Angular Consistency Adaptive Filter

• Angular Consistency Adaptive Filter is a design principle exhibiting selective angular transmission and estimation across wave physics and state-space models.
• It employs mechanisms like resonant phase matching, adaptive weighting, and geometry-preserving updates to enforce angular constraints.
• Implementations range from discrete resonant filters in quantum graphs to circular Kalman and quaternion-based complementary filters in attitude estimation.

Searching arXiv for the cited papers to ground the article in current research. [Tool call omitted in this environment: arXiv search for (Lawrie et al., 2024, Yamagishi et al., 5 Jul 2026, Kutschireiter et al., 2021, Lawrie et al., 7 Oct 2025, Hashim et al., 2022)] “Angular Consistency Adaptive Filter” is best understood as an Editor’s term for a class of filtering and angular-selection mechanisms that preserve, enforce, or exploit angular structure under explicitly modeled constraints. In the literature surveyed here, the phrase does not denote a single standardized algorithm. Instead, it spans at least two distinct technical meanings. In wave physics, it refers to periodic resonant interfaces that transmit only structurally selected incidence angles and reject the rest, with the strongest results obtained in quantum-graph and thin-channel models (Lawrie et al., 2024). In estimation and navigation, it refers to filters that remain consistent with circular or rotational geometry while adaptively weighting information from increments, direct angular observations, inertial sensors, or vector measurements, as in circular projection filtering on $S^1$, quaternion-domain complementary filtering, and geometric filtering on $SO(3)$ (Kutschireiter et al., 2021, Yamagishi et al., 5 Jul 2026, Hashim et al., 2022). A plausible implication is that “angular consistency” names a unifying design principle rather than a single formalism.

1. Scope and terminological status

The surveyed literature supports three closely related but non-identical uses of the concept. First, there is discrete angular filtering in the scattering sense: a periodic interface selects a discrete set of tangential wave numbers $\kappa_y$, and therefore a discrete set of incidence angles, while suppressing other channels (Lawrie et al., 2024). Second, there is circular or attitude filtering in the state-estimation sense: the hidden variable is an angle or orientation, and the filter respects the underlying manifold $S^1$, $\boldsymbol{S}^3$, or $SO(3)$ rather than treating angle as an ordinary Euclidean scalar (Kutschireiter et al., 2021, Hashim et al., 2022). Third, there is adaptive angular fusion in sensor processing: measurements derived from gyroscopes, accelerometers, and magnetometers are blended with weights that vary according to gait phase or magnetic disturbance, with quaternion geometry preserved during fusion (Yamagishi et al., 5 Jul 2026).

These meanings are related by a common structural theme. In each case, the system is designed to admit only angular information that is compatible with a geometry, resonance, or reliability constraint. In the metamaterial setting, compatibility is enforced by Bloch phase and resonance. In circular filtering, compatibility is enforced by periodic statistics and circular discrepancy functions. In inertial attitude filtering, compatibility is enforced by quaternion averaging or Lie-group kinematics. This suggests that the phrase “angular consistency adaptive filter” is most precise when accompanied by its operating domain.

A compact classification of representative realizations is given below.

Paper	Domain	Sense of angular consistency/adaptivity
(Lawrie et al., 2024)	Resonant periodic interface	Discrete angular pass filter with structurally programmable pass angles
(Lawrie et al., 7 Oct 2025)	Magnetic quantum-graph interface	Flux-programmable discrete-angle selector with tunable pass amplitude
(Kutschireiter et al., 2021)	Continuous-time heading estimation	Circular Bayesian filter on $S^1$ with reliability-weighted updates
(Yamagishi et al., 5 Jul 2026)	Foot-mounted AHRS/PDR	Adaptive quaternion complementary filter with gait- and disturbance-dependent weights
(Hashim et al., 2022)	Attitude estimation on $SO(3)$	Geometry-preserving adaptive stochastic filter with bias/noise compensation

The main ambiguity surrounding the term is therefore semantic, not mathematical. The literature does not support a single universal definition; it supports a family resemblance across angularly selective and angularly consistent mechanisms.

2. Resonant discrete-angle filtering in periodic interfaces

The paper “A Non-diffracting Resonant Angular Filter” (Lawrie et al., 2024) gives the clearest wave-physics realization of an angular consistency filter. The device is an infinitely periodic $1$D interface embedded in a surrounding medium, with each interface site connected not only to nearest neighbors but to its $\mu$-th nearest neighbors through resonant links of length $SO(3)$0. These beyond-nearest-neighbor couplings are the nonlocal ingredient that makes the angular response sharply phase-selective.

