Find your next learning step
A fixed detection threshold cannot keep the same false-alarm behaviour when background power changes across range or Doppler. Constant false alarm rate processing estimates a local background from reference cells and scales that estimate into a threshold for the cell under test. The method works only as well as its reference window: target leakage, nearby targets and clutter edges can make the estimate unrepresentative.
Open the chapter contents · Radar root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the decision, or read from Chapter 1. The chapters remain open; these grouped routes are a navigable index.
Route 1 · Define the decision
Route 2 · Work the ledgers
Route 3 · Transfer the method
Route 4 · Check limits and evidence
Route 5 · Ask and continue
CHAPTER 1 OF 12 · DEFINE THE DECISION
1. Separate the cell under test from its neighbours
A CFAR window has one cell under test, guard cells beside it and reference or training cells farther away. The cell under test contains the measurement being judged.
Guard cells reduce the chance that energy from the tested target’s main lobe or sidelobes contaminates the background estimate. Reference cells estimate the background against which the test cell is compared.
CHAPTER 2 OF 12 · DEFINE THE DECISION
2. Define the promised false-alarm rate
The designer chooses a target probability of false alarm under a stated background model. The threshold multiplier is then calibrated so noise-only cells cross the threshold at that rate.
“Constant” is conditional. If the reference samples are not representative or the assumed distribution is wrong, the realised false-alarm rate can depart from the design value.
CHAPTER 3 OF 12 · WORK THE LEDGERS
3. Calculate a CA-CFAR threshold
For 16 independent exponentially distributed reference-cell powers, a standard cell-averaging CFAR multiplier for PFA=10⁻⁴ is α=N(PFA^(−1/N)−1). This gives α=16(10^0.25−1)=12.45.
| Input | Value | Calculation or role |
|---|---|---|
| Reference cells N | 16 | Background sample size |
| Design PFA | 0.0001 | Noise-only exceedance target |
| Multiplier α | 12.45 | 16(PFA^(−1/16)−1) |
| Reference mean | 2.00 power units | Illustrative local estimate |
| Threshold | 24.9 power units | 12.45×2.00 |
A cell-under-test power of 30 exceeds 24.9 and is declared a detection in this simplified test. That declaration is not yet a track or an identified object.
CHAPTER 4 OF 12 · WORK THE LEDGERS
4. See how a nearby target causes masking
Suppose the same 16 reference cells would average 2 units, but one reference cell contains an interfering target that adds 80 units. The reference sum rises from 32 to 112, so the mean rises to 7.
| Window condition | Reference mean | Threshold α×mean | CUT power 35 |
|---|---|---|---|
| Homogeneous background | 2.0 | 24.9 | Detected |
| One strong interferer in reference cells | 7.0 | 87.2 | Missed |
| Interferer properly censored; 15-cell recalibration required | About 2.0 | Model-dependent | Potentially detected |
The strong neighbour raises the adaptive threshold and hides the weaker cell under test. This is target masking, not proof that the weaker return vanished from the data.
CHAPTER 5 OF 12 · WORK THE LEDGERS
5. Size guard bands from the processed response
Guard width should follow the actual range–Doppler response after windowing, pulse compression and other processing. Too few guard cells admit target energy into the reference estimate.
Too many guard cells reduce the number or locality of background samples. A wide quiet moat can make the estimate less responsive to a changing clutter field.
CHAPTER 6 OF 12 · TRANSFER THE METHOD
6. Choose a CFAR family for the environment
Cell-averaging CFAR performs well in a homogeneous model. Greatest-of or smallest-of variants respond differently near clutter edges. Ordered-statistic and censored methods can reduce sensitivity to some interfering targets.
No variant dominates every environment. Robustness to multiple targets, clutter transitions and small sample windows trades against detection loss and threshold variability.
CHAPTER 7 OF 12 · TRANSFER THE METHOD
7. Changed-context transfer: a shoreline clutter edge
A range window straddles low-power water clutter and higher-power land clutter. A symmetric average no longer estimates one homogeneous background.
The transfer check is to split or rank the reference evidence, test both approach directions and compare empirical false alarms on labelled shoreline data. Tuning only on open water would conceal the edge failure.
CHAPTER 8 OF 12 · TRANSFER THE METHOD
8. Connect sidelobes, ambiguity and tracking
Read weak echoes beside strong targets for the point-spread and sidelobe problem that can contaminate guard or reference cells. Use blind speeds and staggered pulses when the ambiguity comes from pulse timing rather than threshold adaptation.
A CFAR detection becomes operationally useful only after localisation, association and track confirmation preserve uncertainty across scans.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
CFAR does not remove clutter; it adapts a threshold to a background estimate. It cannot guarantee constant false alarms under every nonstationary, correlated or heavy-tailed environment.
Guard cells are not additional reference samples, and more guard cells are not automatically safer. A detection also does not prove target identity, intent or persistence.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Practical evidence and checking routine
Record the cell dimensions, coherent processing, window shape, guard count, reference count, estimator, multiplier and assumed background distribution. Keep range and Doppler axes explicit.
Measure false-alarm and detection rates separately on homogeneous regions, clutter edges and multi-target scenes. Recalculate multipliers whenever the effective number or weighting of reference samples changes.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Why not use one fixed threshold?
A fixed value cannot follow spatial or temporal changes in noise and clutter power.
Why can a stronger neighbour hide a weaker target?
If the neighbour enters the reference window, it inflates the estimated background and therefore the threshold.
Does CFAR mean the false-alarm rate is exactly constant in service?
Only under the calibration model. Real backgrounds require empirical validation and monitoring.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Sources and next reading
The research article on multiple-target CFAR performance in sea clutter examines masking when interfering targets occupy reference cells. The article on boundary-adaptive CA-CFAR describes the cell under test, guard cells, training cells and adaptive threshold in an FMCW detection chain. The threshold and masking ledgers here are original teaching calculations.
Continue through the Radar root, the Gold Standard of Education and How X Works.
Top · Contents · Previous · How X Works