Radar systems that use frequency-modulated continuous-wave (FMCW) signals normally form a full range–Doppler image before deciding what to track. This paper proposes a different approach for the tracked mode. The tracker first predicts the target’s next range and radial speed. The new method, called track-conditioned residual estimation (TCRE), removes the predicted motion from the raw samples, summarizes each chirp into a few complex numbers, and then estimates the small remaining frequency errors. That change lets the system work with far less data while focusing on the residual that actually matters for a tracked target.

In the reported design the authors work with a 1-transmit, 3-receive (1-TX/3-RX) radar platform in the 60 GHz band. After association, the tracker hands down a predicted range and velocity. TCRE multiplies the coherent intermediate-frequency samples by the conjugate of the predicted target phase so that the predicted motion is canceled. Each 64-sample chirp is then divided into a few blocks and each block is coherently summed, producing four complex summaries per chirp. The remaining beat-frequency and Doppler-frequency errors are recovered from how these block phases progress over time.

The team also treated an important motion detail that simpler models often ignore: the coupling between fast-time (within a chirp) and slow-time (across chirps) caused by a moving target. They derive a local phase-unwrapping region, which defines how large prediction errors can be before phase ambiguity occurs. They also derive a local information bound for three receivers. According to their analysis, using four coherent sums keeps about 93.77% of the fast-time information you would have with the full chirp, while reducing the data representation by a factor of 16.

The method was tested in controlled, known-truth simulations. In 5,000 matched cluttered trials using a 60 GHz configuration, TCRE reduced range root-mean-square error (RMSE) by about 32.8% to 46.3% and reduced velocity RMSE by about 58.6% to 71.0% when compared to a fully specified residual zoom FFT that used the same prior. The paper also compares TCRE to a strengthened conventional range–Doppler processor and to an uncompressed phase-estimation control, and reports consistency with an ideal local Cramér–Rao bound in some regimes.