Photoplethysmography (PPG) enables unobtrusive car- diovascular monitoring, but motion artifacts, poor con- tact, and low perfusion can corrupt downstream estimates without obvious visual warning. We investigated an un- supervised signal quality assessment framework based on contrastive representation learning and persistent homol- ogy. A SimCLR model with a one-dimensional residual en- coder was trained on 61363 unlabeled 8-second PPG seg- ments from 120 subjects. We evaluated HDBSCAN cluster- ing across learned embeddings, topology-only spaces, and combined feature spaces. Cluster quality was validated post hoc using simultaneous ECG-derived heart rate, with- out using ECG during model training or clustering. In the final validated experiment, standardized 12-dimensional dual-topology features from 50000 synchronized PPG seg- ments produced three clusters and 818 unassigned seg- ments. The dominant cluster contained 48784 segments and had a median absolute ECG–PPG heart-rate error of 0.21 BPM (IQR 0.09–0.49 BPM) across 44341 valid HR pairs. A small artifact cluster contained 137 segments and had a median absolute ECG–PPG heart-rate error of 51.48 BPM (IQR 48.00–54.85 BPM). Grid-search ex- periments showed that high silhouette scores alone were insufficient, because some high-silhouette configurations discarded many segments. These results suggest that a self-supervised topological PPG manifold can isolate se- vere low-quality segments while preserving a large high- quality cluster.