The dynamic gain of a few-mode erbium-doped fiber amplifier (FM-EDFA) is vital for the long-haul mode division multiplexing (MDM) transmission. Here, we investigate the mode-dependent dynamic gain of an FM-EDFA under various manipulations of the pump mode. First, we numerically calculate the gain variation with respect to the input signal power, where a mode-dependent saturation input power occurs under different pump modes. Even under the fixed intensity profile of the pump laser, the saturation input power of each spatial mode is different. Moreover, high-order mode pumping leads to a compression of the linear amplification region, even though it is beneficial for the mitigation of the differential modal gain (DMG) arising in all guided modes. Then, we develop an all-fiber 3-mode EDFA, where the fundamental mode of the pump laser can be efficiently converted to the LP₁₁ mode using the all-fiber mode-selective coupler (MSC). In comparison with the traditional LP₀₁ pumping scheme, the DMG at 1550 nm can be mitigated from 1.61 dB to 0.97 dB under the LP₁₁ mode pumping, while both an average gain of 19.93 dB and a DMG of less than 1 dB can be achieved from 1530 nm to 1560 nm. However, the corresponding signal input saturation powers are reduced by 0.3 dB for the LP₀₁ mode and 1.6 dB for the LP₁₁ mode, respectively. Both theoretical and experimental results indicate that a trade-off occurs between the DMG mitigation and the extension of the linear amplification range when the intensity profile of pump laser is manipulated.

红外辐射在大气中传输会在大气分子、气溶胶粒子的吸收和散射以及大气自身辐射的影响下发生变化，导致红外辐射测量精度的降低。为消除大气在红外目标模拟器校准中的影响，在基于恒定标准源的宽动态红外辐射测量方法的基础上，提出了一种红外目标模拟器的大气传输校准方法。在水平均匀大气近距离的红外目标模拟器校准中，利用卷积神经网络的数据分析能力建立了不同波段、不同温度、不同距离下的大气透过率和大气程辐射的动态模型，将探测器输出电压作为基于编码器-解码器结构的卷积神经网络的输入，按照训练流程对网络进行训练，在实验环境下预测了大气传输对红外辐射的影响。所建模型能够反映大气透过率和大气程辐射的动态变化规律，并通过红外辐射反演对提出的方法进行了验证。实验结果表明：基于编码器-解码器结构的卷积神经网络算法能够较好地预测大气透过率和大气程辐射，在三个波段下的平均误差为3.0783%、3.8186%、5.3452%，低于传统方法，降低了大气透过率和大气程辐射的影响，从而减小了红外辐射的测量误差，提高了校准精度。