1. and hi (1 ≤ i ≤ K) are independent and identically distributed (i. IM ). Theorem 1]. [8].i and βi denote respectively the pilot power and the channel power gain. i.M ). H hH i ti − E h i ti ⎛ xi + 1/2 βi hH i ⎝ K j =i ⎞ tj x j ⎠ + w i . We further assume that certain uplink power control is enabled to make the received signals from UEs be of equal power. z is the additive white Gaussian noise distributed as CN (0. Si is the τ M × M matrix given by Si = si ⊗ IM . σz2 Iτ M ).) small-scale fading channels from UEs to the BS with distribution CN (0. This channel model is commonly assumed in Massive MIMO systems [6].e. ps. the above expansion facilitates defining an ergodic achievable rate log2 (1 + SINRi. and thus a fair estimation performance (5) where the second part can be regarded as the effective noise which is uncorrelated with the first part. part 2 where ps.i. [12].

H 2 E h i ti β i Pi .M = 2 . SINRi.

H K 2 Var hi ti βi Pi + j =i E hH i tj β i P j + σ w (6) .

the case under consideration. i. A group of UEs is said to be admissible in the specified TDD Massive MIMO system if there exists a feasible pilot sequence matrix S and a power allocation vector p = [P1 . In this examination. Lemma 1: The SINRi. A pilot sequence matrix S is claimed to be feasible if S ∈ S = {[s1 . this group is less likely to be admissible as (9) may not hold with a given pilot length. once the UE number is less than or equal to the upper bound. we wonder if there exists a set of S ∈ S and p 0. the following results will show the effects of pilot contamination on the user capacity. we obtain the limit form of the effective SINR SINRi. A. τ K ≤ τ (MSE + 1). · · · .∞ = tr D−1 ST SDST S − K. the symbol ∞ denotes that the antenna size goes to infinity so that the effect of the noise wi can be ignored. The following section will show that the answer to this question is positive. 14. can not be simultaneously eliminated for every UE as non-orthogonal pilot sequences have to be used.M = βi K j =i ρ2ji Pj + β i Pi K 1 j=1 M α j Pj + 1 2 M αi σ w . · · · . x majorizes y if n k=1 x[k] ≥ n k=1 y[k] . · · · . (8) where D = diag(P1 . A lower bound on the sum of MSEi. s2 . Therefore. On the other hand.3186 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. can be jointly satisfied. (9) is always true for K ≤ τ . For instance. D OWNLINK U SER C APACITY C HARACTERIZATION The asymptotic SINR expression. Proposition 3: If K γi ≤ τ.∞ = K Pi j =i = ρ2ji Pj ≥ (10) where (31) and (32) are applied in the second inequality. y ∈ RN . the achievability issue.∞ is given by K MSEi. 6. Proposition 2 provides an upper bound for the user capacity. . referred to as a valid combination of pilot sequence design and transmit power allocation.∞ seen by the ith UE is equivalent to 1/SINRi. Before going on to the second part. (9) 1+ K≤ τ γi i=1 Proof: Please refer to Appendix B. N. a definition and a lemma to be utilized later are provided. if a group of UEs has high SINR requirements. The following lemma is a consequence of this definition. In the LAS regime (M → ∞).∞ /K represents the average of the normalized mean-square errors. can exist. Pi T . The following proposition will specify a condition under which this pair. for n = 1. How the number of admissible UEs is related to the pilot length and SINR requirements will be presented in the proposition below. fulfilling the SINR requirements. NO. Proof: The Cauchy-Schwarz inequality gives K 1 K2 . (12) 1 + γi i=1 1/2 and there exists a valid then K ≤ {τ [ K i=1 (1 + 1/γi )]} combination of pilot sequence design and transmit power allocation. describing how the SINR varies with the BS antenna size. PK ).. i=1 (7) Proof: Please refer to Appendix A. another explanation can be offered.M defined in (6) can be expressed as SIN Ri.e. identified in the previous section. Moreover. K j =i ji Pj . In other words. in the pilot-contaminated regime. VOL. Capacity-Achieving Pilot Sequence Design and Transmit Power Allocation Validating the converse of Proposition 2 requires to identify the pair of S and p with which we have SINRi. This bound can be immediately translated into a bound on K.∞ ≥ γi for 1 ≤ i ≤ K.e. sK ]. and can be viewed as the performance limits due to the antenna size growing limitlessly. si ∈ Rτ ×1 |si 2 = 1}. JUNE 2015 2 H H in which Var[hH i ti ] = E|(hi ti − E[hi ti ])| . tr si SDST si − Pi K2 − K.. then K 1/2 1 . K > τ will be implicitly assumed as the pilot-contaminated regime is of interest.∞ . III. SINRi. As the normalized mean-square error MSEi. i.∞ ≥ γi for 1 ≤ i ≤ K. Definition 4: Given x. τ which is equivalent to (9) and proves the first part of the statement. . The next question to answer is whether K UEs are admissible if the inequality (9) is satisfied. · · · . Proposition 2: If K UEs are admissible in the TDD Massive MIMO system. 1+ ≥ K γi γi i=1 i=1 ≥ 1+γi K2 . the inter 2 ρ ference part. we reach the conclusion that the intrinsic bound on K is the result of inevitable mean-square errors caused by pilot contamination. will be further employed to examine the user capacity in the LAS regime. PK ]T 0 such that the SINR requirements. It is clear from this proposition that the number of admissible UEs is fundamentally limited once the length of pilot sequences and the SINR requirements are given. S = [s1 . s2 . Hence. · · · . sK ]. This SINR expression suggests that the downlink transmission operates in the interference-limited regime because of using a large number of antennas. which is not (11) where MSE = K i=1 MSEi.

