Urban environments, characterized by dense high-rise buildings and narrow streets, present substantial challenges to GNSS positioning due to signal blockages and multipath effects. Conventional fault detection and exclusion (FDE) methods struggle in these settings because the majority of measurements contain multipath or non-line-of-sight (NLOS) errors. To address these challenges, a novel pedestrian dead reckoning (PDR)-aided FDE framework is proposed to enhance GNSS positioning accuracy in urban canyons. In this framework, constraints are first derived from PDR and receiver clock error, and these constraints are then used to enhance GNSS positioning through clustering. Both static and dynamic tests were carried out to evaluate the performance of the proposed system. Results show that the proposed approach achieves accuracies of 1.7-11.5 m, compared to 4.2-57.4 m for chip outputs, while the conventional FDE method is impractical in deep urban canyons because its availability decreases to 3.2%. Kinematic tests in Hong Kong reveal a 52.4% enhancement in root mean square (RMS) accuracy (12.9 m versus 27.6 m for chip outputs) with 100% availability. This work provides a computationally efficient, hardware-independent solution for reliable urban positioning on consumer devices.