Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RiSi: Spectro-temporal RAN-agnostic Modulation Identification for OFDMA Signals

Daulet Kurmantayev, Dohyun Kwun, Hyoil Kim, Sung Whan Yoon

TL;DR

RiSi addresses the challenge of blind modulation identification for OFDMA signals to enable RAN-agnostic underlay coexistence in unlicensed bands. It proposes a semantic segmentation network that operates on OFDMA spectrograms, using flattened convolutions and a three-channel input (I, Q, amplitude) to detect resource-block grid patterns and identify their modulations. On a realistic dataset with channel impairments, RiSi achieves an average identification accuracy of $86\%$ across BPSK, QPSK, 16-QAM, and 64-QAM, with up to $92.9\%$ for BPSK and $75.5\%$ for QAMs, and demonstrates improved generalization via domain generalization methods (e.g., SWAD, MLDG). The results highlight RiSi's potential to enhance inter-RAN coexistence beyond energy-detection approaches, and show avenues for extending to mixed-interference scenarios and reinforcement-learning-based coordination.

Abstract

RAN-agnostic communications can identify intrinsic features of the unknown signal without any prior knowledge, with which incompatible RANs in the same unlicensed band could achieve better coexistence performance than today's LBT-based coexistence. Blind modulation identification is its key building block, which blindly identifies the modulation type of an incompatible signal without any prior knowledge. Recent blind modulation identification schemes are built upon deep neural networks, which are limited to single-carrier signal recognition thus not pragmatic for identifying spectro-temporal OFDMA signals whose modulation varies with time and frequency. Therefore, this paper proposes RiSi, a semantic segmentation neural network designed to work on OFDMA's spectrograms, that employs flattened convolutions to better identify the grid-like pattern of OFDMA's resource blocks. We trained RiSi with a realistic OFDMA dataset including various channel impairments, and achieved the modulation identification accuracy of 86% on average over four modulation types of BPSK, QPSK, 16-QAM, 64-QAM. Then, we enhanced the generalization performance of RiSi by applying domain generalization methods while treating varying FFT size or varying CP length as different domains, showing that thus-generalized RiSi can perform reasonably well with unseen data.

RiSi: Spectro-temporal RAN-agnostic Modulation Identification for OFDMA Signals

TL;DR

RiSi addresses the challenge of blind modulation identification for OFDMA signals to enable RAN-agnostic underlay coexistence in unlicensed bands. It proposes a semantic segmentation network that operates on OFDMA spectrograms, using flattened convolutions and a three-channel input (I, Q, amplitude) to detect resource-block grid patterns and identify their modulations. On a realistic dataset with channel impairments, RiSi achieves an average identification accuracy of across BPSK, QPSK, 16-QAM, and 64-QAM, with up to for BPSK and for QAMs, and demonstrates improved generalization via domain generalization methods (e.g., SWAD, MLDG). The results highlight RiSi's potential to enhance inter-RAN coexistence beyond energy-detection approaches, and show avenues for extending to mixed-interference scenarios and reinforcement-learning-based coordination.

Abstract

RAN-agnostic communications can identify intrinsic features of the unknown signal without any prior knowledge, with which incompatible RANs in the same unlicensed band could achieve better coexistence performance than today's LBT-based coexistence. Blind modulation identification is its key building block, which blindly identifies the modulation type of an incompatible signal without any prior knowledge. Recent blind modulation identification schemes are built upon deep neural networks, which are limited to single-carrier signal recognition thus not pragmatic for identifying spectro-temporal OFDMA signals whose modulation varies with time and frequency. Therefore, this paper proposes RiSi, a semantic segmentation neural network designed to work on OFDMA's spectrograms, that employs flattened convolutions to better identify the grid-like pattern of OFDMA's resource blocks. We trained RiSi with a realistic OFDMA dataset including various channel impairments, and achieved the modulation identification accuracy of 86% on average over four modulation types of BPSK, QPSK, 16-QAM, 64-QAM. Then, we enhanced the generalization performance of RiSi by applying domain generalization methods while treating varying FFT size or varying CP length as different domains, showing that thus-generalized RiSi can perform reasonably well with unseen data.
Paper Structure (18 sections, 4 equations, 5 figures, 2 tables)

This paper contains 18 sections, 4 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Neural Networks for Modulation Identification
  4. Semantic Segmentation for Modulation Recognition
  5. RiSi: Proposed Neural Network Model
  6. Semantic Segmentation: Background
  7. Proposed Architecture of RiSi
  8. Domain-specific Dataset Construction for RiSi
  9. Identification Performance of RiSi
  10. Effect of Flattened Convolutions
  11. 2-Channel vs. 3-Channel
  12. Impact of Dataset Size
  13. Performance with Real-captured Packets
  14. Performance Comparison with Other Existing Methods
  15. Domain Generalization Performance of RiSi
  16. ...and 3 more sections

Figures (5)

  • Figure 1: Architectural diagram of RiSi (parts in orange are proposed revision)
  • Figure 2: Segmentation results: vanilla DeepLab V3+ with 2D convolutions (left), our proposal with 1D convolutions (center), the ground truth (right)
  • Figure 3: Confusion matrices of RiSi trained on datasets with various sizes ('Nothing' means the 'no data' class)
  • Figure 4: A few examples of RiSi's performance with real-captured WLAN packets (vertical axis: frequency, horizontal axis: time)
  • Figure 5: Evaluation for varying FFT size and CP length. In (a), the vertical dashed lines imply the FFT sizes in the train domain. In (b), the lefthand side of the vertical dashed line construct the train domain.