The theory is first developed on a discrete infinitely periodic quantum graph. Vertices are indexed by $SO(3)$1 along the periodic direction, with period $SO(3)$2. Each vertex has left and right semi-infinite leads and up and down bonds connecting to the $SO(3)$3-th neighbors. On each graph edge, the field satisfies the $SO(3)$4D Helmholtz equation

$SO(3)$5

and the Bloch ansatz introduces the tangential quasi-momentum $SO(3)$6. In the surrounding continuous medium, $SO(3)$7 is the tangential component of the incident wavevector, so selecting $SO(3)$8 is equivalent to selecting incidence angle.

The exact lead-to-lead transmission coefficient is

$SO(3)$9

This formula makes the governing competition explicit: resonant phase accumulation $\kappa_y$0, tangential Bloch phase $\kappa_y$1, and nonlocal interaction distance $\kappa_y$2. The central result occurs at bond resonance,

$\kappa_y$3

where the interface becomes an almost binary angular filter: $\kappa_y$4 At exact resonance, the structure is therefore fully reflective for almost all $\kappa_y$5, but perfectly transmissive for a discrete set of tangential wave numbers.

The physical mechanism is described as a resonance-induced switch in effective boundary condition. Off the selected angular values, the resonant bonds support standing waves and the vertex amplitudes vanish, giving effective Dirichlet boundary conditions and suppressing through-transmission. At the selected $\kappa_y$6, the resonance and Bloch phase match, the effective boundary condition switches to a Bloch-compatible one, and transmission becomes perfect. This is the sense in which the device exhibits angular consistency: the interface passes only those directions that satisfy the structural phase condition.

The same paper is explicit about an important limitation. The filter is not adaptive in the dynamic or self-tuning sense. Its pass angles are set structurally by $\kappa_y$7, $\kappa_y$8, $\kappa_y$9, and the operating wavenumber $S^1$0. The supported phrase is therefore “predesigned angularly selective resonant filter,” not active online adaptation (Lawrie et al., 2024).

3. Flux-tunable discrete angular filters

A stronger adaptive interpretation appears in “A Flux-Tunable Discrete Angular Filter” (Lawrie et al., 7 Oct 2025). This work extends the earlier resonant angular filter by replacing the classical wave equation with the magnetic Schrödinger equation on a periodic network of thin channels and by adding tunable $S^1$1-type vertex conditions. The earlier geometry passed a topology-fixed discrete, symmetry-paired set of incidence angles. The magnetic version preserves the discrete filtering mechanism but makes the passed direction and pass amplitude externally controllable.

On each edge, the field satisfies the one-dimensional magnetic Schrödinger equation

$S^1$2

with a Bloch factor $S^1$3 along the periodic interface and an Aharonov–Bohm-type phase accumulation on the internal bonds. The solenoid-induced flux $S^1$4 enters the bond matching conditions, so the phase-sensitive interference term shifts from $S^1$5 to

$S^1$6

The resulting transmission amplitude is

$S^1$7

where $S^1$8 is the $S^1$9-coupling strength at the vertices. In the resonant limit $\boldsymbol{S}^3$0, transmission occurs only at the discrete tangential wave numbers

$\boldsymbol{S}^3$1

and the pass amplitude becomes

$\boldsymbol{S}^3$2

at the selected angle. The non-magnetic case is recovered by $\boldsymbol{S}^3$3, where the pass set reduces to $\boldsymbol{S}^3$4 and is fixed entirely by topology and periodicity.

This is the strongest direct support for the “adaptive” part of the phrase in the wave-filtering literature. The filter is not adaptive in a feedback or self-sensing sense, but it is adaptive in the sense of externally programmable reconfiguration. Magnetic flux $\boldsymbol{S}^3$5 continuously shifts the discrete pass direction, while $\boldsymbol{S}^3$6 independently tunes the transmission level. The authors describe the resulting device as a “programmable steering device” (Lawrie et al., 7 Oct 2025).

The conceptual distinction from the earlier non-diffracting resonant filter is precise. In the earlier device, angular consistency was structural and fixed; in the magnetic device, angular consistency is maintained for fixed control settings, but the selected discrete direction can be reprogrammed. The pass set remains discrete rather than broadband, and the filtering remains resonant and therefore narrowband in frequency. The paper does not provide a detailed robustness analysis for disorder, loss, or fabrication tolerance, so the adaptive claim is strongest at the level of idealized control parameters rather than deployment robustness (Lawrie et al., 7 Oct 2025).

