ECG signal denoising based on deep factor analysis.

•When using the proposed algorithm to reduce the noise of the ECG signal, a good noise reduction effect could be achieved if the training sample contained samples with the morphology similar to that of the test sample. Also, the algorithm had good universality.•In addition, the deep factor analysis was used to remove the complex noise,not just a single one. Simultaneously, the NN constructed using the proposed algorithm did not depend on the frequency domain information and threshold setting. The training process of the network was simple and easy to implement.•After constructing the deep factor analysis model, the algorithm used the gradient descent algorithm to supervise the training of the network. Supervised training could fully learn some small waveform features in the ECG signal and took the advantage of the noise characteristics. The performance of noise reduction was outstanding, and it could cope with the occurrence of arrhythmia.•When our algorithm uses a lot of data during training, then the scope of use will be more extensive when applied.When people encounter new samples, they can also be added to the training samples at any time to update the model. So the algorithm is more universal.