4592
11, NOVEMBER 2016
Analyzing Uplink SINR and Rate in Massive
MIMO Systems Using Stochastic Geometry
Tianyang Bai, Student Member, IEEE, and Robert W. Heath, Jr., Fellow, IEEE
Abstract— This paper proposes a stochastic geometry frame-
work to analyze the signal-to-noise-and-interference ratio (SINR)
and rate performance in a large-scale uplink massive multiple-
input and multiple-output (MIMO) network. Based on the model,
expressions are derived for spatial average SINR distributions
over user and base station distributions with maximum ratio
combining (MRC) and zero-forcing (ZF) receivers. We show
that, using massive MIMO, the uplink SINR in certain urban
marcocell scenarios is limited by interference. In the interference-
limited regime, the results reveal that for MRC receivers,
a superlinear (polynomial) scaling law between the number of
base station antennas and scheduled users per cell preserves
the uplink signal-to-interference ratio (SIR) distribution, while
a linear scaling applies to ZF receivers. ZF receivers are shown
to outperform MRC receivers in the SIR coverage, and the
performance gap is quantified in terms of the difference in
the number of antennas to achieve the same SIR distribution.
Numerical results verify the analysis. It is found that the optimal
compensation fraction in fractional power control to optimize
rate is generally different for MRC and ZF receivers. Besides,
simulations show that the scaling results derived from the
proposed framework apply to the networks, where base stations
are distributed according to a hexagonal lattice.
Index Terms— Massive MIMO, SINR, achievable rate, perfor-
mance evaluation, stochastic geometry.
I. I NTRODUCTION
M ASSIVE multiple-input and multiple-output (MIMO) is
an approach to increase the area spectrum efficiency in
5G cellular systems [2]–[5]. By deploying large-scale antenna
arrays, base stations can use multi-user MIMO to serve a large
number of users and provide high cell throughput [2]–[5].
In this paper, we focus on the defacto massive MIMO systems
operated below 6 GHz, where pilot-aided channel estimation
is performed in the uplink, and pilots are reused across cells
to reduce the training overhead [2]–[5]. Prior work showed
that when the number of base station antennas grows large,
a high throughput is achieved through simple signal process-
ing, and that the asymptotic performance of massive MIMO
Manuscript received October 8, 2015; revised March 8, 2016 and
May 31, 2016; accepted July 2, 2016. Date of publication July 13, 2016;
date of current version November 15, 2016. This work was supported
by the National Science Foundation through the Division of Computing
and Communication Foundations under Grant Nos. 1218338, 1319556, and
1514257. A conference version of this paper was presented at the 2015 IEEE
Global Conference on Communications [1]. The associate editor coordinating
the review of this paper and approving it for publication was M. Di Renzo.
T. Bai was with The University of Texas at Austin, Austin, TX 78701 USA.
He is now with Qualcomm Flarion Technologies, Inc., Bridgewater, NJ 08807
USA (e-mail: tybai@utexas.edu).
R. W. Heath is with The University of Texas at Austin, Austin, TX 78701
USA (e-mail: rheath@utexas.edu).
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/TCOMM.2016.2591007
0090-6778 © 2016 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.
Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO.
(in the limit of the number of base station antennas) is limited
by pilot contamination [2]. In this paper, we propose to study
the uplink signal-to-noise-and-interference ratio (SINR) and
rate performance in a massive MIMO network with maximum
ratio combining (MRC) and zero-forcing (ZF) receivers, for a
random base station topology.
A. Related Work
The performance with MRC and ZF beamforming, in
terms of SINR, spectrum efficiency, and energy effi-
ciency, was examined though in a simple network topology
in [2] and [6]–[10], where the SINR and rate expressions
were conditioned on specific user locations, or equivalently
the received power for each user. The conclusions drawn from
the conditional expressions, however, need not apply to the
system-level performance taking average over different users’
distributions. For example, the linear scaling between the num-
ber of users and antennas examined in [8] need not maintain
the uplink signal-to-interference ratio (SIR) distribution, as
will be shown in our analysis. This motivates the investiga-
tion of the spatial average performance over different base
station and user distributions in large-scale massive MIMO
networks [11]–[14]. While the spatial average performance
can be simulated using prior results conditioning on fixed
base station and user realizations, an analytical framework
approximating the key features of the system is desirable, as
it helps reveal key dependencies of parameters in the system.
Stochastic geometry provides a powerful tool to analyze
system-level performance in a large-scale network with ran-
domly distributed base stations and users. Assuming a single
antenna at each base station, the spatial average downlink
SINR and rate distributions were derived for a network with
Poisson point process (PPP) distributed base stations, which
were shown a reasonable fit with simulations using real base
station data [15]. The stochastic geometry framework in [15]
was further extended to analyze the performance of MIMO
networks: the downlink SINR and rate of multi-user MIMO
cellular system were analyzed, e.g. in [16]–[19] assuming
perfect channel state information (CSI), and in [20] with
quantized CSI from limited feedback. For uplink analysis,
prior work [21]–[23] showed that the uplink and downlink
SINR follows different distributions, due to the difference
in network topology. In [23], a stochastic geometry uplink
model was proposed to incorporate the pairwise correlations
in the user locations, where the SINR distributions derived
based on the analytical model were shown a good fit with the
simulations. The prior results in [15]–[23], however, do not
directly apply to analyze uplink massive MIMO networks, as
BAI AND HEATH: ANALYZING UPLINK SINR AND RATE IN MASSIVE MIMO SYSTEMS
(i) they did not take account for the effects of pilot contam-
ination, which becomes a limiting factor with large numbers
of antennas [2]; (ii) the analysis in [15]–[20] was intended for
downlink performance, which follows different distributions
from the uplink network; and (iii) the results in [16]–[19]
were intended for MIMO networks with a few antennas, where
the computational complexity for the analytical expressions
grows with the number of antennas, and hinders their direct
applications to the massive MIMO scenarios.
Stochastic geometry was first applied to study the asymp-
totic SINR and rate (in the limit of base station antennas)
in a massive MIMO networks in [13] and [14], where the
asymptotic SINR was shown to be approached with impracti-
cally large number of antennas, e.g. 104 antennas. To analyze
the performance with finite antennas, related work in [24]
applied stochastic geometry to compute the uplink interference
in a massive MIMO network. A linear scaling between the
numbers of base station antennas and scheduled users was
found to maintain the mean interference, which need not
preserve the SIR distribution. Our prior work in [1] derived the
uplink SIR distribution for MRC receivers as a function of the
number of antennas, assuming no power control. In this paper,
we extend the results by incorporating thermal noise into the
analysis, assuming a general fractional power control for MRC
receivers, and investigating the performance of ZF receivers.
B. Contributions
1) Stochastic Geometry Framework for Uplink Massive
MIMO Networks: We propose a stochastic geometry frame-
work to derive the uplink SINR and rate distributions in a
large-scale cellular network using multi-user MIMO. To model
the uplink topology, we propose an exclusion ball model
based on prior work [23], which simplifies the computation.
Channel estimation error due to pilot contamination is also
considered in the system model. The proposed framework
also incorporates the fractional power control by compensating
for a fraction of the path loss as in the long term evolution
(LTE) systems [25]. Based on the framework, we derive
analytical expressions for the uplink SINR distribution for
both MRC and ZF receivers in the massive MIMO regime.
Unlike prior work analyzing asymptotic performance with
infinity antennas [13], [14], the SINR coverage is examined
as a function of the number of base station antennas and
scheduled users per cell.
2) Scaling Laws to Maintain SIR Distributions: We apply
the SINR results to investigate the interference-limited case,
as numerical results show that the impact of noise becomes
minor in the urban macro-cell scenario. We derive scaling laws
between the number of base station antennas and scheduled
users per cell to maintain the uplink SIR distributions. Unlike
the linear scaling law examined in prior work [8], [24], we find
that a super-linear scaling is generally required for MRC
receivers to maintain the uplink SIR distributions, due to the
near-far effect from intra-cell interference. For ZF receivers,
we show that a linear scaling law still holds, as the intra-
cell interference is negligible. Simulations verify our analysis,
and indicate that the scaling laws derived from the stochastic
geometry framework also apply to the hexagonal model.
Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
4593
3) Comparison of MRC and ZF Receivers: We use the
scaling law results to quantify the performance gap between
ZF and MRC receivers, in terms of the difference in the
number of antennas to provide the same SIR distribution. The
results show that ZF receivers provides better SIR coverage
than MRC receivers; the performance gap increases with the
number of scheduled users in a cell, and is reduced with a
larger compensation fraction in the fractional power control,
as it mitigates the near-far effect from intra-cell interference.
Numerical results indicate different compensation fractions
in fractional power control to maximize the per-user-rate for
MRC and ZF receivers: the optimal fraction is around 0.5 for
MRC receivers, and 0.2 for ZF receivers.
C. Organization and Notation
This paper is organized as follows. We present the system
model for the network topology and propagation channel
in Section II. We analyze the performance of MRC receivers
in Section III, and that of ZF receivers in Section IV-A,
followed by a performance comparison between two receivers
in Section IV-B. We present numerical results to verify the
analysis in Section V, and conclude the paper in Section VI.
We use the following notation throughout this paper. Bold
lower-case letters x are used to denote vectors, and bold upper-
case letters X are used to denote matrices. We use X[:, k] to
denote the k-th row of matrix X, X∗ as the Hermitian transpose
of X, and X† as the pseudo-inverse of X. We use E to denote
expectation, and P to denote probability.
II. S YSTEM M ODEL
In this section, we introduce the system model for an uplink
massive MIMO cellular network. We focus on the networks
operated in the sub-6 GHz band; the proposed model can be
extended for massive MIMO at millimeter wave (mmWave)
frequencies by incorporating key differences in propagation
and hardware constraints [26]. Each base station is assumed
to have M antennas. In each time-frequency resource block, a
base station can simultaneously schedule K users in its cell.
Let X  be the location of the -th base station, Y(k) be the
location of the k-th scheduled user in the cell of -th base
(k) (k)
station, and h the channel vector from X  to Y .
We consider a cellular network with perfect synchroniza-
tion, and assume the following pilot-aided channel estima-
tion in the uplink. In the uplink channel training stage,
the scheduled users Y(k) send their assigned pilots tk , and
base stations X  estimate the channels by correlating the
corresponding pilots and using an minimum mean square error
(MMSE) estimator; in the uplink data transmission, the base
stations will apply either MRC or ZF receivers, based on the
channel estimates. Further, we assume the pilots {tk }1≤k≤K
are orthogonal and fully reused in the network. Note that the
system model assumption applies to general uplink multi-user
MIMO networks with pilot-aided channel estimation in the
uplink, including but not limited to the time-division duplex
(TDD) massive MIMO [2].
Now, we introduce the channel model assumptions. The
channel is assumed to be constant during one resource block,
4594
and fades independently from block to block. Moreover, we
apply a narrowband channel model, as frequency selectivity
in fading can be minimized by techniques like orthogo-
nal frequency-division multiplexing (OFDM) and frequency
domain equalization [27]. In this paper, we focus on the case of
identically and independently distributed (IID) Rayleigh fading
channels, and defer the incorporation of spatial correlations
to future work. Therefore, we express the channel vector
(k)
hn ∈ C M×1 as
 
