PaperlayerView exact revision on arXiv

Source-linked AI summary

Massive MIMO in Sub-6 GHz and mmWave: Physical, Practical, and Use-Case Differences

Emil Björnson, Liesbet Van der Perre, Stefano Buzzi, Erik G. Larsson

arXiv:1803.11023v2cs.IT

TL;DR

The paper examines why conceptually similar Massive MIMO systems require different designs and uses in sub-6 GHz and mmWave bands. It compares propagation, hardware, signal processing, and use cases, finding attractive but distinct propositions in both bands while identifying interconnects as potential bottlenecks.

  • Problem

    Sub-6 GHz and mmWave mMIMO share a concept but differ fundamentally in propagation and hardware, requiring a comparative understanding of their appropriate exploitation.

  • Method

    The paper reviews differences in propagation mechanisms, transceiver hardware, signal-processing algorithms, and associated 5G use cases.

  • Results

    Both frequency bands offer attractive propositions, but their mMIMO designs and use cases differ; interconnects can be potential bottlenecks despite computational complexity no longer being primary.

  • Takeaways & Limitations

    Efficient mMIMO exploitation requires band-specific signal-processing schemes and use cases rather than treating sub-6 GHz and mmWave systems identically.

  • Takeaways & Limitations

    High-speed interconnects remain a potential bottleneck, and practical sub-6 GHz channel behavior is not always captured by i.i.d. Rayleigh models.

Abstract

from arXiv · show

The use of base stations (BSs) and access points (APs) with a large number of antennas, called Massive MIMO (multiple-input multiple-output), is a key technology for increasing the capacity of 5G networks and beyond. While originally conceived for conventional sub-6GHz frequencies, Massive MIMO (mMIMO) is ideal also for frequency bands in the range 30-300 GHz, known as millimeter wave (mmWave). Despite conceptual similarities, the way in which mMIMO can be exploited in these bands is radically different, due to their specific propagation behaviors and hardware characteristics. This paper reviews these differences and their implications, while dispelling common misunderstandings. Building on this foundation, we suggest appropriate signal processing schemes and use cases to efficiently exploit mMIMO in both frequency bands.

Abstract

Massive MIMO increases capacity through beamforming and spatial multiplexing, but its design and exploitation differ fundamentally between sub-6 GHz and mmWave bands because propagation, hardware, and signal-processing conditions change.

  • mMIMO uses large BS antenna arrays to provide beamforming amplification, high spatial resolution, and multiplexing of many simultaneous users.
  • mmWave offers many GHz of unused spectrum that can complement current sub-6 GHz bands, although path loss and blockage are more severe.
  • The paper compares sub-6 GHz and mmWave mMIMO through differences in propagation phenomena, hardware architecture, and signal-processing algorithms.
  • Channel estimation is resource-demanding at sub-6 GHz while beamforming is straightforward; mmWave reverses this balance, especially under hybrid beamforming.
  • These differences determine how each band should be exploited for distinct 5G and beyond use cases.
  • Understanding electromagnetic propagation is necessary because conventional cellular channel-modeling simplifications can expose weaknesses for mMIMO systems extending toward mmWave.

Sub-6 GHz: favorable propagation and spatial correlation

Sub-6 GHz mMIMO builds on extensively studied propagation and practical testbeds, but real channels depart from i.i.d. Rayleigh assumptions through line-of-sight components and spatial correlation. Increasing antenna count strengthens directivity without requiring more frequent channel estimation under mobility.

  • Measurement and deployment: Sub-6 GHz mMIMO channels have been characterized through measurement campaigns and operational co-located and distributed testbeds.
  • Propagation models: i.i.d. Rayleigh models provide tractable analysis and channel hardening, but practical channels can include strong LoS components and spatially correlated fading.
  • Propagation models: Spatial correlation changes interference between UEs with similar or different correlation characteristics, and weakens channel hardening.
  • Mobility and beamforming: Increasing the antenna count makes beamforming more directive but does not change how frequently the channel must be re-estimated under mobility.
  • Mobility and beamforming: 3 dB is the worst-case beamforming-gain reduction for a UE moving up to one-eighth wavelength, independently of the antenna count M.

mmWave: blessing and curse of attenuation and directivity

mmWave mMIMO can offset severe attenuation through compact, highly directional arrays and broad bandwidth, but blockage, rapid channel variation, power consumption, interconnect loss, and hybrid hardware constraints complicate implementation.

