We proposed a neural network (NN)-based framework for the joint estimation of hardware impairments in wireless devices. We validated the approach using real-world measurements from Bluetooth Low Energy (BLE) and WiFi devices. Experimental results show that the estimated impairments exhibit a stationary distribution after the warm-up phase, highlighting their stability and potential as device-specific fingerprints. Moreover, we demonstrated that these impairments provide a lightweight yet robust alternative to raw IQ samples for RF fingerprinting (RFFP), achieving an average improvement of 54% in classification accuracy across different domains.