(k) (k) 1/2 (k)
hn = βn wn ,
(k)
where βn is the large-scale path loss, wn (k)
∈ C M×1 is a
Gaussian vector with the distribution CN (0, I M ) for Rayleigh
fading.
The large-scale path loss gain βn(k)
is computed as
 
(k) (k) −α
βn = C Rn , (2)
where C is a constant determined by the carrier frequency and
reference distance, α > 2 is the path loss exponent.
Next, we introduce the network topology assumptions based
on stochastic geometry. We assume the base stations are
distributed as a PPP with a density λb . A user is assumed to
be associated with the base station that provides the minimum
path loss signal. The users are assumed to be distributed as
an independent PPP with a sufficiently high density. In a
resource block, each base station randomly schedules K users
that are independently and uniformly distributed in its Voronoi
cell [23], [28]. Without loss of generality, a typical scheduled
(1)
user Y0 is fixed at the origin, and its serving base station X 0
is denoted as the tagged base station in this paper. We will
investigate the SINR and rate performance at this typical user.
Now we focus on modeling the distribution of scheduled
user process in a resource block. For 1 ≤ k ≤ K , the k-th
scheduled user Y(k) in each cell is assigned with the same pilot
(k)
tk . Let Nu be the point process formed by the locations of the
k-th scheduled users Y(k) from all the cells. Note that the
scheduled user process Nu(k) is non-stationary (also non-PPP),
as their locations are correlated with the base station process,
(k)
and one scheduled user Y prohibits the presence of all
(k)
other users in Nu in cell X  [21]–[23]. Unfortunately, the
correlations in the scheduled users’ locations make the exact
analysis intractable. In [23], the authors proposed an uplink
model to account for the pairwise correlations, where the
other-cell scheduled users for base station X 0 in Nu(k) is
modelled as an inhomogeneous PPP with a density function of
 
λu (r ) = λb 1 − e−λb πr ,
2
(3)
where r is the distance to base station X 0 . To further simplify
the analysis, e.g., the computation in (23) and (26), we propose
an exclusion ball approximation, as a first-order approximation
of the model in [23], on the distribution of the scheduled user
process Nu(k) as follows.
Assumption 1: The following assumptions are made to
(k)
approximate the exact scheduled users’ process Nu .
Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 11, NOVEMBER 2016
(k)
1) The distances R from a user to their associated
base stations are assumed to√be IID Rayleigh random
variables with a mean of 0.5 1/λb [21].
2) The other-cell scheduled user process Nu(k) is modeled
by a homogeneous PPP of density λb outside an exclu-
sion ball centered at the tagged base station X 0 with a
radius Re .
3) The scheduled users processes using different pilots
(k) (k  )
Nu and Nu are assumed to be independent
for k = k  .
(1) Note that in the exclusion ball model, we equivalently use a
step function λb (1 − I(r < Re )) to approximate the density
function in (3), where I(·) is the√indicator function.
In this paper, we let Re = 1/ (πλb ) by matching the
average number of the excluded points from a homogenous
PPP of density λb in the  ∞step function and in (3), i.e., by
letting λb π Re2 = 2πλb 0 e−λb πr r dr = 1. An alternative
2
explanation for our choice of Re is to let the size of the
exclusion ball π Re2 equal the average cell size 1/λb [29].
In Section V, we will show in simulations that the proposed
exclusion ball model, when applied together with subsequent
approximations in (23) and (26), causes minor errors in the
SINR distributions. The approximate process generated by
the exclusion ball approximation alone, however, need not be
representative of the exact scheduled user process, in terms of
other properties, e.g. the null probability.
Fractional power control, as used in the LTE systems [25],
is assumed in both the uplink training and uplink data stages:
the user Y(k) transmits with power
 
(k) (k) −
P = Pt β , (4)
(k)
where β is the path loss in the corresponding signal link,
 ∈ [0, 1] is the fraction of the path loss compensation, and
Pt is the open loop transmit power with no power control.
We omit the constraint on the maximum uplink transmit
power for simplicity, which will increase the average transmit
power, and reduce the impact of noise. As shown in Fig. 2(b),
however, the maximum transmit power constraint can have a
secondary impact on the SINR distributions in certain massive
MIMO networks, as due to the large antenna array gains,
mobile stations can transmit with a lower open loop power Pt
to meet the receiver sensitivity requirement. The incorporation
of more complicated power control algorithms is deferred to
future work. The noise power is denoted as σ 2 .
In the uplink training stage, after correlating the received
training signal with the corresponding pilot, base station X 0
(1)
has an observation of the channel h00 as
  
(1) (1) (1) (1) (1)
u00 = P0 h00 + P h0 + nt ,
>0
where nt is the noise
 vector
 in the training stage following the
distribution CN 0, σK I M .
2
We assume for  > 0, the large-scale path losses β0 (1)
are perfectly known to base station X 0 . Since the channels
(1)
are assumed to be IID Rayleigh fading, the channel h00 is
BAI AND HEATH: ANALYZING UPLINK SINR AND RATE IN MASSIVE MIMO SYSTEMS 4595

estimated by an MMSE estimator as by the interference in typical urban macro-cell scenarios.

 We derive a scaling law between the number of users and
(1)
P0(1) β00
(1)
(1) antennas that maintains the uplink SIR distribution for the
h̄00 =  (1) (1) u00 , (5) typical user. Finally, we present a method to compute the per-
σ2
 P β0 + K user achievable rate and cell throughput, based on the SINR
(1) (1) distribution.
where h̄00 is the estimation of h00 . Due to the orthogonality
(1)
principle, the channel vector h00 can be decomposed as
A. SINR Coverage Analysis
(1) (1) (1)
h00 = h̄00 + ĥ00 , (6) Now we investigate the uplink SINR coverage based on
(1) the system model. With MRC receivers, we assume that base
where (k)
 ĥ00 is the estimation error
following the distribution station X 0 applies the combining vector g00 as a scaled version
(1) P0(1) β00
(1)
(1) (1)
CN 0, β00 1 −  (1) (1) σ 2 I . of the channel estimate h̄00 to decode the signal from Y00 :
P β0 +
(k)
 K
(k)  (1) (1) σ 2
Let s be the uplink data symbol for user Y with (1) P β + K (1) (1)

(k)

