PaperlayerView exact revision on arXiv

Source-linked AI summary

Massive MIMO: Ten Myths and One Critical Question

Emil Björnson, Erik G. Larsson, Thomas L. Marzetta

arXiv:1503.06854v2cs.IT

TL;DR

Massive MIMO is surrounded by widespread misunderstandings, while its applicability in FDD operation remains unclear. This overview examines ten common beliefs, explains why they are erroneous, and identifies FDD applicability as a question requiring further research and channel measurements.

  • Problem

    Widespread misunderstandings about Massive MIMO and uncertainty about its applicability in FDD operation remain important issues for practical adoption.

  • Method

    The article reviews ten common beliefs about Massive MIMO and explains why they are erroneous, supported by references to technical papers.

  • Results

    The article concludes that the ten inspected beliefs are erroneous and that Massive MIMO’s applicability in FDD operation remains unclear.

  • Takeaways & Limitations

    Further research and channel measurements are required to determine whether Massive MIMO can be properly applied in FDD operation.

  • Takeaways & Limitations

    The extent to which Massive MIMO can be applied in FDD mode remains unclear.

Abstract

from arXiv · show

Wireless communications is one of the most successful technologies in modern years, given that an exponential growth rate in wireless traffic has been sustained for over a century (known as Cooper's law). This trend will certainly continue driven by new innovative applications; for example, augmented reality and internet-of-things. Massive MIMO (multiple-input multiple-output) has been identified as a key technology to handle orders of magnitude more data traffic. Despite the attention it is receiving from the communication community, we have personally witnessed that Massive MIMO is subject to several widespread misunderstandings, as epitomized by following (fictional) abstract: "The Massive MIMO technology uses a nearly infinite number of high-quality antennas at the base stations. By having at least an order of magnitude more antennas than active terminals, one can exploit asymptotic behaviors that some special kinds of wireless channels have. This technology looks great at first sight, but unfortunately the signal processing complexity is off the charts and the antenna arrays would be so huge that it can only be implemented in millimeter wave bands." The statements above are, in fact, completely false. In this overview article, we identify ten myths and explain why they are not true. We also ask a question that is critical for the practical adoption of the technology and which will require intense future research activities to answer properly. We provide references to key technical papers that support our claims, while a further list of related overview and technical papers can be found at the Massive MIMO Info Point: http://massivemimo.eu

I. INTRODUCTION

Massive MIMO equips each base station with M active antennas to serve K single-antenna terminals on the same time-frequency resource, using coherent signal processing. Its canonical TDD operation exploits channel reciprocity, while antenna-scalable uplink estimation and practical calibration support implementation.

  • System model: Massive MIMO uses M active base-station antennas to communicate coherently with K single-antenna terminals over the same time and frequency band.It is a multi-user MIMO technology whose deployment with more than a handful of service antennas is relatively new.
  • TDD operation: In canonical TDD operation, uplink and downlink share frequency resources but are separated in time, enabling uplink channel estimates to support both receive combining and transmit precoding.This relies on reciprocal physical propagation channels, although transceiver hardware itself is generally non-reciprocal.
  • TDD operation: Hardware calibration is required in practice, but uplink-downlink mismatches change by only a few degrees over one hour and can be mitigated by simple relative calibration.The passage states that this can be done without extra reference transceivers, using mutual information between transceiver chains.
  • Scalability: TDD makes uplink estimation overhead proportional to the number of terminals and independent of M, making the protocol scalable with the number of service antennas.Only the base station needs channel knowledge for coherent antenna processing, and adding antennas does not reduce per-antenna estimation quality.
  • Coherence conditions: Channel estimation and payload transmission must fit within approximately static time-frequency blocks characterized by coherence bandwidth Bc Hz and coherence time Tc s.These dimensions correspond to τ = BcTc transmission symbols, and Massive MIMO can use either single-carrier or multi-carrier modulation.

A. Linear Processing

Massive MIMO payload transmission relies on linear receive combining and transmit precoding, which turn multiuser signals into effective scalar channels while amplifying desired signals and suppressing interference. Adding BS antennas improves judicious combining by providing more channel observations, and TDD reciprocity links uplink combining with downlink precoding.

