TY - JOUR
T1 - LSTM-based equalizer with multi-head attention and gating mechanism for 2 × 2 MIMO photonics-assisted THz wireless system
AU - Pan, Xiaolong
AU - Wang, Siqi
AU - Li, Gang
AU - Zhu, Jianping
AU - Xu, Yuxiao
AU - Zhang, Jie
AU - Zhou, Wen
N1 - Publisher Copyright:
© 2025 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement.
PY - 2025/12/15
Y1 - 2025/12/15
N2 - Photonics-assisted terahertz (THz) communication, with its ultra-wide available bandwidth and good compatibility with fiber optic systems, is considered a key candidate solution to meet the requirements of 6G ultra-high speed, low latency, and large capacity transmission. However, it is susceptible to severe nonlinear damage during actual transmission. In this paper, we propose a long short-term memory (LSTM)-based equalizer with multi-head attention and gating mechanism (MHAG-LSTM). The model combines LSTM’s modeling ability for temporal dependencies with the global modeling advantage of attention mechanism, and dynamically adjusts the weights of both in the equalized output through gating mechanism. The experiment demonstrated the 200 m outdoor wireless transmission of 50 GBaud polarization-division multiplexing (PDM) QPSK signal in a 140 GHz photonics-assisted THz system. The results indicate that under the same input optical power conditions, the MHAG-LSTM equalizer can achieve lower bit error rates (BER) than fully connected neural network (FCNN), convolutional neural network (CNN), LSTM, and Transformer equalizers. When the optical power into uni-traveling-carrier photodiode (UTC-PD) is 16 dBm, the BER of the received signals in the horizontal and vertical polarization directions are 3.64 × 10−3 and 3.70 × 10−3, respectively, which are lower than the 7% hard decision forward error correction (HD-FEC) threshold of 3.8 × 10−3. Compared with CNN, LSTM, and Transformer, MHAG-LSTM reduces the average equalized BER of both polarization signals by 57.07% while reducing the number of real multiplications required to equalize per symbol (RMpS) by an average of 13.5%. MHAG-LSTM not only demonstrates superior nonlinear compensation capability in photonics-assisted THz communication, but also provides important reference for intelligent equalization and low complexity implementation in future 6G high-frequency high-capacity communication systems.
AB - Photonics-assisted terahertz (THz) communication, with its ultra-wide available bandwidth and good compatibility with fiber optic systems, is considered a key candidate solution to meet the requirements of 6G ultra-high speed, low latency, and large capacity transmission. However, it is susceptible to severe nonlinear damage during actual transmission. In this paper, we propose a long short-term memory (LSTM)-based equalizer with multi-head attention and gating mechanism (MHAG-LSTM). The model combines LSTM’s modeling ability for temporal dependencies with the global modeling advantage of attention mechanism, and dynamically adjusts the weights of both in the equalized output through gating mechanism. The experiment demonstrated the 200 m outdoor wireless transmission of 50 GBaud polarization-division multiplexing (PDM) QPSK signal in a 140 GHz photonics-assisted THz system. The results indicate that under the same input optical power conditions, the MHAG-LSTM equalizer can achieve lower bit error rates (BER) than fully connected neural network (FCNN), convolutional neural network (CNN), LSTM, and Transformer equalizers. When the optical power into uni-traveling-carrier photodiode (UTC-PD) is 16 dBm, the BER of the received signals in the horizontal and vertical polarization directions are 3.64 × 10−3 and 3.70 × 10−3, respectively, which are lower than the 7% hard decision forward error correction (HD-FEC) threshold of 3.8 × 10−3. Compared with CNN, LSTM, and Transformer, MHAG-LSTM reduces the average equalized BER of both polarization signals by 57.07% while reducing the number of real multiplications required to equalize per symbol (RMpS) by an average of 13.5%. MHAG-LSTM not only demonstrates superior nonlinear compensation capability in photonics-assisted THz communication, but also provides important reference for intelligent equalization and low complexity implementation in future 6G high-frequency high-capacity communication systems.
UR - https://www.scopus.com/pages/publications/105025151394
U2 - 10.1364/OE.580821
DO - 10.1364/OE.580821
M3 - Article
C2 - 41414471
AN - SCOPUS:105025151394
SN - 1094-4087
VL - 33
SP - 53084
EP - 53097
JO - Optics Express
JF - Optics Express
IS - 25
ER -