(k)
g00 =   0 h̄00 = u00 . (9)
E |s |2 = P . In uplink data transmission, base station (1) (1)
P0 β00
(k)
X  is assumed to use the combiner vector g to decode s(k) Note the scaling on the combining vector is intended to sim-
(k) (k) plify expressions, and will not change the SINR distribution.
from Y , based on the channel estimate h̄ .
Then, at base
station X 0 , the decoded symbol ŝ0(1) for the typical user X 0(1) is Then, using the combining vector in (9), the SINR expression
can be simplified in (10), as shown  at the bottom of this
(1) (1)∗ (1) (1)
= g00 h̄00 s0 (k)  (k) − (k)
β0 + Kσ Pt , and (k)
ŝ0 2
 page, where 1 = >0 β 2 =
(1)∗ (1) (1)
+ g00 ĥ00 s0 +
(1)∗ (k) (k) (1)∗
g00 h0 s + g00 nu , (7)   −2  2
(k) (k)
(,k) =(0,1) >0 β β0 . The derivation to obtain (10) is
   (k) (k)
unknown at base station given in Appendix A. Note that 1 and 2 correspond
to the sum of certain interference terms from other-cell users.
where nu ∈ C M×1 is the thermal noise vector in the uplink Next, we denote the exact SINR distribution for (10) (using
data transmission. Treating the unknown terms at base station the exact scheduled user distribution defined in Section II
X 0 as uncorrelated additive noise, the uplink SINR for the without the exclusion ball approximation) as P(SINR > T ).
typical user Y0(1) is given in (8), as shown at the bottom of Due to pilot contamination, the combining vector g00 (k)
is
this page, where the expectation operators are taken over the correlated with certain interference channel vectors as shown
(1) (k)
channel estimation error ĥ00 and small-scale fading w0 in the in (9). As a result, the denominator in (10) contains cross-
interference links. Note that the SINR given in (8) is a random products of the path losses from different interferers. More-
variable, due to the randomness in the large scale path losses. over, different cross-product terms in the denominator of (10)
We will investigate the SINR distributions for MRC and ZF can be correlated, as they may contains common path loss
receivers in the following sections. terms, which renders the exact derivation of P(SINR > T )
The proposed system model represents a simple multi-user intractable. Therefore, we compute an approximate SINR
MIMO systems in which the SINR expression can be analyzed distribution P̄(SINR > T ), which we argue in Section V is a
using stochastic geometry. In the following sections, we will good match for P(SINR > T ), in Theorem 1.
study the uplink SINR and rate distributions for MRC and ZF Theorem 1 (MRC SINR): In the proposed massive MIMO
receivers, when the number of base station antennas is large. networks, an approximate uplink SINR distribution with MRC
receivers can be computed as
III. P ERFORMANCE A NALYSIS FOR MRC R ECEIVERS N 
N
In this section, we derive an approximate uplink SINR P̄(SINR > T ) = (−1)n+1
n
distribution for MRC receivers, where the approximation n=1
 ∞ α
becomes tight in the massive MIMO regime, e.g. with M ≥ 64 α(1−) −T ηC t 2 (1−)
e−t −T ηC1 t 2 C3 (t)dt,
antennas. Then, we focus on the interference limited case, as 0
numerical results show that the uplink SINR is dominated (11)

P0(1) |g00
(1)∗ (1) 2
h̄00 |
SINR = (1) (1)∗ (1)
 (k) (1)∗ (k) (1)
, (8)
P0 Eĥ(1) |g00 ĥ00 |2 + P Ew(k) |g00 h0 |2 + |g00 |2 σ 2
00 0
(,k) =(0,1)
 
(1) 2(1−)
(M + 1) β00
SINR =     (10)
(1) (1) 1− (1)  K  (k) 1−  K (k)

(1) 1− (1)
M 2 + β00 1 + k=2 β 00 +
k=1 1 β 00 + 1

Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
4596 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 11, NOVEMBER 2016

where N is the number of terms used in the cal- Corollary 1: With  = 1 and σ 2 = 0, the approximate SIR
σ2
culation, η = N(N!)− N , Cσ 2 =
1
α(1−) , distribution can be computed as
 α   K Pt C 1− (λb π) 2
N   
C1 = M+1 K +1 2 ( 2 +1) M α (+1)  C5 K + α (1.5)
(α−2) + Cσ 2 , C2 = (M+1)(α−1) + N n+1 −T η
1
+ α−1
 α  2 P̄(SIR > T ) = (−1) e M+1
,
2 ( 2 +1) n
(α−2) + Cσ 2 ,
K n=1
M+1
(12)
 ∞ α
K −1 4 2α (1.5)+(α 2 −4) α (1.5)
−u−u − 2 (1−) T ηC 4 (t ) where C5 = .
C3 (t) = e du (α−2)2
0
  K −1 Based on Corollary 1, a linear scaling law between the
∞ e−u
≈ 1 − ηT C4 (t) du , number of users and antennas is observed as follows.
α
0 T ηC4 (t) + u − 2 (1−) Corollary 2: With  = 1 and σ 2 = 0, to maintain the
 α    uplink SIR distribution unchanged, the scaling law between
2 ( 2 +1) α
C4 (t) = M+1
1
α−2 + Cσ 2 t α(1−) + t 2 (1−) , and the number of base station antennas M and users per cell K
 ∞ −t α−1 is approximately
(α) = 0 e t dt is the gamma function. 
 α (1.5)
Proof: See Appendix B. (M + 1) ∼ K + ≈ K. (13)
C5
Besides the exclusion ball approximation, the main approx-
Note that when  = 1, the linear scaling law matches prior
imation in Theorem 1 is to replace certain out-of-cell interfer-
results in [8, Sec. IV], where the path loss to all associated
ence terms by their means in (23) and (26). The approximation
users in the typical cell was assumed to be identical. The linear
results in a minor error in the SINR distribution, as (i) with
scaling law, however, does not apply to other cases with  < 1,
K users in a cell, the intra-cell interference dominates the
e.g. in the following case without power control.
out-of-cell interference with high probability; (ii) with large
Case 2 (No Power Control,  = 0): When assuming no
antenna arrays, the ratio of the signal power to certain out-of-
(k) power control and MRC receivers, the uplink SIR can be
cell interference power terms, e.g. the terms in 1 , decays evaluated as follows.
as M . In Section V, using N ≥ 5 terms, the distribution
1
Theorem 2: With  = 0 and σ 2 = 0, an approximate uplink
P̄(SINR > T ) from Theorem 1 is shown to be a good SIR distribution can be calculated as
match with the SINR distribution P(SINR > T ) from Monte N 
Carlo simulations. In addition, the error of the approximation  N
P̄(SIR > T ) = (−1)n+1
becomes more prominent with a smaller noise power, as all n
n=1
the approximations are made with respect to the interference  ∞ nηT α
e−(μ(1−2/α)(nηT ) +1)t − α−1 t dt, (14)
2/α
distribution. The expression is intended for the massive MIMO
regime when M  1, as the error of the approximations 0
1 where N is the number of terms used in the computation, and
decays with M . In simulations, we find that the results
generally applies to the multi-user MIMO networks with not- μ = (M+1)
K
2/α .
so-large M, e.g. the case of (M, K ) = (10, 2). Proof: The proof is similar to that in [1, Appendix A].
In Theorem 1, the noise power is taken account by the We will show in Section V that Theorem 2 provides a tight
σ2 approximation of the exact SIR distribution P(SIR > T ), when
parameter Cσ 2 = α(1−) , which shows that the
K Pt C 1− (λb π) 2
N ≥ 5 terms are used. Moreover, note that in (14), the number
impact of noise on the SINR is reduced with a larger number of antennas M and the number of scheduled users per cell K
of scheduled users per cell K , a higher base station density λb , only affect the value of μ. Therefore, by Theorem 2, in the
a smaller path loss α, and a larger power control parameter . no power control case, we observe the following scaling law
Besides, the impact of noise goes down with larger M, as in to maintain SIR.
the expressions for C1 and C2 , the noise parameter Cσ 2 is Corollary 1: Assuming no power control, the approximate
divided by (M + 1). scaling law to maintain the same uplink SIR distribution is
We will show in Section V that the impact of noise is negli- (M + 1) ∼ K α/2 , which is a superlinear polynomial scaling
gible in certain urban macro-cell cases with M = 64 antennas when α > 2.
at base stations. Therefore, we focus on the performance of In the case of no power control, the difference in the path
interference-limited networks, where the general expression in losses between the typical user and the intra-cell interferers
Theorem 1 can be further simplified in the following special affect the SIR distribution, and thus the scaling law to maintain
cases. the SIR becomes a function of the path loss exponent. The
Case 1 (Full Power Control,  = 1): In this case, super-linearity in the scaling law can be explained by the
the transmitting power at scheduled user Y(k) is adjusted near-far effect of the intra-cell interference from multiple
(k) (k)
to compensate for the full path loss, i.e., P = Pt β , users in a cell. With no power control, the cell edge users
such that a base station receives equal signal powers will receive weaker signals than the cell center user. With
from all of its associated users. When  = 1, the a uniform user distribution in a cell, the typical user will be
SIR distribution can be simplified as in the following more likely to be located at the cell edge. When increasing the
corollary. number of scheduled users K in a cell, the probability that the

Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
BAI AND HEATH: ANALYZING UPLINK SINR AND RATE IN MASSIVE MIMO SYSTEMS 4597

interference from a cell-center interferer dominates the desired We will examine the average cell throughput as a function of
signal increases. Therefore, compared with the linear scaling M and K in Section V. Before that, we continue to present
law with full power control ( = 1) where such near-far effect the results for ZF receivers in the next section.
is mitigated, more antennas will be needed in the no power
control case to reduce the intra-cell interference, and preserve
IV. P ERFORMANCE A NALYSIS W ITH ZF R ECEIVERS
the SIR distribution, when increasing K .
Next, we focus on the scaling law in the general fractional In this section, we first derive the SINR distributions
power control case with  ∈ (0, 1). It is difficult to derive with ZF receivers. Then, we apply the analytical results to
the exact scaling law directly from the expression (11), due to compare the SIR performance of MRC and ZF receivers
the integral form. Since with the fractional power control, the in an interference-limited network. In particular, we aim to
equivalent path loss exponent in the signal link linearly scales answer the question: compared with MRC receivers, how
with , we propose the following conjectured scaling law by many antennas can be saved by applying ZF receivers, while
linearly fitting the exponent s of the scaling law (M +1) ∼ K s , keeping the same uplink SIR distribution.
based on two special cases of : by Corollary 2, when  = 1,
s = 1; and by Theorem 2, when  = 0, s = α2 . Therefore,
for general 0 <  < 1, the linearly fitted exponent s in the A. SINR Analysis of ZF Receivers
conjectured scaling law is given as follows. Now we begin to investigate the performance of
Scaling law 1: With fractional power control and MRC ZF receivers in an uplink massive MIMO network. For
receivers, a conjectured scaling law between M and K is ZF receivers, we still focus on the case of IID fading. To cancel
(M + 1) ∼ K s , where the exponent of the scaling law is the intra-cell interference, base station X  will apply the
(k)
s = α2 (1 − ) + . combining vector g for user Y(k) as
Scaling law 1 indicates that a (superlinear) polynomial
(k)
scaling law between K and M is required to maintain uplink g = H† [:, k], (18)
SIR distribution, for a general  < 1. The conjecture in Scaling

