Related Experiment Video
Updated: Aug 15, 2026
09:43
Transmission of Multiple Signals through an Optical Fiber Using Wavefront Shaping
Published on: March 20, 2017
Transfer-learning-adapted CNN-BiLSTM nonlinear equalizer for DSCM-enabled microwave-band fiber-wireless integration
Optics Express
|August 14, 2026
Summary
A novel nonlinear equalizer using transfer learning (TL) with convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) effectively mitigates distortions in fiber-wireless integration (FWI) systems. This approach significantly improves data recovery across multiple subchannels, even with limited training data.
Area of Science:
- Optical Communications
- Wireless Communications
- Machine Learning in Communications
Background:
- Fiber-wireless integration (FWI) offers high-capacity fronthaul transmission.
- Digital subcarrier multiplexing (DSCM) allows flexible resource allocation.
- Cascaded impairments in FWI systems cause subchannel-dependent distortions, limiting performance after linear equalization.
Purpose of the Study:
- To propose a DSCM-enabled microwave-band FWI transmission scheme.
- To introduce a transfer learning (TL)-adapted CNN-BiLSTM nonlinear equalizer to address residual distortions.
- To reduce training overhead for subchannel equalization using a progressive TL strategy.
Main Methods:
- Developed a CNN-BiLSTM nonlinear equalizer operating on linearly equalized symbols.
- Implemented a progressive TL adaptation strategy with source model training and adaptation to target subchannels.
- Experimentally validated the scheme in a 17.5-GHz FWI system with 25-km SMF-28 fiber.
Main Results:
- The CNN-BiLSTM equalizer significantly reduced average error vector magnitude (EVM): QPSK from 34.1% to 6.3%, 8QAM from 21.7% to 8.7%, and 16QAM from 19.6% to 15.1%.
- TL adaptation with 10% target data reduced QPSK EVM from ~41% to ~22%, nearing full-training performance (~20%).
- Demonstrated consistent performance improvement across different modulation formats.
Conclusions:
- The proposed TL-adapted CNN-BiLSTM nonlinear equalizer effectively compensates for residual distortions in DSCM-enabled FWI systems.
- The TL strategy significantly enhances data efficiency by reducing the required training data for equalization.
- The approach validates a practical solution for high-performance FWI fronthaul transmission.
