Smart wearable and portable healthcare devices are used for continuous patient health monitoring but have limited battery power and on-board memory. Therefore, there is a huge demand for an automated photoplethysmogram (PPG) signal quality assessment (SQA) for energy-constrained smart medical devices. In this article, we propose a novel convolutional neural network (CNN)-based compressive sensing PPG-SQA method for low-power wireless healthcare devices. The main focus of this paper is to find the best compressive sensing matrix to directly classify the compressed PPG segments into noise-free and noisy segments using the optimal one-dimensional CNN architecture. The proposed CS-PPG SQA method was evaluated using 4-layer and 32 filters with a rectified linear unit (ReLU) activation function. Evaluation results showed that the deterministic binary block diagonal (DBBD) sensing matrix with a compression factor of 2 outperforms the existing methods with original PPG signal. The proposed method had an accuracy (ACC) of 99.55% for noise-free PPG (NF-PPG) versus wrist-cup noisy PPG database (MA-DB01), 99.99% for NF versus random noise-added PPG (RN-PPG) segments and 72.71% for NF-PPG versus acceleration corrupted PPG database (MA-DB02). The proposed method can reduce false alarms by discarding noisy PPG segments and reduce model size, and computational time by 50% as compared to other existing SQA methods. © 2023 IEEE.