You are on page 1of 10


qxp_Layout 1/30/14 1:08 PM Page 186


Massive MIMO for Next Generation
Wireless Systems
Erik G. Larsson, ISY, Linköping University, Sweden
Ove Edfors and Fredrik Tufvesson, Lund University, Sweden
Thomas L. Marzetta, Bell Labs, Alcatel-Lucent, United States

ABSTRACT up on our earlier exposition [1], with a focus on
the developments in the last three years; most
Multi-user MIMO offers big advantages over particularly, energy efficiency, exploitation of
conventional point-to-point MIMO: it works excess degrees of freedom, time-division duplex
with cheap single-antenna terminals, a rich scat- (TDD) calibration, techniques to combat pilot
tering environment is not required, and resource contamination, and entirely new channel mea-
allocation is simplified because every active ter- surements.
minal utilizes all of the time-frequency bins. With massive MIMO, we think of systems
However, multi-user MIMO, as originally envi- that use antenna arrays with a few hundred
sioned, with roughly equal numbers of service antennas simultaneously serving many tens of
antennas and terminals and frequency-division terminals in the same time-frequency resource.
duplex operation, is not a scalable technology. The basic premise behind massive MIMO is to
Massive MIMO (also known as large-scale reap all the benefits of conventional MIMO, but
antenna systems, very large MIMO, hyper on a much greater scale. Overall, massive MIMO
MIMO, full-dimension MIMO, and ARGOS) is an enabler for the development of future
makes a clean break with current practice broadband (fixed and mobile) networks, which
through the use of a large excess of service will be energy-efficient, secure, and robust, and
antennas over active terminals and time-division will use the spectrum efficiently. As such, it is an
duplex operation. Extra antennas help by focus- enabler for the future digital society infra-
ing energy into ever smaller regions of space to structure that will connect the Internet of people
bring huge improvements in throughput and and Internet of Things with clouds and other
radiated energy efficiency. Other benefits of network infrastructure. Many different configu-
massive MIMO include extensive use of inexpen- rations and deployment scenarios for the actual
sive low-power components, reduced latency, antenna arrays used by a massive MIMO system
simplification of the MAC layer, and robustness can be envisioned (Fig. 1). Each antenna unit
against intentional jamming. The anticipated would be small and active, preferably fed via an
throughput depends on the propagation environ- optical or electric digital bus.
ment providing asymptotically orthogonal chan- Massive MIMO relies on spatial multiplexing,
nels to the terminals, but so far experiments which in turn relies on the base station having
have not disclosed any limitations in this regard. good enough channel knowledge, on both the
While massive MIMO renders many traditional uplink and the downlink. On the uplink, this is
research problems irrelevant, it uncovers entirely easy to accomplish by having the terminals send
new problems that urgently need attention: the pilots, based on which the base station estimates
challenge of making many low-cost low-precision the channel responses to each of the terminals.
components that work effectively together, The downlink is more difficult. In conventional
acquisition and synchronization for newly joined MIMO systems such as the Long Term Evolu-
terminals, the exploitation of extra degrees of tion (LTE) standard, the base station sends out
freedom provided by the excess of service anten- pilot waveforms, based on which the terminals
nas, reducing internal power consumption to estimate the channel responses, quantize the
achieve total energy efficiency reductions, and thus obtained estimates, and feed them back to
finding new deployment scenarios. This article the base station. This will not be feasible in mas-
presents an overview of the massive MIMO con- sive MIMO systems, at least not when operating
cept and contemporary research on the topic. in a high-mobility environment, for two reasons.
First, optimal downlink pilots should be mutually
GOING LARGE: MASSIVE MIMO orthogonal between the antennas. This means
that the amount of time-frequency resources
Massive multiple-input multiple-output (MIMO) needed for downlink pilots scales with the num-
is an emerging technology that scales up MIMO ber of antennas, so a massive MIMO system
by possibly orders of magnitude compared to the would require up to 100 times more such
current state of the art. In this article, we follow resources than a conventional system. Second,

186 0163-6804/14/$25.00 © 2014 IEEE IEEE Communications Magazine • February 2014

rate) transmitted per Joule per terminal receiv. may come at the cost of more the nearly noise-limited regime of information transmitted power. Interference and energy cost of using pilots to acquire chan- between terminals can be suppressed even fur. The reason that from the terminals to the base station (the down. This means providing each terminal with More quantitatively. The reason that small regions in space (Fig. when using ciency in terms of the total number of bits (sum MRC and operating in the “green” regime (i. b/s/Hz). efficiency). zero-forcing (ZF).qxp_Layout 1/30/14 1:08 PM Page 187 the number of channel responses each terminal must estimate is also proportional to the number Distributed of base station antennas. The fundamental principle that makes a distributed fashion. With the MRC receiver. When serving a single terminal (purple) operating in the 1 b/dimension/terminal regime. l rica ind THE POTENTIAL OF MASSIVE MIMO Cyl Massive MIMO technology relies on phase- coherent but computationally very simple pro- cessing of signals from all the antennas at the base station.. Hence. basic testbeds are gul ar becoming available [3]. the solution is to operate in TDD mode. The underlying MRC works so well for massive MIMO is that physics is coherent superposition of wavefronts.. • A system with 100 antennas serving a single there is also some evidence that intersymbol terminal using conventional beamforming interference can be treated as additional thermal (green) noise [7]. from hardware imperfections tend to be over- The figure illustrates the relation for the uplink. Fig. whelmed by the thermal noise. independently at each the dramatic increase in energy efficiency possible antenna unit. hence offering a way of disposing with • A massive MIMO system with 100 antennas orthogonal frequency-division multiplexing simultaneously serving multiple (about 40 (OFDM) as a means of combatting intersymbol here) terminals (red. and