Indoor positioning systems are attracting more and more attention from the academia and industry recently. Among them, approaches based on WiFi techniques are more favorable since they are built upon the WiFi infrastructures
available in most indoor spaces. However, due to the bandwidth limit in mainstream WiFi systems, the indoor positioning system leveraging WiFi techniques can hardly achieve centimeter localization accuracy under
strong non-line-of-sight conditions, which is common for indoor spaces. In this paper, we present a WiFi-based indoor positioning system that achieves centimeter accuracy in non-line-of-sight scenarios by exploiting the frequency diversity via frequency hopping. During the offline phase, the system collects channel state information from multiple channels at locations-of-interest. Then, the channel state information are post-processed to combat the synchronization errors and interference. The channel state information from multiple channels are then combined into location fingerprints via bandwidth concatenation and in a database. During the online phase, channel state information from an unknown location are formulated into the location fingerprint and is compared against the fingerprints in the database using the time-reversal resonating strength. Finally, the location is determined by the calculated time-reversal resonating strengths. Extensive experiment results demonstrate a perfect centimeter accuracy in an office environment in non-line-of-sight scenarios with only one pair of single-antenna WiFi devices.