TRICE: A Channel Estimation Framework for RIS-Aided Millimeter-Wave MIMO Systems

I. INTRODUCTION

• Passive RIS channel estimation is receiver-only and must recover many coefficients within practical coherence time; existing LS/MMSE methods require at least as many training subframes as RIS elements.Source: §I · p. 1
• MmWave channels’ low-rank angular structure supports sparse recovery with few measurements by representing each channel as a small number of paths.Each path is characterized by DOD, DOA, and complex gain.Source: §I · p. 1
• Earlier sparse-recovery methods can require active RIS elements, assume known or single-path channels, or rely on exact angle grids that trade accuracy against complexity.A prior general formulation also requires a computationally prohibitive 4D sensing matrix, even at low grid resolution.Source: §I · p. 1
• TRICE first estimates BS-to-RIS DODs and RIS-to-MS DOAs, then estimates effective cascaded-channel angles and complex path gains using multidimensional DOA methods.The framework is presented for single-user systems with passive RIS elements and a direct link that is blocked or pre-estimated.Source: §I · p. 1

II. SYSTEM AND CHANNEL MODELS

• The modeled single-user system has a BS, an MS, and a passive RIS aiding the link; the direct BS–MS link is unavailable, for example because of blockage.The BS and MS use ULAs, and the RIS elements are arranged on a rectangular surface.Source: §II · p. 2
• The BS–RIS and RIS–MS mmWave channels follow the Saleh–Valenzuela path model, with each path described by a complex gain and angular spatial frequencies.The BS/MS spatial frequencies represent DODs/DOAs, while the RIS frequencies represent azimuth and elevation angles.Source: §II · p. 2
• Channels are block-fading and are trained at the start of each block over K = KS·KT subframes, using one BS RF chain during training.The RIS phase-shift matrix is designed with a Kronecker structure for a low-complexity estimation method.Source: §II · p. 2
• Each training measurement uses a BS analog training vector, an RIS phase-shift vector, a unit-power pilot, and MS decoding, with additive white Gaussian noise.The stacked measurements form the model used to estimate the channel tensor.Source: §II · p. 2
• A direct least-squares channel estimate requires many training subframes, making it impractical in massive MIMO when channel coherence time is limited.The paper notes that the TRICE framework can be extended straightforwardly to BS and/or MS uniform rectangular arrays.Source: §II · p. 2

• The first stage casts angular-domain estimation as sparse recovery with an L-row-sparse coefficient matrix, exploiting mmWave channel low rank to reduce training overhead.Source: §III · p. 3
• The paper leaves possible sensing-matrix designs to improve X’s recoverability guarantee outside its scope.Source: §III · p. 3
• In stage two, frequencies are estimated sequentially because joint estimation alone does not guarantee pairing with the pre-estimated frequency vectors.The method uses the diagonal structure of G to address the pairing issue.Source: §III · p. 3
• TRICE estimates each diagonal path gain, reconstructs B, and applies LSKRF to recover the two channel matrices up to trivial scaling factors.Source: §III · pp. 3–4
• Any efficient 2D parameter-estimation method can be used in TRICE’s two estimation stages.Source: §III · p. 3

IV. NUMERICAL RESULTS

• The simulations compare TRICE-BES, TRICE-CS, and Joint-CS; the CS methods use OMP, while TRICE-CS uses two 2D dictionaries rather than Joint-CS’s single 4D dictionary.The methods are evaluated using the stated DFT-beamspace ESPRIT and on-grid CS approaches.Source: §IV · p. 4
• With grid resolutions increased as in C.2, TRICE-CS outperforms Joint-CS even in the low-SNR regime, unlike C.1, where it approaches Joint-CS performance only as SNR increases.Fig. 2 reports NMSE versus SNR with NR = 8, KT = 8, and KS = 16 [4×4].Source: §IV · p. 4, Figure 2 · p. 4
• TRICE-BES performs well in medium and high SNR, with satisfactory performance for very sparse channels (L = 2).Fig. 3 shows the very-sparse-channel case.Source: Figure 3 · p. 4, §IV · p. 4
• At 5 dB SNR, estimation accuracy improves by increasing KT and/or KS.This trend is shown in Fig. 4.Source: Figure 4 · p. 4, §IV · p. 4

V. CONCLUSIONS

• TRICE is a two-stage channel-parameter estimation scheme for single-user RIS-aided mmWave MIMO systems.The conclusion presents it as the proposed framework.Source: §V · p. 4
• By exploiting mmWave channels’ low-rank nature and decoupling parameter estimation into two stages, TRICE offers high estimation performance with low training overhead and computational complexity.Source: §V · p. 4
• The authors describe TRICE as appealing in practical applications.This conclusion follows their stated estimation performance, training overhead, and computational complexity.Source: §V · p. 4