Source-linked AI summary
Linear Precoding in Cooperative MIMO Cellular Networks with Limited Coordination Clusters
Chris T. K. Ng, Howard Huang
TL;DR
The paper addresses concave rate-utility maximization in cooperative MIMO downlinks with limited, overlapping coordination clusters and practical linear precoding constraints. It proposes soft interference nulling, computed through sequential convex optimization, and reports performance at least as good as interference-free linear precoding, with gains over full-network ZF or myopic ZF in supported settings.
Problem
Cooperative downlink networks need practical linear precoding for concave rate-utility optimization when coordination clusters are limited and overlapping, while optimal DPC is complex and optimal linear precoding is generally nonconvex.
Method
The paper proposes soft interference nulling, an approximation obtained by convexifying the linear-precoder optimization and solving a sequence of convex problems.
Results
SIN performs better than or equal to any linear interference-free precoding scheme and can outperform full-network ZF at moderate SNRs and myopic ZF under overlapping clusters.
Takeaways & Limitations
SIN is particularly effective for partial coordination with overlapping clusters, where its rates improve with cluster size while myopic-ZF rates improve only marginally.
Abstract
from arXiv · showhide
In a cooperative multiple-antenna downlink cellular network, maximization of a concave function of user rates is considered. A new linear precoding technique called soft interference nulling (SIN) is proposed, which performs at least as well as zero-forcing (ZF) beamforming. All base stations share channel state information, but each user's message is only routed to those that participate in the user's coordination cluster. SIN precoding is particularly useful when clusters of limited sizes overlap in the network, in which case traditional techniques such as dirty paper coding or ZF do not directly apply. The SIN precoder is computed by solving a sequence of convex optimization problems. SIN under partial network coordination can outperform ZF under full network coordination at moderate SNRs. Under overlapping coordination clusters, SIN precoding achieves considerably higher throughput compared to myopic ZF, especially when the clusters are large.
I. INTRODUCTION
The paper studies cooperative downlink MIMO cellular networks, where base-station cooperation addresses interference but practical precoding and coordination constraints remain important.
- Without base-station cooperation, increasing transmit power cannot improve SINR because it also strengthens interference.
- Joint encoding across cooperating base stations can overcome the interference limitation and model the downlink as a broadcast channel.
- Dirty paper coding is theoretically optimal for the broadcast channel but can be too complex for practical implementation.
- The paper considers concave utility maximization using linear precoding, treating interference as noise, with shared CSI and user messages routed only within coordination clusters.The network uses narrow-band flat-fading channels, accurate and timely channel-state knowledge, and per-base-station short-term power constraints.
- Coordination clusters may be limited and overlap, so base stations can participate in multiple users’ transmissions.
III. COOPERATIVE CELLULAR NETWORKS
The cooperative model assigns each user a base-station cluster, restricts message access accordingly, and represents transmission through linear precoding and signal covariance matrices.
- A. Base Station Coordination Clusters: Each user specifies a coordination cluster whose base stations jointly encode that user’s message, while other base stations cannot access it.
- A. Base Station Coordination Clusters: Clusters may overlap, allowing one base station to participate in transmissions for multiple users.
- B. Linear Precoding: Linear precoding expresses each base station’s transmit signal as a sum of precoded user information signals.
- B. Linear Precoding: A user’s information signal may contain multiple spatial streams, up to the total number of transmit antennas in its coordination cluster.
- B. Linear Precoding: Association matrices enforce zero precoding weights at nonparticipating base stations and encode the base-station–user cluster structure.
- B. Linear Precoding: Given aggregate channels and covariance matrices, SVD or MMSE-SIC detection achieves the stated MIMO rate when interference is treated as noise.
C. Optimal Precoder
The optimal linear-precoder design maximizes a concave utility of user rates subject to covariance and power constraints, but the resulting problem is generally nonconvex.
- The objective is a concave utility function of the user-rate vector, including weighted sum rate and sum rate as examples.
- The design variables are users’ covariance matrices, from which the corresponding precoding matrices are recovered.
- The optimization includes rate, covariance, and per-base-station power constraints.
- The optimization is generally nonconvex, making the optimal rates and covariance matrices difficult to compute efficiently.
A. Precoder Optimization
Soft interference nulling (SIN) convexifies the nonconvex linear-precoder design around an operating point and iteratively refines the solution. It performs at least as well as linear interference-free schemes while permitting nonzero interference and automatically determining active users.
- SIN formulation: SIN convexifies the precoder optimization around a given operating point and solves the resulting convex problem.The convex subproblem can be efficiently solved with standard numerical methods such as interior-point methods.
- Iterative optimization: The iterative algorithm updates each operating covariance from the previous solution and stops when utility improvement falls below ε.The iterations are monotonically nondecreasing, and bounded user rates imply convergence to a local optimum.
