TL;DR
This paper tackles the challenge of efficient beam training and hybrid beamforming for Terahertz (THz) multi-user massive MIMO systems augmented with Intelligent Reflecting Surfaces (IRS). They propose a novel, low-complexity cooperative beam training scheme and corresponding hybrid beamforming designs that achieve performance close to optimal fully digital beamforming, even with estimated channel state information.
What Problem Does This Paper Address?
The demand for higher data rates is pushing wireless communication towards higher frequencies, with Terahertz (THz) bands offering vast bandwidth. However, THz signals suffer from severe path attenuation and poor diffraction, limiting their coverage. To overcome this, researchers are combining massive MIMO (for beamforming gains) and Intelligent Reflecting Surfaces (IRS) (for aperture gains). The main hurdles are: 1) efficient channel estimation in the presence of passive IRSs and 2) the high cost and complexity of fully digital beamforming in massive MIMO. Existing solutions for channel estimation and beamforming are often not scalable or practical for this combined system, especially in THz where signal strength is a major concern. This paper aims to bridge this gap by developing a practical and efficient solution for beam training and hybrid beamforming in IRS-assisted THz multi-user massive MIMO.
Methods
The authors propose a two-pronged approach:
- Cooperative Beam Training:
- Partial IRS Search: Instead of searching for exact path angles at the IRS, they focus on searching for angle differences in the sine space, significantly reducing complexity.
- Ternary-Tree Hierarchical Search: For the BS and users, they introduce a ternary-tree hierarchical search, which is shown to be more efficient than traditional binary-tree searches. This involves multiple stages of wide and narrow beams.
- Novel Codebooks: Two codebooks are designed for the ternary-tree search:
- TD (Tree Dictionary) Codebook: Offers better anti-noise capability but requires more RF chains.
- PSD (Phase Shift Deactivation) Codebook: More sensitive to noise but requires fewer RF chains, leading to faster search.
- Beam Distribution: They propose a non-uniform beam distribution strategy that prioritizes coverage-edge gain, leading to lower misalignment probability compared to uniformly distributed beams.
- Hybrid Beamforming (HB) Designs:
- Single-User: A Direct Power Allocation (DPA) scheme is proposed, leveraging the near-orthogonality of massive MIMO channels.
- Multi-User: Block Diagonalization (BD) is applied to eliminate inter-user interference, which is crucial for maximizing sum-rate in multi-user scenarios.
Datasets
Not discussed in the paper. The evaluation relies on simulated channel realizations.
Key Insights and Ideas
- IRS Angle Difference Search: The realization that searching for angle differences at the IRS, rather than absolute angles, dramatically simplifies the problem. This is a key insight for making IRS-assisted systems practical.
- Ternary-Tree Efficiency: Demonstrating that a ternary-tree search is more efficient than a binary-tree search for beam training in this context. This is a novel contribution to hierarchical search algorithms.
- Non-Uniform Beam Coverage: The idea of designing beams with uniform coverage-edge gain rather than uniform beamwidth is crucial for robust beam alignment, especially in scenarios with potential misalignment.
- Codebook Trade-offs: The TD and PSD codebooks offer a clear trade-off between anti-noise performance and hardware complexity (RF chains), allowing system designers to choose based on their priorities.
- HB Design for THz: The proposed DPA and BD designs for hybrid beamforming are tailored for the specific challenges of THz massive MIMO with IRS, showing how to leverage estimated CSI effectively.
Key Results
- Beam Training Accuracy: The proposed ternary-tree beam training schemes (both TD and PSD codebooks) achieve 100% detection probability in high SNR regimes.
- Performance Close to Optimal: The proposed joint beam training and HB scheme achieves performance very close to optimal fully digital beamforming, even when implemented with estimated channel state information.
- IRS Benefits: IRS-assisted schemes show significant performance gains (achievable rate) compared to systems without IRS or with random IRS phase shifts, validating the aperture gain provided by IRS.
- HB Design Effectiveness: In multi-user scenarios, the proposed HB with BD significantly outperforms HB with DPA at higher transmit powers due to its ability to mitigate inter-user interference.
- Codebook Performance: The TD codebook demonstrates better anti-noise performance than the PSD codebook, requiring a lower SNR for 100% correct detection.
- Complexity Reduction: The proposed methods offer substantial reductions in search complexity compared to exhaustive search, making them practical for massive MIMO systems. For example, with N=27, exhaustive search requires over 500,000 tests, while the proposed TD and PSD methods require significantly fewer (519 and 173, respectively).
Limitations and Future Work
- Narrowband Assumption: The proposed DPA and BD designs are dedicated to narrowband beamforming. The authors acknowledge that for wideband communications, the beam squint effect needs to be considered, which is a direction for future work.
- Quantization Error: The performance gap between estimated CSI and perfect CSI increases with antenna number, attributed to quantization errors in AoDs/AoAs. Higher training accuracy (N/Na) reduces this gap.
- LoS Blockage/Idle Users: The current designs assume that all BS-user paths can be estimated. Scenarios with LoS blockage or idle users would require modifications to remove corresponding ARVs.
- Future Work: The paper suggests that the proposed scheme is appealing for future THz massive MIMO networks due to its close-to-optimal performance and low complexity, especially with advancements in large-scale antenna arrays and quantization resolution.
Summary updated