In this paper, we propose an adaptive deep learning-based unmanned aerial vehicle (UAV) receiver design for coded multiple-input multiple-output (MIMO) systems, where the noise in the systems presents some correlation among time domain, which deteriorates the system transmission performance severely. To improve the system performance, we employ the linear convolutional code at the transmitter, and then propose an adaptive deep learning based iterative UAV receiver. The iterative UAV receiver contains three parts: the detector such as zero-forcing (ZF) or minimum mean square error (MMSE) detector, the deep convolutional neural network (DCNN) which can help suppress the noise by capturing the correlation characteristics among noise, and the decoder such as Viterbi decoding. In particular, the cyclic redundancy check (CRC) appended to the code can help control the iteration of the detection, DCNN and decoding, which leads to an adaptive implementation of receiver. Simulation results demonstrate that the proposed UAV receiver can achieve a much better bit error rate (BER) performance over conventional receivers with a reduced computational complexity.

本文针对编码多输入多输出（MIMO）系统提出了一种基于深度学习的自适应UAV接收机设计，该系统中的噪声在时域之间存在一定的相关性，从而使系统性能下降传输性能严重。为了提高系统性能，我们在发射机处采用线性卷积码，然后提出了一种基于自适应深度学习的迭代式UAV接收机。迭代式UAV接收器包含三部分：零强制（ZF）或最小均方误差（MMSE）检测器之类的检测器，深卷积神经网络（DCNN），可通过捕获噪声之间的相关特性来帮助抑制噪声，以及维特比（Viterbi）解码器之类的解码器。特别是，附加在代码上的循环冗余校验（CRC）可以帮助控制检测，DCNN和解码的迭代，从而实现了接收器的自适应实现。仿真结果表明，所提出的UAV接收机可以以比传统接收机更好的比特误码率（BER）性能，并且具有降低的计算复杂度。