Federated learning has emerged as a promising approach for training machine learning models across decentralized edge devices, greatly enhancing the utilization of local data. Despite its advantages, federated learning faces challenges where many effective federated learning schemes suffer from a lack of transparency in aggregation and execution, resulting in issues such as difficulty in tracing malicious data and lack of transparency in central aggregators. Addressing these concerns, this paper proposes a novel integration of federated learning with blockchain technology, termed Blockchain-powered Decentralized Federated Learning (BDFL). Our approach leverages the inherent security and transparency features of blockchain to fortify the privacy and integrity of the federated learning process. We detail the architecture of BDFL and validate its effectiveness by developing and implementing an Ethereum Virtual Machine (EVM)-compatible smart contract, which serves as a decentralized aggregator within our system. The successful implementation of this proof of concept underscores the feasibility and potential of BDFL, paving the way for robust, secure decentralized machine learning solutions.