VK×τ V The approach to constructing Q as well as H is provided in [23. RUC = {γ1∼K = γ ∈ R+ |Kγ/(1 + γ) ≤ τ }. which means that all SINR requirements K are met. These sequences also find applications in wavelet expansions and Grassmannian frames [25].SHEN et al. Δ (13) where Σ = diag(λ1 . Making use of (14). and the orthogonal matrix Q can be presented as ˜ K×(K−τ ) . By exploiting the previous result. However.e. 1) The valid combination of pilot sequence design and power allocation is also referred to as the capacityachieving scheme or alternatively the GWBE scheme. Both GWBE and WBE sequences have been extensively applied for constructing optimal CDMA signature sequences. The valid set of S and p used in Proposition 3 is summarized here.∞ = = Pi . 0 · · · . N Given x. Λ = diag(λ1 . The transmit power allocated to UEi is given by γi /(1 + γˆi ) where γˆi ≥ γi and K γi /(1 + γˆi )] = τ . i. this characterization will be utilized to evaluate different joint pilot sequence design and power allocation schemes. 0]T majorizes p and τi=1 λi = K i=1 Pi . (17) γ Unlike (9). = γi .2]: jorizes y and k=1 x[k] = k=1 y[k] . Given that the 1 × K vector of eigenvalues e= [λ1 . λτ . we intend to characterize the user capacity as the admissible region RUC = {γ1∼K ∈ R+ | K i=1 [γi /(1 + γi )] ≤ τ }. In the next section. · · · . · · · . the pilot sequences in use are called Welch bound equality (WBE) sequences. 0 · · · . Now we turn to the case of i=1 [γi /(1 + γi )] < τ . K UEs are admissible in the TDD Massive MIMO system if and only if 1 K ≤ 1+ τ. the right-hand side of (17) does not depend on K. different combinations of S and p. Later on. the proposed pilot sequence design and transmit power allocation can be applied to meet the target SINRs using the minimum pilot length. there exists a real symmetric matrix H = QΛQT . . It means that once the SINR requirements are located within RUC . Moreover. Pi = cˆ i=1 [ˆ In the case of identical SINR requirements. we can find a corresponding valid set of S ∈ S and p 0. Define S = Σ1/2 VT D−1/2 . IV-A]. λτ . Overall. tr sTi Σsi − Pi γi c 1+γ i γi c − c 1+γ i . S ∈ S is true as the diagonal entries of ST S = D−1/2 HD−1/2 are equal to 1. it will be shown that other non-capacityachieving schemes can not guarantee this.B. ∀i = 1. First consider the case of K i=1 [γi /(1 + γi )] = τ . λτ ). τ = c. · · · . we have SINRi. for 1 ≤ k ≤ K} . the pilot sequence assigned to this UE should be orthogonal to others so that it will receive less interference. Corollary 6: Given the identical SINR requirement γ. When a UE has a high SINR requirement. Then. the corresponding K UEs are admissible. K ≤ {τ [ K i=1 (1 + 1/ˆ holds. this constraint indicates that the desired pilot length is lower bounded for the given K and γi . So the system can roughly admit τ UEs with high SINR requirements. In the general case. satisfying SDST = cIτ for some c ∈ R+ [24]. a comparative evaluation of several sequence-construction algorithms is presented. 1 + γi (16) It can be verified that e majorizes p since for 1 ≤ i ≤ τ . 0). this region maintains the same structure. · · · .e. for some c > 0. In [27]. According to the present analytical results. In addition. Next we will check if the SINR requirements are satisfied by using such pilot sequence design and transmit power allocation. where the vector of diagonal entries of H is equal to p. An explanation of the constraint. (15) and Pi = c γi . At the same time. only τ such assignments are allowed in the system. Sec. where the second equality is due to the case under consideration.: DOWNLINK USER CAPACITY OF MASSIVE MIMO UNDER PILOT CONTAMINATION where x[k] and y[k] are respectively the elements of x and y in decreasing order. As f (x) = x/(1 + x) is monotonically increasing for x > 0. (9) and (12) do not provide upper bounds of this kind. if x ma NTheorem 9. and thus it provides an explicit upper bound of admissible UEs. this is true only for identical SINR requirements. following Lemma 5. then there exists a real symmetric matrix H with diagonal elements y[k] and eigenvalues x[k] . A corollary which follows from Propositions 2 and 3 is provided below. The matrix S results from the generalized Welch bound equality (GWBE) sequences. y ∈ R . > max {Pk . (14) Let us specify that K λ1 = · · · = λτ = i=1 τ Pi . N Lemma 5 [22. there γi /(1 + exists a set {ˆ γi ≥ γi for 1 ≤ i ≤ K} such that K i=1 [ˆ 3187 γi )]}1/2 γˆi )] = τ . the following remarks can be made. K. When it comes to identical SINR requirements. γi c K i=1 1+γi λi = . i. From another viewpoint. K i=1 [γi /(1 + γi )] ≤ τ . [26]. is presented as follows. we have SDST = Σ. To have a consistent interpretation of the user capacity in the general case... It means that any K UEs having the SINR requirements within RUC can be admitted by using this scheme.