(1) (2) (K )
law 1 are verified by numerical simulations in Section V. where H = u , u , . . . , u ∈ C M×K is the matrix of
all estimated channels to the associated users in cell X  , and
B. Rate Analysis  (1) (1) 2
(k)  P β +σ (k)
u =    K h̄ is a scaled version of the channel
In this section, we apply the SINR results to compute (k) (k)
P β
the achievable rate. First, we define the average achievable
estimate. The scaling in the channel estimates will not change
spectrum efficiency at a typical user as
  the uplink SINR distribution, as it will only cause certain
τ0 = E log2 (1 + min{SINR, Tmax }) , (15) scaling in the corresponding combining vector. Similar to the
where Tmax is a SINR distortion threshold determined by case of MRC receivers, the exact uplink SINR distribution is
limiting factors like distortion in the radio frequency front-end. difficult to derive, as due to pilot contamination, the combining
By [30, Sec. III-C], given the SINR distribution P(SINR > T ), vector is correlated with certain interference channel vectors.
the average achievable spectrum efficiency can be computed Therefore, applying the same approximations in (23) and (26),
 Tmax P(SINR>x) we derive an approximate distribution for the uplink SINR
as τ0 = ln(2)1
0 1+x dx. To take account for the expression in (8) for the typical user X 0(1) in the following
overhead, let ψ be the fraction time for overhead. In this theorem.
paper, for simplicity, we only consider the overhead due to Theorem 3: With M  K and ZF receivers, an approx-
uplink channel training, and compute the overhead fraction ψ imate uplink SINR distribution for the typical user can be
as ψ = TTct = TKc , where Tt and Tc are the length of channel calculated by
training period and coherent time, in terms of symbol time.
N 
The length of channel training is assumed to be equal to the N
number of scheduled users in a cell, as we assumed full reuse P̄(SINR > T ) = (−1)n+1
n
of orthogonal pilots throughout the network. Then the average n=1
 ∞  α 
achievable rate with the overhead penalty τ̄0 equals −nηT C 6 t 2 (1−) +C 7 t α(1−) −t
e dt, (19)
0
τ̄0 = (1 − ψ)τ0 , (16)
where the constant
Note that when ignoring thermal noise, the scaling law to 
maintain SIR distribution also maintains the average achiev- 1 1 M(K − 1)
C6 = C9 + +
able rate τ0 . When taking account for the training over- M −K +1 M + 1 (M − K + 1)2
head penalty ψ, however, the scaling law will not keep τ̄0 M(K − 1)C8
unchanged, as 1 − ψ linearly decreases with K , unless ψ is + ,
(M − K + 1)2
negligible, e.g. when the coherence time Tc  K . Next, we 
M  α ( + 1) 1 (K − 1)M
define the average cell throughput τcell , in terms of spectrum C7 = + + C92
M +1 α−1 M + 1 (M − K + 1)2
efficiency, as
(K − 1)M
τcell = K (1 − ψ)τ0 . + C8 C9 ,
(17) (M − K + 1)2

Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
4598
2 α ( 2 +1)+(α−2)C σ 2 2 α ( 2 +1)
C8 = (α−2)(1+C σ 2 )+2 α ( 2 +1)
, C9 = α−2 + Cσ 2 ,
σ2
Cσ 2 = α(1−) , N is the number of terms used
K Pt C 1− (λb π) 2 Based on Scaling law 1 and Scaling law 2, we have the
1
in the computation, and η = N(N!)− N .
Proof: See Appendix C.
Note that when K = 1, the SINR distribution in (19) for
ZF receivers is the same as that for MRC receivers in (12).
We will verify the tightness of the approximation P(SINR >
T ) ≈ P̄(SINR > T ) by numerical simulation in Section V.
We have the following remark on the applicable regime for
Theorem 3.
Remark 1: We need the condition M  K in the proof,
as the error in the approximation in (28) decays as M−K 1 MRC(M, K ) denote the uplink SIR distributions with ZF and
+1 .
In numerical simulations, we find that the approximate SINR
distribution in Theorem 3 shows a good match with the
simulations when M K ≥ 3 and M ≥ 10. The same comment
applies to Scaling law 2 as below.
Next, we focus on the interference-limited case. Based on
Theorem 3, we can derive an approximate scaling law between
M and K to maintain the SIR distribution in the region of
M  K as follows.
Scaling law 2: With ZF receivers and σ 2 = 0, the uplink
SIR distribution of the typical user remains approximately
unchanged when the number of antennas M linearly scales
with the number of users per cell K as (M + 1) ∼ K .
Proof: Note that when σ 2 = 0, Cσ = 0. The dependence
on M and K in (19) only occurs in the constants C6 and C7 .
Therefore, it is sufficient to show that a linear scaling between
M and K (approximately) maintains the values of C6 and C7 .
Note that when M → ∞, and M  K , the following limits
hold: M+11
→ 0, M−K 1
+1 → 0, and M+1 → 1. Therefore, it
M
follows that when keeping M+1 K
=t, lim M→∞ C6 = (C81−t +C 9 )t
,
 α (+1) 4 2α ( 2 +1) 2C 8  α ( 2 +1)
and lim M→∞ C7 = α−1 + 1−t
t
(α−2)2
+ α−2
which are invariant when (M + 1) linearly scales with K .
Compared with MRC receivers, the near-far effect for users
in a cell becomes minor with ZF receivers, as the intra-cell
interference is largely suppressed. Therefore, a linear scaling
law applies for ZF receivers even without power control. Based
on the SINR coverage results, the achievable rate per user and
sum throughput can be computed following the same line as
in Section III-B. In the next section, we will use the derived
results to compare the SIR coverage performance between
MRC and ZF receivers in an interference-limited network.
B. Comparison of SIR Coverage Performance
Now assuming the network is interference-limited, we
compare the SIR coverage between ZF and MRC receivers.
Prior work [8] showed that ZF and MRC receivers have the
same asymptotic performance, both which are limited by the
pilot contamination. The analysis in [24] showed that by
suppressing intra-cell interference, which turns to be more
dominant than the out-of-cell interference, the ZF receivers
suffers from less interference than MRC receivers. In this
section, we make a quantitative comparison by answering the
following question: in IID fading channels, how many base
station antennas MZF is needed for ZF receivers to provide the
Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 11, NOVEMBER 2016
same uplink SIR distribution as MRC receivers with MMRC
antennas?
following proposition to determine MZF to match the SIR
coverage with MRC receivers.
Proposition 1: Assuming MZF  K , ZF receivers with
(MZF +1) = ξ(MMRC +1) antennas approximately provide the
same uplink SIR distribution as MRC receivers with MMRC
antennas in a massive MIMO networks, where the scaling
α
factor ξ = K −( 2 −1)(1−), and K is the number of scheduled
users in a cell.
Proof: For the ease of notation, let ZF(M, K ) and
MRC receivers of M antennas, when serving K users in a
cell. By Scaling law 2, when MZF  K , ZF(MZF , K ) ≈
ZF( MZFK+1 , 1). Next, note that when K = 1, i.e., with a single
scheduled user in a cell, MRC and ZF receivers provide the
same SIR coverage. Thus, it follows that ZF(MZF , K ) =
ZF( MZFK+1 , 1) = MRC( MZFK+1 , 1). Last, by Scaling law 1,
ZF(MZF , K ) ≈ MRC( MZFK+1 , 1) ≈ MRC((MZF +
α
1)K ( 2 −1)(1−) − 1, K ).
The condition MZF  K in the proposition is required
to ensure the applicability of Scaling law 2. In numerical
simulations, the result is found to be a good approximation
with MKZF > 3. Note that the exponent of the scaling factor
−( α2 − 1)(1 − ) is non-positive, which indicates we need
MMRC ≥ MZF to provide the same SIR coverage. Further, the
scaling factor ξ increases with the number of the scheduled
user K , which reveals that the performance gap between
MRC and ZF receivers grows with K . When K increases, the
mitigation of the intra-cell interference from (K − 1) users
by ZF receivers becomes more prominent to improve SIR
coverage. In addition, Proposition 1 also shows that in terms of
, the SIR distribution, the performance gap reduces with larger
 in the power control scheme, as the scaling factor ξ is a
decreasing function of . Simulations show that with  = 1,
only a minor gap exists between the SIR coverage curves for
ZF and MRC receivers.
Last, we note that Proposition 1, which is drawn based on
the SIR distribution, need not extend to a general SINR dis-
tribution that is not dominated by interference; prior work [6]
showed that when the noise is not negligible, MRC receivers
would provide a comparable or even better SINR coverage,
compared with the ZF receivers. In the following section, we
will present numerical results to validate our analytical results.
V. N UMERICAL R ESULTS
In this section, we verify our analytical results with numeri-
cal simulations, which follow the procedure as: (1) generating
the base station process as a PPP of density λb ; (2) generating
the overall user process as a PPP of density λu,o , where we
use λu,o = 60λb , unless otherwise specified; (3) associating
the points in the overall user process to base stations, based
on the minimal path loss rule, and then randomly scheduling
K out of the associated users in each cell as their scheduled
users; (4) picking the base station closest to the origin as the
(1)
tagged base station X 0 , and its first scheduled user Y0 as the
BAI AND HEATH: ANALYZING UPLINK SINR AND RATE IN MASSIVE MIMO SYSTEMS
Fig. 1. Comparison of SINR and SIR distributions. In the figures, we use
markers to represent SINR curves, solid lines for SIR. We assume K = 10
users per cell,  = 0, and α = 4 in all cases. The gap between the SIR
and SINR distributions becomes minor when ISD=500 meters, which is the
typical size for the urban macro cells [31].
typical user; (5) generating channel vectors as IID Gaussian
vectors, and computing the SINR for the iteration; (6) repeat-
ing the step (1)-(5) for 10,000 iterations, and computing the
(1)
empirical distribution of the SINR at Y0 . For the simulations
using hexagonal grids, we follow the same procedure except
that the base station process is generated as a 19-cell hexagonal
grid, and the tagged base station is the center cell. In addition,
we will use N = 5 terms when evaluating the analytical
expressions.
A. Impact of the Noise
To begin with, we examine the impact of noise by com-
paring the SINR and SIR distributions for both MRC and ZF
receivers in different scenarios in Fig. 1. In the simulations,
we assume Pt =23 dBm, and the bandwidth is 20 MHz as
in the current LTE standards [31]. We examine the case of
 = 0, which maximizes the impact of the noise. We sim-
ulate with two inter-site distances (ISDs): an average ISD
of 500 meters in Fig. 1(a), and 1000 meters in Fig. 1(b).
Note that a typical ISD of 500 meters is assumed for urban
macro-cells in the 3GPP standards [31]. In Fig. 1(a), the
Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
4599
Fig. 2. SINR coverage for MRC receivers. In the simulations, we assume
α = 4. In (a), the analytical curves are drawn based on Theorem 1, which
are shown a good fit with simulation. The difference in the curves for
(M, K , ) = (64, 10, 0) and (M, K , ) = (128, 20, 0) indicates that linear
scaling between M and K does not generally preserve SIR for MRC receivers.
In (b), we compare the SINR distributions with and without the maximum
transmit power constraint. We assume (M, K ) = (128, 10), the average
ISD= 500 m, the maximum uplink transmit power is 23 dBm, the system
bandwidth is 20 MHz. The open loop transmit power Pt is chosen, such that
the user located at the edge of an average-size cell has 0 dB uplink SNR.
network with ISD=500 meters is shown to be interference-
limited with M = 64 antennas for both MRC and ZF receivers,
as the SIR curves almost coincide with the SINR curves,
which justifies the interference-limited assumption in urban
marco cells. In the sparse network with ISD=1000 meters,
however, simulations show that even with M=64 antennas,
notable gaps exist between the SINR and SIR distributions,
especially for ZF receivers. In addition, the results in Fig. 1(b)
shows that when the noise power is high, even with no
power control, ZF and MRC receivers have a similar SINR
coverage performance, which indicates that the SIR com-
parison results in Proposition 1 need not extends to general
SINR comparisons.
B. SINR Coverage for MRC Receivers
In Fig. 2(a), we verify the analytical results for the SIR
distribution with MRC receivers. Numerical results show that
4600 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 11, NOVEMBER 2016

