Modular Magnetic Encoder System

• The Modular Magnetic Encoder System is a reconfigurable hardware platform that employs programmable magnetic pixels and MR sensors for high-resolution position and orientation tracking.
• It leverages advanced methods such as Hadamard matrix-based patterning, precise magnetization techniques, and robust calibration algorithms to achieve sub-millimeter accuracy.
• Applications include self-assembling robotics, precision encoders, and adaptive interfaces, though challenges remain with near-field limitations and reprogramming speeds.

A Modular Magnetic Encoder System is a reconfigurable hardware platform employing spatial patterns of magnetic bits or pixels in combination with electronic readout and/or mechanical self-assembly. This system enables encoding, decoding, and real-time tracking of positions, orientations, and selective mechanical interactions. Modern realizations exploit advances in thin-film permanent magnets, programmable magnetization, precise magnetoresistive (MR) sensors, and robust calibration/inversion algorithms for sub-millimeter spatial resolution, selective force-based self-assembly, and scalable modular readout architectures. The system finds use in passive self-assembling robotics, high-precision positional encoders, and adaptive tangible interfaces (Nisser et al., 2022, Abrunhosa et al., 2019, Biancalana et al., 2020).

1. Magnetic Pixel Encoding and Programmability

The foundation of modular magnetic encoding centers on arranging a grid of magnetizable pixels (typically soft or hard magnetic material) on each module face or substrate. Each pixel’s polarity is set to one of two binary states (+1/North, −1/South), effectively creating a programmable 2D magnetic "barcode." In mechanically self-assembling modules, each face is typically covered with an $N \times N$ grid (e.g., $N=8$) of low-coercivity soft magnetic pixels. These can be re-magnetized individually with a CNC-based magnetic plotter equipped with both N- and S-poled NdFeB tips; the system rasterizes toolpaths to overwrite arbitrary patterns, facilitating rapid erasure and rewriting for full reconfigurability. Each programmed pattern is retained until intentionally rewritten (Nisser et al., 2022).

Modular encoder tracks using thin-film Co₆₆Cr₁₆Pt₁₈ elements may be microfabricated with widths down to $w = 1000\,\mu$m and controlled separations $s$ spanning $63$–$3000\,\mu$m. These non-periodic patterns are particularly suited for absolute encoding, Vernier tracks, and codewords with high information density (Abrunhosa et al., 2019).

2. Mathematical Framework for Selectivity and Sensing

Magnetic force-driven encoding relies on pairing each binary matrix $A \in \{\pm1\}^{N \times N}$ with a unique mate $A' = -A$. When $A$ and $A'$ are aligned (pixel-for-pixel registration), every north face meets a south face, yielding maximal attractive force, mathematically quantified by a normalized overlap score $S(A, A'; 0, 0, 0) = -1$. For maximal selectivity, all other alignments—i.e., non-mate faces or misaligned pairs—must exhibit negligible net force ("agnostic" interaction). This is enforced by generating $A$ from mutually orthogonal Hadamard matrices, ensuring $S(A, B; u,v,\theta) = 0$ for B not a mate of A, upon full registration (Nisser et al., 2022).

For thin-film encoders, the stray magnetic field at height $z$ for a stack of $N$ bits is modeled as a sum over closed-form field contributions: $$H_z(x,z) = \frac{M_0}{2\pi} \sum_{k=0}^{N-1} \left[ \ln \frac{r_4^k}{r_2^k} - \ln \frac{r_3^k}{r_1^k} \right]$$ with $M_0$ the in-plane remanent magnetization and $r_{1\ldots4}^k$ geometric distances from observer to the corners of each bit (Abrunhosa et al., 2019). The vertical magnetic induction $B_z$ is then sensed by TMR (Tunnel Magnetoresistive) devices or arrays of MR sensors.

3. Hardware Architecture and Modularization

Modular encoder systems are physically realized in several forms:

• Self-Assembling Cubic Modules: Each face (25 mm$^2$) is overlaid with an $8 \times 8$ magnetic pixel array (pixel $\sim$3 mm), fabricated from fridge-magnet-style sheets ($\sim$26 mil thick), within PLA-printed cubes. Magnetization patterns are rewritten by automated CNC plotter; six faces require roughly 12 minutes to completely reprogram (Nisser et al., 2022).
• Irregular Thin-Film Bit Scales: Sputtered CoCrPt scales on glass, with precisely defined bit geometries and separations, processed by photolithography. Array extension is achieved by stacking multiple tracks side-by-side, enabling high-resolution "Vernier" encoding or modular plug-in architectures with unified bit geometry (Abrunhosa et al., 2019).
• MR Sensor Arrays: Arrayed 3-axis MR sensors (e.g., IST8308, ±200–500 µT range), deployed on stackable PCBs with sub-mm inter-sensor precision, connected to master microcontroller units via I²C or direct lines. Arrays with $K=8$ (24 channels) are typical, allowing flexible arrangements—planar, linear, ring, or cubic—for scalable localization and readout. The communication protocol employs robust binary packet streams; sample rates up to 200 Sa/s are achievable (Biancalana et al., 2020).

