Source-linked AI summary
Structured Compressive Sensing Based Spatio-Temporal Joint Channel Estimation for FDD Massive MIMO
Zhen Gao, Linglong Dai, Wei Dai, Byonghyo Shim, Zhaocheng Wang
TL;DR
FDD massive MIMO needs accurate channel estimation without the unaffordable pilot overhead caused by large BS antenna arrays. The paper combines CS-based non-orthogonal pilots, structured joint recovery, and adaptive pilot reuse, achieving accurate estimation with reduced overhead and performance approaching oracle least squares.
Problem
FDD massive MIMO requires CSI for operation, but conventional channel estimation incurs prohibitively high pilot overhead as the number of BS antennas grows.
Method
The scheme uses non-orthogonal CS pilots, an ASSP algorithm exploiting spatio-temporal common sparsity, and a space-time adaptive pilot design.
Results
The proposed ASSP algorithm approaches oracle least-squares performance for ηp ≥19.04% and achieves a 9 dB gain over the scheme in.
Takeaways & Limitations
Spatio-temporal common sparsity enables reliable FDD massive-MIMO channel estimation with significantly reduced pilot overhead.
Abstract
from arXiv · showhide
Massive MIMO is a promising technique for future 5G communications due to its high spectrum and energy efficiency. To realize its potential performance gain, accurate channel estimation is essential. However, due to massive number of antennas at the base station (BS), the pilot overhead required by conventional channel estimation schemes will be unaffordable, especially for frequency division duplex (FDD) massive MIMO. To overcome this problem, we propose a structured compressive sensing (SCS)-based spatio-temporal joint channel estimation scheme to reduce the required pilot overhead, whereby the spatio-temporal common sparsity of delay-domain MIMO channels is leveraged. Particularly, we first propose the non-orthogonal pilots at the BS under the framework of CS theory to reduce the required pilot overhead. Then, an adaptive structured subspace pursuit (ASSP) algorithm at the user is proposed to jointly estimate channels associated with multiple OFDM symbols from the limited number of pilots, whereby the spatio-temporal common sparsity of MIMO channels is exploited to improve the channel estimation accuracy. Moreover, by exploiting the temporal channel correlation, we propose a space-time adaptive pilot scheme to further reduce the pilot overhead. Additionally, we discuss the proposed channel estimation scheme in multi-cell scenario. Simulation results demonstrate that the proposed scheme can accurately estimate channels with the reduced pilot overhead, and it is capable of approaching the optimal oracle least squares estimator.
I. INTRODUCTION
FDD massive MIMO requires accurate CSI, but conventional pilot overhead grows prohibitively with the number of BS antennas. The paper addresses this challenge by exploiting sparse, spatially common, and temporally persistent delay-domain channels.
- Accurate CSI is essential for massive MIMO, yet estimating channels across hundreds of BS transmit antennas creates prohibitively high pilot overhead in FDD systems.
- Conventional approaches often rely on TDD reciprocity, but calibration errors and limited coherence time make this assumption problematic for CSI acquisition.
- Delay-domain broadband wireless channels are sparse because relatively few significant scatterers dominate the channel energy.
- Co-located BS antennas observe similar path delays from common scatterers, producing spatial common sparsity across delay-domain MIMO channels.
- The proposed SCS-based scheme combines non-orthogonal pilots, adaptive structured pursuit, and temporal correlation to reduce pilot overhead while improving channel estimation.
- Path delays vary more slowly than path gains, so the spatial sparsity pattern remains nearly unchanged across successive OFDM symbols.
III. PROPOSED SCS-BASED SPATIO-TEMPORAL JOINT CHANNEL ESTIMATION SCHEME
The proposed scheme uses CS-based non-orthogonal pilots and structured joint recovery to estimate high-dimensional FDD massive-MIMO channels from limited measurements. ASSP exploits common sparsity across antennas and OFDM symbols while adaptively determining sparsity.
- The scheme combines non-orthogonal BS pilots, user-side ASSP recovery, and a space-time adaptive pilot strategy for progressively lower pilot overhead.
