Source-linked AI summary
Massive MIMO in Real Propagation Environments: Do All Antennas Contribute Equally?
Xiang Gao, Ove Edfors, Fredrik Tufvesson, Erik G. Larsson
TL;DR
The paper asks whether massive-MIMO systems can reduce costly RF chains without substantial performance loss. It evaluates antenna selection on measured 2.6 GHz channels using linear and cylindrical 128-element arrays, benchmarking convex optimization against received-power selection. Real channels support substantial RF-chain reductions, while the simple power-based scheme performs close to the near-optimal benchmark.
Problem
Massive MIMO improves efficiency but requires many energy-consuming RF chains, and antenna selection is less promising in theoretical i.i.d. Rayleigh channels than in real propagation channels.
Method
The paper evaluates sum-rate-maximizing antenna selection on measured 2.6 GHz channels using 128-element linear and cylindrical arrays, with convex optimization as a benchmark and received-power selection as a simple scheme.
Results
Measured channels allow substantial RF-chain reductions without significant performance loss, while received-power selection gives performance close to convex optimization.
Takeaways & Limitations
Antenna selection may reduce massive-MIMO implementation complexity, cost, and hardware energy consumption.
Abstract
from arXiv · showhide
Massive MIMO can greatly increase both spectral and transmit-energy efficiency. This is achieved by allowing the number of antennas and RF chains to grow very large. However, the challenges include high system complexity and hardware energy consumption. Here we investigate the possibilities to reduce the required number of RF chains, by performing antenna selection. While this approach is not a very effective strategy for theoretical independent Rayleigh fading channels, a substantial reduction in the number of RF chains can be achieved for real massive MIMO channels, without significant performance loss. We evaluate antenna selection performance on measured channels at 2.6 GHz, using a linear and a cylindrical array, both having 128 elements. Sum-rate maximization is used as the criterion for antenna selection. A selection scheme based on convex optimization is nearly optimal and used as a benchmark. The achieved sum-rate is compared with that of a very simple scheme that selects the antennas with the highest received power. The power-based scheme gives performance close to the convex optimization scheme, for the measured channels. This observation indicates a potential for significant reductions of massive MIMO implementation complexity, by reducing the number of RF chains and performing antenna selection using simple algorithms.
I. INTRODUCTION
Massive MIMO improves efficiency through many antennas but faces complexity and hardware-power challenges. The paper examines antenna selection as a way to exploit unequal antenna contributions in measured real propagation channels.
- Motivation: Massive MIMO uses tens to hundreds of base-station antennas to serve multiple users in the same time-frequency resource.Its expected benefits include substantially improved spectral and transmit-energy efficiency and near-optimal simple signal processing.
- Motivation: High system complexity and hardware power consumption arise from the large number of antennas and associated transceiver chains.
- Research question: The paper investigates whether all massive-MIMO antennas contribute equally, using 2.6 GHz measurement campaigns to demonstrate that they often do not.
- Study approach: The study proposes and analyzes antenna-selection algorithms for architectures with fewer activated RF chains than physical antennas.It compares measured-channel performance under different operating conditions.
- Real propagation channels: In ideal i.i.d. Rayleigh channels, antennas are expected to contribute equally, whereas real channels can exhibit substantial power variation across arrays.Measured arrays used linear omni-directional and cylindrical patch antennas, each with 128 elements.
- Antenna selection: Measured real-channel variation enables selecting stronger antennas while reducing active antennas and RF transceivers.This can simplify base-station design and potentially reduce energy and cost, although switching and selection algorithms add implementation complexity.
III. APPROACH
The study evaluates adaptive antenna selection in measured massive MIMO channels, focusing on user geometry, propagation conditions, and the number of active antennas. Selection maximizes downlink capacity across subcarriers and is evaluated with both DPC and ZF sum-rate criteria.
- Study focus: The study examines how user number, user separation, and LOS or NLOS propagation affect antenna-selection performance.Measured channels at 2.6 GHz are used with linear and cylindrical arrays, each containing 128 elements.
- Selection schemes: The study compares a near-optimal convex-optimization scheme with a low-complexity scheme based only on received power.The power-based method selects antennas with the highest received power averaged over users and subcarriers, while the benchmark also accounts for channel correlation.
- System model: Antenna selection chooses N active antennas from M available antennas, with N ranging from K to M while M ≫ K.The selected antennas and corresponding transceiver chains are activated; the remaining M−N chains are switched off.