blue. The figure shows times higher than in conventional MIMO is that the trade-off for three cases: many tens of terminals are served simultaneous- • A reference system with one single antenna ly. ear link channels. that all wavefronts collectively emitted by all The prediction in Fig. and Lin rely on reciprocity between the uplink and down. The capacity increase results from the aggressive spatial multiplexing used in massive responses — but also that it can be performed in MIMO. stimulating much research particularly in random matrix theory Rec tan and related mathematics.e. consider an array consist- ing (MRC) compared with ZF is not only its ing of 6400 omnidirectional antennas (total form computational simplicity — multiplication of the factor 6400 × (l/2)2  40 m2) transmitting with a received signals by the conjugate channel total power of 120 W (i. and initial channel mea- surements have been performed [4. MRC generally does not. multiuser interference and effects mitted per unit of radio spectrum consumed. To understand the scale of the capacity gains The attractiveness of maximum ratio combin. radiated energy efficiency on the order of 100 times. 2. as illustrated in Fig. 5]. intracell interference. the base station can make sure number of base station antennas is large. the overall spectral efficiency still can be 10 link performance is similar). 3 is based on an infor- antennas add up constructively at the locations mation-theoretic analysis that takes into account of the intended terminals. theory. but destructively (ran. In a massive MIMO system. Generally. and spectral efficiency without seriously affecting the overall spectral in terms of total number of bits (sum rate) trans.EDFORS_LAYOUT. using ZF). Some possible antenna configurations and deployment scenarios for times or more and simultaneously improve the a massive MIMO base station. energy a conventional or moderately sized MIMO sys- can be focused with extreme sharpness into tem. each antenna radiat- IEEE Communications Magazine • February 2014 187 . using maximum ratio interference. scaling down the power as much as possible ing service of energy spent. the uplink resources needed to inform the base station of the channel responses would be up to 100 times larger than in conventional systems. the channel responses associated with different By appropriately shaping the signals sent out by terminals tend to be nearly orthogonal when the the antennas. combining. as well as the bandwidth domly) almost everywhere else. nel state information in a high-mobility environ- ther by using. 2). 3 (from [6]) depicts a rate of about 1 b/complex dimension (1 the fundamental trade-off between energy effi. however. ment [6]. we operate in This. for example. Some specific benefits of a massive MU-MIMO system are: •Massive MIMO can increase the capacity 10 Figure 1. While ZF also works fairly well for is that with a large number of antennas. although frequency-division duplext (FDD) operation may be possible in cer- tain cases [2]. While the concepts of massive MIMO have been mostly theoretical so far. in the same time-frequency resource. massive MIMO offers.e.

With massive MIMO. The array serves 1000 fixed minals and therefore equal throughput. allowing an alternate view of the spatial focusing. and the height of the ter. the overhead of pilot transmis. 188 IEEE Communications Magazine • February 2014 . gy with regard to theory.EDFORS_LAYOUT. Relative field strength around a target terminal in a scattering environment of size 800 l × 800 l when the base station is placed 1600 l to the left. The receivers have a 9 Mb/s broadband service to each of 1000 homes. for example. sufficient accuracy. resulting in a range. and it is assumed that the chan. Overall.52. tions and the shadow fading shows that 95 per- ing an 8 dB gain antenna. which confers an equal signal-to-inter- the personal communications services (PCS) ference-plus-noise ratio on each of the 950 ter- band (1900 MHz). tation.2 Mb/s/terminal. This There is also log-normal shadow fading with 8 would be enough. example will offer the 1000 terminals a total we find that the path loss is 127 dB at 1 km downlink throughput of 20 Gb/s. vice simultaneously to 950 terminals. and the range-decay exponent is 3. Left: pseudo-color plots of average field strengths. Massive MIMO is a game changing technolo- which accounts for receiver noise. We use optimal max-min power systems are replaced by hundreds of low-cost 5 100-element (dB) λ/2-spaced 0 linear array 5 0 800 λ [dB] Narrow beam -5 -5 -10 400 λ -10 ≤ -15 400 λ Area with 400 random scatterers ≤ −15 Incoming narrow beam 1600 λ 800 λ a) MRT precoding 5 100-element (dB) λ/2-spaced 0 linear array 5 0 800 λ [dB] Wide beam -5 -5 -10 400 λ -10 ≤ -15 400 λ Area with 400 random scatterers ≤ 15 Incoming wide beam 1600 λ 800 λ b) ZF precoding Figure 2. are excluded from service. 21.qxp_Layout 1/30/14 1:08 PM Page 188 ing about 20 mW) over a 20 MHz bandwidth in control. counterpart MRT precoding (for the downlink) forming combined with power control. Other types nel estimation. with target user posi- tions at the center () and four other users nearby (). dB noise figure. One-quarter of the time is spent The max-min power control provides equal ser- on transmission of uplink pilots for TDD chan.000 random placements of 400 scatterers when two different linear precoders are used: a) MRT precoders. Average field strengths are calculated over 10. low-power components. channel esti. Downlink data is transmitted The MRC receiver (for the uplink) and its via maximum ratio transmission (MRT) beam. terminals randomly distributed in a disk of radius Numerical averaging over random terminal loca- 6 km centered on the array. expensive sion. systems. and the imperfections of MRT ultra-linear 50 W amplifiers used in conventional beamforming. Right: average field strengths as surface plots. and implemen- mation errors. the array in this minals is 5 m. to provide 20 dB standard deviation. each terminal hav. where the are also known as matched filtering (MF) in the 5 percent of terminals with the worst channels literature. b) ZF precoders. near/far effects and power control. We use a capacity • Massive MIMO can be built with inexpen- lower bound from [8] extended to accommodate sive. of power control combined with time-division nel is substantially constant over intervals of 164 multiplexing could accommodate heterogeneous ms in order to estimate the channel gains with traffic demands of a larger set of terminals. The height of the cent of the terminals will receive a throughput of antenna array is 30 m. Using the Hata-COST231 model. slow fading. sum-spectral efficiency of 1000 b/s/Hz.

