Spectrum sensing in cognitive radio paradigm can be categorized as model/analytical and data-based approaches. However, the former is sensitive to model inaccuracies in evolving network environments, while the latter (machine learning/deep learning approach) suffers from high computational complexity and black box-like behaviour. For devices with low computational abilities, such approaches could be rendered less useful. In this context, we propose a deep unfolding architecture namely Primary User-Detection Network (PU-DetNet) that harvests the strength of both: model and data-based approach. In particular, a technique is described that reduces computational complexity. It involves binding loss function such that each layer of proposed architecture possesses its own loss function whose aggregate is optimized during training. Compared to state-of-the-art, experimental results demonstrate that probability of detection is significantly improved as compared to long short term memory (LSTM) scheme (between 39% and 56%), convolutional neural network (CNN) scheme (between 45% and 84%), and artificial neural network (ANN) scheme (between 53% and 128%) over empirical, 5G-simulated, DeepSig, satellite-communications, and radar dataset. Additionally, computational complexity reduces by 91.69% and 93.15% w.r.t. LSTM and ANN schemes respectively. Moreover, the proposed scheme also shows improvement in throughput by 87% and 130% w.r.t. LSTM and ANN schemes, respectively.