among these three schemes will be provided in Section IV. More comparisons. The transmit power Pi allocated to the ith UE is given by cγi /(1 + γi ) for some c > 0. 3) Generally. (19). The power allocation is the same as the one in the WBE scheme. VOL. Each pilot sequence si is used by a collection Ei of UEs. r ∈ Z and 0 ≤ r < τ . it follows that the FOS scheme imposes more constraints.i ≥ 1−ν . there might exist some superior power allocation scheme other than the proposed one. when GWBE sequences are employed. the region specified by (9) and K i=1 [γi /(1 + γi )] > τ has not been characterized.e. This section moves on to examine the performance in the non-LAS regime. Therefore. With the aid of Lemma 1. However. C. However. we are able to answer the question “How many BS antennas are required to achieve a significant fraction of the ultimate SINRi. the FOS scheme potentially can admit fewer UEs. 4) Following the previous remark.i ≥ (22) (K − τ )βi (Ptotal − Pi ) 1−ν ν . 2) The Finite Orthogonal Sequence (FOS) Scheme: Given a pilot sequence length τ . which implies potentially fewer admissible UEs as the allowable combinations of γ1 ∼ γK become much less. in which the existence of a valid set of S and p is unknown. For example. 6.3188 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. making this uncharacterized region empty. The following lemmas will show the potential reduction of the user capacity when the WBE and FOS schemes are applied to the case of general SINR constraints. when all the SINR requirements are the same. and Ei ∩ Ej = ∅ for i = j. (19) 1 + γk k∈Ei and K γi ≤ τ. we can not remove the possibility that there exists a better scheme which leads to an admissible region larger than RUC . Lemma 7: The general SINR requirements γi are satisfied by using the WBE scheme if and only if K γi γmax ≤ min τ. for the GWBE scheme. How Many BS Antennas Are Required? So far this paper has focused on the LAS regime. This pilot design is commonly adopted in TDD Massive MIMO. on the admissibility of UEs. Assume that K = qτ + r where q.. When γmax 1.M /SINRi. Comparison With Non-Capacity-Achieving Schemes The superiority of the proposed capacity-achieving scheme over other existing schemes will be presented in this section. i.∞ ≥ νi . the comparisons made below largely point out the benefit of using the GWBE sequences. From Lemma 8. Let card(Ei ) = q + 1 for 1 ≤ i ≤ r. 1 ≤ i ≤ K}. MW BE. NO. and the FOS scheme. compared with the GWBE scheme. The main difference of these two schemes from the GWBE scheme lies in the pilot sequence design. 1). It is possible to have an alternative scheme such as Pi = cγi . taking account of the admissible region. 1) The WBE Scheme: Pilot sequences in use are the WBE 2 sequences. f or 1 ≤ i ≤ τ. Lemma 8: The general SINR requirements γi are satisfied by using the FOS scheme if and only if γk ≤ 1. and the previous discussion applies provided the BS antenna size goes to infinity. having properties: SST = K τ Iτ . Let us first define two non-capacity-achieving schemes as follows. Proof: Please refer to Appendix C. and ρij = (K − τ )/[(K − 1)τ ] for i = j[28]. the second term in the curly brackets becomes less than or equal to τ as γmax ≥ τ /(K − τ ). 2 βi K j =i ρji Pj ν 2 2 (K − 1) K + τ σz βi Ptotal + σw . 14. the WBE scheme. Hence. Lemma 9: To have SINRi. only τ orthogonal pilot sequences will be repeatedly used in the pilot-contaminated regime. card(Ei ) = q for r + 1 ≤ i ≤ τ .∞ ?” The lemma below presents the required antenna sizes for three joint pilot sequence design and transmit power allocation schemes. κ − (κ − 1) . how to identify the optimal power allocation corresponding to the WBE sequences remains an open question. the converse of Proposition 3 can be shown to be true. In Lemma 7. we need antenna sizes 2 βi K j=1 (αj Pj ) + αi σw (21) MGW BE. (9) turns into K i=1 [γ/(1 + γ)] ≤ τ . (18) turns into K i=1 [γi /(1 + γi )] ≤ 1 + ∀ > 0. 1 + γi i=1 (20) Proof: Similar to the proof of Lemma 7. where νi ∈ (0. JUNE 2015 2) When using the capacity-achieving pilot sequence and power allocation. 1 + γi 1 + γmax i=1 (18) where κ = (K−1)τ (K−τ ) and γmax = max{γi . B. Note that the power allocation in the WBE scheme is adopted for ease of comparison.

2 card(Ek ) + σz2 βi j∈Ek Pj + σw . There are important implications of this result. (23) MF OS. and i ∈ Ek is assumed for the FOS scheme. In addition. the number of the required BS antennas becomes larger whichever scheme is applied. (22) and (23) are basically the same as (21) except that the cross-correlations ρji are replaced by the actual values. With a higher ν.∞ when the transmit power allocated to a UE is .i ≥ 1−ν βi j∈Ek \{i} Pj ν respectively. Proof: (21) results from Lemma 1. the BS antenna size has to be significantly large in order to achieve 100ν% of the limit SINRi. where Ptotal = K i=1 Pi .