ZF processing MRC processing accuracy and linearity of each individual amplifi. In particular. All that matters is their com- bined action. systems is normally limited by fading. based platforms for a few thousand dollars. •Massive MIMO simplifies the multiple access straints. which a total output RF power two orders of magni. using extremely simple penalty in terms of increased total radiated signal processing. With OFDM. The techniques in [9. and tially less electromagnetic interference. no beams are render the received signal strength very small at formed. MIMO relies on the law of large numbers and nals see. the effective channels layer. These degrees of freedom Spectral efficiency (b/s/Hz) can be used for hardware-friendly signal shap- ing. Fading can envelope multiuser precoding. paths before it reaches the terminal. renders most of the physical layer control sig- tude less than with current technology. waveforms satisfy the desired envelope con. and hence easily deployed where tems is a growing concern and a serious cyber- no electricity grid is available. facilitates the use of extremely cheap and power- efficient RF amplifiers. and the public. Notwithstanding. naling redundant. tion of latency on the air interface. In particular. With (near) constant. components can be put beamforming in order to avoid fading dips. It is this fading that makes it nals we want them to see. 102 er and RF chain. is significant. Simple jammers can be bought off therefore the base station will generate substan. Half the power — twice the force (from [6]): Improving uplink spec- nals with very small peak-to-average ratio [9] or tral efficiency 10 times and simultaneously increasing the radiated power effi- even constant envelope [10] at a very modest ciency 100 times with massive MIMO technology. more than 4 cm in diameter!) 100 antennas. Figure 3. of magnitude less power could be powered by Intentional jamming of civilian wireless sys- wind or solar. This distinction The performance of wireless communications is explained in Fig. If the property of the massive MIMO channel that terminal is trapped in a fading dip. a from a base station travels through multiple wavefield is created such that when this wave. and hardware imperfections aver. (The typical coaxial cables 103 Relative energy efficiency (b /J)/(b/J) used for tower-mounted base stations today are 100 antennas. with 200 10–1 antennas serving 20 terminals. can be eliminated altogether. because the energy consumption of cellular •Massive MIMO increases the robustness base stations is a growing concern worldwide.EDFORS_LAYOUT. Such (near-constant) envelope signaling obtaining channel state information. Rather. As a bonus. base stations that consume many orders and intentional jamming.qxp_Layout 1/30/14 1:08 PM Page 189 amplifiers with output power in the milli-Watt range. and the field is sampled at the spots where the terminals waves resulting from these multiple paths inter- are located. Due to the law of large numbers. each antenna can transmit sig. fere destructively. This mat. 101 age out when signals from a large number of 100 antennas. This happens when the signal sent are not formed by weighing a symbol. Several expensive and bulky items. The fundamental hard to build low-latency wireless links. Massive this nullspace without affecting what the termi. such as large coaxial cables. 4. put together using off-the-shelf software radio- ing electromagnetic exposure. antenna unit(s). The contrast to classical array designs. between the base station and each of the termi. single terminal A massive MIMO system has a large surplus of degrees of freedom. This is equipment that used to be military-grade can be important due to the increased concerns regard. Each ter- enables massive MIMO systems to operate with minal can be given the whole bandwidth. weight beamforming techniques. massive MIMO relies on the law of large numbers to make sure that noise. so into this nullspace that make the transmitted fading no longer limits latency. In a way. Massive MIMO reduces the constraints on multiple terminals. The same property that makes massive MIMO resilient against fading also makes the technology 100 extremely robust to failure of one or a few of the Single antenna. the Internet for a few hundred dollars. subcarrier in a massive MIMO system will have The drastically improved energy efficiency substantially the same channel gain. fading. IEEE Communications Magazine • February 2014 189 . ters. For example. single terminal antennas are combined in the air. 180 degrees of 0 10 20 30 40 50 60 70 80 90 freedom are unused. taking into account the energy and bandwidth costs of power. the security threat that seems to be little known to total emitted power can be dramatically cut. multiple terminals. In against both unintended man-made interference addition. the terminals see precisely the sig. each (PSK) modulation. it has to wait makes this possible is that the channel has a until the propagation channel has sufficiently large nullspace: almost anything can be put into changed until any data can be received. and the signals emitted by each antenna certain times. 10] must not be confused with conventional •Massive MIMO enables a significant reduc- beamforming techniques or equal-magnitude. the chan- nals can take any signal constellation as input nel hardens so that frequency domain schedul- and do not require the use of phase shift keying ing no longer pays off. 104 which use few antennas fed from high-power amplifiers.