Fig. 4. Comparison of SIR coverage with MRC and ZF receivers. We assume

α = 4. As the double arrays display, when fixing  = 0, the performance gap
in SIR coverage is shown to increase with K ; when fixing K = 20, the gap
Fig. 3. SIR distributions with ZF receivers. We assume α = 4, and IID fading diminishes when  → 1.
channel. The analytical curves are plotted based on Theorem 3. Simulations
verify the analytical results, and show that when both M and K double, the
SIR curves remain almost unchanged.

the SIR coverage is sensitive to the compensation fraction 

in the fractional power control: a large compensation fraction
 improves the SIR coverage in the low SIR regime at the
expense of sacrificing the coverage in the high SIR regime.
Besides, a comparison of the curves for (M, K ) = (64, 10)
and (M, K ) = (128, 20) shows that the linear scaling
law does not maintain the SIR distribution when  = 0.
In Fig. 2(b), we consider the case when the uplink transmit
power after the fractional power control is clipped by a
maximum power constraint. In the simulations, the maximum
power constraint is assumed to be 23 dBm [31]; the open
loop transmit power Pt is determined, such that the uplink
SNR at a cell-edge user is 0 dB in a cell of the average
size. The results indicate that ignoring the maximum transmit
power constraint has a minor impact on the SINR distribution
in massive MIMO networks; as with large antenna arrays,
mobile stations can apply a lower transmit power than the
single-antenna case, to meet the target SNR.
C. SIR Coverage for ZF Receivers
We verify the analysis for ZF receivers in Fig. 3. The analyt-
ical curves generally match well with numerical simulations.
A comparison of the curves for (M, K , ) = (64, 10, 0) and
(M, K , ) = (128, 20, 0) shows that unlike the case of MRC
receivers, a linear scaling law between M and K maintains
the SIR distribution, even when there is no fractional power
control implemented.
D. SIR Comparison Between MRC and ZF Receivers
We compare the uplink SIR distributions for MRC and
ZF receivers in Fig. 4. Simulations show that ZF receivers
provide better SIR coverage, due to the suppression of intra-
cell interference. Moreover, for the same , the performance
gap between MRC and ZF receivers increases with the number
of scheduled users K , as the strength of total intra-cell
interference also increases with K . When fixing M and K ,
Fig. 5. Verification of Proposition 1. In the simulation, α = 4. In the
simulation, we use the SIR curve of MRC(64, 5) as a baseline for comparison.
We use Proposition 1 to compute the required number of antennas for ZF
receivers that provides the same SIR distribution as the baseline curve.
the performance gap decreases with ; when  = 1, the
SIR coverage gap becomes minimal between MRC and ZF
receivers. With full compensation of path loss in power
control, linear scaling laws between M and K apply to both
MRC and ZF receivers, as the near-far effect for users in
a cell is mitigated. When M  K , the difference in the
average (residue) intra-cell interference between MRC and ZF
1
receivers becomes minor, as it decays with M .
In Fig. 5, we verify our theoretical results in Proposition 1.
In the simulation, we fix the number of antennas for the MRC
receivers to be MMRC = 64, and use Proposition 1 to calculate
the required MZF , to maintain the same SIR distribution.
Numerical results show a good match with our analysis; the
minor mismatch in the case  = 0 is because Proposition 1
theoretically requires MKZF  1, while we use MKZF = 13 5 in
the simulation.
E. Verification With Hexagonal Grid Model
We verify the scaling laws derived from stochastic geometry
with the hexagonal grid model in Fig. 6. In the simulations,

Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
BAI AND HEATH: ANALYZING UPLINK SINR AND RATE IN MASSIVE MIMO SYSTEMS
Fig. 6. Verification of the scaling laws in the hexagonal model. We use
(M, K ) = (32, 5) as the baseline curves. When increasing the number of
users to K =10, we compute the required M to preserve the SIR distribution
as baseline curves, according to Scaling law 1 and Scaling law 2. Simulations
indicates that the scaling law results apply to the hexagonal model.
we use a layout of 19 hexagonal cells with inter-site distance
of 300 meters; only the scheduled users in the central cell are
counted for the SIR statistics, to avoid edge effect. In Fig. 6(a),
for MRC receivers, we use a (M, K ) = (32, 5) as the baseline
curve for comparison. When doubling the number of scheduled
users to K = 10, we use Scaling law 1 to compute the required
M to maintain the same SIR distribution. Scaling law 1 is
shown to be almost accurate with extensive combinations of
the system parameters in the hexagonal grid model. Similarly,
results in Fig. 6(b) verifies the linear scaling law for ZF
receivers in Scaling law 2. This indicates that the stochastic
geometry model provides reasonable predictions even for the
hexagonal model.
F. Comparison of Scaling Laws
We compare scaling laws to maintain the uplink SIR dis-
tribution in different scenarios in Fig. 7. We plot the required
number of antennas to maintain the same SIR distribution as
Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
4601
Fig. 7. Comparison of different scaling laws. We plot the required number
of antennas to provide the same SIR as that of the case (M, K ) = (16, 5) as
a function of K with different system parameters.
Fig. 8. Average spectrum efficiency per user in an interference-limited
network. In the simulation, we assume Tmax = 21 dB, which sets the
maximum spectrum efficiency per data stream as 7 bps/Hz. Training overhead
is not taken account in this figure.
that in the case of (M, K ) = (16, 5), as a function of K .
As shown in the plot, for MRC receivers, given the path loss
exponent α, the slope of the scaling law is determined by
the fraction of path loss compensation : the linear scaling
law proposed in prior work [8], [24] is only achieved when
 = 1. Although the choice of  = 1 makes the system with
MRC receivers linearly scalable, it need not maximize the per-
user rate, as will be shown in Fig. 8. On the contrary, for ZF
receivers, the linear scaling applies for all  ∈ [0, 1].
G. Rate Performance
We illustrate the results on the average spectrum efficiency
per user in Fig. 8. In the simulation, the average ISD is 500
meters, and K = 10, which is shown to be interference-
limited in Fig. 1. Consistent with the SIR results, in a
interference-limited network, ZF receivers provide a higher
spectrum efficiency per user. Numerical results also show that
the average spectrum efficiency is sensitive to the fraction of
the path loss compensation ; the optimum  for per user
4602
Fig. 9. Uplink cell throughput as a function of K . The overhead due
to channel training is taken account in the simulations. We simulate an
interference-limited network with ISD=500 meters. We use  = 0.5 for MRC
receivers, and  = 0.2 for ZF receivers, which are shown to optimize the per
user rate.
TABLE I
C OHERENCE T IME Tc IN THE E XAMPLES
rate is generally around 0.5 for MRC receiver, and 0.2 for
ZF receivers. In addition, we also observe that there is a
minor performance gap in rate between ZF and MRC receivers
under full channel compensation power control, as predicted
by Proposition 1.
Last, we examine the cell throughput in a system operated
at 2 GHz in Fig. 9. As an example, we consider an OFDM
system, where the symbol time is 66.7 μs. We consider
two cases with different mobilities as listed in Table I; the
coherence time Tc is computed as Tc = 4 1fD [32], where
f D is the maximum doppler frequency. In this simulation, we
assume the density of overall users to be 100 times the base
Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 11, NOVEMBER 2016
station density, to simulate the case with large K . In Fig. 9(a),
in the high mobility case, when Tc < M, the optimal K ∗
for cell throughput is limited by the duration of Tc , and the
optimal value generally is K ∗ ≤ T2c . In the example of low
mobility case, when Tc > M, the results in Fig. 9(b) show
that the optimal K ∗ depends much on M: for ZF receivers,
K
the cell throughput drops fast when M approaches to 1, and the
∗ M
optimal K is around 2 for maximum throughput; for MRC
receivers, the cell throughput becomes saturated approximately
when K > M3 . In addition, ZF receivers generally achieve
better cell throughput than MRC receivers; the only exception
is the case of M K ≈ 1, where the cell throughput of ZF
receivers drops below that of MRC. In addition, compared
with the single user per cell case (K = 1), the results confirm
that massive MIMO improves the cell throughput by serving
multiple users simultaneously.
VI. C ONCLUSIONS
In this paper, we proposed a stochastic geometry framework
to analyze the spatial average SINR coverage and rate in mas-
sive MIMO networks. We applied the analysis and numerical
results to draw the following important system design insights.
• The uplink massive MIMO networks can be interference-
limited in urban marco cells (ISD=500 meters) with
M = 64 antennas at base stations.
• With MRC receivers, the number of antennas M should
scale super-linearly with the number of scheduled users
α
per cell K as (M + 1) ∼ K 2 (1−)+ , to maintain the
uplink SIR distribution; a linear scaling law only applies
to the case of full path loss compensation in the power
control, i.e., when  = 1.
• With ZF receivers, a linear scaling between the number
of antennas M and users per cell K maintains the uplink
SIR distribution in massive MIMO.
• When noise is negligible, ZF receivers provide better SIR
coverage rate than MRC receivers. The performance gap
increases with K , and decreases with path loss compen-
sation faction . The gap becomes minor when  = 1.
• The SIR coverage and rate are sensitive to the fraction 
of path loss compensation in power control. Larger 
improves coverage in the low SIR regime while reducing
coverage probability at high SIR. Numerical results show
that the optimal  for rate is around 0.5 for MRC,
and 0.2 for ZF receivers in certain cases.
A PPENDIX
A. Derivation of (10):
With the combining vector in (9), the SINR expression
equals (20), as shown at the top of the next page
In the numerator, the signal power can be computed as
 