  • Propagation and directivity: Shorter wavelengths allow fixed-area or array antennas to recover mmWave path loss through increasingly directional gains while retaining beamforming flexibility.
  • Propagation and directivity: The total link beamforming gain is the product of transmitter and receiver gains, allowing smaller arrays at both ends to replace a much larger array at one side.
  • Blockage and Fresnel zones: At 38 GHz and 100 m, the Fresnel zone has approximately 0.5 m radius, so small obstructions can cause abrupt channel variations.
  • Blockage and attenuation: Humans can fully block mmWave signals, while reduced diffraction, oxygen absorption near 60 GHz, foliage, and rain increase coverage and attenuation challenges.
  • Channel variation: 10 times faster channel variations occur at 30 GHz than at 3 GHz for equal speed, requiring 10 times more frequent channel estimation, although small cells may favor low-mobility users.
  • Hardware implementation: High-speed interconnects, rather than digital processing or data converters alone, are identified as a bottleneck in integrated mMIMO systems.
  • Hardware implementation: Full digital beamforming offers the highest theoretical performance, whereas hybrid analog-digital beamforming reuses hardware across antenna paths.

DSP and data converters: lean processing suits the system

mMIMO processing can be made efficient by separating modem, central, and antenna-signal DSP, while data-converter requirements differ sharply between sub-6 GHz and mmWave. Hardware-scalable implementations are feasible, but mmWave bandwidth makes ADC power a key trade-off.

  • DSP partitioning: mMIMO DSP separates outer-modem coding, central decoding and beamforming, and antenna-signal processing with complexity tied mainly to full RF chains.Central matrix operations can exploit nearly orthogonal user channels, while antenna-signal processing may dominate operations per second but can use low resolution.
  • Implementation feasibility: Efficient DSP implementations are feasible in both sub-6 GHz and mmWave using scaled CMOS technology and system-level opportunities.
  • Data converters: ENOB = 5 is expected to suit realistic mMIMO, allowing a 128-antenna BS to consume less ADC power than a conventional 8-antenna system.The comparison is subject to practical converter overheads, including voltage regulation, buffering, and calibration.
  • Data converters: Multi-GHz mmWave bandwidth requires ADCs consuming an order of magnitude more power than comparable sub-6 GHz converters at similar linearity.This power difference directly affects hybrid-beamforming trade-offs when UEs use antenna arrays.

Generating, phase shifting, and amplifying RF signals: divide and conquer?

RF generation and amplification expose different implementation pressures across frequency bands: mmWave phase shifting is difficult and power-intensive, while PA efficiency is reduced by power combining and lower gain. Compact mmWave integration helps realize small antenna modules but does not remove these constraints.

  • RF generation: mmWave frequency synthesis faces tougher phase-noise and EVM requirements than sub-6 GHz, with oscillator efficiency affected by operating frequency, channel spacing, and resonator Q-factor.
  • Phase shifting: Hybrid beamforming reuses RF hardware through analog phase shifting, but mmWave phase shifters must settle quickly when the LoS path is disrupted.Precise high-frequency phase shifting can incur considerable power overhead, making analog-baseband implementation preferable in some cases.
  • Phase shifting: A 5.4 x 9.2 mm mmWave module co-integrates a four-antenna transceiver IC with patch antennas, using two patches per antenna path.
  • Power amplification: mMIMO reduces required output power for the array and each antenna, with combined PA complexity and power decreasing as antenna count grows but with diminishing returns.
  • Power amplification: At 6 dB back-off, sub-6 GHz PAs can achieve 18% PAE, whereas CMOS mmWave PAs achieve PAE< 10% under the same back-off.mmWave power combining adds losses, and lower gain requires higher DC drive currents.
  • Power amplification: Operating PAs closer to saturation may improve power efficiency, but coherent nonlinear distortion in a few dominant mmWave beams can violate EVM and out-of-band-radiation specifications.