Next. WBE. 2) If SINRi. E2 = {UE2 . to whom similar powers are assigned. For ζ < 1. if the ratio Pi /Ptotal approaches 1 in the WBE scheme.: DOWNLINK USER CAPACITY OF MASSIVE MIMO UNDER PILOT CONTAMINATION relatively higher. 3189 Fig.SHEN et al. (26) for the GWBE. for 1 ≤ i ≤ 3 . comparisons between the capacity-achieving and non-capacity-achieving schemes are made. Some supplementary remarks are added below. it is obvious that the GWBE scheme outperforms the other two in terms of achieving higher SINRs for ζ ranging from 0.i ≥ νi SINRi. making overall requirements of 6 UEs more or less diverse. It can be observed that the boundary surface of RGWBE lies well above those of RWBE and RFOS . an extra restriction γ3 = min{γ3 . WBE. Compared to the FOS scheme. Upper boundaries of admissible regions for the GWBE. then using M ≥ max{MGWBE. IV. N UMERICAL R ESULTS In this section.1 to 1. 1 ≤ i ≤ K} should guarantee that all downlink SINR targets are satisfied when the GWBE scheme is applied. we consider a pilot-contaminated Massive MIMO system with K = 6 and τ = 3. a sensible alternative to the computation in Lemma 9 is to evaluate the required number of antennas to achieve a fraction of the ergodic rate log2 (1 + SINRi. For the GWBE scheme. As shown in this figure. then the required anPtotal . 5} is placed as the admissible γ3 can go to infinity. is given by E1 = {UE1 . This conclusion can be drawn from Lemma 9.∞ ≥ γi holds for any index i. the GWBE scheme is more . 3. so more general SINR constraints γ1∼3 can be met in the GWBE scheme. (25) and RFOS = RGWBE ∩ γ1∼3 ∈ R+ | γi ≤ 1. Additionally.∞ ). Note that it is implicitly assumed that E1 = {UE1 . UE4 }. UE4 }. A similar argument also holds for the other two schemes. UEs with similar SINR requirements. Fig. UE6 } for the FOS scheme.MGWBE . we investigate how target SINRs are asymptotically achieved and how many BS antennas are used to attain the desired performance. Achievable SINR γ versus the factor ζ for the GWBE. This implies that RGWBE contains more admissible points than RWBE and RFOS . 2. and FOS schemes given a fixed SINR-requirement pattern. 1) If there is a shortfall in the transmission power budget 2 . WBE. for 1 ≤ i ≤ 3 . then the corresponding MWBE increases without limit. we look into admissible regions of the remaining SINR requirements given by 3 + RGWBE = γ1∼3 ∈ R | [γi /(1 + γi )] ≤ 3/2 . The first set of analyses examine the impact of different schemes on the size of the admissible region of SINR requirements. leading to βi Ptotal σw tenna size to achieve a significant fraction of SINRi.∞ will become very large. The upper boundaries of these regions in the positive orthant are plotted in Fig. E2 = {UE3 . that is {γ1 = γ2 = γ3 = γ. Admissible Region Characterization in the LAS Regime To verify the results in Proposition 3 and in Lemmas 7 and 8. and FOS schemes. the advantage of the GWBE scheme remains significant over the FOS scheme while it becomes negligible over the WBE scheme. The UE grouping. 3 presents the achievable SINR for different schemes when the pattern of the SINR requirements is specified. UE6 }. 2.i . UE2 }. and FOS schemes. (24) i=1 RWBE = RGWBE 3 3 ∩ γ1∼3 ∈ R+ | [γj /(1 + γj )] j=1 j=1 ≤ (7/2 − 4γi /(1+γi )). Hence. γ4 = γ5 = γ6 = ζγ}. UE5 }. By fixing certain SINR requirements {γ4 = γ5 = γ6 = 1}. and Fig. E3 = {UE3 . A. The factor ζ adjusts a subset of the SINR requirements. and E3 = {UE5 . This observation suggests that if the BS antenna size is limited. should be clustered together and be served cluster by cluster. there is a clear trend of decreasing the achievable SINR with increasing ζ. For example. assumed for the FOS scheme. 3) Knowing the achievable rates that are guaranteed is intrinsically important.

