RF signal leakage occurs in the coaxial distribution system due to cable damage, loose connections, and other issues. FCC regulations limit allowable leakage field strengths and require cable providers to routinely survey the plant for RF signal leakage. In standard-split plants, the primary spectrum of interest falls in the 108 MHz to 137 MHz VHF aeronautical band, and leaks are detected by having headend equipment generate a test signal that is easily detected by specially designed meters. In high-split and full duplex (FDX) plants, critical bands now fall in the upstream, and the cable modems must generate the probe signal. The upstream OFDMA signal is noise-like, making it difficult to detect directly. The OFDMA frame contains a known preamble pilot pattern that can be detected with a traditional matched filter approach, but since the pilot is a small portion of the overall frame, an equally small portion of the leak energy can be captured for detection. In this paper, we present a new upstream leak detection technique built around a specially designed DOCSIS ODFMA upstream data profile (OUDP) testing burst. By sending two identical pulses in a row, with known spacing between them, we can build a correlation-based detector that captures the energy of the full leak pulse. As a result, we have demonstrated a sensitivity increase of greater than 20 dB over the preamble detection approach. The output of the correlation-based detector has both magnitude (indicating the strength of the leak) and a phase (indicating the relative velocity of the detector with respect to the leak) components. We demonstrate in the paper how this Doppler velocity measurement can be used to localize leaks in space using particle filtering techniques borrowed from the radar and image processing fields.