Scholarlink

<p>Beamforming offers a powerful means to concentrate RF energy, extend read range, and suppress interference in chipless RFID systems deployed at the wireless edge of 5G/IoT networks. Deep learning, in parallel, provides a flexible decoding framework that can learn robust features directly from distorted multi-notch spectra, overcoming the limitations of hand-crafted thresholds and shallow machine-learning models under realistic noise, bending, and angular mismatch. This work integrates both capabilities into a unified chipless RFID platform that combines a miniaturized RCS-based tag with a beamforming-enabled, deep-learning reader operating over the 4–6&#xa0;GHz band. The tag is implemented as a ground-backed array of T-shaped microstrip resonators on RO4350B, spectrally distributed using a coupling-aware guard-band and spatial reordering strategy. This design yields a dense multi-notch code within a 55 × 32&#xa0;mm² footprint, with an average spacing of approximately 80&#xa0;MHz between notches and maintains stable resonance behavior under bending and bistatic angular variations. Full-wave simulations and bistatic RCS measurements in an anechoic chamber show good agreement, confirming that the high-density spectral encoding is practically realizable and providing a calibrated reference for system-level evaluation. On the reader side, a conventional bistatic single-horn setup models a 1 × 4 Tx/Rx beamforming reader and narrowband digital beamforming on an FPGA/SoC platform. The modeled (array-factor-based) beamforming gain is about 10–12 dB at broadside and substantially improves the link budget and extends the interrogation range relative to the single-horn baseline. On the same edge platform, a compact one-dimensional convolutional neural network operates on baseline-corrected RCS spectra, learning to decode the entire multi-notch pattern without hand-crafted features. Compared with a probabilistic machine-learning decoder, the proposed deep-learning approach, especially when combined with beamforming, significantly improves bit-wise and whole-code accuracies (e.g., from ≈ 0.93 to ≈ 0.98 and from ≈ 0.73 to ≈ 0.86, respectively), with clear gains under low-SNR, rotated, or slightly detuned conditions. The resulting beamforming-aware, AI-assisted chipless RFID system demonstrates that co-design of tag physics, beamforming reader, and deep-learning inference is a promising route toward scalable, spectrally efficient, and robust identification in 5G-enabled wireless edge IoT environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Smart beamforming and deep-learning reader for RFID systems in edge IoT

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月19日

查看全文
  • Kawther Mekki,
  • Marwa Ben Slimene,
  • Syrine Neffati,
  • Nadia Ghezaiel,
  • Khaled Kefi,
  • Hatem Rmili,
  • Ali Gharsallah

摘要

Beamforming offers a powerful means to concentrate RF energy, extend read range, and suppress interference in chipless RFID systems deployed at the wireless edge of 5G/IoT networks. Deep learning, in parallel, provides a flexible decoding framework that can learn robust features directly from distorted multi-notch spectra, overcoming the limitations of hand-crafted thresholds and shallow machine-learning models under realistic noise, bending, and angular mismatch. This work integrates both capabilities into a unified chipless RFID platform that combines a miniaturized RCS-based tag with a beamforming-enabled, deep-learning reader operating over the 4–6 GHz band. The tag is implemented as a ground-backed array of T-shaped microstrip resonators on RO4350B, spectrally distributed using a coupling-aware guard-band and spatial reordering strategy. This design yields a dense multi-notch code within a 55 × 32 mm² footprint, with an average spacing of approximately 80 MHz between notches and maintains stable resonance behavior under bending and bistatic angular variations. Full-wave simulations and bistatic RCS measurements in an anechoic chamber show good agreement, confirming that the high-density spectral encoding is practically realizable and providing a calibrated reference for system-level evaluation. On the reader side, a conventional bistatic single-horn setup models a 1 × 4 Tx/Rx beamforming reader and narrowband digital beamforming on an FPGA/SoC platform. The modeled (array-factor-based) beamforming gain is about 10–12 dB at broadside and substantially improves the link budget and extends the interrogation range relative to the single-horn baseline. On the same edge platform, a compact one-dimensional convolutional neural network operates on baseline-corrected RCS spectra, learning to decode the entire multi-notch pattern without hand-crafted features. Compared with a probabilistic machine-learning decoder, the proposed deep-learning approach, especially when combined with beamforming, significantly improves bit-wise and whole-code accuracies (e.g., from ≈ 0.93 to ≈ 0.98 and from ≈ 0.73 to ≈ 0.86, respectively), with clear gains under low-SNR, rotated, or slightly detuned conditions. The resulting beamforming-aware, AI-assisted chipless RFID system demonstrates that co-design of tag physics, beamforming reader, and deep-learning inference is a promising route toward scalable, spectrally efficient, and robust identification in 5G-enabled wireless edge IoT environments.