- Selection criterion: The selection objective is the average DPC capacity over all L subcarriers, using one common antenna set across subcarriers.Different antenna combinations may be optimal on individual subcarriers, but practical OFDM operation requires a common selection.
- Evaluation: The resulting selection is also evaluated using ZF sum-rate, although the DPC-optimal antenna set need not be optimal for ZF.ZF provides a more practical linear-precoding performance measure than DPC.
B. Antenna Selection Using Convex Optimization
The convex-optimization method approximates capacity-maximizing antenna selection by relaxing binary antenna decisions, producing a tractable near-optimal benchmark.
- Optimization procedure: The method first fixes equal user power allocation, selects antennas maximizing average capacity, then optimizes power allocation on each subcarrier.This decomposition provides a lower bound on adaptive-selection performance because it does not guarantee the global optimum.
- Convex relaxation: Relaxing binary antenna variables to continuous values between 0 and 1 converts the selection problem into a convex optimization problem solvable in polynomial time.The N largest relaxed variables determine the selected antenna indices.
- Computational complexity: The original binary antenna-selection problem is NP-hard because each antenna decision variable must be integer-valued.The relaxation trades exact integrality for computational tractability.
C. Antenna Selection Based on Received Power
The received-power method selects antennas using a simple per-antenna power ranking, offering very low complexity and potentially near-optimal performance when channel correlation is low.
- Selection rule: The scheme selects the N antennas with the highest received power, averaged over all K users and L subcarriers.Uplink power measurements can support downlink selection by exploiting channel reciprocity.
- Complexity: The power-based scheme has much lower complexity than the convex-optimization scheme because it uses only received-power measurements.It avoids explicitly incorporating antenna-channel correlation into the selection decision.
- Performance condition: The power-based method may become near-optimal in NLOS scenarios with rich scattering and relatively low antenna-channel correlation.In general, it performs worse than the convex-optimization method, but measured channels show fairly good performance.
V. MEASURED CHANNELS
The evaluation uses outdoor 2.6 GHz channel measurements from linear and cylindrical 128-element arrays in a semi-urban environment with multiple LOS and NLOS user sites.
- Array configurations: Measurements span 50 MHz at 2.6 GHz using cylindrical and linear base-station arrays, each containing 128 antenna elements.Both arrays use half-wavelength adjacent-element spacing; the cylindrical array has 128 ports from dual-polarized patch antennas.
- Cylindrical array: The cylindrical array is physically compact, with diameter and height of about 30 cm.It consists of four stacked circles, each containing 16 dual-polarized directional patch antennas.
- Measurement environment: Outdoor measurements were conducted around Lund University’s E-building, with both base-station arrays placed on the same roof.The user antenna was moved among eight measurement sites around the building.
- Propagation conditions: The measurement sites include three LOS sites, four NLOS sites, and one site with array-dependent LOS conditions.At MS 4, the LOS component is blocked for the linear array, while diffraction produces LOS characteristics there.
- Array comparison: With random selection of the same number of antennas, the linear array achieves higher average sum-rates than the cylindrical array.Its larger aperture provides higher angular resolution and better spatial separation of closely located users.
VI. PERFORMANCE RESULTS IN MEASURED CHANNELS
The study applies antenna selection to measured channels, selecting the best N of 128 antennas across 161 subcarriers for 4, 16, and 40 users.
- The evaluation uses measured channels and compares antenna selection with random selection while varying the number of active RF transceivers.The convex-optimization scheme is treated as near-optimal, and performance is assessed as N grows from K to 128.
A. Performance of Convex-Optimization Selection Scheme
The experiments begin with four users and then increase the user count to sixteen and forty to assess antenna selection across propagation scenarios and system loads.
- Four users are evaluated first, followed by sixteen and forty users, reflecting massive MIMO’s ability to serve more simultaneous users.
- The four-user study combines user separation with LOS or NLOS propagation to define two reference scenarios.
1) Four users, K = 4:
For four users, adaptive antenna selection is far more beneficial in measured channels than in i.i.d. Rayleigh channels, while substantial RF-chain reductions remain possible across LOS and NLOS settings.
- In i.i.d. Rayleigh channels, adaptive selection provides only a very small gain over random antenna combinations because all antennas are equally good.
- In closely located LOS measurements, 40 RF transceivers yield gains of 11% in DPC capacity and 18% in ZF sum-rate for the linear array versus random selection.
- In the same LOS setting, the cylindrical array achieves gains above 30% in both DPC capacity and ZF sum-rate at 40 RF transceivers.
