Source-linked AI summary
Tracking Angles of Departure and Arrival in a Mobile Millimeter Wave Channel
Chuang Zhang, Dongning Guo, Pingyi Fan
TL;DR
Millimeter-wave channel estimation must support beamforming despite severe attenuation, while sparse channels permit focusing on dominant paths. This paper models abrupt path changes and slow angle variations, then proposes Kalman-filter tracking and abrupt-change detection. Simulations report better tracking efficiency than the compared adaptive algorithm and successful detection with moderate resources.
Problem
Millimeter-wave systems need channel state information for beamforming and combining, while their channels involve dominant paths undergoing abrupt changes and slow AoD and AoA variations.
Method
The paper uses a dual-timescale channel model and proposes a Kalman-filter tracking algorithm plus an abrupt-change detection method.
Results
The proposed approach requires lower SNR and fewer pilots for comparable tracking performance, while the change detection method detects abrupt changes with moderate pilots and SNR.
Takeaways & Limitations
Kalman-filter tracking outperforms the compared adaptive algorithm, and both proposed schemes work well when AoDs and AoAs vary at a reasonable speed.
Abstract
from arXiv · showhide
Millimeter wave provides a very promising approach for meeting the ever-growing traffic demand in next generation wireless networks. To utilize this band, it is crucial to obtain the channel state information in order to perform beamforming and combining to compensate for severe path loss. In contrast to lower frequencies, a typical millimeter wave channel consists of a few dominant paths. Thus it is generally sufficient to estimate the path gains, angles of departure (AoDs), and angles of arrival (AoAs) of those paths. Proposed in this paper is a dual timescale model to characterize abrupt channel changes (e.g., blockage) and slow variations of AoDs and AoAs. This work focuses on tracking the slow variations and detecting abrupt changes. A Kalman filter based tracking algorithm and an abrupt change detection method are proposed. The tracking algorithm is compared with the adaptive algorithm due to Alkhateeb, Ayach, Leus and Heath (2014) in the case with single radio frequency chain. Simulation results show that to achieve the same tracking performance, the proposed algorithm requires much lower signal-to-noise-ratio (SNR) and much fewer pilots than the other algorithm. Moreover, the change detection method can always detect abrupt changes with moderate number of pilots and SNR.
I. INTRODUCTION
Millimeter-wave systems need accurate channel state information for directional beamforming and combining, but limited RF chains and sparse scattering shape channel estimation. The paper proposes a dual-timescale framework with Kalman-filter tracking and abrupt-change detection.
- Millimeter-wave communication requires accurate channel state information to support beamforming and combining against high attenuation loss.
- Limited RF chains constrain feasible beamforming and combining, while sparse scattering makes estimating dominant path parameters sufficient.
- Exhaustive angle search has limited estimation accuracy even with many pilots, motivating adaptive and sparse-recovery alternatives.
- The proposed framework separates abrupt channel changes from slow AoD and AoA variations, using acquisition after changes and tracking otherwise.
- The paper develops a Kalman-filter tracking algorithm and an abrupt-change detection method, while treating channel acquisition separately.
A. Structure of transmitter and receiver
The system uses uniform linear arrays with one RF chain at each transmitter and receiver, so only analog beamforming and combining are applied. Phase shifters provide constant-modulus vectors with quantized phases.
- The transmitter and receiver use uniform linear arrays with n_t and n_r antennas, respectively.
- Both sides use one RF chain, restricting the system to analog beamforming and combining while allowing extension to hybrid processing.
- Beamforming and combining vectors have constant-modulus elements whose phases can be varied.
- Each phase shifter is modeled with a limited number of quantization levels at the transmitter and receiver.
B. Channel model
The paper adopts an L-scatterer millimeter-wave channel model in which each path has an AoD, AoA, gain, attenuation, and distance. The path gains and angles together determine the channel.
- The L-scatterer model assigns each path l an angle of departure φ_l and angle of arrival ψ_l.
- Transmit and receive array responses are represented by angle-dependent steering vectors for the corresponding antenna arrays.
- The channel is expressed through the contributions of the L scattering paths.
- For each path, α_l denotes path gain, ρ_l attenuation, and d_l the distance between the reference transmit and receive antennas along that path.
- The path-gain vector α and angle vector θ fully determine the channel.
C. Channel variation model
The channel model combines infrequent abrupt path changes across blocks with slower AoD and AoA variations within slots. The paper therefore tracks existing paths between changes and detects when a new block begins.
- Abrupt changes can make dominant paths disappear or appear, while noise causes slower AoD and AoA variations.
- The model divides time into variable-length blocks between abrupt changes and shorter slots within each block.
- Within a block, existing paths vary slowly in angle while their gains remain invariant, requiring slot-by-slot channel tracking.
- At block boundaries, the system detects abrupt changes and acquires dominant paths; within blocks, it tracks those paths from an initial CSI estimate.
- The angle vector follows a linear state model with known matrix A and Gaussian process noise u(n) distributed as N(0, Q_u).
III. TRANSMISSION SCHEME
The transmission scheme sends a symbol through a transmit beamforming vector and receives it using a combining vector, producing an observation at the receiver.
- The transmitter sends symbol x using beamforming vector f, while the receiver combines it with vector w to obtain the received observation.
A. Beamforming and combining vectors
The scheme restricts beamforming and combining vectors to direction-parameterized forms suited to the channel structure, then searches quantized AoD and AoA directions.
- Beamforming and combining vectors: Beamforming and combining vectors are restricted to channel-compatible forms whose directions are each determined by one parameter.