  • Linear Processing: In the uplink, the BS applies linear receive combining to distinguish each terminal’s signal from interfering signals.Maximum-ratio combining coherently adds a terminal’s signal components using its channel estimate.
  • Linear Processing: ZF suppresses inter-cell interference but reduces the array gain to M −K + 1, while MMSE balances signal amplification and interference suppression.These are alternatives to maximum-ratio combining.
  • Linear Processing: Adding more BS antennas improves judicious receive combining by providing more channel observations, while residual interference can be treated as extra additive noise.The resulting effective scalar channels enable conventional single-user detection, and small-scale fading averages out over the combining process.
  • Linear Processing: In TDD, channel reciprocity creates uplink-downlink duality, allowing MR, ZF, or MMSE linear precoding to focus signals at intended terminals and potentially mitigate interference.The downlink precoding is strongly connected to uplink receive combining.
  • Linear Processing: For i.i.d. Rayleigh fading with MR processing, the achievable spectral-efficiency expression increases with M through array gain, while its denominator represents interference plus noise.The channel-estimate quality is represented by cCSI, with cCSI = 1 denoting perfect CSI.

Myth 1: Massive MIMO is only suitable for millimeter wave bands

Massive MIMO is not restricted to millimeter-wave bands: research has focused on cellular frequencies below 6 GHz, where hardware is mature, and the concept also applies at millimeter waves. Antenna-array size depends on wavelength, while a sub-6-GHz testbed demonstrates that substantial arrays can fit in practical panels.

  • Antenna-array scaling: Antenna spacing is typically at least λc/2, so higher carrier frequencies enable smaller antenna-array form factors because wavelength decreases as fc increases.Larger spacings can provide less-correlated channel responses and more spatial diversity, but Massive MIMO primarily requires distinct spatial channel characteristics for each terminal.
  • Practical sub-6-GHz array: 3.7 GHz carrier frequency yields λc = 8.1 cm in the LuMaMi testbed, whose 60 × 120 cm panel contains 160 dual-polarized antennas and can accommodate additional elements.The panel is equivalent to a 53 inch flatscreen TV and could be deployed on a building facade.
  • Operating frequencies: Research has focused on cellular frequencies below 6 GHz, where transceiver hardware is very mature, while the same concept can also be applied in millimeter wave bands.Millimeter-wave systems might require many antennas because each antenna has a much smaller effective area, although implementation may differ from existing Massive MIMO designs.

Myth 2: Massive MIMO only works in rich-scattering environments · Myth 3: Massive MIMO performance can be achieved by open-loop beamforming techniques

Massive MIMO remains effective beyond rich-scattering environments because both isotropic scattering and line-of-sight propagation can provide favorable propagation. It also outperforms open-loop beamforming in scalability and channel exploitation by using estimated channel responses rather than predetermined beams.

  • Myth 2: Massive MIMO only works in rich-scattering environments: Favorable propagation means users’ channel vectors are mutually orthogonal, allowing interference-free multi-user transmission with simple MR and ZF processing.Without favorable propagation, advanced processing such as dirty paper coding is needed to suppress interference and achieve sum capacity.
  • Myth 2: Massive MIMO only works in rich-scattering environments: In rich-scattering non-LoS environments, normalized channel-vector inner products approach zero as M increases, yielding increasingly orthogonal channels.Rayleigh fading satisfies the stated sufficient condition.
  • Myth 2: Massive MIMO only works in rich-scattering environments: 10 % risk of LoS performance loss exceeding 10 % occurs in the stated M = 100, K = 12, SNRu = −5 dB example.The LoS sum capacity is otherwise similar to isotropic scattering; dropping two worst terminals reduces the issue caused by similar angles.
  • Myth 2: Massive MIMO only works in rich-scattering environments: Because isotropic and LoS propagation are both favorable, real environments between these extremes are also expected to support favorable propagation.This expectation explains favorable-propagation characteristics observed consistently in measurement campaigns.
  • Myth 3: Massive MIMO performance can be achieved by open-loop beamforming techniques: Massive MIMO precoding and combining use measured or estimated terminal channels, providing array gain in any propagation environment without requiring particular array geometry or calibration.The base station estimates uplink channels from K mutually orthogonal pilots, so pilot resources scale with K rather than M.
  • Myth 3: Massive MIMO performance can be achieved by open-loop beamforming techniques: Open-loop beamforming uses L predetermined beams, requiring L downlink pilots and log2(L) feedback bits per terminal.L should be proportional to M to explore all channel dimensions, while limited feedback prevents accurate channel learning for spatial multiplexing.
  • Myth 3: Massive MIMO performance can be achieved by open-loop beamforming techniques: With K = 12 and SNRu = −5 dB, Massive MIMO achieves array gain ≈0.79M in both isotropic-scattering and LoS cases.The comparison uses MR processing and contrasts Massive MIMO with open-loop beamforming for the same propagation scenarios.
  • Myth 3: Massive MIMO performance can be achieved by open-loop beamforming techniques: Conventional open-loop beamforming offers decent gains only for small LoS arrays, is not scalable in overhead or array tolerance, and cannot handle isotropic fading.Its codebook must explore all possible array directions rather than adapt to each terminal’s statistical spatial properties.

