Short-Range Ground Surveillance Radar: A Reference Architecture

The problem this radar class solves

A short-range ground surveillance radar (GSR) sits between thermal cameras and full air-defense systems in the perimeter security stack. Cameras give you identification but are degraded by weather, smoke, and cover. Air-defense radars give you range but their economic and SWaP profiles are wrong for fixed-site perimeter use. A GSR with one to ten kilometer range, sub-meter range resolution, and the ability to discriminate humans from animals from vehicles, fills the middle.

For DoD-relevant perimeter use cases — forward operating bases, expeditionary airfields, distributed depot security, unmanned watch towers — the practical envelope is:

• Range: 200 m (human, walking) to 8 km (vehicle, moving)
• Range resolution: 0.5–1 m
• Update rate: 1–4 Hz volume scan, faster for tracked targets
• Power: under 100 W mean for a fixed-site unit
• Form factor: a man-portable tripod for an expeditionary unit

FMCW vs pulsed

Two waveform families dominate this class. Pulsed radars send short bursts and measure time-of-flight; FMCW (frequency-modulated continuous-wave) sweeps a chirp and measures the beat frequency between transmit and receive. For short-range GSR, FMCW wins on several axes that matter for the field:

• Lower peak power. FMCW spreads transmit power continuously, so the peak-to-average ratio is roughly 1:1 versus 100:1 or more for pulsed. This dramatically eases the PA, the heat-sink design, and the antenna isolation budget.
• Better range resolution per dollar. FMCW range resolution is bandwidth-limited, not pulse-width-limited. A 200 MHz chirp gives ~0.75 m resolution with cheap, narrow-band analog parts.
• Cleaner Doppler. Continuous transmission means the Doppler resolution is set by total observation time, which can be hundreds of milliseconds without giving up range. Walking humans show up cleanly against ground clutter.

The trade-off is transmit-receive isolation. FMCW radars transmit and receive simultaneously, so the antenna isolation between TX and RX paths must be 60+ dB. This is usually solved with separate antennas and careful PCB layout.

Signal chain

The reference back-end uses a multi-core DSP for the heavy math and an FPGA for the time-critical I/Q acquisition and matched filtering:

RF front-end ──→ ADC ──→ FPGA dechirp + windowing ──→ Range FFT
                                                          │
   Doppler FFT ←── Slow-time buffer ←─────────────────────┘
        │
        ├── CFAR detection
        ├── Tracker (Kalman, IMM)
        └── Classifier (target type)

• Dechirp in fabric. Multiplying RX by a delayed conjugate of TX produces baseband samples whose frequency encodes range. Fabric does this well at line rate.
• Range FFT per chirp. A 1024-point FFT sized for the desired range gate.
• Doppler FFT across chirps. A 128- to 256-point FFT across the slow-time dimension produces the range-Doppler map.
• CFAR + tracker on the DSP. Constant-false-alarm-rate detection thresholds the range-Doppler map; detections feed a multi-target tracker — typically an Interacting Multiple Model (IMM) Kalman filter to handle regime transitions from "stationary" to "walking" to "vehicle."
• Classifier on the DSP. A small neural classifier — or a hand-tuned feature classifier when explainability matters — labels each track.

A multi-core C66x DSP comfortably runs CFAR + tracker + classifier at update rates well above what the radar produces.

What changes for counter-UAS

The same architecture serves counter-UAS with a few changes:

• Higher elevation coverage (the antenna has to look up, not just at the horizon).
• Higher pulse-repetition or chirp rate to handle the higher Doppler velocities of small drones.
• A different classifier — drones look very different from walking humans in micro-Doppler signatures.

The back-end signal chain is reusable; only the antenna, the chirp parameters, and the classifier change.

Where this fits in a sustainment posture

A ground surveillance radar is one of the most reliable sources of "what just moved" telemetry in a contested environment. Cameras fail in weather and at night without IR. Acoustic sensors fail in wind. GSRs are unaffected by both. Integrating GSR tracks into the sustainment decision graph — "an unauthorized vehicle approached the fuel depot at 23:14 from heading 270, speed 45 km/h, 3.2 km" — lets the platform reason about supply integrity events without depending on whether a human happened to be looking at a camera at the right moment.