What question did this study set out to answer?

To reformulate Fourier Holographic Reduced Representation (FHRR) using pure integers for efficient processing in low-power systems.

June 28, 2026Open Access

Pure Integer Hyperdimensional Computing (HDC) for Low-Power Embedded Systems

Key Points

To reformulate Fourier Holographic Reduced Representation (FHRR) using pure integers for efficient processing in low-power systems.
Proposed a pure integer version of FHRR for microcontrollers and digital signal processors.
Implemented quantized phase angles as 8-bit integers using modular ring Z_256.
Used CORDIC algorithm for trigonometric resolution without floating-point units.
Successfully demonstrated integer-based hyperdimensional computing suitable for low power, with significant energy savings.
Achieved effective vector binding through modular addition and coordinate voting, improving processing speed.
Eliminated the need for floating-point hardware, reducing system complexity.

Abstract

Hyperdimensional Computing (HDC) models cognitive association using high-dimensional random vectors. A primary implementation of HDC is Fourier Holographic Reduced Representation (FHRR), which employs continuous phase angles represented as floating-point complex numbers. In this paper, we propose a pure integer reformulation of FHRR designed for low-power microcontrollers and digital signal processors. Phase angles are quantized as 8-bit integers mapped over a modular ring Z₂56. Vector binding is implemented as modular addition, and bundling is solved through majority coordinate voting. Furthermore, we use the CORDIC (Coordinate Rotation Digital Computer) algorithm and integer lookup tables to resolve trigonometric functions without floating-point units.

Read Full Paperexternally

Ask AI

Helpful

Bookmark

View Full Paper