Experimental End-to-End Optimization of Directly Modulated Laser-based IM/DD Transmission

TL;DR

This work tackles the nonlinear, chirp-prone dynamics of directly modulated lasers in IM/DD short-reach links by training a differentiable, data-driven surrogate model of the end-to-end channel with an LSTM architecture. Using this surrogate, the authors perform offline end-to-end optimization of transmitter DSP, geometric constellation shaping, pulse shaping, and laser driving parameters ($I_ ext{bias}$ and $P_ ext{RF}$), alongside receiver DSP including MLSE-based detection. Experimental results across multiple baud rates and fiber lengths demonstrate that the proposed E2E scheme achieves lower SER than RX-side equalizers, while reducing modulation RF power by ~2 dB and shrinking bandwidth usage by over 24%, illustrating the practical benefits of joint TX-RX optimization in DML-based IM/DD systems. The approach also provides insights into learnable pulse shaping and filter designs, and validates the surrogate’s ability to capture the interaction between chirp and chromatic dispersion in real hardware.

Abstract

Directly modulated lasers (DMLs) are an attractive technology for short-reach intensity modulation and direct detection communication systems. However, their complex nonlinear dynamics make the modeling and optimization of DML-based systems challenging. In this paper, we study the end-to-end optimization of DML-based systems based on a data-driven surrogate model trained on experimental data. The end-to-end optimization includes the pulse shaping and equalizer filters, the bias current and the modulation radio-frequency (RF) power applied to the laser. The performance of the end-to-end optimization scheme is tested on the experimental setup and compared to 4 different benchmark schemes based on linear and nonlinear receiver-side equalization. The results show that the proposed end-to-end scheme is able to deliver better performance throughout the studied symbol rates and transmission distances while employing lower modulation RF power, fewer filter taps and utilizing a smaller signal bandwidth.

Figures (10)

• Figure 1: a) Measured light-current (L-I) curve and b) small-signal modulation response $H(f)$ of the utilized NEL NLK1551SSC DML.
• Figure 2: Structure of the surrogate model. The two numbers in each layer represent the input and output size of the layer, respectively. The vertical arrows represent the recurrence of LSTM layers. Acronyms: LSTM: Long-Short Term Memory; FFNN: feedforward neural network.
• Figure 3: Block diagram of the proposed experimental modeling approach. The digital domain (a) is designed to build a data-driven model resembling the dynamics of the experimental setup (b). Acronyms and symbols: 4PAM: 4-Pulse Amplitude Modulation; AWG: arbitrary waveform generator; $\mathbf{P_\mathrm{RF}}$: modulation RF power; $\mathbf{G}$: RF amplifier gain; $\mathbf{I_\mathrm{bias}}$: bias current; DML: directly modulated laser; $\mathbf{L}$: standard single-mode fiber length; PD: photodetector; DSO: digital storage oscilloscope; LSTM: Long-Short Term Memory; MSE: mean squared error.
• Figure 4: Block diagram of the E2E scheme parameter optimization. The elements optimized by the scheme are highlighted in red. Acronyms and symbols: 4PAM: 4-Pulse Amplitude Modulation; $\mathbf{P_\mathrm{RF}}$: modulation RF power; $\mathbf{I_\mathrm{bias}}$: bias current; LSTM: Long-Short Term Memory; MLSE: maximum likelihood sequence estimation.
• Figure 5: Structure of the maximum likelihood sequence estimation (MLSE) within the end-to-end (E2E) optimization. The two numbers in each layer represent the input and output size of the layer, respectively. Acronyms: FFNN: feedforward neural network; ReLU: rectified linear unit.