Interconnect is the main implementation challenge

Interconnects, rather than digital processing or data converters, are the principal hardware obstacle to exploiting many antenna signals, especially at mmWave. Sub-6 GHz systems can mitigate this through partly distributed processing, while mmWave requires compact chip–antenna–package co-design.

  • Sub-6 GHz implementation: Sub-6 GHz systems can use partly distributed processing to bring individual antenna signals to the DSP level and circumvent interconnect bottlenecks.
  • mmWave implementation: At mmWave, micro-strip lines become extremely lossy, with losses of several dB/cm at 60 GHz, and component matching is challenging.
  • mmWave implementation: mmWave systems benefit from additional antennas only when antennas, chips, and packages are integrated very compactly.The required approach is illustrated by the co-integrated transceiver-and-patch-antenna module in Figure 2.
  • mmWave implementation: Hybrid beamforming does not remove the need to connect mmWave signals to antennas, and oscillator distribution also remains severely challenging.Interconnects therefore bottleneck the integration of many small antennas and exploitation of high bandwidth.
  • Design implication: Propagation and hardware differences fundamentally determine the algorithms required for channel estimation, beamforming, and resource allocation.

Opportunities for efficient channel estimation

Channel-estimation strategies should reflect propagation structure: sub-6 GHz systems estimate many coefficients with manageable hardware parallelism, whereas mmWave can estimate a few angular parameters and potentially use FDD reciprocity.

  • Estimation burden: A 200-BS-antenna, 20-UE OFDM example contains 3.4·10^5 complex scalar coefficients.With 1024 subcarriers, 12-subcarrier channel constancy, and 50 ms coherence, this corresponds to 6.8·10^6 estimates/second.
  • Sub-6 GHz: Sub-6 GHz multipath channels have antenna-correlated coefficients, but exploiting that correlation only marginally improves estimation while substantially increasing complexity.TDD operation keeps estimation overhead small, and the process can be implemented or parallelized in hardware.
  • mmWave: mmWave channels with a line-of-sight path and few one-bounce reflections can be represented through a small number of angular channel coefficients instead of individual coefficients.For a single data stream, estimating the dominant angle of arrival or departure can suffice.
  • Duplexing: In mmWave, FDD may perform as well as TDD because channel-describing angular parameters remain reciprocal over a wide bandwidth.

Choosing between analog, hybrid, and digital beamforming

Beamforming architecture determines how flexibly mMIMO can handle propagation and frequency selectivity. Full digital beamforming is flexible at sub-6 GHz, while hybrid or analog approaches simplify mmWave hardware but constrain multiplexing or adaptation.

  • Analog beamforming: Analog beamforming creates one beam across the frequency band, which is sufficient for line-of-sight beamforming but cannot adapt to multipath or frequency-selective fading.The transmitted signal is phase-shifted identically across antennas to direct the beam toward a particular angle.
  • Hybrid beamforming: Hybrid beamforming uses a few analog beams whose digital superpositions provide limited flexibility for non-line-of-sight scenarios.The number of multiplexed UEs is limited by the number of phase-shifter sets and baseband inputs.
  • Digital beamforming: Full digital beamforming lets each antenna transmit any signal, supporting multipath, frequency-selective fading, and multiplexing many UEs on shared resources.Digital precoding can form superpositions of multiple beams per UE and different beams per subcarrier.
  • Hybrid and analog mmWave use: At mmWave, analog beamforming can suffice for single-user wideband links, despite lower user multiplexing capability with hybrid implementations.The example considers a 60 GHz channel with five reflections and 32 × 32, 64 × 64, and 128 × 128 planar arrays.
  • Analog beamforming: 80-90% of maximum beamforming gain is achievable across a 400 MHz band with analog beamforming, but wider bandwidths and larger arrays suffer beam-squinting losses.More than 75% is maintained over 2 GHz for a 32 × 32 array, whereas gain drops quickly for a 128 × 128 array.