Myth 4: The case for Massive MIMO relies on asymptotic results

Massive MIMO does not depend solely on asymptotic analysis: closed-form spectral-efficiency expressions apply to finite antenna and terminal counts, arbitrary SNR, and pilot signaling. A finite-system example with estimated channels closely approaches the predicted zero-BER threshold at moderate codeword lengths.

  • Finite-system analysis: Closed-form achievable spectral-efficiency expressions apply to any number of antennas and terminals, any SNR, and any pilot signaling choice.They use worst-case assumptions about channel acquisition and signal processing rather than perfect CSI.
  • Finite-system analysis: 30 bit/s/Hz in total for the cell is achieved with M = 100 antennas, K = 30 terminals, QPSK, rate-1/2 LDPC coding, and estimated channels.Each terminal achieves a net spectral efficiency of 1 bit/s/Hz using one pilot per terminal.
  • Finite-system validation: −13.94 dB is the SNR threshold where zero BER is achievable as codeword length approaches infinity.BER curves drop rapidly with increasing codeword length, and performance close to this bound is reached at moderate lengths.

Myth 5: Too much performance is lost by linear processing · Myth 6: Massive MIMO requires an order of magnitude more antennas than users

Linear processing has a rapidly shrinking performance gap relative to DPC/SIC as the antenna count grows, while Massive MIMO has no strict required M/K ratio. The suitable ratio depends on the performance metric, propagation environment, and coherence-block length.

  • Myth 5: Too much performance is lost by linear processing: Nonlinear DPC/SIC achieves sum capacity with perfect CSI, whereas linear processing suppresses interference through projections such as ZF.The comparison concerns optimal interference-aware encoding/decoding versus linear interference rejection.
  • Myth 5: Too much performance is lost by linear processing: With K = 20 terminals and SNRu = SNRd = −5 dB, the DPC/SIC–ZF performance gap reduces quickly as M increases.The example assumes i.i.d. Rayleigh fading, perfect CSI, and a single cell.
  • Myth 5: Too much performance is lost by linear processing: Nonlinear processing provides a large gain mainly when M ≈K, while its gain is small in Massive MIMO cases with M/K > 2.The channels decorrelate as M grows, bringing the curves closer to favorable propagation.
  • Myth 6: Massive MIMO requires an order of magnitude more antennas than users: Adding service antennas always improves spectral efficiency through larger array gain and favorable propagation, but M/K > 10 is not a general requirement.M is typically fixed in deployment, whereas K is the design parameter.
  • Myth 6: Massive MIMO requires an order of magnitude more antennas than users: With M = 100 antennas per cell, a multi-cellular deployment at SNR −5 dB per terminal and coherence block τ = 200 symbols supports a wide range of K-values with almost identical sum performance.The results apply to uplink and downlink when power control provides the stated SNR.
  • Myth 5: Too much performance is lost by linear processing: ZF with around M + 10 antennas achieves performance equivalent to capacity with M antennas.This result is stated for the Figure 4a comparison of DPC/SIC and linear processing.
  • Myth 6: Massive MIMO requires an order of magnitude more antennas than users: Massive MIMO has no strict M–K relationship; it instead uses unconventionally many active antennas to serve an unconventionally large number of terminals.The appropriate ratio depends on the performance metric, propagation environment, and coherence-block length.