- In well-separated NLOS measurements, adaptive selection at 40 RF transceivers improves both DPC capacity and ZF sum-rate by 10% for the linear array and 20% for the cylindrical array.
- With four users, 50–60 RF transceivers can be switched off while losing only 10% of full MIMO performance, depending on array and propagation scenario.
2) Sixteen and forty users, K = 16 and K = 40:
With sixteen or forty users, antenna selection gains depend strongly on array geometry, while increased loading requires more active transceivers to maintain performance.
- With more users, ZF sum-rate can decrease when the number of active transceivers is close to the user count because of high interuser interference.
- At 60 RF transceivers, selection gains for sixteen users are 6% for the linear array and 14% in DPC capacity and 17% in ZF sum-rate for the cylindrical array.
- At 60 RF transceivers, selection gains for forty users are 4% for the linear array and 16% in DPC capacity and 50% in ZF sum-rate for the cylindrical array.
- For sixteen users, 90% of full MIMO performance requires more than 80 linear-array and more than 70 cylindrical-array RF transceivers; for forty users, it requires 90 and 80, respectively.
- The cylindrical array performs better with relatively few active transceivers, whereas the linear array becomes superior as user count and active-transceiver count increase.High angular resolution helps the linear array spatially separate larger numbers of users.
- Across propagation conditions and user counts, many RF transceivers can be switched off to save energy and simplify massive MIMO systems.The study uses convex-optimization selection as a benchmark before evaluating a simpler power-based scheme.
FULL MIMO PERFORMANCE, WITH THE CONVEX-OPTIMIZATION
The “mixed” scenario combines co-located and widely separated users, with equal proportions of LOS and NLOS conditions.
- “Mixed” scenarios include users co-located at the same site and users at different sites with large spacing.
- Half of the users are in line-of-sight (LOS) conditions.
- Half of the users are in non-line-of-sight (NLOS) conditions.
B. Performance of Power-Based Antenna Selection
Power-based antenna selection approaches convex optimization closely in many measured-channel scenarios, but performance depends on user separation, propagation conditions, array type, SNR, and the number of active transceivers.
- Antenna correlation creates a trade-off with SNR for spatial multiplexing, which convex optimization considers but power-based selection ignores.Power-based selection is expected to work better when users are well separated under NLOS conditions, where antenna-channel correlation is lower.
- Below 1% loss requires about 70 RF transceivers for four co-located LOS users with DPC and 90 with ZF, for both arrays.With fewer transceivers, performance losses are relatively high in this difficult scenario.
- Below 1% loss occurs with 20 RF transceivers for four well-separated NLOS users, in both DPC capacity and ZF sum-rate.The comparison is relative to the convex-optimization scheme.
- For sixteen users, 60 transceivers suffice for below 1% DPC loss, whereas more than 80 are needed with ZF for both arrays.
- For forty users, below 1% loss requires about 70 and 90 transceivers with the linear array, versus up to 100 and 120 with the cylindrical array, for DPC and ZF respectively.With forty users and cylindrical-array ZF precoding, almost all transceivers are needed.
- The power-based scheme selects high-gain antennas, while convex optimization can choose lower-gain antennas to avoid high correlation.On the cylindrical array, convex optimization spreads selections across circles rather than selecting only antennas facing user directions.
- Overall, power-based selection gives competitive performance, and many RF transceivers can generally be switched off while retaining 90% of full-MIMO performance.The most difficult cases involve closely spaced LOS users and many users served by the cylindrical array.
- At low SNR, high channel gains favor array gain; at high SNR, user separation becomes more important for spatial multiplexing.The paper leaves measuring base-station hardware consumption and optimizing transmit power and active-transceiver count for future work.
VII. SUMMARY AND CONCLUSIONS
Measured real propagation channels show unequal antenna contributions, enabling antenna selection to reduce RF-chain requirements with limited performance loss. The results also motivate channel models beyond i.i.d. Rayleigh fading.
- Unlike i.i.d. Rayleigh channels, real propagation channels can make some antennas contribute more than others.Large-scale fading across arrays and differences in antenna patterns produce this unequal contribution.
- Measured 2.6 GHz channels from linear omni-directional and cylindrical patch-element arrays demonstrate gains from adaptive antenna selection over random selection.
- A substantial number of RF transceiver chains can be turned off without significant performance loss.
- Convex-optimization selection is near-optimal, while received-power-only selection remains competitive in the measured channels.
- Antenna selection may reduce massive-MIMO implementation complexity, cost, and hardware energy consumption.The contrast between theoretical and measured channels also highlights the importance of developing new massive-MIMO channel models.