- Performance guarantee: SIN performs at least as well as any linear precoding scheme that completely eliminates interference, including zero-forcing beamforming.The guarantee follows because an interference-free solution is feasible for the SIN optimization and the surrogate objective is a global under-estimator.
- Soft interference nulling: Unlike interference-free precoding, SIN relaxes the interference constraint and allows nonzero interference rather than imposing an infinite interference penalty.This relaxation is useful when strict interference elimination is infeasible or overly restrictive.
- Operational properties: SIN remains well-defined when transmit antennas are fewer than total receive antennas, and its algorithm determines which users have nonzero rates.A separate user-selection step is therefore unnecessary under SIN precoding.
B. Clustering Algorithms
The paper uses clustering based on long-term channel conditions because exhaustive search over cluster combinations is too complex. It evaluates nearest-bases and nearest-interferers clustering rules.
- Clustering assumptions: Users choose coordination clusters from long-term channel conditions and do not adapt them to fast Rayleigh fading.The cluster-selection rules use shadowing realizations rather than instantaneous fading.
- Nearest Bases Clustering: Nearest-bases clustering assigns each user the |Bi| bases with the greatest signal strength.The selected bases form the user’s coordination cluster.
- Nearest Interferers Clustering: Nearest-interferers clustering selects a strongest home base and adds bases associated with the user’s most-interfered neighbors.The cluster contains the home base plus bases serving the |Bi| −1 neighbors suffering the most interference.
A. Line Network
The line-network model places one base station and one mobile user per site along a wraparound line, with equal inter-site and user-base distances. It uses K = B users.
- Network structure: The network contains B base stations positioned along a line, with each base station serving one mobile user.Thus, the total number of users is K = B.
- Network geometry: Each user is located dy away from its serving base station, while neighboring base stations are spaced dx apart.Wraparound is used to reduce boundary effects.
1) Network Geometry:
The line-network geometry uses wraparound indexing and distance-based Rayleigh fading, with unit spacing, path-loss exponent four, and single antennas at every base station and user.
- Network Geometry:: Wraparound distance indexing defines d(i, j) as the modular separation between User i and Base j.The separation takes values from 0 through B −1.
- Channel model: Channels use independent Rayleigh fading with distance-based power attenuation governed by path-loss exponent η.Each channel entry is modeled as an i.i.d. complex Gaussian variable whose variance depends on distance.
- Network Geometry:: The line network sets dx = dy = 1 and η = 4.These parameters specify the geometry and propagation loss used in the line-network evaluation.
- SISO specialization: Every base station and user has one antenna, so the channel from Base j to User i is represented by a scalar hij.This is the single-input single-output specialization of the MIMO model.
- Full cooperation baseline: With perfect cooperation, the cellular downlink can be modeled as a broadcast channel with an B-antenna transmitter under per-antenna power constraints.Dirty paper coding achieves the Gaussian MIMO broadcast-channel capacity region in this fully cooperative case.
- ZF baseline: Zero-forcing beamforming uses the channel pseudoinverse to eliminate interference at each mobile user when all base stations jointly encode messages.Unlike dirty paper coding, it performs no interference pre-subtraction at the transmitter.
3) Full-Network Zero-Forcing:
In the line network, full-network ZF overcomes the interference-limited bottleneck but requires every base station to cooperate. SIN can outperform full-network ZF at moderate SNRs with limited coordination, although partial coordination becomes interference-limited at high SNR.
- Without cooperation, average user rate saturates at approximately 1.6 bps/Hz, while cooperative ZF achieves increasing throughput as SNR rises.ZF and DPC show similar scaling trends, although ZF remains below DPC.
- Full-network ZF requires all base stations to cooperate to satisfy the interference-free condition.
- At low SNRs, SIN outperforms ZF because ZF suffers from noise amplification.
- At SNR P = 18 dB, SIN with coordination cluster size 7 outperforms ZF with full network coordination.
- As SNR increases, SIN with partial network coordination becomes interference-limited again.
B. Cellular Network
In the hexagonal three-sectored network, limited coordination is evaluated using different clustering strategies and compared with myopic ZF. SIN achieves higher throughput and benefits more from larger coordination clusters.
- Network model: The cellular network contains 19 wraparound hexagonal cells, 57 sectors, one transmit antenna per sector, and one user per sector.Users are randomly populated, with propagation modeled using path loss and shadow fading.
- Clustering: Nearest-interferers clustering outperforms nearest-bases clustering, indicating greater benefit from mitigating interference than boosting users’ signal strength.
- Performance comparison: For the same coordination cluster size, SIN achieves considerably higher throughput than myopic ZF.
- Performance comparison: SIN rates improve as coordination clusters grow, whereas myopic-ZF rates improve only marginally.
- Mechanism: SIN outperforms myopic ZF by optimizing transmit-power allocation and balancing beamforming against interference nulling.