(1) (1) 2
P β
(1) (1)∗ (1) (a) 0 00 (1)
P0 |u00 h̄00 |2 =  |u |4
 (1) (1) σ 2 2 00
 P β0 + K
 
(1) (1) 2
(b) P0 β00
(1) 4
≈  E|u00 |
 (1) (1) σ 2 2
P
  β 0 + K
BAI AND HEATH: ANALYZING UPLINK SINR AND RATE IN MASSIVE MIMO SYSTEMS 4603

 
 (1)∗ (1) 2
P0(1) u00 h̄00 
SINR =     2 . (20)
(1)  (1)∗ (1) 2  (k)  ∗ (k)  (1) (1)∗ (k)  (1) 2 2
P0 E u00 ĥ00  + (,k)=(0,1) P E nt h0 +  ≥0 P h0 h0  + |u00 | σ

 
(c) (1) 2
= P0(1) β00 (M 2 + M) Therefore, we can express (22) as
   σ2  
(k) − (k)
(1) 2(1−) M Pt β β0
= Pt2 β00 (M 2 + M),
K
(,k) =(0,1)
where (a) follows from the MMSE estimator in (5), 


(1) M→∞ (1) (k) (1) (1)∗ (k)
(b) follows from the fact that |u00 |4 → E|u00 |4 , and the + P P E |h0 h0 |2

approximation error decays as M12 [33], and (c) follows from
 2 ⎛  >0 ⎞
(1) 4 (1) (1) σ2   −
the fact that E|u00 | = (M 2 +M)  P β0 + K . Next, = M Pt2 ⎝ (k)
β (k) ⎠
β0
we compute the first term in the denominator of (20) as (k,) =(1,0)
⎛ ⎞
P0(1) E|u00
(1)∗ (1) 2
ĥ00 | =
1 (1) (1) 2
P0 E|u00 (1) 2
| E|ĥ00 | σ2   (1) − (1)
M ×⎝ + β    β0 ⎠
  K Pt
(a) 1 (1)  (1) (1) σ 2 (1)
 ≥0
= P0 M P β0 + Mβ00   (1) −2  (1) 2
M

K + M 2 Pt2 β β0 .
  >0
P0(1) β00
(1)
× 1 −  (1) (1) Last, the thermal noise term in the denominator can be
σ2
 P β0 + K simplified as
   
    (1)− (1)
(1) 1− σ2   (1)− (1) σ2
= M Pt β00
2
β β0 + , (1) 2 2
|u00 | σ = Pt Mσ 2
β β0 + .
Pt K Pt K
>0 
where (a) follows from the fact that the channel Then the expression in (10) is obtained through algebraic
(1)
estimation
  error ĥ00 follows
the distribution manipulation in the denominator.
(1) (1)
(1) P0 β00
CN 0, β00 1 −  (1) (1) σ 2
IM .
 P β0 + K B. Proof of Theorem 1:
Next, we simplify the second term in the denominator. Note To allow for tractable computation and decouple the cor-
that unless (,  , k) = (n, n  , m), h(1)∗ (k) (1)∗ (m)
 0 h0 and hn  0 hn0 are related terms in the denominator of (10), we propose the
uncorrelated zero-mean random variables. Therefore, we can following approximations on the out-of-cell interference terms:
simplify the second term in the denominator of (20) as for k ∈ [1, K ],
   (1) (1)∗ (k)  
∗ (k) (k) (a)
  (k) − (k) σ2
E|nt h0 + P h0 h0 |2 (21) 1 ≈ E β β0 +
(,k) =(0,1)  ≥0 K Pt
     >0 
σ2 (k) − (k)   (k) − (k) σ2
= M Pt β β0 = E E β β0 +
K K Pt
(,k) =(0,1) >0
  

  (k)
+
(k) (1)
P P E
(1)∗ (k)
|h0 h0 |2 . (22)
(b) − − α α  σ2
≈ C (λb π) 2  ( + 1)E β0 +
 >0 2 K Pt
>0
Note that for k = 1, =  = 0, the expression is simplified (c) 2C
1− (λ π) α(1−)
2   σ2
α
b
as = +1 + , (23)


 α−2 2 K Pt
(k) (1) (1)∗ (k) (1)2 (1) (k)
P P E |h0 h0 |2 = P E |h0 |4 where in (a), we approximate 1 by its mean; step (b)
  follows from
(1)2 (1) 2   
  
= (M 2 + M)P β0 (k) − (k) α α
    E β = C − E R = (λb π)− 2  α +1 ;
(1) −2 (1) 2 2
= Pt2 (M 2 + M) β β0 ; (24)
for k > 1 or k = 1,  =  > 0, it follows that and step (c) follows from the exclusion ball model in

 Approximation 1 and the Campbell’s theorem [29] as
P(k) P(1)
 E |h(1)∗ (k) 2
h
0 0 | = M P(1) (k) (1) (k)
 P β0 β0
   ∞
 (k) 2C(λb π)α/2
  E β0 = 2πλb C x −α xdx = . (25)
(k) − (1) (k)
= M Pt2 β(1)
  β β0 β0 . Re α−2
>0

Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
4604 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 11, NOVEMBER 2016

Similarly, we can approximate (1)

2 by its mean as: where in (a) C1 , C2 and C4 (x) are defined in Theorem 1;
(k) C 2(1−) (λ π)α(1−) in (b) we use a “dummy” gamma variable g with unit
2 ≈ α−1
b
 α ( + 1), (26) mean and shape parameter N to approximate the con-
  
α
(k) −2 (k)2 stant number one, and the approximation follows from the
where follows from E β = C −2 E R =
fact that g converges to one when N goes to infinity,
(λb π)−α  α ( + 1). n n−1 −nx
i.e., lim n→∞ n x (n)e = δ(x − 1) [34], where δ(x) is
Next, applying the approximation in (23) and (26) and
(1) the Dirac delta function; in (c), the approximation follows
conditioning on R00 = x, we simplify the approximate
conditional uplink SINR as from Alzer’s inequality [30], [35, Appendix A], where
 (k)
η = N(N!)− N ; (d) follows from the fact that R00 is
1
α
SINR ≈ T C1 (λb π x 2 )α(1−) + C2 (λb π x 2 ) 2 (1−) assumed to be IID Rayleigh random variable; in (e) we change
−1
  α
(k)2 − 2 (1−)
variable as u = λb πs 2 ; in (f), we apply the approximation
+ C4 (λb π x )
2
λb π R00 , (27) exp(−x) ≈ 1+x 1
inside the integral, to allow for faster numer-
k>0 ical evaluations. Last, we obtain the uplink SIR distribution
(1)
where in (a) C1 , C2 and C4 (x) are defined in Theorem 1. Next, by de-conditioning on R00 = x, which√is assumed to be a
(1)
conditioning on R00 = x, the approximate SINR distribution Rayleigh random variable with mean 0.5 1/λb , and changing
can be computed as the variable as t = πλb x 2 .
 
