PaperlayerView exact revision on arXiv

Source-linked AI summary

Channel Estimation and Hybrid Precoding for Millimeter Wave Cellular Systems

Ahmed Alkhateeb, Omar El Ayach, Geert Leus, Robert W. Heath

arXiv:1401.7426v1cs.IT

TL;DR

MmWave large-array systems require channel estimation and precoding methods that accommodate directional beamforming and constrained analog hardware. The paper develops sparse adaptive estimation, a hierarchical multi-resolution training codebook, and hybrid analog/digital precoding, with simulations showing performance comparable to exhaustive training and near perfect-channel digital solutions.

  • Problem

    MmWave large-array systems require specialized channel estimation and precoding because directional beamforming and analog hardware constraints complicate access to channel knowledge and digital precoding.

  • Method

    The paper combines sparse adaptive channel estimation, a hierarchical multi-resolution hybrid training codebook, and hybrid analog/digital precoding for mmWave channels.

  • Results

    The proposed low-complexity algorithms achieve precoding gains comparable to exhaustive search and near-optimal performance relative to unconstrained digital solutions.

  • Takeaways & Limitations

    The approach provides low-overhead mmWave channel estimation and hybrid precoding that can approach digital-solution performance under practical hardware constraints.

Abstract

from arXiv · show

Millimeter wave (mmWave) cellular systems will enable gigabit-per-second data rates thanks to the large bandwidth available at mmWave frequencies. To realize sufficient link margin, mmWave systems will employ directional beamforming with large antenna arrays at both the transmitter and receiver. Due to the high cost and power consumption of gigasample mixed-signal devices, mmWave precoding will likely be divided among the analog and digital domains. The large number of antennas and the presence of analog beamforming requires the development of mmWave-specific channel estimation and precoding algorithms. This paper develops an adaptive algorithm to estimate the mmWave channel parameters that exploits the poor scattering nature of the channel. To enable the efficient operation of this algorithm, a novel hierarchical multi-resolution codebook is designed to construct training beamforming vectors with different beamwidths. For single-path channels, an upper bound on the estimation error probability using the proposed algorithm is derived, and some insights into the efficient allocation of the training power among the adaptive stages of the algorithm are obtained. The adaptive channel estimation algorithm is then extended to the multi-path case relying on the sparse nature of the channel. Using the estimated channel, this paper proposes a new hybrid analog/digital precoding algorithm that overcomes the hardware constraints on the analog-only beamforming, and approaches the performance of digital solutions. Simulation results show that the proposed low-complexity channel estimation algorithm achieves comparable precoding gains compared to exhaustive channel training algorithms. The results also illustrate that the proposed algorithms can approach the coverage probability achieved by perfect channel knowledge even in the presence of interference.

I. INTRODUCTION

The paper develops mmWave-specific channel estimation and hybrid precoding methods for large-array systems constrained by analog hardware and limited RF chains. Its adaptive sparse-channel approach and hybrid precoding achieve performance close to exhaustive-search, perfect-channel-knowledge, and unconstrained-digital baselines in simulations.

  • Motivation: mmWave systems need directional precoding with large antenna arrays, while high mixed-signal cost motivates dividing precoding between analog and digital domains.These hardware and propagation conditions require channel estimation and precoding algorithms tailored to mmWave systems.
  • Limitations of prior approaches: Prior analog beam-training methods generally converge toward one communication beam, preventing multiplexing gains from multiple parallel streams.Analog strategies can also be sub-optimal because of constant-amplitude phase shifters and potentially low-resolution phase control.
  • Channel estimation: The paper formulates mmWave channel estimation to capture channel sparsity and designs an adaptive compressed-sensing algorithm requiring few iterations with high success probability.The single-path analysis derives an upper bound on parameter-estimation error probability and conditions for allocating training power across adaptive stages.
  • Training design: A hierarchical multi-resolution training codebook uses hybrid analog/digital processing to generate beamforming vectors with different beamwidths for adaptive estimation.The codebook is designed to support the adaptive channel estimation procedure.
  • Hybrid precoding: The proposed hybrid analog/digital precoder directly approximates dominant channel singular vectors using quantized beamsteering directions, while accounting for hardware constraints.This design avoids assuming known arrival and departure angles and generalizes more easily to arbitrary antenna arrays.
  • Evaluation: Simulation results show comparable precoding gains to exhaustive search and comparable spectral efficiency and coverage probability to perfect channel knowledge with unconstrained digital solutions.The estimated channel also supports multi-stream multiplexing in multipath channels, unlike prior single-beam training work.