4. Calibration, Signal Processing, and Decoding

High-precision operation requires multilevel calibration:

• MR Sensor Calibration: Each axis is modeled as $V^{(k)} = M^{(k)} B^{(k)} + O^{(k)}$; calibration involves ellipsoidal fitting to data collected during arbitrary orientation sweeps in a uniform calibration field. Remaining systematic errors are further minimized via per-sensor alignment procedures referencing a common Cartesian frame (Biancalana et al., 2020).
• Bit Pattern Decoding: Signal readout is based on mapping measured $B_z(x)$ to theoretical fingerprints using cross-correlation or direct peak detection. For non-periodic bit patterns, system resolution is ultimately set by the minimum feature separation $s_\mathrm{min}$, constrained by noise floor ($n_H \approx 0.1\,\mu$T/$\sqrt{\mathrm{Hz}}$ at 10 Hz for TMR), bit geometry, and reading distance (lift-off RD). For secure discrimination, $s_\mathrm{min} \approx 3z$; e.g., at $z = 200\,\mu$m, $s_\mathrm{min} \approx 250\,\mu$m (Abrunhosa et al., 2019).
• Real-Time Position and Orientation Reconstruction: Field-mapping arrays solve a $9$-dimensional nonlinear least-squares (NLLS) inversion, fitting dipole location $\vec{r}$, moment $\vec{m}$, and background field $\vec{B}_0$ via Levenberg–Marquardt optimization. With $K \geq 6$ MR sensors, sub-0.3 mm RMS position errors and latency under 10 ms per cycle are routinely achieved on commodity hardware (Biancalana et al., 2020).

5. Experimental Validation and Performance Metrics

Force measurement setups for self-assembling modules involve fixing one cube to a micro-balance and stepping or rotating the other using the CNC end-effector. Experimental data confirm quantitative agreement between the predicted and measured cross-correlation forces; for checkerboard-programmed faces, RMS error between model and data reaches 1.4% (after accounting for soft-magnetic realignment) (Nisser et al., 2022).

Self-assembly trials of eight cubes pre-programmed with selective Hadamard codes demonstrate that correct bond formation (S = −1) generates ~160 mN (256 Pa), while any mis-bond is restricted to no more than 36% of full mate pull (S ≈ −0.36). In stochastic fluidic agitation (random turbulence in a 200 mm tank), the system successfully formed the programmed octree meta-cube over ~32 hours with zero persistent mis-assemblies; performance remained stable upon face reprogramming (Nisser et al., 2022).

For encoder readout, spatial resolutions below $100\,\mu$m are achievable at optimized lift-off ($\mathrm{RD} < 0.2s_\mathrm{min}$). Bit shift detection down to $\sim$25 µm is verified with $B_z$ slope and MR sensor SNR analysis (Abrunhosa et al., 2019). In tracker implementations, dynamic accuracy of 0.27 mm (RMS) is achieved at typical working distances (Biancalana et al., 2020).

6. Applications, Advantages, and System Limitations

Applications of modular magnetic encoder systems include passive and reprogrammable self-assembling robots, rapid-configurable mechanical fixtures or jigs for manufacturing, deployable fluidic or underwater structures, and tangible user interaction interfaces. Advantages include full reprogrammability (via overwritable soft-magnetic pixels or modular thin-film tracks), purely passive operation (no onboard power), and sub-millimeter spatial resolution with robust error compensation (Nisser et al., 2022, Abrunhosa et al., 2019, Biancalana et al., 2020).

Limitations identified include restricted near-field interaction range (modules must enter a narrow attraction basin), uncontrolled self-assembly ordering (potentially mitigated by hierarchical schemes), moderate pull forces (typically <250 Pa for demonstrated soft-magnetic modules; higher-coercivity or permanent magnet alternatives are required for increased strength), and bottlenecks in sequential reprogramming (currently minutes per face; parallel actuation is a plausible improvement). In large coding systems, the exponential search for orthogonal Hadamard pairs or bit configurations presents a combinatorial challenge, particularly beyond $N=8$ or more than 12 distinct modules (Nisser et al., 2022).

7. Modularity, Scalability, and Design Trade-offs

The modular nature of both the mechanical assemblies and sensor arrays enables plug-and-play system extension. Additional sensor modules or thin-film tracks can be incorporated to increase resolution, coding capacity, or geometric coverage, with array sizes up to $K=32$ sensors feasible within standard MCU I/O constraints. However, increased complexity raises calibration overhead, demands careful bus architecture design, and requires strict mechanical tolerances (inter-sensor spacing $\gtrsim$5× sensor chip size). For optimal SNR and conditioning, sensor–magnet spacing should be ~50× magnet size. Absolute encoding is facilitated by stacking multiple tracks with permutational or Vernier patterning.

The table below summarizes dominant trade-offs for modular magnetic encoder architectures:

Design Choice	Benefit	Drawback / Limitation
Smaller bits, tighter spacing	Higher code density	Lower SNR, needs smaller RD
Larger bits, wider spacing	Robust reading, higher SNR	Lower spatial/code resolution
More sensors or tracks	Better accuracy, redundancy	More complex calibration
Serial vs. parallel programming	Simple hardware	Serial: slow reprogramming; Parallel: complex

Overall system modularity and trade-offs are governed by application-specific requirements on resolution, robustness, code capacity, real-time constraints, and environmental conditions (Abrunhosa et al., 2019, Biancalana et al., 2020, Nisser et al., 2022).