channel estimation. [3] treats reciprocity cal- jammer located in a nearby apartment. Sweden. every terminal in a massive MIMO sys- tion is affected by materials with strange mag. obtain the full beamforming gains of massive so the only way of improving robustness of MIMO: if the base station equipment is properly wireless communications is to use multiple calibrated. especially in public does not seem to constitute a serious problem. ibration for a 64-antenna system in some detail ing critical phases of riots. Massive MIMO offers many excess coherent beam to the terminal. In 190 IEEE Communications Magazine • February 2014 .) Absolute calibration within the array power. smart jammers could cause the overhead for these supplementary pilots is harmful interference with modest transmission very small. channel reciprocity. antenna and each of the other antennas to derive a compensation factor for that antenna. Calibration of the hardware chains interval divided by the channel delay spread. It may be possible to entirely forgo reciprocity calibra- LIMITING FACTORS OF tion within the array. Numerous recent incidents. Instead. but this can be handled by trans- MIMO is implemented using uplink pilots for mitting pilots through the beam to the terminal. for example if the maxi- MASSIVE MIMO mum phase difference between the uplink and downlink chains were less than 60˚. should be able to substantially diminish that and signals can be traded between the reference problem. If massive the terminal. the maximum number of in the base station and terminal transceivers may orthogonal pilot sequences that can exist is not be reciprocal between the uplink and the upper-bounded by the duration of the coherence downlink. in 2001. more clever implementations is not required. During the EU summit in Gothen. Ideally. and dur. Conventional MIMO beamforming contrasted with per-antenna constant envelope transmission in massive MIMO. tem is assigned an orthogonal uplink pilot netic properties. where the signal emitted by each antenna has a large dynamic range. coherent CHANNEL RECIPROCITY beamforming would still occur (at least with MRT beamforming). have already been tested to some extent in prac- burg. Specifically. the chief commander and claims a successful experimental implemen- could not reach any of the 700 police officers tation. However. However. illustrate the magnitude of and there are calibration-based solutions that the problem. albeit with a possible 3 dB Time-division duplexing operation relies on reduction in gain. Note that calibration of the terminal uplink Due to the scarcity of bandwidth. one using joint channel estimation and decoding of the antennas can be treated as a reference.H) uk αm envelope uk envelope e jφ m uK uK αM e jφ M fc fc Figure 4. engaged [11]. demonstrators used a tice [3. the array will indeed transmit a antennas. as proposed in [3]. where each antenna sends out a signal with a constant envelope. Left: con- ventional beamforming. 12]. There appears to be a rea- sonable consensus that the propagation channel PILOT CONTAMINATION itself is essentially reciprocal unless the propaga. safety applications. Right: per-antenna constant envelope transmission. spreading and downlink chains is not required in order to information over frequency just is not feasible. the hardware chains sequence. (There will still degrees of freedom that can be used to cancel be some mismatch within the receiver chain of signals from intentional jammers. However.EDFORS_LAYOUT.qxp_Layout 1/30/14 1:08 PM Page 190 Per-antenna Per-antenna varying constant envelope envelope α1 e jφ 1 u1 u1 Phase-only encoder Combined Combined Linear encoder α = W (H)u varying varying φ = f (u.

Similar inter. rate as the desired signal [13]. RADIO PROPAGATION AND ORTHOGONALITY OF mum number of orthogonal pilot sequences in a CHANNEL RESPONSES Massive MIMO (and 1 ms coherence interval is estimated to be about especially MRC/MRT 200. effect on massive MIMO appears to be much This array consists of 16 dual-polarized patch more profound than in classical MIMO [13. polarized circular array. environment called specifically. The most and largest singular values for the different mea- promising direction seems to be blind techniques sured (narrowband) frequency points in the dif- that jointly estimate the channels and the pay. While this argu. To study the behavior of massive favorable propaga- its received pilot signal with the pilot sequence MIMO systems. This directed interference grows with also true for physically smaller arrays with the number of service antennas at the same directional antenna elements pointing in sufficiently different. array also provides the possibility to resolve scat- out bound. channel responses to different terminals are is to sive frequency reuse factor for the pilots (but look at the spread between the smallest and not necessarily for the payload data). at least with receivers that rely on terers at different elevations. largest singular values of the matrix that contains This pushes mutually contaminating cells farther the channel responses. this when the number of antennas is increased with. Simply stated. 3 or 7.i. Figure 6 illustrates this apart. may mitigate or elimi. On the left is a compact circu- not really specific to massive MIMO. Practical number is a measure of the fading margin. which requires estimates of the actual measurements. this is ferent terminals are of data. serve all users with a reasonable received signal IEEE Communications Magazine • February 2014 191 . Cur. often used account the network structure. the median of the ination precoding requires only the corre. say. antenna elements arranged in a circle. The center frequency was 2. the pilot contamination precoding remains to be additional power that has to be used in order to developed. channel sounder was employed. As a reference. the directed interfer. worse resolution in azimuth due to its limited ment has been contested recently [15]. single-polarized linear array or a compact dual- •Clever channel estimation algorithms [15].EDFORS_LAYOUT. share the same pilot sequence. it actual. This can tions in an otherwise static environment to emu- be done in several ways: late a real array with the same dimensions. but its lar massive MIMO array with 128 antenna ports. the RUSK Lund ence that results from pilot contamination. One possibility is to use a less aggres. ferent cases. the maxi. a large extent on a The effect of reusing pilots from one cell to tion. channel measurements have to associated with a particular terminal.) channel matrices.qxp_Layout 1/30/14 1:08 PM Page 191 [13]. To the right is a physically large linear under some specific assumptions on the power (virtual) array. Of course. when the service array correlates ly different. at least aperture. The measurements were contamination precoding [18]. at least partially. the or even blind techniques that circumvent the use figure shows the cumulative density function of pilots altogether [17]. configured as either a