Myth 7: A new terminal cannot join the system since there is no initial array gain · Myth 8: Massive MIMO requires high precision hardware

Massive MIMO can admit previously inactive terminals through broadcast control signaling despite the absence of an initial array gain. Its coherent processing also suppresses many additive hardware distortions, allowing gains with lower precision than contemporary systems.

  • Myth 7: A new terminal cannot join the system since there is no initial array gain: Coherent processing improves effective SNR by a factor c_CS I_M, where 0 < c_CSI ≤ 1 represents channel-state-information quality.The base station estimates the current channel response from uplink pilots before exploiting the array gain.
  • Myth 7: A new terminal cannot join the system since there is no initial array gain: An inactive terminal can select an unused pilot sequence when it wishes to send or request data.This provides a straightforward mechanism for initiating channel acquisition and system access.
  • Myth 7: A new terminal cannot join the system since there is no initial array gain: Broadcast control information can contact inactive terminals without requiring an array gain.The base station occasionally uses the combined transmit power for cell-wide control signaling instead of precoded signals to K terminals.
  • Myth 7: A new terminal cannot join the system since there is no initial array gain: c_CSI M/K times weaker is the broadcast signal’s strength relative to user-specific precoded signals when transmit power is held constant.The passage also notes M/K < 10 at many practical operating points.
  • Myth 7: A new terminal cannot join the system since there is no initial array gain: Control data rates can be comparable to individual precoded payload rates, but broadcasting one signal sacrifices the multiplexing gain of K separate signals.The comparison is attributed to the lack of intra-cell interference and concentration of transmit power.
  • Myth 8: Massive MIMO requires high precision hardware: Coherent processing amplifies desired signals through M service antennas, while uncorrelated undesired signals combine noncoherently.Receiver noise and other terminals’ data signals are examples of undesired additive quantities mitigated by this processing.
  • Myth 8: Massive MIMO requires high precision hardware: Hardware impairments generally create additive distortions substantially uncorrelated with desired signals, alongside desired-signal power loss and phase rotation.Examples include amplifier nonlinearities, oscillator phase noise, quantization errors, mixer I/Q imbalance, and non-ideal analog filters.
  • Myth 8: Massive MIMO requires high precision hardware: Massive MIMO gains do not require high-precision hardware because additive distortions are suppressed, while low-order modulations support high spectral efficiency across many terminals.Contemporary systems instead use high-order modulations for a few terminals and therefore require higher precision.

Myth 9: With so many antennas, resource allocation and power control is hugely complicated

Massive MIMO simplifies resource allocation because channel hardening removes most frequency-domain channel variation, leaving admission control and long-term power control as the main tasks. The resulting power-control problem can be formulated as a linear optimization for fixed worst-terminal performance, rather than requiring highly complex frequency-selective allocation.

  • Simplified resource allocation: Channel hardening makes frequency-domain channel variations negligible, so conventional frequency-selective resource allocation becomes unnecessary.The whole spectrum can be allocated simultaneously to each active terminal.
  • Simplified resource allocation: 100–1000 times slower: large-scale fading varies much more slowly over time than small-scale fading, enabling long-term rather than rapidly changing power control.The channel variations mainly depend on large-scale fading in the time domain.
  • Power control: Max-min fairness equalizes the performance of all terminals by maximizing the worst-terminal performance.The problem is non-trivial when terminals have different average channel conditions and can be addressed through user-specific power-control coefficients.
  • Power control: For fixed worst-terminal performance R, the epigraph formulation has linear constraints in the power-control coefficients, making it a linear optimization problem.The constraints involve η1, . . . , ηK linearly.
  • Simplified resource allocation: Resource allocation reduces mainly to admission control and long-term power control, with power-control complexity scaling with the number of terminals.Admitted terminals may use the full bandwidth when there is no frequency-selective fading.

