The paper proposes a massive digital AirComp (MD-AirComp) scheme to enhance the communication efficiency of federated edge learning (FEEL). MD-AirComp leverages an unsourced massive access protocol to improve compatibility with current and future wireless networks. It uses vector quantization to reduce uplink communication overhead and employs shared quantization and modulation codebooks. At the receiver, an approximate message passing-based algorithm is proposed to estimate the model aggregation results from the superposed sequences, focusing on estimating the number of devices transmitting each code sequence rather than decoding individual messages. The proposed scheme is applied to FEEL, showing significant acceleration in convergence compared to state-of-the-art methods while using the same amount of communication resources. The code for the proposed scheme is available at https://github.com/liqiao9MD-AirComp.The paper proposes a massive digital AirComp (MD-AirComp) scheme to enhance the communication efficiency of federated edge learning (FEEL). MD-AirComp leverages an unsourced massive access protocol to improve compatibility with current and future wireless networks. It uses vector quantization to reduce uplink communication overhead and employs shared quantization and modulation codebooks. At the receiver, an approximate message passing-based algorithm is proposed to estimate the model aggregation results from the superposed sequences, focusing on estimating the number of devices transmitting each code sequence rather than decoding individual messages. The proposed scheme is applied to FEEL, showing significant acceleration in convergence compared to state-of-the-art methods while using the same amount of communication resources. The code for the proposed scheme is available at https://github.com/liqiao9MD-AirComp.