92 for the WBE scheme and SINR1. NO. The noise 2 . This SINR set is located in the admissible region for the GWBE scheme. 1/2.1 or 10.M curves of different UEs for the GWBE scheme and WBE scheme. VOL. This result means that as the set γ1 ∼ γ6 falls outside the admissible region for the WBE scheme.e. This linear relationship directly The results discussed in Section III-C will be numerically evaluated in this section. the GWBE scheme. 1/3}. γ(2l+1)∼3l = 3}. Also shown in the same figure. 5. at the BS and UEs.M for a given number of BS antennas when using different joint pilot sequence and power allocation schemes.M reach their own target SINRs respectively for the WBE scheme. We can see that the curves of SINR3. Performance Comparison in the Non-LAS Regime Fig. only partial SINR requirements can be satisfied by this . Achievable SINR versus the number K of UEs for the GWBE. that is {γ1∼l = 1/3. can achieve the target SINR with a finite number of BS antennas. but does so for K ≥ 7. The system of interest still operates in the pilot-contaminated regime with K = 6 > τ = 3. The transmit power allocated to the ith UE is Pi = cγi /(1 + γi ) for the WBE γi /(1 + γˆi ) for the GWBE scheme and FOS scheme and Pi = cˆ scheme. i. the grouping among UEs for any given K is assumed to be optimal in the sense of maximizing the achievable SINR.∞ = 1. 2/3. how many UEs are admissible for a given pilot length is depicted in Fig. 4. favorable when we have diverse SINR requirements. and all channel power powers. Three UE sets for the FOS scheme are the same as those specified in the previous section. however. demonstrates how the user capacity is limited by the pilot length. σz2 and σw gains βi are set to 1.19 for the FOS scheme are less than γ1 . It can be seen that the simulated (dashed.∞ = 1. and FOS schemes given a fixed SINR-requirement pattern. γ(l+1)∼2l = 1. 5. It is because that both SINR1. 4 plots the achievable SINR versus the number of UEs when fixing τ = 3. and FOS schemes. Our proposed GWBE scheme. WBE. substantially outperforms the other two schemes in terms of admitting more UEs. the GWBE scheme. 6 presents the SINRi. 2/3. 2. It can be observed that the number of admissible UEs scales almost linearly with the pilot length whatever scheme is adopted.3190 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. By specifying the SINR-requirement pattern of K = 3l UEs. Number of admissible UEs versus pilot sequence length for the GWBE. Interestingly.M of UE1 achieved by different schemes for a given antenna size M . where c = 10 and γˆ1∼6 = {3. JUNE 2015 Fig. 7 shows the SINRi. This highlights that the GWBE scheme exhibits a consistent superiority over the FOS scheme compared with the WBE scheme. making pilot contamination more serious. 2/9}. 3/8. attains relatively higher SINRs compared with the WBE and FOS schemes. the WBE scheme does not always outperform the FOS scheme for K < 7. 3/2. γ4 = · · · = γK = γ/2}. Fig. 6. 1. given a fixed SINR-requirement pattern.∞ = 3 greater than γ1 . Achieved SINR1. two non-capacityachieving schemes exhibit comparable user capacities especially with short pilot lengths. In contrast. given by {5/2. 14. Fig. WBE.M curves of the WBE scheme and FOS scheme can not approach the target SINR by increasing the antenna size. 1. dash-dot) curves due to the average over 100 000 channel realizations fit well with the analytical (circlemarker) values obtained from (7). without doubt. For the FOS scheme. 6. It can be observed that increasing K.M and SINR6. Fig. To explore the effects of having different numbers of UEs. We first consider a set of SINR requirements γ1 ∼ γ6 . Another observation is that the SINR1. B. having SINR1. Fig. the result in Lemma 1 is verified. In addition. when ζ approaches 0. dotted. while outside those for the WBE scheme and FOS scheme. Hence. that is {γ1 = γ2 = γ3 = γ. will decrease achievable SINRs for all three schemes.

has been proposed and compared with the non-capacity-achieving WBE and FOS schemes.M is greater than the target SINR. Next. which can achieve the identified user capacity and satisfy the SINR requirements.6% of the limit SINR3. the corresponding capacity-achieving pilot sequence design and transmit power allocation. the simulated SINRi. we need to achieve 66. the capacity-achieving GWBE scheme. WBE. Then. The necessary condition for admitting a group of UEs with general SINR requirements has been provided. From then onwards.and inter-cell pilot contamination. 7.∞ .∞ is achieved. For UEs with lower SINR constraints. C ONCLUSION This paper has investigated the user capacity of downlink TDD Massive MIMO systems in the pilot-contaminated regime. H ρjm ρjn hH m hi hi hn + m=1 n=1 K K H H ρjm hH m hi hi Sj z m=1 H H H ρjn zH Sj hi hH i hn + z Sj hi hi Sj z.M and target γi curves intersect at the points around M = 196 for UE1 and M = 38 for UE3 . If 83. The SINR limits for this scheme are given by SINRi. The current investigation has only examined the singlecell scenario. More study is needed to better understand the user capacity in the multicell scenario. the GWBE scheme is taken as an example for demonstrating the accuracy of results in Lemma 9. where the simulated SINRi. 7. and FOS schemes. the corresponding SINR requirements γ1 will be met.: DOWNLINK USER CAPACITY OF MASSIVE MIMO UNDER PILOT CONTAMINATION 3191 A PPENDIX A P ROOF OF L EMMA 1 First. resulting in MGWBE ≥ 38 for UE3 . + n=1 Fig. Meanwhile.∞ = γˆi > γi for 1 ≤ i ≤ 6. scheme. " # H ˆ 2 E hi hj K K = . we have H ˆ 2 h i h j K = × = !H H H ρjm hH i hm + hi Sj z m=1 K ! ρjn hH i hn n=1 K K + H hH i Sj z . Based on (22). it can be observed that the difference between the results due to two schemes becomes less significant.SHEN et al. such as UE3 and UE6 .3% of the limit SINR1. and an efficient method for performing intercell resource allocation. Future research might explore the capacity region under intra. This analytical inference is verified in Fig. It has shown an intrinsic capacity upper bound due to the limited length of pilot sequences. V. the required antennas size is expected to be MGWBE ≥ 186 for UE1 . The results of this study indicate that the capacity-achieving GWBE scheme is superior in terms of enhancing higher user capacity. Similarly. Average received SINRs at UEs versus the number of BS antennas for the GWBE.

H .

H H H ρjm ρjn E hH m hi hi hn + E z Sj hi hi Sj z . m=1 n=1 K .

.

H H = ρ2ji E hH + h h h ρ2jm E hH i i i i m hi hi hm m=n =i + K .

ρjm ρjn E hH m hi hi hn +E tr Sj hi hi Sj zz . H H H H .

