<p>With the rapid development of Internet of Things (IoT) technology, the massive interconnection of devices has imposed requirements on wireless communication in terms of low latency, high throughput, and low power consumption. Adaptive Modulation and Coding (AMC) technology, as a core mechanism of link adaptation, enables dynamic adjustment of the Modulation and Coding Scheme (MCS) based on channel states, thereby maximizing throughput while ensuring reliability. To address these challenges, an AMC method based on the ConvNeXt V2 was proposed for IoT with the application on IEEE 802.11ah protocol. In the proposed method, an attention module was first introduced for convolution blocks to enhance the model’s ability to focus on fine-grained features of wireless channels. Then, a Unified Progressive Depth Pruner (UPDP) mechanism was applied to prune the ConvNeXt blocks, thereby reducing model size and computation costs. Finally, a Reptile meta-learning algorithm was incorporated to fine-tune the model, which can improve the generalization in different channel conditions and few-shot learning performance. Experiment results show that, coupled with the CBAM module, the channel classification accuracy can be improved by 9.7%. After the pruning operation was applied, the number of model parameters was reduced by 43%, while the throughput decreased by only 1.5%. When the channel changed from TGah B to TGah C/D/E/F, the throughput increased by 11%, 10%, 11% and 8% respectively, after fine-tuning by the Reptile algorithm with only 18% additional new-channel samples. These results indicate that the proposed method can effectively address AMC challenges and significantly improve network throughput across different channel conditions in internet of things.</p>