II. SYSTEM MODEL

The system uses hybrid RF/baseband beamforming at the BS and MS over a geometric, limited-scattering mmWave channel. Channel estimation precedes hybrid precoding, with simulations also considering out-of-cell interference.

  • Architecture: The BS and MS use hybrid architectures with fewer RF chains than antennas, combining RF and baseband precoding or combining.The BS transmits NS data streams with NS ≤ NRF ≤ NBS; analogous constraints apply at the MS.
  • Architecture: The RF precoder uses analog phase shifters whose entries have constant modulus and are normalized under a total transmit-power constraint.The combined precoder is FT = FRFFBB.
  • Channel model: The paper adopts a narrowband block-fading geometric channel model with L scatterers, each contributing one propagation path.Path gains are modeled as Rayleigh distributed, with azimuthal path angles and array response vectors at both ends.
  • Channel model: The model focuses on horizontal two-dimensional beamforming with uniform linear arrays in simulations, while allowing arbitrary antenna arrays in principle.Elevation is neglected, and extensions to three-dimensional beamforming are noted as possible.
  • Processing scope: The algorithms assume no prior channel knowledge, estimate the channel, and then use the estimate to construct hybrid precoding.The algorithms are developed for a BS-MS link without interfering BSs but are also evaluated with out-of-cell interference.

III. FORMULATION OF THE MMWAVE CHANNEL ESTIMATION PROBLEM

The channel-estimation problem is formulated as sparse recovery of path angles and gains, enabling compressed-sensing measurements that avoid probing the full beam dictionary. The formulation assumes quantized angular grids and motivates adaptive training design.

  • III. FORMULATION OF THE MMWAVE CHANNEL ESTIMATION PROBLEM: Estimating the mmWave channel reduces to estimating each path’s AoAs, AoDs, and complex gain with low training overhead.The BS and MS must jointly design training precoders and combiners for accurate estimation.
  • III. FORMULATION OF THE MMWAVE CHANNEL ESTIMATION PROBLEM: The poor-scattering channel is represented as a sparse problem, allowing compressed-sensing ideas to guide training precoder and combiner design.The proposed formulation uses the hybrid analog/digital architecture.
  • A. A Sparse Formulation of the MmWave Channel Estimation Problem: Training measurements are formed by transmitting beamformed symbols across successive BS slots and combining the received signals with MS measurement vectors.The resulting matrices collect the beamforming vectors, measurement vectors, transmitted symbols, and noise.
  • A. A Sparse Formulation of the MmWave Channel Estimation Problem: A uniform N-point angular grid converts the channel into a dictionary representation whose sparse vector contains path gains at quantized AoA/AoD pairs.Nonzero dictionary columns identify dominant path directions, while the corresponding sparse-vector entries provide path gains.
  • A. A Sparse Formulation of the MmWave Channel Estimation Problem: The formulation quantizes continuous AoAs and AoDs, leaving off-grid refinement for future work and evaluating quantization effects numerically.The paper notes sparse regularization, continuous basis pursuit, and Newton refinement as possible approaches to reduce quantization error.
  • A. A Sparse Formulation of the MmWave Channel Estimation Problem: Because L ≪ N^2, compressed sensing requires fewer measurements than exhaustive probing of every dictionary direction.The BS need not transmit along every dictionary vector, and the MS need not observe with its entire codebook.
  • A. A Sparse Formulation of the MmWave Channel Estimation Problem: The sensing design seeks reliable recovery with few measurements, with RIP offered as one criterion for the sensing matrix.Standard compressed-sensing theory gives an O(L log(N/L)) measurement order for an L-sparse vector, but application-specific implementation requires further work.

PFT A∗

The paper develops adaptive compressed-sensing training and a hybrid analog/digital multi-resolution codebook. Successive bisection narrows angular partitions, while digital processing addresses limitations of analog-only beamforming.

  • PFT A∗: The channel is decomposed into sparse AoD and AoA vectors supported by dictionary matrices built from array-response columns.This separates the directional supports at the BS and MS for training design.
  • PFT A∗: Adaptive compressed sensing divides training into stages whose precoders and combiners depend on earlier received signals.Adaptive CS is selected because prior work reports better low-SNR performance than standard CS, relevant before mmWave beamforming.
  • B. Adaptive Compressed Sensing Solution: The proposed estimator uses successive bisection: it senses angular partitions, retains likely nonzero regions, and recursively refines them to the required resolution.With K precoding vectors per stage, the number of stages is S = log_K N under the paper’s integer-resolution assumption.
  • IV. HYBRID PRECODING BASED MULTI-RESOLUTION HIERARCHICAL CODEBOOK: A hierarchical multi-resolution codebook supplies training beams with different beamwidths for the adaptive estimator.The codebook is essential because each stage requires beamforming vectors matched to its angular partitioning.
  • IV. HYBRID PRECODING BASED MULTI-RESOLUTION HIERARCHICAL CODEBOOK: The hybrid analog/digital construction accommodates constant-amplitude quantized-phase RF hardware while remaining applicable to ULAs and non-ULAs.The additional digital layer gives the proposed codebook very low complexity and reported advantages over analog-only codebooks.
  • IV. HYBRID PRECODING BASED MULTI-RESOLUTION HIERARCHICAL CODEBOOK: Analog-only multi-resolution codebooks face phase-quantization and beam-pattern design difficulties, especially for non-ULA arrays.For non-ULAs, the lack of intuitive beam-pattern structure makes prescribed beamwidths hard to design.

