Reciprocity calibration of hydrophones in a bounded water tank is biased by boundary reflections that prevent free-field conditions. A method combining complex analytic signals with physically adaptive coherent smoothing is proposed to recover the free-field sensitivity. Sine and cosine linear sweeps are transmitted in a V-shaped (down/up) sequence; the complex analytic signal is mapped into the frequency domain to remove the time–frequency ambiguity. The smoothing window width is set from the geometry of the shortest reflection path in the tank, and the transfer impedance is smoothed by a three-stage cascaded complex moving weighted average (CMWA). A mirror-source correction factor compensates for the deterministic first-order reflections, and a modal coherence enhancement factor corrects the low-frequency bias where the tank modes are sparse and CMWA's incoherent-superposition assumption fails. The equivalent absorption coefficient αeq(f) of the tank is obtained by simulating the frequency-dependent reflection coefficient of the anechoic wedges, anchoring the enhancement factor to measurable physical quantities. The uncertainty is evaluated following the GUM from sensitivity coefficients derived from the measurement model. Dual-tank experiments (small research tank, large verification tank) over 1 kHz–10 kHz show a maximum deviation of 0.67 dB from single-frequency calibrations on a free-field standard device (at 6.3 kHz) and an inter-tank deviation below 1 dB. The combined standard uncertainty is 0.40 dB–0.45 dB, the expanded uncertainty (k = 2) is 0.81 dB–0.90 dB, and the results agree with the standard device within its stated uncertainty, confirming the feasibility of reciprocity calibration in bounded water tanks.