4. Circular filtering and angular consistency on $\boldsymbol{S}^3$7

A different, but equally rigorous, interpretation appears in “Projection Filtering with Observed State Increments with Applications in Continuous-Time Circular Filtering” (Kutschireiter et al., 2021). Here the target is not a scattering direction but a hidden heading angle $\boldsymbol{S}^3$8. The central problem is angular path integration from noisy angular increments and optional direct angular observations. The paper argues that this task is inherently nonlinear because the state lives on a circle rather than on $\boldsymbol{S}^3$9, and therefore standard linear Gaussian filters do not preserve circular consistency.

The framework extends projection filtering to the case of observed state increments and then specializes to a von Mises approximation for the posterior. The circular state evolves as

$SO(3)$0

and in the increment-only model

$SO(3)$1

The posterior is projected onto the von Mises family

$SO(3)$2

with mean direction $SO(3)$3 and concentration $SO(3)$4. This family preserves periodicity and wrap-around by construction.

For the increment-only case, the projected dynamics are

$SO(3)$5

$SO(3)$6

with

$SO(3)$7

With direct angular observations $SO(3)$8, the paper proposes a quasi-continuous-time von Mises observation model and derives the full circular Kalman filter: $SO(3)$9

$S^1$0

These equations show precisely why the method can be described as angularly consistent and adaptive. It is angularly consistent because the hidden state is represented directly on $S^1$1, the posterior remains von Mises, and the corrections use $S^1$2 and $S^1$3 rather than Euclidean angular differences. It is adaptive because observation reliability enters explicitly through $S^1$4 and $S^1$5, while the current certainty $S^1$6 modulates how strongly the mean moves in response to new data. The confidence update can increase or decrease according to observation agreement, since $S^1$7 changes sign (Kutschireiter et al., 2021).

The paper reports that the circular Kalman filter is analytically accessible, interpretable, and more than ten times faster than a particle filter in one benchmark, while closely tracking particle-filter performance and outperforming a Gaussian approximation in both mean and certainty estimation (Kutschireiter et al., 2021). Within the surveyed literature, this is the most direct estimator-level realization of an angular consistency adaptive filter.

5. Quaternion and Lie-group adaptive attitude filters

The same principle reappears in $S^1$8D orientation estimation, but the manifold is now $S^1$9 or $SO(3)$0 rather than $SO(3)$1. Two papers illustrate complementary approaches.

“Quaternion-Averaging-Based Adaptive Complementary Filter for Pedestrian Dead Reckoning With a Foot-Mounted AHRS” (Yamagishi et al., 5 Jul 2026) proposes a two-stage adaptive complementary filter in quaternion space for foot-mounted PDR. The method propagates orientation from angular velocity, derives auxiliary quaternions from acceleration and magnetic field measurements, and fuses them with Markley’s quaternion averaging rather than linear interpolation. The averaged quaternion is defined as the minimizer of a sum of squared DCM differences,

$SO(3)$2

which is solved via the dominant eigenvector of

$SO(3)$3

For the two-quaternion case, the paper derives a closed-form fusion rule and applies it twice: first to $SO(3)$4 and $SO(3)$5, then to $SO(3)$6 and $SO(3)$7.

Adaptivity enters through smooth rule-based weights. The accelerometer weight is

$SO(3)$8

so accelerometer correction is trusted more during low-angular-velocity, midstance-like intervals. The magnetometer weight is

$SO(3)$9

where

$1$0

Thus the magnetometer is trusted only when angular velocity is low and field magnitude is close to nominal. The paper states that the method is not explicitly framed as “angular consistency,” but it improves temporal reliability and geometric coherence of attitude estimates by respecting quaternion geometry and suppressing unreliable corrections (Yamagishi et al., 5 Jul 2026).

The reported performance is specific. Average Euler-angle RMSEs are $1$1 for roll, $1$2 for pitch, and $1$3 for yaw, lower than the compared EKF, FKF, RMr-GDALKF, KCKF, FCF, Madgwick, and Mahony filters. Average per-sample runtime is $1$4 on a MacBook Pro (M1 Pro) and $1$5 on a Raspberry Pi 4, slower than the fastest simple complementary filter but faster than the Kalman filters in the benchmark (Yamagishi et al., 5 Jul 2026).