- Non-Orthogonal Pilot Design: Non-orthogonal pilots let different transmit antennas share pilot subcarriers, replacing the high overhead of conventional orthogonal pilot allocation.
- SCS-Based Channel Estimation at the User: Because the aggregate CIR is high-dimensional and underdetermined from limited pilots, the method reformulates channel recovery as structured sparse signal estimation.
- SCS-Based Channel Estimation at the User: Jointly processing R OFDM symbols exploits a CIR matrix whose structured sparsity reflects common delay support across antennas and time.
- Adaptive Structured Subspace Pursuit: ASSP extends subspace pursuit by exploiting structured sparsity and adaptively acquiring the channel sparsity level instead of requiring it as prior information.
C. Space-Time Adaptive Pilot Scheme
The space-time adaptive pilot scheme adapts pilot reuse to antenna geometry and user mobility. It groups nearby antennas for spatial common sparsity and interpolates channels between pilot-bearing OFDM symbols.
- Nearby BS antennas are partitioned into NG groups so spatial common sparsity is preserved within each group even when the full array is widely spaced.
- Pilots are non-orthogonal within an antenna group but orthogonal across groups in time or frequency, balancing pilot reuse with group separation.
- The pilot-sharing interval fd is selected using temporal channel correlation, allowing larger fd values to reduce pilot overhead.
- Channels of OFDM symbols without pilots are estimated by interpolating between channel estimates from adjacent pilot-bearing symbols.
- The scheme considers both antenna-array geometry and user mobility while applying ASSP separately to antenna groups.
D. Channel Estimation in Multi-Cell Massive MIMO
The scheme is extended to multi-cell FDD massive MIMO, where pilot contamination creates a trade-off between overhead and estimation performance. TDM preserves single-cell pilot overhead with slight performance loss relative to FDM.
- The proposed channel estimation scheme is extended from single-cell to multi-cell FDD massive MIMO.
- FDM can perfectly mitigate pilot contamination when training fits within channel coherence time, but increases pilot overhead by L times.
- TDM keeps multi-cell pilot overhead equal to the single-cell case by transmitting adjacent-cell pilots in different time slots.
- TDM is presented as viable because it substantially reduces overhead while causing only slight performance loss compared with FDM.
IV. PERFORMANCE ANALYSIS
The performance analysis designs a CS sensing matrix through pilot placement and sequences, using uniformly spaced subcarriers and random pilot phases to obtain favorable column correlations with limited pilots.
- A. Non-Orthogonal Pilot Design Under the Framework of CS Theory: The analysis designs the sensing matrix Ψ through pilot placement ξ and pilot sequences {p_m}^M_m=1.
- A. Non-Orthogonal Pilot Design Under the Framework of CS Theory: The proposed pilot sequences use i.i.d. uniform random phases, giving each sensing-matrix column a constant l2-norm.
- A. Non-Orthogonal Pilot Design Under the Framework of CS Theory: The proposed pilot sequences achieve good cross-correlation among columns of Ψ_l for any l according to random matrix theory.
- A. Non-Orthogonal Pilot Design Under the Framework of CS Theory: The design analyzes cross-correlation for equal and unequal transmit antennas and delay indices to establish asymptotic orthogonality conditions.
- A. Non-Orthogonal Pilot Design Under the Framework of CS Theory: The pilot placement uses uniformly spaced subcarriers, selected because N_p > L under the stated antenna and delay-spread considerations.
- A. Non-Orthogonal Pilot Design Under the Framework of CS Theory: Compared with random pilot placement, uniformly spaced placement is easier to implement and supports backward compatibility with existing cellular networks.
B. Convergence Analysis of Proposed ASSP Algorithm
The ASSP convergence analysis covers both correctly specified and mismatched sparsity levels for structured sparse matrices. Its support initialization and stopping criteria address recovery efficiency and noise-dominated paths.
- Unlike classical and model-based SP analyses for single sparse vectors, ASSP analyzes reconstruction of a structured sparse matrix.
- For s = P, the ASSP convergence guarantee follows from Theorem 1 under conditions characterized by SRIP constants.