A. Codebook Structure

The hierarchical codebook uses multiple levels and subsets to provide beamforming vectors with progressively refined beamwidths and predefined steering directions. These vectors are approximated under hybrid analog/digital hardware constraints using sparse approximation and orthogonal matching pursuit.

  • The codebook contains S levels, with each level providing beamforming vectors associated with different beamwidths and predefined directions.
  • At level s, vectors are divided into K^(s−1) subsets, each covering a unique AoD range that is partitioned into K sub-ranges.
  • Each beamforming vector is designed for nearly equal projection within its assigned sub-range and near-zero projection elsewhere.
  • The hybrid training precoders are designed through a sparse approximation problem that selects RF precoder columns from a candidate analog beamforming matrix.
  • Orthogonal matching pursuit provides a lower-complexity iterative solution, selecting up to N_RF beamforming vectors and calculating the corresponding baseband precoder and normalization constant.
  • The approximation introduces spectral leakage and nonuniform beamforming gain, producing main-lobe ripples that are evaluated through an error matrix.

T + EBS

The beamforming design uses non-overlapping precoding and measurement beams so each quantized direction is associated with one beam pair. Forward and backward gains then quantify desired and undesired directional responses.

  • The matrix G(s,kBS,kMS) represents precoding/measurement pairs against quantized AoA/AoD directions, with each column containing one nonzero value equal to 1.
  • Because the beamforming vectors are non-overlapping, a direction cannot lie in the main lobes of more than one precoding/measurement pair.
  • Forward gain measures the response in direction d for the selected codebook subsets, while backward gain requires specifying the competing beam pair.
  • The forward-to-backward gain ratio captures the directional quality of a selected precoding/measurement pair.
  • Main-lobe ripples arise from the approximate solution and finite design directions, but small side lobes reduce their impact for sparse channel estimation.
  • The proposed estimation performance depends mainly on the backward-to-forward gain ratio.

V. ADAPTIVE ESTIMATION ALGORITHMS FOR MMWAVE CHANNELS

The paper formulates mmWave channel estimation as a sparse recovery problem and develops adaptive algorithms based on a hierarchical codebook. It treats single-path channels first and then extends the approach to multipath channels.

  • The proposed algorithms adaptively use the hierarchical codebook to estimate sparse mmWave channel parameters.
  • The analysis first addresses rank-one channels with a single path.
  • The algorithm is then extended to general multipath mmWave channels.

A. Adaptive Channel Estimation Algorithm for Single-Path MmWave Channels

For single-path channels, the adaptive estimator progressively refines the AoA/AoD range by selecting the strongest hierarchical beam pair. The paper derives error bounds and training-power conditions, then extends the approach to multipath channels.

  • A single-path channel corresponds to a vector z with one nonzero element, whose location and value determine the path direction and gain.
  • At each stage, the receiver compares K^2 received-signal powers, selects the strongest partition, and feeds back the corresponding transmitter subset.
  • Successive codebook levels provide higher resolution, adaptively narrowing the AoA/AoD range until the target resolution is reached.
  • The required training effort is K^2 log_K N steps, with further reduction possible when N_RF receiver chains combine measurements simultaneously.
  • For single-path channels, the paper derives an upper bound on the average AoA/AoD estimation error probability.
  • Under the stated asymptotic gain conditions, distributing training power across adaptive stages can guarantee an average error probability no greater than δ.

B. Adaptive Channel Estimation Algorithm for Multi-Path MmWave Channels

The multi-path estimator exploits channel sparsity through adaptive hierarchical beam refinement and sequential path detection. It estimates dominant path parameters while reducing training requirements relative to exhaustive search, though early-stage path-gain interference remains a limitation.

