Ambulatory human health monitoring using wireless communication technology is a prominent area of modern research. This paper describes a wavelet based intelligent electrocardiogram (ECG) compression scheme under fluctuating communication link scenario that can be applied for personal healthcare applications. It adopts an autoencoder based feature extractor and slope delineator, followed by a multilayer perceptron neural network for optimal quantization of the wavelet coefficients in accordance with the current link speed, which may vary between 0.5 and 4.5 kbps. The proposed technique was validated using a window size of 64 samples with four types of abnormal beat arrhythmia data (mitdb) from Physionet, resulting an average CR and PRDN of 7.52 and 4.41 respectively. All the reconstructed ECG records were clinically validated by qualified cardiologists. © 2020 IEEE.