DeepSC is a deep learning-based semantic communication system that aims to improve communication efficiency and robustness by focusing on semantic information rather than bit or symbol errors. Inspired by advances in natural language processing (NLP) and deep learning, DeepSC uses a Transformer-based architecture to extract and encode semantic information from text. The system is designed to maximize system capacity and minimize semantic errors by recovering the meaning of sentences, rather than just transmitting bits or symbols. Transfer learning is employed to ensure the system can adapt to different communication environments and accelerate model training. A new metric, sentence similarity, is introduced to evaluate the performance of semantic communication. DeepSC consists of a semantic encoder, channel encoder, channel decoder, and semantic decoder. The semantic encoder extracts semantic information from text, while the channel encoder ensures successful transmission over the physical channel. The channel decoder and semantic decoder work together to recover the transmitted sentence. The system uses two loss functions: cross-entropy to minimize semantic differences and mutual information to maximize data rate. The proposed DeepSC outperforms traditional communication systems, especially in low signal-to-noise ratio (SNR) environments, as demonstrated by simulation results. The system is trained using an end-to-end approach, with the mutual information estimation model trained first to estimate data rate, followed by training the entire network. Transfer learning is used to adapt the system to different communication scenarios, allowing it to reuse knowledge from previously trained models. The DeepSC is tested under various channel conditions, including AWGN and Rayleigh fading channels, and shows improved performance in terms of both BLEU score and sentence similarity. The system is also evaluated under different background knowledge scenarios, demonstrating its adaptability and effectiveness in dynamic environments. The results show that DeepSC achieves higher accuracy and robustness compared to traditional communication systems, particularly in low SNR regimes.DeepSC is a deep learning-based semantic communication system that aims to improve communication efficiency and robustness by focusing on semantic information rather than bit or symbol errors. Inspired by advances in natural language processing (NLP) and deep learning, DeepSC uses a Transformer-based architecture to extract and encode semantic information from text. The system is designed to maximize system capacity and minimize semantic errors by recovering the meaning of sentences, rather than just transmitting bits or symbols. Transfer learning is employed to ensure the system can adapt to different communication environments and accelerate model training. A new metric, sentence similarity, is introduced to evaluate the performance of semantic communication. DeepSC consists of a semantic encoder, channel encoder, channel decoder, and semantic decoder. The semantic encoder extracts semantic information from text, while the channel encoder ensures successful transmission over the physical channel. The channel decoder and semantic decoder work together to recover the transmitted sentence. The system uses two loss functions: cross-entropy to minimize semantic differences and mutual information to maximize data rate. The proposed DeepSC outperforms traditional communication systems, especially in low signal-to-noise ratio (SNR) environments, as demonstrated by simulation results. The system is trained using an end-to-end approach, with the mutual information estimation model trained first to estimate data rate, followed by training the entire network. Transfer learning is used to adapt the system to different communication scenarios, allowing it to reuse knowledge from previously trained models. The DeepSC is tested under various channel conditions, including AWGN and Rayleigh fading channels, and shows improved performance in terms of both BLEU score and sentence similarity. The system is also evaluated under different background knowledge scenarios, demonstrating its adaptability and effectiveness in dynamic environments. The results show that DeepSC achieves higher accuracy and robustness compared to traditional communication systems, particularly in low SNR regimes.