  • The multi-path channel estimation problem is formulated as sparse compressed sensing, targeting the Ld non-zero channel components with maximum power.
  • Algorithm 3 adaptively estimates Ld dominant paths by repeatedly refining promising AoA/AoD ranges and projecting out previously estimated path contributions.Each outer iteration detects one additional path, and final path gains are estimated using a linear least squares estimator.
  • The modified hierarchical codebook begins with KLd beamforming vectors and recursively partitions selected angular ranges for multi-path refinement.
  • Early-stage summation of path gains can cause destructive interference, although the impact is smaller for sparse mmWave channels.
  • The proposed algorithm requires K2L3 training steps, reduced to KL2 when multiple RF chains combine measurements at the MS.

VI. HYBRID PRECODING DESIGN

The hybrid precoding design seeks to maximize mutual information while respecting analog and digital RF-processing constraints. It frames the task as rate maximization over feasible hybrid processing matrices.

  • The design jointly selects transmitter and receiver hybrid precoders and combiners to maximize mutual information with Gaussian signaling.
  • The hybrid precoding problem directly maximizes the rate expression over feasible analog and digital processing matrices.

n WBBHWRFHHFRFFBBFBBHFRFHHHWRFWBB

The proposed precoding procedure first estimates the channel, then constructs hybrid matrices that approximate unconstrained singular-vector precoders. It uses sparse optimization and matching pursuit to handle analog constraints.

  • Precoding is split into channel-parameter estimation followed by construction of the downlink channel matrix and hybrid precoders.
  • At the MS, basis pursuit computes hybrid combining matrices whose combined effect approximates the dominant eigenvectors of the uplink channel.
  • The BS uses estimated steering matrices and path gains to reconstruct the downlink channel before designing hybrid data precoders.
  • The combined hybrid precoder approximates the dominant singular vectors of the estimated channel, represented by the unconstrained precoder Fopt.
  • The BS designs FRF and FBB with an iterative matching-pursuit procedure by substituting Fopt for the residual and target matrices.

VII. SIMULATION RESULTS

Simulations evaluate the proposed codebook, channel estimation, and hybrid precoding methods in point-to-point and cellular settings. Results show near-exhaustive-search performance with substantially less training, small quantization loss at sufficient resolution, and near-optimal rates under adequate RF hardware.

  • The simulations evaluate the training codebook, adaptive channel estimation, and hybrid precoding algorithms first on a single BS–MS link and then in a mmWave cellular model.
  • With K << N, the adaptive algorithms achieve performance very close to exhaustive search while using much smaller numbers of iterations.
  • For Ld = 3 and K = 2, 96 training steps versus 2048 for exhaustive search produce less than 1 bps/Hz spectral-efficiency degradation.
  • For Ld = 3 and K = 2, more than 90% of exhaustive-search gain is achieved with 70 iterations, while increasing to 140 yields only 1 bps/Hz improvement.
  • The performance loss from the AoA/AoD quantization assumption is very small when the resolution parameter N is sufficiently large.

B. Performance Evaluation with MmWave Cellular System Setup

The paper evaluates its low-complexity channel-estimation and hybrid-precoding algorithms in a mmWave cellular setting with interference, comparing them against perfect-CSI, analog-only, and digital baselines. Results show reasonable hybrid-precoding gains, near-optimal performance relative to unconstrained digital solutions, and limited interference impact despite low complexity.

  • System model: The cellular evaluation models out-of-cell interference using a Poisson point process, with the nearest base station treated as the desired transmitter.The desired cell has radius Rc = 100m, while interfering base stations use density λ = 1/(πR_c^2).
  • Evaluation metric: Coverage probability is defined relative to user rate, with an outage occurring when the rate falls below threshold η.This metric is chosen because the evaluation targets multiplexing many streams per user.
  • Compared cases: The coverage study compares estimated-channel operation with interference, estimated-channel operation without interference during transmission, perfect-CSI hybrid precoding, and analog-only beam steering.The interference cases distinguish effects during both estimation and transmission from effects during estimation only.
  • Evaluation procedure: Algorithm 3 estimates channel parameters under interference before the proposed hybrid precoders are designed for data transmission.Interference affects the maximum-power detection problem at every estimation stage.
  • Results: The proposed hybrid precoding achieves reasonable gains by managing inter-stream interference while overcoming RF hardware constraints.The comparison includes analog-only beamforming directed toward dominant channel paths.
  • Results: The simulations indicate that cellular interference has a non-critical effect on the channel-estimation and precoding performance despite the algorithms' low complexity.The broader conclusions report comparable gains to exhaustive search and near-optimal performance relative to unconstrained digital solutions.
  • Scope and future work: The study considers a single-user mmWave system with fixed, known array structures and identifies random blockage and time-varying arrays as future extensions.The paper calls for robust estimation under random blockage and adaptive estimation for random or time-varying array manifolds.
Loading 1401.7426v1…