A geometric counterpart on $1$6 is given by “Adaptive Neural Network Stochastic-Filter-based Controller for Attitude Tracking with Disturbance Rejection” (Hashim et al., 2022). The filter estimates attitude, gyro bias, and a stochastic-noise-related NN weight matrix directly from vector observations and angular-rate measurements. Its core kinematics are

$1$7

with innovation built from direct vector measurements,

$1$8

The design is adaptive because it updates $1$9 and $\mu$0 online, and it is angularly consistent in the geometric sense because the state remains on $\mu$1, the error is multiplicative, and the discrete implementation uses Rodrigues’ formula

$\mu$2

thereby preserving the rotation-group structure (Hashim et al., 2022).

The filter and the coupled controller are shown to be semi-globally uniformly ultimately bounded. The paper is explicit that this is not a covariance-consistency result in the statistical sense; it is a geometric and Lyapunov-bounded consistency result on the attitude manifold. This distinction is central when comparing it to circular Bayesian filtering or quaternion complementary filtering (Hashim et al., 2022).

6. Conceptual synthesis, misconceptions, and adjacent analytical frameworks

Across these papers, “angular consistency” has at least three non-equivalent meanings. In periodic wave filters, it means deterministic selectivity in tangential wave number or incidence angle under resonance and Bloch matching (Lawrie et al., 2024, Lawrie et al., 7 Oct 2025). In circular Bayesian filtering, it means respecting the topology of $\mu$3 and using periodic statistics rather than Euclidean approximations (Kutschireiter et al., 2021). In attitude estimation, it means preserving the geometry of unit quaternions or $\mu$4, and adapting sensor trust without violating rotation structure (Yamagishi et al., 5 Jul 2026, Hashim et al., 2022).

A common misconception is to treat all of these as instances of the same adaptive mechanism. They are not. The resonant angular filter of Lawrie and collaborators is structurally programmable, not dynamically self-tuning (Lawrie et al., 2024). The flux-tunable extension is adaptive only in the sense of external control of $\mu$5 and $\mu$6, not in the sense of online sensor-reliability inference (Lawrie et al., 7 Oct 2025). The circular Kalman filter is adaptive through reliability-weighted recursive inference and concentration dynamics, but it is not a beam-steering or metamaterial device (Kutschireiter et al., 2021). QAACF is adaptive through gait- and magnetic-disturbance-dependent weighting, but it does not define “angular consistency” as a theorem or formal metric (Yamagishi et al., 5 Jul 2026). The $\mu$7 neural stochastic filter is geometry-preserving and adaptive, yet it is embedded in a filter-controller architecture and does not provide covariance consistency diagnostics (Hashim et al., 2022).

A second misconception is that angular consistency necessarily implies dynamic adaptation. The literature does not support that equivalence. The non-diffracting resonant angular filter is strongly angularly selective and structurally consistent while explicitly lacking real-time retuning (Lawrie et al., 2024). Conversely, QAACF is strongly adaptive while only indirectly matching the phrase “angular consistency” (Yamagishi et al., 5 Jul 2026). The flux-tunable discrete angular filter occupies an intermediate position by making the pass direction programmable without making the device self-adjusting (Lawrie et al., 7 Oct 2025).

An adjacent but non-equivalent line of work is “Convergence Analysis of $\mu$8-RLS Adaptive Filter” (Das et al., 2017). That paper does not address angular variables, circular manifolds, or directional selectivity. Its relevance is methodological rather than direct: it shows how a structurally regularized adaptive filter can be analyzed through mean and mean-square deviation recursions. A plausible implication is that similar analytical machinery could be adapted to future angular-consistency regularizers, but the paper itself remains a sparsity-aware RLS study rather than an angular filter (Das et al., 2017).

Taken together, the surveyed literature supports a stable encyclopedic characterization. An angular consistency adaptive filter is not a single canonical object. It is a family of techniques in which the admissible or estimated angle is constrained by a geometry-aware, resonance-aware, or reliability-aware mechanism. The most exact estimator-level instantiation is the circular Kalman filter on the von Mises family (Kutschireiter et al., 2021). The most exact wave-selection instantiations are the non-diffracting resonant angular filter and its flux-tunable extension (Lawrie et al., 2024, Lawrie et al., 7 Oct 2025). The most practically engineered attitude-estimation realizations are QAACF and the adaptive stochastic $\mu$9 filter (Yamagishi et al., 5 Jul 2026, Hashim et al., 2022). The phrase is therefore most useful when treated as a cross-domain descriptor of methods that preserve angular structure while selectively transmitting, fusing, or estimating direction-dependent information.