(1)
P SIR > T |R00 =x

(a) T  α
C. Proof of Theorem 3
≈ P 1> C1 (λb π x 2 )α(1−) + C2 (λb π x 2 ) 2 (1−) Before proving the theorem, we present a useful lemma on
M +1  (1)
  α
(k)2 − 2 (1−)
the distribution of the combining vector g00 as follows.
+ C4 (λb π x )
2
λb π R00 Lemma 1 (From [36]): The square norm of the
(1) 2
 k>0 combining vector |g00 | follows the distribution of
(b) T  α   −1
≈ P g> C1 (λb π x 2 )α(1−) + C2 (λb π x 2 ) 2 (1−) χ2(K −M+1) (1) (1) σ2
 P β0 + K , where χ2(K
2 2
M +1  −M+1)
  α
(k)2 − 2 (1−) represents a Chi-square random variables with 2(K − M + 1)
+ C4 (λb π x )
2
λb π R00
degrees of freedom.
 k>0 χ2
(c) ηT  Note that when (M K + 1) → ∞, M−K 2(K −M+1)
+1 → 1. Therefore,
≈ 1 − E 1 − exp − C1 (λb π x 2 )α(1−) by Lemma 1, when M  K , the following approximation
M +1
α
+ C2 (λb π x 2 ) 2 (1−) holds as
 N ⎤
  α
(k)2 − 2 (1−)
(1)
|g00 |2 ≈
1
, (28)
+ C4 (λb π x )
2
λb π R00 ⎦ (M − K + 1)S (1)
k>0 
N   
α(1−)  α (1−)  where for ease of notation, we define S (k) =  P(1) (1)
 β +
N −nT η C 1 λb π x 2 +C 2 λb π x 2 2  
= (−1)n+1 e σ2 (k) 1− (k)
n K = β00 + 1 , and the approximation error decays
n=1
  K −1 (1) (1) 2
+1 . Noting that |g00 u00 | = 1, the nominator of the
  α 1
 (k)2 − 2 (1−) as M−K
−nT ηC 4 λb π x 2 λb π R00
×E e SINR expression in (8) can be computed as

N   
α(1−)  α (1−) 
(1) (1)∗ (1) P0(1)2 β00
(1)2
(1) (1)
(d)
=
N
(−1)n+1 e
−nT η C 1 λb π x 2 +C 2 λb π x 2 2 P0 |g00 h̄00 |2 =  2
|g00 u00 |2
S (1)
n
n=1
 ∞ K −1    2
(1) 2(1−)

−nT ηC 4 λb π x 2 s −α(1−) −λb πs 2 = β00 / S (1) . (29)
× e λb πsds
0
N    
2 α(1−) +C λ π x 2
α (1−)  Applying the results in (28), the first term in the denomi-
(e) N n+1 −nT η C 1 λb π x 2
= (−1) e 2 b nator of (8) is computed as
n
P0(1) E|g00
(1)∗ (1) 2
n=1
 ∞  α
K −1 ĥ00 |
−nT ηC 4 λb π x 2 u − 2 (1−) −u  
e du × P β
(1) (1)
1
0 ≈ P0(1) β00
(1)
1 − 0 (1)00
N 
  
α(1−)  α (1−)  S (M − K + 1)S (1)
(f) N n+1 −nT η C 1 λb π x +C 2 λb π x 2
2 2
≈ (−1) e  1−
n (1) (1)

n=1 1 β00
  =  2
. (30)
× 1 − nT ηC4 λb π x 2 (M − K + 1) S (1)

  K −1 Now we simplify the second sum in the denominator of (8):

∞ e−u du
×  , for k = 1, and  > 0, as (31), shown at the top of the next
α
0 1 + nT ηC4 λb π x 2 u − 2 (1−) page.

Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
BAI AND HEATH: ANALYZING UPLINK SINR AND RATE IN MASSIVE MIMO SYSTEMS 4605

(1)∗ (1) 2 (1) −4 (1)∗ (1) (1)∗ (1) 2 (1) −4 (1)∗ (1) 2
E|g00 h0 | = E|u00 | |g00 u00 u00 h0 | = E|u00 | |u00 h0 |
(1)2 (1)2  (1) (1) (1) (1)
(M + M)P β0 + M  = P P β0 β0 +
2 Mσ 2
K P(1) β0
(1)
=  2
(31)
(M 2 + M) S (1)

For k > 1, it follows that [6] H. Yang and T. L. Marzetta, “Performance of conjugate and zero-
  forcing beamforming in large-scale antenna systems,” IEEE J. Sel. Areas
(1) 2 (k) 2 (k) −2 (k)∗ (k) 2
E|g00 | E |h0 | − |u00 | |u00 h0 | Commun., vol. 31, no. 2, pp. 172–179, Feb. 2013.
(1)∗ (k) 2
E|g00 h0 | = [7] N. Krishnan, R. D. Yates, and N. B. Mandayam, “Uplink linear receivers
 M −K +1 
for multi-cell multiuser MIMO with pilot contamination: Large system
(k) (k) (k)∗ (k) analysis,” IEEE Trans. Wireless Commun., vol. 13, no. 8, pp. 4360–4373,
E |h0 |2 − |u00 |−2 |u00 h0 |2 Aug. 2014.
= [8] J. Hoydis, S. ten Brink, and M. Debbah, “Massive MIMO in the UL/DL
(M − K + 1)2 S (1)
 of cellular networks: How many antennas do we need?” IEEE J. Sel.
(k) (k) P(k) β0
(k) Areas Commun., vol. 31, no. 2, pp. 160–171, Feb. 2013.
M P β0 1 − S (k) [9] Y. Li, Y.-H. Nam, B. L. Ng, and J. Zhang, “A non-asymptotic throughput
≈ . for massive MIMO cellular uplink with pilot reuse,” in Proc. IEEE
(M − K + 1)2 S (1) Globecom Conf., Dec. 2012, pp. 4500–4504.
[10] H. Q. Ngo, M. Matthaiou, T. Q. Duong, and E. G. Larsson, “Uplink per-
Then, in the case of k > 1, for  = 0, we approximate the formance analysis of multicell MU-SIMO systems with ZF receivers,”
residue intra-cell interference of ZF receivers as IEEE Trans. Veh. Technol., vol. 62, no. 9, pp. 4471–4483, Nov. 2013.
[11] E. Björnson, E. G. Larsson, and M. Debbah. (2014). “Massive MIMO
α(1−) (k)
(1)∗ (k) MC 1− (λb π) 2 1 for maximal spectral efficiency: How many users and pilots should be
E|g00 h00 |2 ≈  α(1−)
 , allocated?” [Online]. Available: http://arxiv.org/abs/1412.7102
(k)
(M − K +1)2 C 1− (λb π) 2 + 1 S (1) [12] I. Atzeni, J. Arnau, and M. Debbah, “Fractional pilot reuse in mas-
sive MIMO systems,” in Proc. IEEE ICC Workshops, Jun. 2015,
(32) pp. 1030–1035.
[13] P. Madhusudhanan, X. Li, Y. Liu, and T. X. Brown, “Stochastic geo-

