💬 NLP

Arbitrary Polygon Oscillator: Generalizing Polygonal Synthesis to Arbitrary Shapes, Morphing, and Three-Dimensional Polyhedra

This paper presents a generalized polygonal synthesis system that utilizes constant arc-length velocity to generate audio from arbitrary 2D and 3D polygonal shapes, enabling smooth morphing between irregular geometries and efficient real-time implementation with advanced antialiasing.

Original authors: Antonio Argentieri, Francesco Scagliola

Published 2026-08-26

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Antonio Argentieri, Francesco Scagliola

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Sound is often thought of as a wave moving through the air, but in the world of digital music, it begins as a simple line drawn on a graph. For decades, musicians and engineers have used a technique called polygonal synthesis to create these waves. Imagine a shape, like a triangle or a square, sitting on a piece of paper. A marker moves around the edge of that shape at a steady speed. As the marker travels, its horizontal and vertical positions are recorded to create a sound wave. The sharper the corners of the shape, the richer and more complex the sound becomes. A circle produces a pure, simple tone, while a jagged star creates a bright, buzzy timbre. This method has long been limited to perfect, regular shapes that can be described by a single mathematical rule, such as a five-sided star or a ten-sided polygon. If a musician wanted to change the sound, they had to switch from one perfect shape to another, leaving a vast universe of irregular, hand-drawn, or strange shapes completely out of reach.

A team of researchers from the Conservatorio Niccolò Piccinni in Italy has now broken these limits. They have built a new system that allows any closed shape, no matter how irregular or complex, to generate sound. Instead of relying on a mathematical formula to define the shape, their system simply reads a list of points—the corners of the shape—drawn by a user. This means a musician can draw a lopsided, jagged, or curving shape on a computer, and the system will turn it into a musical instrument. The researchers found that by moving the marker around the shape at a constant speed along the edge, rather than at a constant speed around the center, they could create a much more stable and predictable sound. This approach, which they call arc-length traversal, ensures that the pitch of the note does not wobble even if the shape has sides of very different lengths.

The power of this new system lies in its ability to morph, or smoothly transform, one shape into another. In the past, changing from a triangle to a star was difficult because they have different numbers of corners. The researchers solved this by inventing a clever way to match the corners of one shape to the corners of another, even when the counts do not match. They do this by adding invisible "sleeping" corners to the shape with fewer points. These extra corners start right on top of the existing ones and slowly separate out as the shape changes, allowing the triangle to grow into a star without losing its sharp edges or creating a glitch in the sound. This process works for any combination of shapes, whether they are simple, concave, or even self-intersecting, opening up a new region of sound that was previously unreachable.

The team took this concept even further by moving into three dimensions. They imagined a solid object, like a cube or a pyramid, floating in space. By rotating this object and slicing it with a flat, horizontal plane, they could see the cross-section of the shape at that moment. As the object spins, the cross-section changes from a square to a hexagon and back again, creating a continuously evolving sound. The researchers found that this rotation acts like a physical gesture, where the sound grows complex and then simplifies as the slice passes over the corners of the 3D object. This allows a musician to control the timbre of the sound simply by turning a knob that rotates the object in three directions, mapping the movement of the shape directly to the stereo sound field.

To ensure the sound is clean and free of digital distortion, the researchers developed a method to smooth out the sharp corners where the marker changes direction. In digital audio, these sharp turns usually create unwanted high-frequency noise. The team calculated the exact angle of the turn at every corner and applied a precise correction to the sound wave, effectively smoothing the path without changing the shape of the sound. They combined this with a technique that temporarily increases the speed of the sound processing to catch any remaining noise, ensuring the final output is clear and professional. The entire system runs efficiently on modern computer chips, meaning it can be used in real-time music software without slowing down the computer.

This work represents a significant shift in how digital instruments are designed. It moves the focus from shapes that can be described by a single number to shapes that can be drawn freely. The researchers have demonstrated that the geometry of a shape is not just a static picture but a dynamic source of sound that can be stretched, twisted, and morphed into anything the user can imagine. By treating the shape as a drawing rather than a formula, they have expanded the palette of digital synthesis, giving musicians a new way to sculpt sound with their hands. The system is already available for use, allowing anyone to explore the sonic possibilities of the infinite variety of shapes that exist between the perfect circle and the jagged star.

Technical Summary: Arbitrary Polygon Oscillator

Problem Statement
Traditional polygonal synthesis generates audio by traversing the perimeter of a polygon with a phasor. Existing implementations, such as those by Chapman [2] and Hohnerlein et al. [1], rely on parametric definitions of regular or star polygons (often using Schläfli symbols or continuous order parameters) and typically employ constant angular velocity. These approaches limit the synthesis space to specific families of shapes and produce timbral transitions only within those parametric constraints. Furthermore, traversing polygons with unequal edge lengths at constant angular velocity introduces pitch fluctuations and timbral instability. While previous work has hinted at freely drawn shapes (e.g., the proposed Zykloid [7]), no formal implementation or generalization to arbitrary, non-parametric polygons with robust morphing and anti-aliasing has been published.