Rhi and Rz respectively denote H the covariance matrices of hi and z. K 2 2 = ρji (M + M ) + M ρ2jm + M σz2 . By definition. + 2 σw 2 = βi M αi M αi Pi + K M M αi βi Pi M j =i M αi . i 2 M ρ2ji + αj Pj + α βi + M σw β i Pi = K αj αi 2 . and Var[hi hi ] = M αi . m =n = ρ2ji (M 2 + M ) + K ρ2jm [tr (Rhi Rhm )] m =i + tr Rhi SH j Rz Sj . K 2 βi ρ P + j =i ji j j=1 M Pj + M σw = αi M Pi K 1 j =i M Hˆ ˆ where E[hH i hi ] = M . SINRi. m =i (27) = M 2 ρ2ji + M αj .M ˆ ( E [h H i hi ]) = βi 2 βi Pi M αi ˆ i −E[hH h ˆ i ])| Pi E|(hH h i i M αi 2 + K ˆ j | Pj E |h H h i j =i M αj 2 . and E[hH i hi hi hi ] = 2 H (M + M ) because of hi hi being gamma-distributed Γ(M. 2 2 where αj = K m=1 ρjm + σz . (28) . 2 M ρ2ji + αj Pj + σw β i Pi . 1).

Gao. D. 2013. 31. no. we can expand the trace in (29) and obtain its lower bound as below tr D−1/2 Gs DGs D−1/2 K K Pi Pj + =K+ ρ2ij . no. and S. Yin. pp. no. Istanbul. ··· ··· ··· . “Noncooperative cellular wireless with unlimited numbers of base station antennas. Wu and W. USA. 1436–1449. V. no. 2640–2651. “Massive MIMO in the UL/DL of cellular Networks: How many antennas do we need?” IEEE J.. 1999. 52. IEEE WCNC. ! . 2014. 2014. Commun. pp. vol. Marzetta. ρ1K ρ2K ρ3K ··· ⎤ ρ1K ρ2K ⎥ ⎥ ρ3K ⎥ . ⎥ ⎦ . Feb.. [7] Y. = tr (Gs Gs ) . 242–247. Q.. and τi=1 di = K. USA. we have tr(Gs Gs ) = tr UD2G UT . 1 + γj 1 + γj j=1 where ≤ Δ ρ12 1 ρ23 .. pp. Jose. 2010. Zhang et al. 30. Aug. 2013. Sel. and D. [12] H. dK ) with d1 ∼ dτ > 0. “Downlink performance analysis with enhanced multiplexing gain in multi-cell large-scale MIMO under pilot contamination. no. P i=1 i T −1 T = tr(D S SDS S). Sel. “Scaling up MIMO: Opportunities and challenges with very large arrays. Apr. 31. 8.. [9] H. γj ≤ κ − (κ − 1) 1 + γj K K (29) i=1 j=1 K γj γj =K . F. Marzetta. USA. CA.e.. IEEE VTC. Areas Commun. 61. Piscataway. 2006. 160–171.. pp. Mag. Xu. S. 2013. 45. P P j i i=1 j>i=1 K K γi 1 + γi Let us sum up (34) over i = 1. no. 1–5. Marzetta. IEEE Online Conf. “Massive MIMO for next generation wireless systems. ≤ 1 + γj K +κ−1 (35) The desired result follows from (34) and (35).” IEEE Signal Process. R EFERENCES [1] F. 2. 11. .. ten Brink. vol. Topics Signal Process. Rusek. Jan. Debbah. “Optimal stochastic coordinated beamforming for wireless cooperative networks with CSI uncertainty. Marzetta. = tr D−1/2 Gs DGs D−1/2 . NJ. pp. user capacity of synchronous CDMA systems with linear MMSE multiuser receivers. “Pilot contamination and precoding in multi-cell TDD systems. which gives (31) 1 ≥ τ γi 1 + γi i=1 (30) i=1 j>i=1 i=1 κ − (κ − 1) = Kκ − (κ − 1) 2ρ2ij .” in Proc.” IEEE Trans. 2014. . and F. (33) the desired inequality follows. Tufvesson.3192 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. Turkey. and M. 264–273. pp. Letaief. 8. (34) ≥ γi . 2013. CA. “Optimal sequences. E. A PPENDIX C P ROOF OF L EMMA 7 The SINR requirement K Pi 2 j =i ρji Pj Pi 1 j=1 κ Pj = K = c κ − κ1 Pi γi c 1+γ i γj K j=1 1+γj .. [13] J. . 10. Wireless Commun. 4. [10] R. 2. “Energy and spectral efficiency of very large multiuser MIMO systems. and T.” in Proc. vol. no. O.” IEEE J. Then. 1968– 1983.” in Proc. . Hoydis.” IEEE Trans.. · · · . Sel. Larsson. Theory. “How much training is required for multiuser MIMO. pp. K. Shi. pp. 186–195. San Francisco. “Linear pre-coding performance in measured very-large MIMO channels. vol. Oct. no. “Blind pilot decontamination. pp. Mag. and K. 2015. Topics Signal Process. τ !2 di = i=1 K2 . 960–973. Nov. pp. “Total energy efficiency of cellular large scale antenna system multiple access mobile networks. pp. Tufvesson. no. pp. 8. pp. 2011. J.. Apr. The Gram matrix Gs has an eigendecomposition UDG UT .” IEEE Trans. no. [14] X. “Pilot contamination elimination for large-scale multipleantenna aided OFDM systems. Vehkapera. Edfors. vol. F.. Sep. vol. ρ13 ρ23 1 . [5] T. i.. and M. Sep. vol. Anantharam. vol. we have K 1 + SINRi.. [11] J. Feb. [2] H. · · · . 63. vol. Pacific Grove. Oct. T. 5. [4] X. pp. 359–363. 27–32. . GreenCom. Mueller.. JUNE 2015 A PPENDIX B P ROOF OF P ROPOSITION 2 can be recast as K Making use of (8).. 6.∞ SINRi. 1. Filippou. j=1 γj Kκ = τ.∞ i=1 = j=1 K 1 T tr si SDST si . Inf. Vishwanath. A. Areas Commun. Cottatellucci. 14. M. Zhang. Gesbert.. 4. Marzetta. Rusek et al. “A coordinated approach to channel estimation in large-scale multiple-antenna systems. [3] E. Ngo. 2014. . L.∞ i=1 SINRi. VOL. 40th ACSSC. and Y. 6. 2. [15] P. 3590–3600. vol. ⎡ 1 ⎢ ρ12 ⎢ ρ =⎢ ⎢ 13 ⎣ . Viswanath.” IEEE Trans. τ = d2i . vol. where U is a unitary matrix and DG = diag(d1 . [6] T. 2011. [8] J. Wireless Commun. 9. pp. and T.” IEEE J. 773–786. 2013.” IEEE J. Larsson. Ashikhmin. Signal Process. Yang and T. 1 ≥K+ K where the inequality is due to (Pi /Pj + Pj /Pi ) ≥ 2. no. Marzetta. Feb. τ (32) As K 1 + SINRi. power control.∞ ≤ K 1 + γi γi i=1 . Gs = ST S.. Oct. Liu. NO. − c γi κ 1+γi . B.” in Proc.” IEEE Trans. Edfors. 759–772. Feb. dτ +1 ∼ dK = 0. Oct. Sel. 5. Tse. 40–60.” IEEE Commun. K # γi 1 + γi . O. K " i=1 Also.