 α(1−)
(k)2 − 2 metric modeling and interference analysis for massive MIMO systems,”
which follows from P0(k) β00
(k)
≈ E R00 = in Proc. Modeling Optim. Mobile, Ad Hoc Wireless Netw. (WiOpt),
α(1−) May 2013, pp. 15–22.
(λb π) ; for  > 0, the out-of-cell interference is upper
2
[14] T. Bai and R. W. Heath, “Asymptotic coverage and rate in mas-
bounded (and approximated) as sive MIMO networks,” in Proc. IEEE Global Conf. Signal Inf.
Process. (GlobalSIP), Atlanta, GA, USA, 2014, pp. 602–606.
(k)
(1)∗ (k) M P1 β0 [15] J. G. Andrews, F. Baccelli, and R. K. Ganti, “A tractable approach to
E|g00 h0 |2  . (33) coverage and rate in cellular networks,” IEEE Trans. Commun., vol. 59,
(M − K + 1)2 S (1) no. 11, pp. 3122–3134, Nov. 2011.
[16] H. S. Dhillon, M. Kountouris, and J. G. Andrews, “Downlink MIMO
The noise term in the denominator is HetNets: Modeling, ordering results and performance analysis,” IEEE
(1) σ2 Trans. Wireless Commun., vol. 12, no. 10, pp. 5208–5222, Oct. 2013.
|g00 |2 σ 2 = . (34) [17] V. Chandrasekhar, M. Kountouris, and J. G. Andrews, “Coverage
(M − K + 1)S (1) in multi-antenna two-tier networks,” IEEE Trans. Wireless Commun.,
vol. 8, no. 10, pp. 5314–5327, Oct. 2009.
Next, substituting (23), (26) and (30)-(34) for (8), conditioning [18] Y. Wu, Y. Cui, and B. Clerckx, “Analysis and optimization of
(1)
on R00 = x, and after some algebraic manipulation, the SINR inter-tier interference coordination in downlink multi-antenna HetNets
expression is simplified as with offloading,” IEEE Trans. Wireless Commun., vol. 14, no. 12,
pp. 6550–6564, Dec. 2015.
 α
−1 [19] C. Li, J. Zhang, and K. B. Letaief, “Performance analysis of SDMA
SINR = C6 (λb π x 2 ) 2 (1−) + C7 (λb π x 2 )α(1−) . in multicell wireless networks,” in Proc. Globecom, Dec. 2013,
pp. 3867–3872.
The rest of the proof follows the same line as [20] M. Kountouris and J. G. Andrews, “Downlink SDMA with limited feed-
back in interference-limited wireless networks,” IEEE Trans. Wireless
in Appendix B. Commun., vol. 11, no. 8, pp. 2730–2741, Aug. 2012.
[21] T. D. Novlan, H. S. Dhillon, and J. G. Andrews, “Analytical modeling
of uplink cellular networks,” IEEE Trans. Wireless Commun., vol. 12,
R EFERENCES no. 6, pp. 2669–2679, Jun. 2013.
[1] T. Bai and R. W. Heath, Jr., “Uplink massive MIMO SIR analysis: [22] H. ElSawy and E. Hossain, “On stochastic geometry modeling of
How do antennas scale with users?” in Proc. IEEE Globecom Conf., cellular uplink transmission with truncated channel inversion power
Dec. 2015, pp. 1–6. control,” IEEE Trans. Wireless Commun., vol. 13, no. 8, pp. 4454–4469,
[2] T. L. Marzetta, “Noncooperative cellular wireless with unlimited num- Aug. 2014.
bers of base station antennas,” IEEE Trans. Wireless Commun., vol. 9, [23] S. Singh, X. Zhang, and J. G. Andrews, “Joint rate and SINR
no. 11, pp. 3590–3600, Nov. 2010. coverage analysis for decoupled uplink-downlink biased cell associa-
[3] E. G. Larsson, O. Edfors, F. Tufvesson, and T. L. Marzetta, “Massive tions in HetNets,” IEEE Trans. Wireless Commun., vol. 14, no. 10,
MIMO for next generation wireless systems,” IEEE Commun. Mag., pp. 5360–5373, Oct. 2015.
vol. 52, no. 2, pp. 186–195, Feb. 2014. [24] N. Liang, W. Zhang, and C. Shen, “An uplink interference analysis for
[4] L. Lu, G. Y. Li, A. L. Swindlehurst, A. Ashikhmin, and R. Zhang, massive MIMO systems with MRC and ZF receivers,” in Proc. WCNC,
“An overview of massive MIMO: Benefits and challenges,” IEEE J. Sel. 2015, pp. 310–315.
Topics Signal Process., vol. 8, no. 5, pp. 742–758, Oct. 2014. [25] W. Xiao, R. Ratasuk, A. Ghosh, R. Love, Y. Sun, and R. Nory, “Uplink
[5] F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski, power control, interference coordination and resource allocation for
“Five disruptive technology directions for 5G,” IEEE Commun. Mag., 3GPP E-UTRA,” in Proc. IEEE 64th Veh. Technol. Conf., Sep. 2006,
vol. 52, no. 2, pp. 74–80, Feb. 2014. pp. 1–5.

Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
4606
[26] T. Bai and R. W. Heath, “Asymptotic SINR for millimeter wave massive
MIMO cellular networks,” in Proc. IEEE 16th Int. Workshop Signal
Process. Adv. Wireless Commun. (SPAWC), Stockholm, Sweden, 2015,
pp. 620–624.
[27] A. Goldsmith, Wireless Communications. Cambridge, U.K.: Cambridge
Univ. Press, 2005.
[28] E. Björnson, L. Sanguinetti, and M. Kountouris. (2015). “Deploying
dense networks for maximal energy efficiency: Small cells meet massive
MIMO.” [Online]. Available: http://arxiv.org/abs/1505.01181
[29] F. Baccelli and B. Błaszczyszyn, Stochastic Geometry and Wireless
Networks: Volume I Theory. Boston, MA, USA: NOW Publishers, 2009.
[30] T. Bai and R. W. Heath, Jr., “Coverage and rate analysis for millimeter-
wave cellular networks,” IEEE Trans. Wireless Commun., vol. 14, no. 2,
pp. 1100–1114, Feb. 2015.
[31] 3GPP, “Further advancements for E-UTRA physical layer aspects
(release 9),” 3rd Generat. Partnership Project, Cedex, France,
Tech. Rep. TR 36.814, Mar. 2010.
[32] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
Cambridge, U.K.: Cambridge Univ. Press, 2005.
[33] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, “Aspects of favorable
propagation in massive MIMO,” in Proc. 22nd Eur. Signal Process.
Conf. (EUSIPCO), Sep. 2014, pp. 76–80.
[34] R. Aris, Mathematical Modeling: A Chemical Engineer’s Perspective.
San Diego, CA, USA: Academic, 1999.
[35] H. Alzer, “On some inequalities for the incomplete gamma function,”
Math. Comput., vol. 66, pp. 771–778, Apr. 1997.
[36] M. Kiessling and J. Speidel, “Analytical performance of MIMO zero-
forcing receivers in correlated Rayleigh fading environments,” in Proc.
SPAWC, Jun. 2003, pp. 383–387.
Tianyang Bai (S’11) received the B.Eng. degree
from the Harbin Institute of Technology, Harbin,
China, in 2011, and the M.S.E. and Ph.D. degrees
from The University of Texas at Austin, Austin, TX,
USA, in 2013 and 2016, respectively, all in electrical
engineering. He is currently a Senior System Engi-
neer with Qualcomm Flarion Technologies, Inc.,
Bridgewater, NJ, USA. His research interests include
applications of stochastic geometry, mmWave com-
munications, and massive MIMO.
Authorized licensed use limited to: Nanjing University. Downloaded on January 12,2021 at 00:26:06 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 11, NOVEMBER 2016
Robert W. Heath, Jr. (S’96–M’01–SM’06–F’11)
received the B.S. and M.S. degrees from the Univer-
sity of Virginia, Charlottesville, VA, USA, in 1996
and 1997, respectively, and the Ph.D. degree from
Stanford University, Stanford, CA, USA, in 2002,
all in electrical engineering. From 1998 to 2001,
he was a Senior Member with the Technical Staff
and a Senior Consultant with Iospan Wireless Inc.,
San Jose, CA, USA, where he worked on the design
and implementation of the physical and link layers of
the first commercial MIMO-OFDM communication
system. Since 2002, he has been with the Department of Electrical and
Computer Engineering, The University of Texas at Austin, where he is
currently a Cullen Trust for Higher Education Endowed Professor and a
member of the Wireless Networking and Communications Group. He is also
the President and CEO of MIMO Wireless Inc., and the Chief Innovation
Officer of Kuma Signals LLC. His research interests include several aspects of
wireless communication and signal processing: limited feedback techniques,
multihop networking, multiuser and multicell MIMO, interference alignment,
adaptive video transmission, manifold signal processing, and millimeter wave
communication techniques. He has co-authored the book entitled Millimeter
Wave Wireless Communications (Prentice Hall, 2014).
He has been an Editor for the IEEE T RANSACTIONS ON C OMMUNICATION,
an Associate Editor for the IEEE T RANSACTIONS ON V EHICULAR T ECH -
NOLOGY , and a Lead Guest Editor for the IEEE J OURNAL ON S ELECTED
A REAS IN C OMMUNICATIONS of the Special Issue on Limited Feedback
Communication, and for the IEEE J OURNAL ON S ELECTED T OPICS IN
S IGNAL P ROCESSING of the Special Issue on Heterogenous Networks. He was
on the Steering Committee for the IEEE T RANSACTIONS ON W IRELESS
C OMMUNICATIONS from 2011 to 2014. He was a member of the Signal
Processing for Communications Technical Committee in the IEEE Signal
Processing Society and was the Chair of the IEEE COMSOC Communications
Technical Theory Committee. He was the Technical Co-Chair for the 2007
Fall Vehicular Technology Conference, the General Chair of the 2008 Com-
munication Theory Workshop, the General Co-Chair, Technical Co-Chair, and
Co-Organizer of the 2009 IEEE Signal Processing for Wireless Communica-
tions Workshop, the Local Co-Organizer for the 2009 IEEE CAMSAP Con-
ference, the Technical Co-Chair for the 2010 IEEE International Symposium
on Information Theory, the Technical Chair for the 2011 Asilomar Conference
on Signals, Systems, and Computers, the General Chair for the 2013 Asilomar
Conference on Signals, Systems, and Computers, the Founding General
Co-Chair for the 2013 IEEE GlobalSIP conference, and the Technical
Co-Chair for the 2014 IEEE GLOBECOM conference.
Dr. Heath was a co-author of best paper awards at the IEEE VTC 2006
Spring, WPMC 2006, the IEEE GLOBECOM 2006, the IEEE VTC 2007
Spring, the IEEE RWS 2009, and the IEEE GLOBECOM 2015, and a
co-recipient of the Grand Prize in the 2008 WinTech WinCool Demo
Contest. He was a co-recipient of the 2010 and 2013 EURASIP Journal
on Wireless Communications and Networking Best Paper Awards, the 2012
Signal Processing Magazine Best Paper Award, a 2013 Signal Processing
Society Best Paper Award, the 2014 EURASIP Journal on Advances in
Signal Processing Best Paper Award, the 2014 Journal of Communications
and Networks Best Paper Award, the 2016 IEEE Communications Society
Fred W. Ellersick Prize, and the 2016 IEEE Communications Society and
Information Theory Society Joint Paper Award. He was a Frontiers in
Education New Faculty Fellow in 2003. He is also a Licensed Amateur Radio
Operator and a Registered Professional Engineer in Texas.