physically large rently. be performed using realistic antenna arrays. with 4 In [13] it was argued that pilot contamination such circles stacked on top of each other. Simply stated. singular value spread is about 23–18 dB. the optimal strategy is unknown.d. •The allocation of pilot waveforms can be One way of quantifying how different the optimized. performed outdoors in the Lund University cam- erative transmission over a multiplicity of cells pus area. This tion. it appears likely that pilot contami. This sponding slow-fading coefficients. antenna element is moved to 128 different posi- nation must be dealt with in some way. can utilize coop. More base station to different terminals are sufficient. we also show load data. It is also possible to coordinate the use of for a case with 4 user terminals and a base sta- pilots or adaptively allocate pilot sequences to tion having 4. MIMO arrays used for the measurements report- Pilot contamination as a basic phenomenon is ed in this article. ly obtains a channel estimate that is contami. tively. is so because the channel behavior using large favorable propaga- nated by a linear combination of channels with arrays differs from that usually experienced other terminals that share the same pilot using conventional smaller arrays. (CDF) of the difference between the smallest nate the effects of pilot contamination. Besides constitutes an ultimate limit on performance having the advantage of being compact. The first results from the cam- channels between the terminals and the service paign were presented in [4]. Even partially various directions. simulated results for ideal independent identical- •New precoding techniques that take into ly distributed (i. Downlink beamforming based on important differences are that: propagation channel the contaminated channel estimate results in • There might be large-scale fading over the responses from the interference directed at those terminals that array. 14]. that the propagation channel responses from the property of the radio quences is termed pilot contamination. the radio environment called favorable propaga. where a single omnidirectional control used.6 GHz and — outside of the beamforming operation — to the measurement bandwidth 50 MHz. using the cylindrical array. favorable propagation means another and the associated negative conse. while a network Unlike coordinated beamforming over multiple analyzer was used for the synthetic linear array cells. 32. respec- the different terminals in the network [16]. It is easy to exhaust the available supply of Massive MIMO (and especially MRC/MRT pro- orthogonal pilot sequences in a multicellular sys. arrays of the contaminating cells. For the 4-element array. More specifically. When nullify. • The small-scale signal statistics may also base station to dif- ference is associated with uplink transmissions change over the array. correlated pilot sequences result in directed Figure 5 shows pictures of the two massive interference. for a typical operating scenario. and 128 antenna ports. cessing) relies to a large extent on a property of processing) relies to tem. pilot contam. such as pilot in theoretical studies. but it suffers from pilot-based channel estimation. The most tion means that the sequence.

It should be mentioned here that when using 0. due to the fact 0. The shaded areas in red (for spread larger than 3 dB anywhere over the mea. it seems that the large array is able to separate the users to a reasonable extent using the different Figure 6. With the massive linear array. ment campaign is described and analyzed. The m between users at a distance of about 40 m slow decrease can. tion due to the enhanced spatial resolution.i. 192 IEEE Communications Magazine • February 2014 . be attributed from the base station) in a non line-of-sight to the shadow fading occurring across the arrays: (NLOS) scenario when using MRT as precoding. measured channels with linear and cylindrical array structures at the base sta. As before. another aspect worth mentioning is that 0 10 20 30 40 50 Singular value spread (dB) for a very tricky propagation scenario.qxp_Layout 1/30/14 1:08 PM Page 192 Figure 5. The compact circular array has inferior 1 performance compared to the linear array due to 0.i. the convergence to the i. at least partially.d. while the other two cases are would not be possible with conventional MIMO. see [4] for further details. In addition. but slowly for the measured channels. The performance of the physi- cally large array approaches that of the theoreti- cal i.6 horizontal angle. and for the average. 32. the spread As can be seen in Fig. This oretical i. we plot in Fig. by directive antenna elements pointing in the wrong direction.EDFORS_LAYOUT. but still linear. Also.d. channel is shown as a reference. 7. the variance of the sum and antenna configuration.5 butions to the sum rate in this scenario. where another outdoor measure- channel for four base stations.8 its directive antenna elements sometimes point- ing in the wrong direction. the interference-free signal-to-noise. note that none of approaches that of the theoretical interference- the curves has any substantial tail.i. Massive MIMO antenna arrays used for the measurements.i. vations in [5]. The the.d. Linear array nel performance is faster and the variance of the 0. cylindical array in the form of shadowing caused ratio at the terminals is 10 dB.9 its smaller aperture — it cannot resolve the scat- terers as well as the physically large array — and 0.2 i. the linear array) and blue (for the circular array) sured bandwidth is essentially negligible. Note that the curve for the linear array coincides with that of the i.3 coding methods such as ZF or minimum mean square error.4 somewhat more complex. CDF of the singular value spread for MIMO systems with 4 terminals spatial signatures the users have at the base sta- and three different numbers of base station antennas: 4. and 128. 7 the rate decreases as the number of antennas increas- sum rate for 4 closely spaced users (less than 2 es. the sum rate is less than 3 dB.7 that most of the scatterers are seen at the same 0.d. This means free case as the number of antennas at the base that the probability of seeing a singular value station increases.i.1 Cylindrical array sum rate lower as the number of base station # base station antennas: 4 32 128 antennas is increased. such as closely spaced users in line-of-sight conditions. 0 Also. pre- 0. channel coeff. These conclusions are also in line with the obser- tion. for the linear array in the form of shadowing by The transmit power is normalized so that on external objects along the array. power. shows the 90 percent confidence intervals of the To further illustrate the influence of different sum rates for the different narrowband frequency numbers of antenna elements at the base station realizations. case when the number of antennas grows large. the possibility to resolve scatters at different elevations gives only marginal contri- CDF 0. chan- 0.d.

