This paper proposes a deep-learning-based MAC protocol, DL-MAC, designed to integrate channel access, rate adaptation, and channel switch in unlicensed wireless networks. Traditional MAC protocols, such as CSMA/CA in Wi-Fi, face performance degradation due to increased collisions and extended backoff times in dense and heterogeneous networks. DL-MAC leverages spectrum sensing data from energy detection modules to achieve MAC functionalities. It incorporates a deep neural network (DNN) for channel selection and a recurrent neural network (RNN) for joint channel access and rate adaptation. The protocol is validated using real-world data collected in the 2.4 GHz frequency band, demonstrating superior performance over traditional algorithms in both single and multi-channel environments. DL-MAC also outperforms single-function approaches in terms of overall performance. The protocol remains robust even with channel switch overhead. The paper also discusses related works, system scenario, dataset processing, and experimental results showing DL-MAC's effectiveness in enhancing spectrum efficiency and network performance. The DL-MAC framework is designed to operate in multi-channel environments, enabling adaptive channel selection and improved transmission efficiency. The protocol is evaluated using various performance metrics, including throughput and mean delay, and compared against traditional algorithms and deep-learning-based methods. The results show that DL-MAC achieves higher throughput and lower delay compared to traditional algorithms, demonstrating its effectiveness in complex and heterogeneous wireless networks.This paper proposes a deep-learning-based MAC protocol, DL-MAC, designed to integrate channel access, rate adaptation, and channel switch in unlicensed wireless networks. Traditional MAC protocols, such as CSMA/CA in Wi-Fi, face performance degradation due to increased collisions and extended backoff times in dense and heterogeneous networks. DL-MAC leverages spectrum sensing data from energy detection modules to achieve MAC functionalities. It incorporates a deep neural network (DNN) for channel selection and a recurrent neural network (RNN) for joint channel access and rate adaptation. The protocol is validated using real-world data collected in the 2.4 GHz frequency band, demonstrating superior performance over traditional algorithms in both single and multi-channel environments. DL-MAC also outperforms single-function approaches in terms of overall performance. The protocol remains robust even with channel switch overhead. The paper also discusses related works, system scenario, dataset processing, and experimental results showing DL-MAC's effectiveness in enhancing spectrum efficiency and network performance. The DL-MAC framework is designed to operate in multi-channel environments, enabling adaptive channel selection and improved transmission efficiency. The protocol is evaluated using various performance metrics, including throughput and mean delay, and compared against traditional algorithms and deep-learning-based methods. The results show that DL-MAC achieves higher throughput and lower delay compared to traditional algorithms, demonstrating its effectiveness in complex and heterogeneous wireless networks.