- Beamforming and combining vectors: The beamforming and combining gain of path l is calculated for selected transmit and receive directions.
- Beamforming and combining vectors: The AoD and AoA search ranges can both be reduced to [0, π] because the relevant cosine response is symmetric around π.
- Beamforming and combining vectors: The AoD range [0, π] is quantized into Nt bins by uniformly quantizing cos φl over [−1, 1].
- Beamforming and combining vectors: The transmitter cycles through Nt beams and the receiver through Nr combiners, searching Nr × Nt direction pairs and collecting the corresponding received symbols.
B. Observation
The observation is organized as a matrix of received symbols over beamforming and combining directions, then vectorized into a nonlinear function of the channel parameters plus Gaussian noise.
- Observation: Each beamforming–combining pair produces one received symbol, forming the entries of the observation matrix.
- Observation: The transmitted symbol is set to x = 1 in the observation formulation relative to Eqn. (7).
- Observation: The observed data can be written using combining matrix W and beamforming matrix F, with V representing additive noise.
- Observation: The observation matrix is vectorized by concatenating Y's columns, producing y = vec(Y) and v = vec(V).
- Observation: The resulting vector g(θ) is nonlinear with respect to the channel-parameter vector θ.
IV. KALMAN FILTER BASED CHANNEL TRACKING
The channel tracker applies a Kalman-filter framework after linearizing the nonlinear observation model, using recursive estimates of the evolving angle vector. The implementation stores only current observations and the previous estimate, while converting complex quantities to real vectors when needed.
- IV. KALMAN FILTER BASED CHANNEL TRACKING: Because g(θ) is nonlinear, direct classical estimation can require high-complexity grid search, whereas the channel evolution and observation models fit a Kalman-filter framework.
- IV. KALMAN FILTER BASED CHANNEL TRACKING: The observation is linearized around the predicted angle estimate so that Kalman filtering can track the channel parameters.
- IV. KALMAN FILTER BASED CHANNEL TRACKING: The tracker maintains prediction and updated MMSE matrices, together with the Kalman gain and initial estimator values.
- IV. KALMAN FILTER BASED CHANNEL TRACKING: Algorithm 1 recursively estimates the channel using the current observation and the previous estimate, rather than retaining the full observation history.
- IV. KALMAN FILTER BASED CHANNEL TRACKING: Complex observations and intermediate quantities are converted into real and imaginary components so the angle estimate remains real-valued.
V. ABRUPT CHANGE DETECTION
The method detects abrupt channel changes by comparing a Kalman-filter-based residual statistic against a threshold, with the threshold calibrated through false-alarm probability.
- The detector uses Kalman-filter tracking to test whether the channel has undergone an abrupt change.It evaluates hypotheses H0, no abrupt change, and H1, abrupt changes.
- Without abrupt changes, the test statistic remains small because the residual is primarily noise; abrupt changes introduce dominant-path gains and enlarge it.
- The method decides H1 when the test statistic exceeds a predefined threshold γ.
- The threshold γ can be selected from a target false-alarm probability using an approximate Chi-Squared distribution for 2L(y(n)).The approximation uses degree 2N_tN_r under relatively accurate channel acquisition and tracking.
- Figures 4 and 5 evaluate NMSE against SNR and quantization levels, respectively, under the stated simulation conditions.
- The proposed threshold approximation works well in simulations.
VI. SIMULATION RESULTS
Simulations evaluate Kalman-filter tracking and abrupt-change detection under varying SNR, quantization, channel variation, and path dynamics. The Kalman filter outperforms the adaptive baseline, while its tracking degrades at sufficiently high variation speeds.
- Simulation setup: The simulations measure channel-estimation performance using NMSE of H, with H the true channel matrix and bH its estimate.The evaluation includes tracking over SNR, quantization levels, and channel variations.
- Tracking comparison: Both algorithms achieve higher tracking accuracy as SNR or the number of quantization levels increases.
- Tracking comparison: The Kalman filter outperforms the adaptive algorithm, with an estimation-accuracy improvement always larger than 10 dB without acquisition error.It still performs much better when acquisition error is present.
- Channel variation: Kalman-filter tracking deteriorates as channel variation speed increases because the nonlinear observation can cause accumulated estimation error and eventual loss of track.The algorithm therefore has a channel-variation-speed threshold that is difficult to determine analytically.
- Channel variation: At σ2u corresponding to πσu ≈ 0.798, the Kalman filter tracks accurately, and this represents a very fast angle variation speed in the simulations.With 1 ms slots, the corresponding speed is 798°/s, equivalent to rotating a mobile device more than two rounds per second.
- Abrupt-change detection: The change-detection experiment models path appearance and disappearance independently across slots using probabilities papp and pdis.The setup uses generated paths whose gains and angles are initialized independently before subsequent path-state changes.
VII. CONCLUSION
The paper models millimeter wave channels with abrupt changes and slow AoD/AoA variations, then proposes Kalman-filter tracking and abrupt-change detection. Simulations show both methods work well when angles vary at a reasonable speed, with graceful degradation under acquisition error.
- The proposed framework models both abrupt channel changes and slow variations in AoDs and AoAs.
- A Kalman filter based tracking algorithm and an abrupt change detection method are proposed.
- Both proposed methods work well when AoDs and AoAs vary at a reasonable speed.
- The proposed schemes degrade gracefully with acquisition error, supporting extensions that jointly consider acquisition, tracking, and abrupt change detection.