e. when using massive MIMO instead of conven- The challenge of low-cost hardware: Building tional MIMO. probably some form of space-time block coding IEEE Communications Magazine • February 2014 193 . in terms of time. haps dedicated. research must be invested in highly parallel. It is also important to Hardware impairments: Massive MIMO develop more sophisticated analytical channel relies on the law of large numbers to average out models. massive MIMO must be built with ture the essential behavior of the channel. the total tion has no way of knowing the channel response power consumed must be considered. phase noise may it? What is the cost. Such models need not necessarily be noise. this is good in many cases (Fig.i. There is great tamination imposes much more severe limita- potential to get around the phase noise problem tions on massive MIMO than on traditional by design of smart transmission physical layer MIMO systems. with about 10 times more base station antennas than the num- Sum rate (b/s/Hz) 10 ber of users. However. How often must low-cost phase locked loops or even free-running this be done. and what is the best way of doing oscillators at each antenna. is feasible Cost of reciprocity calibration: TDD will with a large excess of transmitter antennas. However. We have discussed some of the schemes and receiver algorithms. A similar very relaxed linearity requirements will necessi. it analog-to-digital (A/D)-digital-to-analog (D/A) is necessary to have channel models that reflect converters. This model is not an exact representation power-efficient A/D converters yield higher lev. 3). interference. quency resources needed to do the calibration. However. in practice. there is enormous nal processing. but they must cap- In reality. Much research needs be invested in the design of optimized algorithms and their imple. up/down converters. This means that no array beam- the cost of baseband signal processing. interval While massive MIMO renders many traditional 2 i. Linear array Cylindrical array it uncovers entirely new problems that need 0 research. channel coefficient problems in communication theory less relevant. Power amplifiers with types of analysis despite its limitations.qxp_Layout 1/30/14 1:08 PM Page 193 Overall.and fre- become a limiting factor. it seems that it is possible to get stable performance not far from the theoretically ideal performance also under what are normally 8 considered very difficult propagation conditions. Figure 7. Some Channel characterization: There are addi- examples of recent work in this direction include tional properties of the channel to consider [19]. This is likely to mean that example. Pilot contamination: It is likely that pilot con- bol is received at each antenna. Number of base station antennas ing: Massive MIMO arrays generate vast amounts of baseband data that must be pro. be simple. signaling. what ulti. fading and to some extent. including to the terminal. the scale in manufacturing comparable to what we propagation channel including effects of realistic have seen for mobile handsets. and so forth. To facilitate a realistic perfor- hundreds of RF chains. will require economy of the true behavior of the radio channel (i. way of thinking could probably be adopted for tate the use of per-antenna low peak-to-average massive MIMO channel modeling. With require reciprocity calibration. mately matters is how much the phase will drift and in terms of additional hardware components between the point in time when a pilot symbol is needed? received and the point in time when a data sym. mance assessment of massive MIMO systems. 6 MASSIVE MIMO: A GOLD MINE OF 4 RESEARCH PROBLEMS Average and 90 percent conf.8 b/s/Hz practice. there is compelling evidence that the assumptions on favorable propagation underpin. Depending on the exact configuration 14 of the large array and the precoding algorithms used. per. For low-cost components.d. issues in detail and outlined some of the most Internal power consumption: Massive MIMO relevant research directions earlier. potential for ingenious precoding schemes. 16 ning massive MIMO are substantially valid in Interference free → 13. as already noted. Much forming gain can be harnessed.. Achieved downlink sum rates using MRT precoding. which. in conventional MIMO the Kronecker hardware imperfections are larger: in particular. hardware for the baseband sig- mentation. of reality.EDFORS_LAYOUT. and simple means linear or nearly lin- ear. correct in every fine detail. 0 20 40 60 80 100 120 Fast and distributed coherent signal process. Low-cost and tion. offers the potential to reduce the radiated power Non-CSI@TX operation: Before a link has 1000 times and at the same time drastically scale been established with a terminal. In this case. Fundamentally. model is widely used to model channel correla- phase noise and I/Q imbalance. the convergence toward the ideal perfor- mance may be faster or slower as the number of 12 antennas is increased. with 4 single- cessed in real time. On the downlink. antenna arrangements). the base sta- up data rates. but provides a useful model for certain els of quantization noise. This processing will have to antenna terminals and between 4 and 128 base station antennas.

