PPG-Based Affect Recognition with Long-Range Deep Models: A Measurement-Driven Comparison of CNN, Transformer, and Mamba Architectures

A new arXiv study benchmarks CNN, CNN-LSTM, Transformer, and Mamba architectures for PPG-based affect recognition, classifying arousal, valence, and relaxation from wrist-worn sensors under subject-independent 5-fold cross-validation. Contrary to expectations, long-range models like Mamba and Transformers show no consistent advantage over CNNs on small, noisy physiological datasets. Wearable AI and affective computing engineers should note this before defaulting to heavier architectures for biosignal classification pipelines.