Position Estimation Baselines for RF (when
DF/TDOA is absent)
When direction finding (DF) or time-difference-ofarrival (TDOA) is absent, teams sometimes adopt a frequencyproxy (FP) position estimate: mapping observed frequency shifts to
range and projecting onto a kinematic prior. We show this is a poor
baseline: it is systematically biased under oscillator offsets, CFO
jitter, and multipath, and it degrades sharply with modest carrier
frequency instability. Across a mixed-motion suite, FP trails a
simple constant-velocity (CV) prior and a fused learned+kinematic
upper bound by 18.4% in ADE@1, with higher signed bias (2.1 m).
We release a tiny, reproducible baseline harness and advocate for
stronger ablations than FP in RF tracking work.