Methodology
The authors propose an "Arbitrary Polygon Oscillator" that decouples waveform generation from parametric shape definitions, instead utilizing a vertex buffer to define the geometry. The system operates on three primary pillars:

Arc-Length Parameterization: Unlike prior angular-velocity baselines, the proposed system traverses the polygon at a constant arc-length (perimeter) velocity. This ensures the phasor spends time on each edge proportional to its length, eliminating pitch fluctuations caused by varying edge lengths. The traversal point $P(\phi)$ is calculated by mapping the phase $\phi$ to a distance $s$ along the perimeter, identifying the containing edge, and evaluating a quadratic Bézier curve defined by the edge's vertices and a control point.
Geometric Flexibility and Curvature: The system accepts $N$ vertices from an external buffer, supporting regular, irregular, convex, concave, and star-shaped polygons. Edges can be curved via a quadratic Bézier control point displaced from the edge midpoint, allowing for the redistribution of harmonic energy without altering harmonic positions.
Hybrid Shape Interpolation: To morph between polygons with unequal vertex counts ( $N_A \neq N_B$ $N_{A} \neq = N_{B}$ ), the system employs a hybrid algorithm:
- Convex Polygons: Uses angular correspondence, expanding the lower-count polygon by duplicating vertices at target angular positions relative to the centroid.
- Concave Polygons: Uses perimeter midpoint expansion, distributing "sleeping" (initially coincident) vertices along edges proportional to edge length to preserve concavities and corner sharpness.
- Continuity: A cyclic shift aligns the vertex buffers to minimize distance, and a write-back mechanism ensures the first vertex ( $v_0$ ) remains consistent across pair switches to prevent DC offsets.
3D Extension: The paradigm extends to three dimensions by rotating a convex polyhedron and slicing it with a fixed horizontal plane. The resulting cross-section polygon is fed into the 2D arc-length engine. A "sleeping vertex" scheme maintains a constant vertex count during topological transitions (e.g., when the plane crosses a vertex), ensuring continuous traversal.
Anti-Aliasing: The system addresses the derivative discontinuities at vertices using a four-point polyBLAMP correction. Instead of closed-form derivatives derived for parametric families, the correction is computed at runtime using the tangent vectors of the incoming and outgoing Bézier edges. This is combined with adaptive oversampling (2x to 6x) to suppress residual aliasing.

Key Contributions

Arbitrary Polygon Oscillator: A generalization of polygonal synthesis where any closed polygon defined by an external vertex buffer acts as a waveform generator, traversed via constant arc-length velocity.
Hybrid Interpolation: A method for smooth morphing between polygons of different vertex counts that preserves corner sharpness and handles both convex and concave geometries without parametric descriptions.
Runtime PolyBLAMP: A generalization of the polyBLAMP anti-aliasing technique to arbitrary vertex configurations by deriving correction jumps from local Bézier tangents, eliminating the need for per-shape analytical derivation.
3D Planar Cross-Section: An extension where rotating a convex polyhedron yields a continuously variable 2D polygon, controlled by spatial orientation.
Implementation: A real-time implementation in RNBO (Cycling '74) utilizing geometry caching to avoid per-sample recomputation of the polygon structure.

Results and Performance

Spectral Characterization: The harmonic content is determined by the rotational symmetry order $M$ of the polygon, not the vertex count $N$ . The active harmonics follow the lattice $H(M) = \{(mM \pm 1)f_0\}$ .
Approximation of Angular Velocity: By combining edge curvature and radial weighting, the arc-length output can visually and spectrally approximate the angular-velocity baseline, though the two methods produce different harmonic amplitude distributions.
Anti-Aliasing Efficacy: Measurements on an equilateral triangle show that the combination of adaptive oversampling and the four-point polyBLAMP correction (BL4) achieves the highest Signal-to-Noise Ratio (SNR), consistently outperforming either method alone. The BL4 correction alone provides approximately 22 dB of alias reduction relative to the uncorrected signal.
Computational Cost: The geometry processing (pair building, expansion) is $O(N^2)$ but runs only on parameter changes. The audio path is $O(OS \cdot N)$ , where $OS$ is the oversampling factor, making it suitable for real-time DSP.

Significance
The paper claims to shift the paradigm of polygonal synthesis from "shape as a parameter" to "shape as a drawing." By removing the constraint of parametric families, the system opens a new region of synthesis space accessible only through geometric interpolation between arbitrary configurations. The work demonstrates that complex, non-parametric shapes and 3D-derived polygons can be synthesized with high fidelity and topological stability, providing a native stereo source where the phase relationship between channels evolves with the geometry. The authors note that while the morphing trajectory is defined geometrically, a perceptual model for the resulting timbral evolution remains a subject for future work.

More like this