Design and Optimization of Multi-Symbol Digital Over-the-air Computation
Time: Tue 2026-03-03 13.15
Location: NSE seminar room, Teknikringen 33
Video link: https://kth-se.zoom.us/j/69251234910
Language: English
Subject area: Computer Science
Doctoral student: Xiaojing Yan , Nätverk och systemteknik
Opponent: Assistant Professor Panagiotis Diamantoulakis, Aristotle University of Thessaloniki, Thessaloniki, Greece
Supervisor: Professor Carlo Fischione, Nätverk och systemteknik; Adjunct professor Gabor Fodor, Reglerteknik
QC 20260208
Abstract
The rapid growth of large-scale Internet of Things (IoT) systems and emerging 6G networks has increased the demand for communication architectures capable of supporting distributed data generation, real-time sensing, and collaborative intelligence. Overthe-Air Computation (AirComp) offers a promising solution by exploiting the natural superposition property of the wireless multiple access channel (MAC) to compute functions directly over the air. While analog AirComp achieves this in principle, its sensitivity to channel distortion, lack of error protection, and incompatibility with modern digital transceivers limit its practical deployment.
Recent digital frameworks, such as ChannelComp, overcome the challenge of analog AirComp by designing finite-alphabet modulations that ensure distinct function values map to separable points in the aggregated constellation. However, ChannelComp relieson single-symbol modulation, which limits the geometric degrees of freedom available for shaping the received constellation and restricts the achievable computation accuracy. Moreover, it treats each quantized value as uniformly important, leaving the bit-level significance structure of digital data unexploited. These limitations motivate the development of multi-symbol and bit-aware digital AirComp, where input values are encoded into sequences of symbols, potentially with power adaptation, coding, or bit-aware protection. By expanding the design space into multiple symbol transmission and leveraging structured redundancy, it becomes possible to improve robustness against noise and fading, and to adapt modulation to the significance of individual bits.
The first part of the thesis develops a unified theoretical and algorithmic foundation for these multi-symbol digital AirComp frameworks. It formalizes the challenges of constellation overlap in single-symbol designs, establishes design principles for reliable multi-symbol aggregation, and introduces scalable optimization tools, such as semidefinite relaxation (SDR), mixed-integer programming (MIP), successive convex approximation (SCA), and concave–convex procedures (CCP), to construct modulation sequences that enhance separability and robustness under realistic channel conditions.
The second part of the thesis consists of four appended papers that implement and evaluate the proposed concepts. Paper A introduces a computation-oriented coding framework that jointly designs modulation and repetition coding. Paper B proposes the repetition for multiple access computing (ReMAC), which selectively repeats the modulated symbol over multiple time slots to avoid overlap of the aggregated symbols. PaperC develops sequential modulation for AirComp (SeMAC), which encodes each input into a sequence of symbols with distinct constellation diagrams across multiple timeslots. Paper D incorporates bit-partitioning and bit-significance weighting into multi-symbol digital AirComp, enabling better protection of more critical bits and achieving substantial gains in computation reliability. Collectively, these works demonstrate that multi-symbol modulation designs form a practical evolution of digital AirComp, compatible with modern wireless systems while enabling accurate computation of general nonlinear functions.