Measurement-driven Evaluation of All-digital Many-antenna Full-duplex Communication

Evan Everett, Clayton Shepard, Lin Zhong, Ashutosh Sabharwal

Introduction

Full-duplex (FD) wireless communication—the ability to transmit and receive on the same frequency simultaneously—has long been considered a "holy grail" for increasing spectral efficiency. Theoretically, it can double the capacity of wireless links. However, the practical implementation of FD is hindered by self-interference (SI), where a radio's own high-power transmission overwhelms the significantly weaker incoming signal. Traditionally, researchers believed that complex and bulky analog cancellation circuitry was mandatory to prevent receiver saturation.

The SoftNull Digital Architecture

SoftNull proposes an alternative: an all-digital approach that leverages the high dimensionality of massive MIMO (many-antenna) arrays to suppress SI. By utilizing the spatial degrees of freedom available in these large arrays, SoftNull can reduce interference to manageable levels without the need for dedicated analog hardware.

SoftNull architecture overview, showing the two-stage digital precoding and the self-interference channel between antenna sets.

The system employs a two-stage digital precoding strategy. First, a standard Multi-User MIMO (MU-MIMO) precoder handles downlink beamforming to specific users. Second, the SoftNull Precoder maps these signals onto the physical transmit antennas. This stage uses a projection based on the Singular Value Decomposition (SVD) of the self-interference channel ($H_{Self}$). Instead of attempting perfect nulling, which would consume excessive spatial resources, SoftNull performs "soft" nulling. It projects signals onto the subspace of the smallest singular values of the SI channel, reducing interference just enough to stay within the receiver's dynamic range while preserving dimensions for data transmission.

Measurement-Driven Evaluation

To validate this approach, the authors conducted extensive measurements using the ArgosV2 platform. This experimental setup features a 72-element 2.4 GHz patch antenna array interfaced with WARP v3 software-defined radios.

The ArgosV2 platform used for measurements, featuring a 72-element antenna array and WARP v3 software-defined radios.

The study utilized over 12 million wideband channel measurements collected across three distinct environments: an anechoic chamber, an outdoor open field with low scattering, and a highly scattered indoor environment. These traces, totaling 40 GB, have been made publicly available to the research community.

Key Insights and Performance

The effectiveness of SoftNull is closely tied to the "many-antenna" regime ($M \gg K$), where the abundance of base station antennas ($M$) relative to users ($K$) provides the necessary spatial dimensions for SI suppression.

Experimental results demonstrate that environment plays a critical role. In low-scattering outdoor environments, the SI channel is highly correlated, allowing for significant interference reduction with minimal loss of spatial dimensions. Indoors, scattering decorrelates the channel, requiring more "effective antennas" to be sacrificed for the same level of suppression. Specifically, in an outdoor setup with 36 transmit and 36 receive antennas, SoftNull achieved 50 dB of SI reduction by sacrificing only 12 transmit dimensions. Indoors, achieving the same reduction required 20 dimensions.

Furthermore, the spatial arrangement of antennas matters. The authors found that partitioning the array into contiguous blocks for transmission and reception (e.g., left half vs. right half) outperformed interleaved patterns by limiting the angular spread of interference.

Results and Conclusion

In terms of throughput, SoftNull shows significant gains for small-cell scenarios. For a 4-user setup at 85 dB path loss, the system achieved a rate improvement of over 40% compared to traditional half-duplex systems. However, a "crossover" point exists: as user density increases or path loss becomes extreme, the overhead of nulling SI outweighs the benefits, and half-duplex operation becomes more efficient.

While SoftNull demonstrates the feasibility of all-digital full-duplex in massive MIMO, future work remains. This includes managing inter-node interference between uplink and downlink users, developing adaptive antenna partitioning strategies, and exploring joint optimization of user SNR and SI suppression.

Summary updated

BibTeX
@article{everett2016measurement,
  author = {Evan Everett and Clayton Shepard and Lin Zhong and Ashutosh Sabharwal},
  title = {{Measurement-driven Evaluation of All-digital Many-antenna Full-duplex Communication}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2016},
  doi = {10.1109/twc.2016.2612625},
}
Figure