- For s ≠ P, the analysis decomposes D into its s-sparse component and residual, absorbing the residual into an effective noise term W′.
- When s ≠ P, the P-sparse signal D may not be reliably reconstructed, even when the s-sparse signal is considered.
- An appropriate SRIP enables the estimated support to overlap the true support, reducing iterations needed for convergence at sparsity level s + 1.
- The stopping criteria use residual behavior and path dominance by AWGN to avoid selecting a sparsity level that reconstructs noise.
C. Computational Complexity of ASSP Algorithm
The ASSP scheme adaptively exploits structured spatial and temporal sparsity for channel recovery, achieving near-oracle estimation with reduced pilot overhead while supporting complexity reductions through joint processing.
- Static-channel performance: For ηp ≥19.04%, ASSP and oracle ASSP achieve similar MSE performance and approach the oracle LS estimator.This indicates reliable acquisition of the channel sparsity level and support set in that pilot-overhead range.
- Pilot overhead: Np_avg = 12.18 approaches 2P = 12, confirming near-optimal sparse recovery with the proposed non-orthogonal pilots.The result is reported as the average pilot overhead per transmit antenna.
- Temporal processing: Joint processing across R OFDM symbols improves MSE, and sharing the matrix inversion reduces the relevant complexity to 1/R.The comparison uses a time-varying ITU-VA channel at 60 km/h.
- Adaptive pilot placement: The space-time adaptive pilot scheme substantially reduces pilot overhead without obvious performance loss, including with fd = 5 under the reported conditions.At SNR ≤20 dB, fd = 5 performs better than fd = 1 because interpolation reduces effective noise; at 30 dB it incurs negligible loss.
VI. CONCLUSIONS
The proposed SCS-based spatio-temporal joint scheme exploits common sparsity in wireless MIMO channels to reduce pilot overhead in FDD massive MIMO. Simulations show substantially reduced overhead with negligible loss relative to the performance bound.
- The scheme exploits spatio-temporal common sparsity in wireless MIMO channels to reduce pilot overhead.
- Non-orthogonal pilots and the ASSP algorithm reliably estimate channels with significantly reduced pilot overhead.
- The space-time adaptive pilot scheme further reduces pilot overhead according to user mobility.
- The scheme achieves much better channel estimation performance than its counterparts with substantially reduced pilot overhead.
- The scheme incurs only negligible performance loss compared with the performance bound.
A. Proof of Theorem 1
The proof bounds the ASSP reconstruction error under structured sparsity and the structured restricted isometry property, using support-set decompositions and iterative residual analysis.
- The SRIP condition controls structured sparse matrices through the measurement matrix Ψ and sparsity level P.
- The theorem analyzes the upper bound of the difference between the structured sparse signal and its estimate.
- The proof distinguishes the estimated support set from the correct support set and decomposes their differing elements.
- The analysis decomposes the residual into contributions from selected and unselected support components during ASSP iterations.
- When the stopping condition is met, the estimated support is identified and the P-sparse signal estimate is obtained.
- The proof concludes the stated error bounds by combining the intermediate inequalities and the SRIP parameter δP.
B. Proof of (25)
This proof establishes inequality (25) by applying the SRIP relation to structured sparse matrices with disjoint support sets and using the Cauchy–Schwarz inequality.
- The proof considers structured sparse matrices with disjoint support sets and normalizes them by their Frobenius norms.
- The real part of the matrix inner product is used to relate the SRIP expression to the normalized structured sparse matrices.
- The Cauchy–Schwarz inequality bounds the inner product, with equality only when the matrices are complex scalar multiples.
- These bounds yield the inequality involving δ and the structured support sizes, proving (25).
C. Proof of (26)
This proof derives the right and left inequalities in (26) from the preceding SRIP-based relations and an auxiliary equality.
- The first inequality in (60) follows from relation (56), while its third equality follows from an auxiliary equality.
- The right inequality in (26) is obtained from relation (61).
- The left inequality in (26) follows from the same preceding relation (61).