realization of dis- energy efficiency. Aug. could provide the backhaul for base stations that ciency. G. and SSF for their funding of ering factors like mobility. vol. rural areas. no.. no.. pp.” Proc. vol. vol. Scale Antenna Systems. 2013. WARP platform). 2012. vol. 303–13. nologies. wireless network appears to be the availability of doctoral student at Lund University. 46th Annual Asilo- sively determine which approach is superior. Istanbul.” IEEE JSAC. Hoydis et al.” IEEE Sig.” IEEE Int’l. Q. One possibility is to intro- duce dedicated applications of massive MIMO concept with even technology that do not require backward com- CONCLUSIONS AND OUTLOOK larger numbers of patibility. S. 1. transmission of multiple individual base stations. robustness. tion with Alcatel-Lucent. pp. Signals. The technology issues that urgently this application. Commun. hardware is possible with wireless technology. mar Conf. Jan. spectral efficiency. France. and shows the basic vol. phase parts of this work. a billboard-sized array could provide potential of massive MIMO systems as a key er potentially new 20 Mb/s service to each of 1000 homes using enabling technology for future beyond fourth special equipment that would be used solely for generation (4G) cellular systems.” 46th Annual Conf. Feb. a competitor to LTE. 276–79. the testbed shows that TDD operation relying [9] C. One of the Multi-User MIMO-OFDM Downlink.. Alternatively. and E. The Safety Applications. ment base station array to serve up to 18 users Continued testbed ed and sent a pilot.EDFORS_LAYOUT. Apr. basic prototyping work Spectral Efficiency of Very Large Multiuser MIMO Systems. 61. CA. Very-Large MIMO at 2. 2012. tions ELLIIT. “Channel Measurements for Large will comprise a combination of all available tech. Nam et al. “Argos: Practical Many-Antenna Base small cell system are also likely to be very differ. ly and drastically increase data rates and overall computational complexity. vol. Mohammed. pp. full potential of the technology. Proc. for her good enough channel state information (CSI) to analysis of the channel measurements presented facilitate phase-coherent processing at multiple in Fig. “Performance of Conjugate and feasibility of the massive MIMO concept using Zero-Forcing Beamforming in Large-Scale Antenna Sys- 64 coherently operating antennas. Doppler shifts.” ACM Int’l.. [5] J. G.. heterogeneous network solutions: The driving There are still challenges ahead to realize the motivation of massive MIMO is to simultaneous. Larsson. 2012. a distributed array or Mag. even larger numbers of antennas and discover both prove the New deployment scenarios: It is considered potentially new issues that urgently need extraordinarily difficult to introduce a radical research. and clock synchronization. and the Swedish organiza- antennas or multiple access points [20]. small cell solution may offer substantial path loss [2] J. virtues of the Argos testbed in particular is that [10] S. or coordination of the entirely new research problems to tackle. tems. reaping the to both prove the massive MIMO concept with highly desired to power gains offered by having a very large array. 61.” IEEE Trans. Pitarokoilis. 1436–49. it can be an enabler for At the base station the use of expensive and something that was just never before considered powerful. Mar. “Scaling Up MIMO: Opportunities and nas are distributed over a large geographical Challenges with Very Large Arrays.. 4. pp. Mobile Computing and Net- working. 2013. Consid. bility. Shepard et al. [1] F. Larsson. Once the terminal has been contact.qxp_Layout 1/30/14 1:08 PM Page 194 is optimal. on various aspects of the technology is going on [7] A. vol. In particular. This gives researchers deployment of small cells.. pp. Other potential ways of reach. “Per-Antenna it is entirely modular and scalable. Pacific However.6 GHz. Commun. Paris. VR. “Joint Spatial Division and Multiplexing: Realizing Massive MIMO Gains with Limited Channel gains and would also provide some diversity State Information. Systems. Ngo. the base station can learn simultaneously with true spatial multiplexing. L. 40–60. Other test systems around the 2013.” IEEE Wireless Commun. and Computers. tributed processing algorithms. L. 1059–71. G. both communication-theoretic and [4] X. Marzetta. the channel response and operate in coherent Continued testbed development is highly desired development is MU-MIMO beamforming mode. in both academia and industry a gold mine of geneous architecture. “An Early-Warning Service for world have also demonstrated the basic feasibili. 2. but power-inefficient. But at the same time. Aug. Aug. pp. “On in different parts of the world. noise. K. Larsson. pp. a massive array offers huge advantages in terms of energy effi- need research. 186–92. no..” IEEE Trans. “Measured Propagation Characteristics for techno-economic studies are needed to conclu. E. massive MIMO new wireless standard. rather than thinking of massive MIMO as at both the base station and the mobile unit side. ACKNOWLEDGMENTS mately limiting factor of the performance of any The authors would like to thank Xiang Gao. 5. 51. and built Constant Envelope Precoding for Large Multi-User around commercially available hardware (the MIMO Systems.” IEEE Commun. Symp. The Argos testbed the Optimality of Single-Carrier Transmission in Large- [3] was developed at Rice University in coopera. and T. Antenna Arrays. Larsson. Systems. resulting in a hetero. 31. Studer and E. 172–79.. 31. Mohammed and E. ent. Stations. The deployment costs of ences and Systems. [8] H. It allows for the use of low-cost hardware Thus. From a purely fundamental perspective.” IEEE JSAC. Rusek et al. Stenumgaard et al. and synchroniza- ing this goal are network densification by the tion of the antenna units. Nov. in In this article we have highlighted the large antennas and discov. 2012. for example. Hence. Yang and T. “PAR-Aware Large-Scale on channel reciprocity is possible. Emerging Communication Problems in Security and ty of scaling up the number of antennas. 6 and Fig. a massive MIMO array and a distributed or [3] C. as discussed earlier. Ngara testbed in Australia [21] uses a 32-ele. 7. 2013. Information Sci- against shadow fading. 2013. 2012.. acquiring high- quality CSI seems to be easier with a collocated REFERENCES massive array than in a system where the anten. K. G. replaced by massive use of parallel low-cost low- System studies and relation to small-cell and power units that operate coherently together. Mar. 2013. Prototype development: While massive [6] H.. Wireless Commun. Marzetta. Gao et al. the ulti. Mag. Conf. Feb. 30. Lett. and relia- serve small cells in a densely populated area. For example. area. it is likely that the winning solution Grove. Turkey. [11] P. 194 IEEE Communications Magazine • February 2014 . “Energy and MIMO is in its infancy.