Network deployment for good coverage

Sub-6 GHz deployment emphasizes broad coverage, mobility, and interference management, whereas mmWave deployment is more sensitive to blockage and coverage holes. Use-case suitability therefore depends on propagation, traffic concentration, and the required coverage geometry.

  • Sub-6 GHz coverage: Sub-6 GHz base stations provide outdoor and indoor coverage and support high-mobility users, making interference management especially important.Advanced digital beamforming can address inter-cell interference and pilot contamination.
  • mmWave coverage: At mmWave, blockage requires careful deployment planning, and wide-area coverage without holes may require many base stations.Interference is less important than ensuring coverage continuity.
  • Deployment pressure: 30-40% annual traffic growth and densification toward roughly 100 m inter-base-station distances suggest that simultaneous-user counts may grow rapidly.Further densification in traffic-intense areas is described as questionable from practical and cost perspectives.
  • Sub-6 GHz macro-cells: 120 Mbit/s per UE and 2.4 Gbit/s cell throughput are achievable with 40 MHz bandwidth, 3 bit/s/Hz spectral efficiency, and 20 multiplexed UEs.The beamforming gain improves cell-edge and outdoor-to-indoor conditions by increasing useful signal power while average interference remains unchanged.
  • mmWave coverage boundary: 10-100 times more mmWave bandwidth comes with a 10-20 dB link-budget reduction under the stated power and antenna-area assumptions.Limited outdoor-to-indoor propagation and blockage can require many base stations, relays, or reflective surfaces, while fixed wireless access can preserve LoS-like conditions.
  • Hotspots: mmWave mMIMO is particularly suitable for LoS-dominated hotspots, where short distances enable extreme throughput and pedestrian mobility support.With 1 GHz bandwidth, 1 bit/s/Hz yields 1 Gbit/s per data stream; 10 Gbit/s is within reach with more spectrum or higher efficiency.

What if extra hardware came at no cost?

With ideal hardware and processing, both sub-6 GHz and mmWave arrays can support enormous aggregate rates, but they favor different multiplexing strategies. Sub-6 GHz serves many more users through longer coherence, whereas mmWave provides higher per-user rates through greater bandwidth.

  • 3,000 homes can be served using a 3,200-antenna sub-6 GHz BS while keeping per-UE rate and total BS radiated power constant as antennas and UEs increase.The cited case uses a roughly 4 × 4 m array at 2 GHz.
  • The number of spatially multiplexed UEs depends on channel coherence-block samples and BS antennas, with fewer samples available as carrier frequency increases.At sub-6 GHz, a few hundred samples per coherence block allow orthogonal channel-estimation resources for a few hundred UEs in high-mobility outdoor networks.
  • 1.38 Tbit/s at 3 GHz and 1.44 Tbit/s at 60 GHz are achieved, but the peaks multiplex 14,000 and 870 UEs, respectively.The scenarios use 50 MHz at 3 GHz and 1 GHz at 60 GHz, with 35% and 44% of coherence blocks allocated to uplink pilots.
  • The mmWave setup delivers 1.66 Gbit/s per UE, while the sub-6 GHz setup delivers 99 Mbit/s per UE by serving far more UEs.The paper attributes the contrast to mmWave bandwidth versus the longer sub-6 GHz coherence time.
  • Data-traffic characteristics determine whether high per-user rates or massive user multiplexing is preferable, supporting deployment of both frequency bands.The paper also identifies cloud-RAN fronthauling and environmental sensing as additional mMIMO applications.
  • A 100,000-antenna simulation yields arrays of 31 m×31 m at 3 GHz and 1.58 m×1.58 m at 60 GHz, but signal interconnect and processing become bottlenecks.The paper concludes that both bands are attractive, while noting unresolved implementation and deployment questions.
Loading 1803.11023v2…