2007. “Signature sequence and training design for overloaded CDMA systems. Anal. Lee.K. vol. the 2011 IEEE Wireless Communications Technical Committee Recognition Award. no. and S. pp. cognitive radio. Varanasi. Jr.S. 257–275. Jul. vol. 1728–1742. Kim. Theory. from Purdue University in 1984.” IEEE Trans.. and S. 11. Ulukus and R. W. Marshall. no. Lett. From 1990 to 1993. Heath. Heath. “Finite-step algorithms for constructing optimal CDMA signature sequences. 2006. Theory. Commun. [19] J. Kishore. Anantharam. 2307–2309. pp. degree in electronic engineering from the University of Science and Technology of China. 2003. 2007. [17] A. 6. 1986 and 1990. vol. 2009. Inf. and K.-K. vol. Poor. TTM’11 in Hong Kong. He has been involved in organizing a number of major international conferences. I. Dec. spectrum sensing. Greenstein. vol. the 2011 IEEE Communications Society Harold Sobol Award. with expertise in wireless communications and networks. C. 1989–1998. and R. 45. pp. Dr.S. he taught courses in communications and electronics. 5. 5. He is a co-recipient of IEEE PIMRC 2014 Best Paper Award.” IEEE Trans. Jung. USA: SIAM. South Korea. and the PIMRC 2014 Best Paper Award. ICC’10 in Cape Town. pp. 2004. vol. Australia. 49. 1984–1991. 2010). compressed sensing. Letaief (S’85–M’86–SM’97–F’03) received the B. no. pp. Daubechies. “Grassmannian frames with applications to coding and communication. 49. and IEEE ComSoc VicePresident for Conferences. and signal processing. Kapur and M. Comput. and is a member of the IEEE Fellow Committee. Sep. pp. Guess.” Appl. the 2010 Purdue University Outstanding Electrical and Computer Engineer Award. ICC’08 in Beijing. no. vol. 50. May 2003. Aug. A. and the Michael Gale Medal for Distinguished Teaching (Highest University-wide Teaching Award at HKUST). He is also the recipient of many other distinguished awards including the 2007 IEEE Communications Society Publications Exemplary Award. 4. Waldron. He has served as an elected member of the IEEE Communications Society (ComSoc) Board of Governors. 14. In these areas. Wireless Commun. Hong Kong University of Science and Technology (HKUST). Jun. 11. NY. G. Hong Kong University of Science and Technology (HKUST)./Jul. 1999. “Uplink user capacity in a multicell CDMA system with hotspot microcells. Yates. degrees.” IEEE Trans. “Optimal sequences and sum capacity of synchronous CDMA systems. 2916–2921. and 11 IEEE Best Paper Awards. no. in 2009. Ponnaluri and T.. in 2013. 2010.” IEEE Commun. Theory. [22] A. he joined the Department of Electronic and Computer Engineering. L. His research interests include wireless communications and networking. W. Jul. He co-authored the book Fundamentals of LTE (Prentice-Hall. and random matrix theory. Arnold.Phil. the M. he was a faculty member at the University of Melbourne. all in electrical engineering. “Pilot contamination problem in multi-cell TDD systems. Wireless Commun.. [25] I. “Iterative construction of optimum signature sequence sets in synchronous CDMA systems. These include WCNC’07 in Hong Kong. Jun. 6. Theory. [18] S. Inf. vol. He has served on the editorial board of other prestigious journals including the IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS —W IRELESS S ERIES (as Editor-in-Chief). 15. Professor Letaief has been a dedicated teacher committed to excellence in teaching and scholarship. Harmon. Letaief serves as a consultant for different organizations and is the founding Editor-in-Chief of the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS. Ashikhmin. Juei-Chin Shen (S’10–M’14) received the Ph. 7.” IEEE Trans. and the M. I. Khaled B. the HKUST Engineering Teaching Excellence Award. PA. respectively. massive MIMO systems. 2003. pp. pp. Theory. Inf. Choi. Inf. [28] S. 362–369. Dhillon. New York. Apr. green communications. 1333–1342. no. 1337–1345. Netw.. 2013. T. In the same year. and S. Vishwanath. 4. 3193 Jun Zhang (S’06–M’10) received the B. Viswanath and V. Seoul. degree (with distinction). Dr. and the Ph.. he has over 500 journal and conference papers and has given invited keynote talks as well as courses all over the world. . as a Research Associate. no. [23] P. degree in Electrical and Computer Engineering from the University of Texas at Austin. Sep.D.” IEEE Trans. “Feasibility conditions of linear multiuser MIMO systems in the asymptotic regime. pp. Since 1993. U. [27] J. He served as a MAC track co-chair for IEEE WCNC 2011. 2184–2188. Nov. millimeter-wave communications. [21] M. Marzetta. [20] S. He has 13 patents including 11 US patents. 12. USA: Springer-Verlag.: DOWNLINK USER CAPACITY OF MASSIVE MIMO UNDER PILOT CONTAMINATION [16] C. 61. Strohmer and R. Ten Lectures on Wavelets. and Ph. and B. J. IEEE ISIT. Letaief is a Fellow of HKIE. 6. 3. Wong. no. From January 1985 and as a Graduate Instructor at Purdue. Inf. [26] T. Jose. He is currently serving as the IEEE ComSoc Vice-President for Technical Activities as a member of the IEEE Product Services and Publications Board.. degree in communication engineering from University of Manchester. [24] S. 1992. Schwartz. “Generalized Welch bound equality sequences are tight frames.D. in 2004.D. 2001. 979–981. Inequalities: Theory of Majorization and Its Applications. Olkin. Zhong. pp. IEEE ComSoc Treasurer.. He received the 2014 Best Paper Award for the EURASIP Journal on Advances in Signal Processing. vol. He is also recognized by Thomson Reuters as an ISI Highly Cited Researcher. degree in information engineering from the Chinese University of Hong Kong. 47. he has been with the Hong Kong University of Science and Technology (HKUST) where he is currently Chair Professor and the Dean of Engineering. no. His research interests include statistical signal processing. Tropp. pp. He is currently a Research Assistant Professor in the Department of Electronic and Computer Engineering.” J. “Optimal number of users in zeroforcing based multiuser MIMO systems with large number of antennas. He received the Mangoon Teaching Award from Purdue University in 1990.” IEEE Trans. Y. in 2006. “Multiuser detection for overloaded CDMA systems. as an IEEE Distinguished lecturer. H. Philadelphia. no.SHEN et al. the 2009 IEEE Marconi Prize Award in Wireless Communications.” in Proc. and ICCC’12 in Beijing. vol. 9.” IEEE Trans. Zheng.Eng. vol.