Myth 10: With so many antennas, the signal processing complexity will be overwhelming

Massive MIMO baseband processing becomes more demanding with more antennas and terminals, but its complexity remains within the practical realm. Most computations are standard, scalable, and largely parallelizable, while coherence-block processing limits the difference between MR and ZF/MMSE.

  • Myth 10: With so many antennas, the signal processing complexity will be overwhelming: Baseband complexity increases with the numbers of service antennas M and terminals K, making its growth rate the key practical question.Except for FFTs, the listed processing tasks also increase with K.
  • Myth 10: With so many antennas, the signal processing complexity will be overwhelming: FFT, pilot-based channel estimation, payload precoding/combining, and matrix computation are standard operations whose required flops can be computed straightforwardly.MR matrix computation scales linearly with K, whereas ZF/MMSE scale faster because they involve matrix inversions.
  • Myth 10: With so many antennas, the signal processing complexity will be overwhelming: Longer coherence blocks reduce complexity by requiring precoding/combining matrices to be computed less frequently, with the largest gain for ZF/MMSE at very many antennas and terminals.The gain is barely visible for MR but can be substantial for ZF/MMSE because matrix inversion becomes costly.
  • Myth 10: With so many antennas, the signal processing complexity will be overwhelming: Massive MIMO baseband complexity is well within the practical realm, and MR versus ZF/MMSE differs relatively little because matrices are computed only once per coherence block.Most complexity comes from FFTs and per-symbol matrix-vector multiplications; most computations can be parallelized and distributed over antennas.

THE CRITICAL QUESTION · Can Massive MIMO work in FDD operation?

Massive MIMO conventionally relies on TDD because channel reciprocity reduces CSI-acquisition overhead, whereas FDD requires additional downlink pilots and uplink feedback. The critical open question is whether Massive MIMO can be made practical in FDD despite antenna-terminal tradeoffs and demanding signaling overhead.

  • Can Massive MIMO work in FDD operation?: TDD is the canonical Massive MIMO protocol because channel reciprocity greatly reduces the CSI-acquisition overhead required for base-station processing.Many contemporary networks nevertheless operate in FDD, where uplink and downlink use different frequency bands and reciprocity cannot be exploited.
  • Can Massive MIMO work in FDD operation?: The coherence-block length is τ = BcTc symbols; TDD uses K uplink pilot symbols, while basic FDD additionally requires M downlink pilots and feedback of M channel coefficients per terminal.In TDD, channel hardening eliminates the need for downlink pilots.
  • Can Massive MIMO work in FDD operation?: FDD limits the number of antennas, whereas TDD can support any number; with τ = 200, M = 100 and K = 25 is supported only in TDD operation.This figure illustrates the signaling-overhead difference between the duplexing modes.
  • Can Massive MIMO work in FDD operation?: The FDD overhead analysis in Figure 6 varies with τ and is categorized by the percentage of overhead required.This provides the basis for comparing signaling demands across coherence-block lengths.
  • Can Massive MIMO work in FDD operation?: FDD overhead is a tradeoff between antennas and terminals, becoming critical at τ = 200 but potentially less important at τ = 5000.The passage associates τ = 5000 with low mobility at low frequencies and τ = 200 with high mobility or higher frequencies.
  • Can Massive MIMO work in FDD operation?: Proposed methods reduce FDD CSI overhead by exploiting channel sparsity, such as strong spatial correlation or impulse responses that are sparse in time.At millimeter wave frequencies, channel responses may indeed be sparse.
  • Can Massive MIMO work in FDD operation?: It remains unclear to what extent Massive MIMO applies in FDD mode, motivating research and channel measurements to test sparsity hypotheses or identify other ways to reduce overhead signaling.The paper characterizes this unresolved issue as requiring intensive future research activities.

AUTHORS

The authors received recognition from IEEE ComSoc EMEA, the 2015 Ingvar Carlsson Award, and several best conference paper awards.

  • The authors received the 2015 Ingvar Carlsson Award, IEEE ComSoc EMEA recognition, and best conference paper awards in 2009, 2011, 2014, and 2015.
Loading 1503.06854v2…