Fine-tuning of a DFB laser and deep learning-based NH₃ identification in high-resolution infrared spectra

This study proposes and validates a hybrid gas identification framework that implements a high-resolution wavelength tuning strategy based on dual-parameter control of a distributed feedback (DFB) laser. In addition, the impact of spectral resolution on identification performance is investigated using deep neural network models. By exploiting the intrinsic dual-dependence of the DFB laser wavelength on temperature and injection current, a continuous tuning range of up to 5.18 nm was achieved, with a minimum tuning step below 10 pm under a temperature increment of 0.2 °C. The system demonstrated excellent linear tuning characteristics and spectral stability under various current settings. To systematically assess the influence of spectral resolution on recognition performance, a series of simulated gas absorption spectra were generated using the HITRAN database, including ammonia(NH 3 ) and nine interfering species. Convolutional neural network, multilayer perceptron, and bidirectional long short-term memory (BiLSTM) models were trained and evaluated using datasets with spectral resolutions of 0.01 nm, 0.05 nm, and 0.10 nm. The results show that only at a resolution of 0.01 nm were all models able to achieve recognition accuracies exceeding 92%, with BiLSTM achieving the highest accuracy of 95.09%. These findings underscore the critical role of high spectral resolution in preserving key absorption features and enhancing model discrimination capabilities. The proposed approach provides technical validation and a practical pathway for integrating high-precision wavelength tuning with intelligent gas recognition in advanced spectroscopic sensing systems.