- Research Journal Article Critique 2 Romel Espinosa V2Uploaded byRomel Ivan Espinosa
- 25DownlinkMultiUserMIMOcapacityUploaded byrmeneguc
- VLSI Implementation of OFDM Transceiver for 802.11n systemsUploaded byAnonymous 7VPPkWS8O
- RNL2 LTEUploaded byarjain75
- Advanced 3G and 4G Wireless MobileUploaded byApoorva Aakarsh
- 31 ijecsUploaded byRamya R
- Performance Analysis of Rayleigh Fading Channel Over Mimo Wireless Communication SystemUploaded byijteee
- MIMO in LTE Operation and MeasurementUploaded bypetersote
- 1707.00291Uploaded bymalli gadu
- 05 Keysight Agilent Hewlett Packard MIMO in LTE Operation and Measurement - Excerpts 5990-4760ENUploaded byJohn Bofarull Guix
- NSN Smart Scheduler for LTEUploaded bylinhhy
- lteUploaded bySatish Jadav
- Advanced telecommunication systemsUploaded byYoussof
- roze2016.pdfUploaded byhendra
- Wp Beamforming Revb 11 2010 LrUploaded byFahrudin Fajar
- Dm Qm Background Mathematics Lecture 9Uploaded bycharles luis
- EventsUploaded byBalan BaLan
- Thesis-Realistic Multi-Cell Interference Coordination InUploaded byharunyilmaz62
- mwoh-samos07Uploaded byarpitmags_1
- 212817433-3G-Events.pdfUploaded bymehran_4x
- RRU3965&RRU3965d Description 01(20151130)Uploaded byercanaktas
- LTE -1- OverviewUploaded byDreamboy Dreamboy
- TECHNICAL.pdfUploaded byUlianov Sinforoso
- 4G and BeyondUploaded byRico Rindra
- LN4_diagonalizationUploaded byNikhilesh Prabhakar
- Sci Lab QuickiesUploaded bydedyhelong725_388971
- SynopsisUploaded byRohit Mathur
- 3G_EventsUploaded byAnonymous kC8oFmS
- Thuraya XT Lite BrochureUploaded byfendy
- Www Teletopix Org Gsm How 26 and 51 Multiframes in GsmUploaded bydawit.selesh