6. CW, FMCW, and MTI: Fighting the Clutter War

6.1 The clutter problem

Real radar lives in a sea of unwanted reflections. Buildings, terrain, sea waves, vegetation moving in the wind, rain, snow, and bugs all return energy to the radar. The total clutter power dwarfs the targets you actually want, often by 60 to 80 dB. A modern fighter has an RCS of 1 m². A square kilometer of vegetated land has a clutter cross section of perhaps $10^4$ m². The radar receives a thousand times more power from the ground than from the airplane.

Without clutter rejection, radar would be useless for detecting aircraft over land or sea. The clever solution: clutter is mostly stationary, while targets move. If we can distinguish moving from stationary, we can suppress the clutter and reveal the targets.

6.2 MTI: the delay-line canceller

The classic moving-target indicator (MTI) processor is breathtakingly simple. Take the echo from one pulse and subtract the echo from the previous pulse. Stationary targets give the same echo on consecutive pulses; the subtraction cancels them. Moving targets give different echoes (because their Doppler phase advances pulse-to-pulse), and the subtraction leaves a residual.

     Pulse N echo:  ─┬─•──•───•─    (clutter at fixed positions)
                     │
                  Delay 1 PRI
                     │
     Pulse N+1 echo: ─┴─•──•───•   (clutter same; moving target shifted)
                                │
                            ────┴───  output of canceller:
                                      stationary clutter cancels,
                                      moving target leaves a residual

In equation form, the output is

$y[n] = x[n] - x[n-1]$

a simple high-pass filter in the slow-time (pulse-to-pulse) dimension. The frequency response is

$|H(f_d)|^2 = 4 \sin^2(\pi f_d T_{PRI})$

It has a zero at $f_d = 0$ (kills DC clutter) and at integer multiples of PRF (the dreaded blind speeds), with a $\sin^2$ shape between. Two-pulse cancellers work; three-pulse cancellers have sharper notches and better clutter rejection.

     |H(f_d)|^2 ↑
                │     ╱╲   ╱╲   ╱╲
                │    ╱  ╲ ╱  ╲ ╱  ╲
                │   ╱    V    V    ╲
                │  ╱                ╲
                │ ╱                  ╲
                │╱                    ╲
                └─────────────────────────►
                0   PRF/4  PRF  2·PRF  3·PRF   f_d
                ↑      ↑           ↑
              null     pass       null (blind speeds)

6.3 MTD: Moving Target Detection (the modern version)

Instead of one or two cancellers, a modern radar uses an FFT across many pulses to produce a fine-grained Doppler spectrum for each range bin. Stationary clutter falls in the zero-Doppler bin and a few neighbors; everything else is target. This is MTD or, more generally, pulse-Doppler processing, and it is what every modern surveillance radar does.

6.4 STAP and adaptive clutter cancellation

For airborne radars, the platform is moving, so clutter itself has Doppler spread over a range: stationary ground returns at the heading have Doppler $-2v_{platform}/\lambda$, sideways returns have zero, and the curve between is complex. Standard pulse-Doppler MTI does not work well here.

Space-time adaptive processing (STAP) is the modern answer: a 2D filter in slow time and array-element space. The radar samples each element per pulse, computes a covariance matrix across both dimensions, and places nulls along the line in the angle-Doppler plane that corresponds to ground clutter while passing targets elsewhere. STAP requires substantial computation but lets airborne radars like the AN/APY-2 in AWACS spot low-flying aircraft against ground clutter.