面向物联网低功耗、低计算与存储资源应用场景，开展低信噪比下的轻量级无线自适应传输理论与技术研究。（1）低信噪比下轻量级自适应调制：通过简单权重系数加权叠加，生成调制符号，实现简单，发送功率小，能耗低，BP解调计算量小；（2）极化码编译码联合设计：将BP译码迭代中子信道上的置信度演化特性作为信道选取的重要依据，构建的极化码与BP译码的匹配性好，误码率低；（3）机器学习辅助的极化码译码：通过提取深度网络中对译码性能占主导作用的深度网络参数，以很小的误码率性能损失为代价，显著降低深度学习的训练时间和译码算法的复杂度；（4）通过对接收数据和训练样本数据进行变换，提升机器学习辅助的通信处理算法对抗攻击的鲁棒性。通过深度神经网络的修剪与量化，得到的神经网络更适用于能耗和计算资源受限的物联网应用场景。研究成果将降低通信带宽、能耗和存储资源消耗，为低能耗物联网应用提供理论与技术支撑。

Facing the challenge of low power, low computation capability and limited memory available in IOT application, the project will seek to investigate the theory and technology of wireless adaptive transmission under very low signal-to-noise ratio (SNR). (1) A light weight adaptive modulation scheme under very low SNR is explored. It is implemented by weighted summation of transmitted bits with very simple coefficients and has the advantages of low complexity of modulation and demodulation, and low power consumption. (2) An innovative scheme of polar codes construction that has a high performance under BP decoding is addressed. By introducing the message evolution behavior in BP iterations of decoding as a metric to choose the subchannels for data transmission in code construction, the codes constructed match well with BP decoding and a low bit error rate of decoding is therefore achieved. (3) AI aided polar codes decoding and its low complexity implementation is explored. By taking the advantages of DNN for performance enhancement and by extracting the parameters that have a dominant role on the performance, a high performance of AI aided decoding may be achieved while maintaining a low complexity of training and implementation. (4) Techniques to enhance the robustness of AI aided algorithms for communications are explored. We seek to ensure the algorithms more robust to adversary attack by transforming the data used for training or received from wireless channels for data recovery. By pruning and quantizing the DNN, low complexity AI aided algorithms are explored that will find a wide application in wireless communications where limited bandwidth and limited memory are available and power consumption is constrained, e.g. IOT applications.