Sequential Beamforming for Multiuser MIMO With Full-Duplex Training

Xu Du, John Tadrous, Ashutosh Sabharwal

TL;DR

This paper introduces a novel "sequential beamforming" strategy for multiuser MIMO systems that leverages full-duplex technology at the base station. By allowing the base station to transmit data to some users while simultaneously receiving CSI from others, the strategy significantly reduces training overhead and boosts spectral efficiency, especially in scenarios with limited training resources.

What Problem Does This Paper Address?

Multiuser MIMO systems promise higher spectral efficiency by allowing a base station to communicate with multiple users simultaneously. However, achieving this requires accurate Channel State Information (CSI) at the transmitter. Obtaining this CSI typically involves a training phase, which consumes valuable time and resources, thereby reducing the overall spectral efficiency. This training overhead is particularly problematic in systems with many users or time-varying channels. Existing methods for acquiring CSI, like closed-loop (user sends CSI) and open-loop (base station infers CSI from uplink), both incur significant overhead. This paper addresses the challenge of reducing this CSI acquisition overhead to improve spectral efficiency in multiuser MIMO downlink systems.

Methods

The core innovation is a sequential beamforming strategy. Instead of waiting for all users to complete their CSI training, the base station initiates downlink data transmission to users as soon as their CSI is acquired. This is enabled by a full-duplex base station, which can transmit and receive simultaneously. The strategy involves pre-scheduled users sending CSI in orthogonal time slots. As CSI for a user is received, the base station updates its precoding matrix and starts transmitting data to that user and any previously trained users. This process continues until all users have been trained. The paper then analyzes the spectral efficiency of this strategy, considering both inter-beam interference (due to imperfect CSI) and inter-node interference (from simultaneous transmission and reception). They derive analytical expressions for optimal training durations and spectral efficiency in both finite and high SNR regimes, comparing it to traditional half-duplex systems.

Datasets

The paper uses experimental data from [1] and a 3GPP 3D channel model [2] for validation.

Key Insights and Ideas

  • Full-duplex as an efficiency booster: The paper demonstrates that full-duplex capability isn't just for simultaneous uplink/downlink traffic, but can be cleverly used to reduce overhead in downlink-only scenarios by enabling concurrent data transmission and CSI acquisition.
  • Sequential training is key: The "sequential" aspect is crucial. By not waiting for all CSI, the system can start utilizing the channel for data transmission much earlier, leading to significant gains.
  • Trade-off between training and data: The optimal training duration is a critical parameter. Too little training leads to poor CSI and high inter-beam interference, while too much training reduces the time available for actual data transmission. The paper analytically finds this sweet spot.
  • Inter-node interference is manageable: While full-duplex introduces inter-node interference, the analysis shows that with proper optimization of training duration, this interference can be managed, and the spectral efficiency gains are still substantial.
  • Closed-loop vs. Open-loop: The paper highlights distinct performance characteristics for closed-loop and open-loop training in conjunction with sequential beamforming, offering guidance on which to choose based on system parameters like the number of antennas and block length.

Key Results

  • Significant spectral efficiency improvement: For an 8x8 MIMO system with a block length of 500 symbols, the proposed sequential beamforming strategy achieved a spectral efficiency improvement of 130% with closed-loop training and 8% with open-loop training compared to its half-duplex counterpart.
  • Optimal training duration analysis: The paper provides analytical approximations for the optimal training duration, showing that it scales with the number of antennas (quadratically for closed-loop, √M for open-loop) and block length (log T for closed-loop, √T for open-loop).
  • High SNR analysis: In the high SNR regime, the sequential beamforming strategy can double the spectral efficiency of unidirectional downlink communication under certain conditions.

Limitations and Future Work

  • User power constraints: The analysis considers limited user training power, which is realistic but adds complexity.
  • Symmetric CSI acquisition: The current strategy assumes symmetric CSI acquisition time for all users. The authors suggest that asymmetric allocation of training symbols could further improve performance.
  • MAC layer designs: Aspects like deafness effects and user selection for orthogonality are not explicitly considered, representing avenues for future work.
  • Full-duplex base station model: The paper assumes a full-duplex MIMO array where each antenna is used for both transmission and reception. Future work could explore systems with separate transmit and receive antennas.
  • Inter-node interference regimes: The paper identifies different regimes for inter-node interference impact in the high SNR analysis, suggesting further investigation into these specific scenarios.

Summary updated

BibTeX
@article{du2016sequential,
  author = {Xu Du and John Tadrous and Ashutosh Sabharwal},
  title = {{Sequential Beamforming for Multiuser MIMO With Full-Duplex Training}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2016},
  doi = {10.1109/twc.2016.2616338},
}
Figure