Symp. 2013.” applications in wireless communications. FREDRIK TUFVESSON received his Ph. T HOMAS L. Ngo and E.” Proc. Ashikhmin and T. surements and modeling for wireless communication. MA. 2013.EDFORS_LAYOUT. After almost two years at a startup company. 9. Cambridge. and L. Future Net- ERIK G. M. Smart Anten.” IEEE Trans. Commun. challenges ahead to lished some 100 journal papers on signal processing and Trans. pp. Sept. ITG Wksp. his research focus is on how realistic prop- [17] H. and L. Feb. His research interests include statistical [16] H. realize the full communications. 2.D. 2013. Hoydis. [20] A. 2010. Feb. M ARZETTA [F’03] received his Ph. He has pub. no. “Relative Channel Reciprocity BIOGRAPHIES Calibration in MIMO/TDD Systems. Suzuki et al. vol. Lund July 2012. Stuttgart.” less Society. Müller. He joined Bell Laboratories in 1995. Lozano. Debbah. Cottatellucci. S. Heath Jr. “Funda- mental Limits of Cooperation. “Massive Associate Editor for IEEE Transactions on Communications. Information Theory. “Hermitian Precoding University. Lund Uni- nas. and J.. Marzetta. State Information. Department of Electrical and Information Technology. “Highly Spectrally Efficient Ngara gy. “Pilot Contamination versity. His main research interests are channel mea. he also works on distributed antenna sys- 2013. Communication Systems in the Department of Electrical [13] T. Besides these. 31. vol. In the context of complexity. for Distributed MIMO Systems with Individual Channel including channels for both MIMO and UWB systems.” IEEE Engineering at Linköping University. LARSSON is a professor and head of the Division for There are still work and Mobile Summit. G. G. Yin et al. and Precoding in Multi-Cell Large Scale Antenna Systems.” Proc. and Information Technolo- the Communications and Statistical Sciences Department. Mathematical Sciences Research Center he was director of IEEE Int’l. Andrews. 31. Info. for Antennas Do We Need?. Theory. He received the 1981 ASSP Paper Award and received the 2013 IEEE Guglielmo Marconi Best Paper Award. 241–50. 59. 2.” IEEE JSAC. 3590–600. L. when it OVE E DFORS is a professor of radio systems in the Depart- Pilot Decontamination.” Proc. Vehkaperä. massive MIMO. ten Brink. pp. pp. Ping. 2010. 31. Mar. He is an [14] J. 264–73. 2013. 2012. Time Block Coding for Wireless Communications. no. Marzetta. IEEE JSAC. 5213–26. 11..” IEEE JSAC. “Noncooperative Cellular Wireless with Unlimited Numbers of Base Station Antennas. “Blind example. Proc. vol. which can provide huge improvements in wireless spectral efficiency and energy efficiency over 4G technologies. 2. “EVD-Based Channel Esti- mations for Multicell Multiuser MIMO with Very Large agation characteristics influence system performance and realization of baseband processing complexity. Column Award 2012. Wireless Commun. versity.. Conf. no.W. in 2000 from Lund Uni- [18] A.qxp_Layout 1/30/14 1:08 PM Page 195 [12] F. He was an early proponent of massive MIMO. Nov. Sweden. gies. and M. the antenna units. Sweden. X. “A Coordinated Approach to Channel signal processing and low-complexity algorithms with computational Estimation in Large-Scale Multiple-Antenna Systems. pp. Feb. Symp. Yuan. R. in electrical vol. Larsson. Kaltenberger et al. tems and radio-based positioning. 160–71. engineering from the Massachusetts Institute of Technolo- [21] H. no. [19] J.D. Zhang. Fiber- ing algorithms. comes to ment of Electrical and Information Technology. L. Mar. potential of the MIMO in the UL/DL of Cellular Networks: How Many and he received the IEEE Signal Processing Magazine Best technology. he is now an associate professor in the synchronization of IEEE Int’l. IEEE Int’l.” Proc. vol. IEEE Communications Magazine • February 2014 195 . distributed process- Speed and Sig.. Acoustics. Within the former Rural Wireless Broadband ACCESS Demonstrator.. and is a co-author of the textbook Space- pp. Q. [15] R. Antenna Arrays. Oct. 2012. 2010.