Find your next learning step
Counting pulses is not the same as counting independent evidence: noncoherent integration gains depend on the detector, operating probabilities and correlation carried by the samples.
Open the chapter contents · Radar root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the question, or read from Chapter 1 for the complete argument. The teaching chapters remain open; these grouped routes form the index.
Route 1 · Define the integrator
Route 2 · Work the correlation ledger
Route 3 · Transfer the pulse count
Route 4 · Check limits and evidence
Route 5 · Ask and continue
CHAPTER 1 OF 12 · DEFINE THE INTEGRATOR
1. What noncoherent integration adds
Radar often receives several echoes from the same resolution cell during a dwell. When phase is sufficiently stable, coherent processing can add complex samples before detection. Noncoherent integration instead combines detected magnitudes or powers, discarding the relative phase.
The operation still pools evidence, but its gain is not simply N for N pulses. It depends on the detector, threshold, required probability of detection, false-alarm probability and target fluctuation model. A square-law integrator that sums pulse powers has a different null distribution from a single-pulse detector.
The first disciplined question is therefore “What statistic is being summed?” Counting pulses without naming the statistic invites an optimistic range budget. The companion article on Swerling target fluctuations explains why target-state correlation must be carried with the integration rule.
CHAPTER 2 OF 12 · DEFINE THE INTEGRATOR
2. Independent pulses and degrees of freedom
Under a simple noise-only model, squared magnitudes from independent complex Gaussian samples have exponential distributions. Summing N independent powers produces a gamma-family statistic with more degrees of freedom. The threshold for a stated false-alarm probability changes with N.
Under signal-plus-noise, detection probability follows a corresponding noncentral distribution. The gain is therefore defined by comparing the single-pulse and N-pulse signal-to-noise ratios needed to achieve the same Pd and Pfa. It is not obtained by adding an arbitrary “pulse gain” after the detector calculation.
Independence is doing real work. It lets the variance of the average fall and gives the sum N distinct pieces of evidence. If clutter, target amplitude or receiver processing correlates the pulses, nominal count N overstates that benefit.
CHAPTER 3 OF 12 · WORK THE CORRELATION LEDGER
3. Worked example: an effective-count ledger
Use a transparent approximation for equally correlated pulse powers. If each pair has correlation ρ, a common effective sample-size expression is Neff = N/[1 + (N − 1)ρ]. It matches N at ρ = 0 and tends toward one as ρ approaches one. It is a teaching diagnostic, not a universal radar detector formula.
| Nominal pulses N | Pair correlation ρ | Effective count Neff | Interpretation |
|---|---|---|---|
| 8 | 0.00 | 8.00 | Eight independent contributions |
| 8 | 0.20 | 3.33 | Some diversity remains |
| 8 | 0.60 | 1.54 | Most evidence is repeated |
The calculation shows why “eight pulses” is incomplete. The receiver still sums eight values, but their information content can resemble far fewer independent observations. A production detector should use its actual covariance or validated detection model rather than this scalar shortcut.
CHAPTER 4 OF 12 · WORK THE CORRELATION LEDGER
4. Correlation loss belongs in the operating point
Correlation can enter through receiver filtering, overlapping processing, coherent clutter, target fluctuation or MTI cancellation. These mechanisms need not share the same covariance. A single estimated ρ is useful only when its definition and estimation window are stated.
The loss should be evaluated at the required Pd and Pfa. A pulse set that looks efficient near Pd = 0.5 may require a different signal level near Pd = 0.95. Noncoherent gain is operating-point dependent because the thresholded distributions are nonlinear.
Do not convert Neff directly into a universal dB credit. Use it to select or interpolate an appropriate detection curve, or compute the correlated statistic numerically. The valid question is the required single-pulse SNR for the same detection specification.
CHAPTER 5 OF 12 · WORK THE CORRELATION LEDGER
5. Fluctuation correlation and noise correlation differ
Noise samples may be nearly independent while target amplitudes remain correlated across a dwell. In a slow-fluctuation Swerling case, several pulses share one target-state draw. Summing noise still helps, yet target diversity does not grow as if every pulse sampled a new radar cross section.
The reverse can also occur: target states decorrelate while processing introduces noise correlation. Treating both effects as one generic loss can hide which mitigation is available. Frequency diversity may change target correlation; waveform or filter design may change noise correlation.
Maintain two ledgers where possible: covariance under the noise-only hypothesis and covariance or fluctuation structure under signal-plus-noise. Detection probability depends on both.
CHAPTER 6 OF 12 · TRANSFER THE PULSE COUNT
6. Changed context: two correlation blocks
Suppose twelve pulses arrive as three groups of four. Pulses inside a group have ρ = 0.5, while groups are independent. Each four-pulse block has effective count 4/[1 + 3(0.5)] = 1.6. Three independent blocks give an approximate total effective count of 4.8.
| Architecture | Nominal count | Correlation structure | Approximate effective count |
|---|---|---|---|
| One eight-pulse dwell | 8 | All pairs ρ = 0.20 | 3.33 |
| Three four-pulse blocks | 12 | ρ = 0.50 within; independent between | 4.80 |
This is not permission to add effective counts under every covariance pattern. The addition works here because the blocks were assumed independent and the approximation was applied within each block. With cross-block correlation, use the full covariance matrix.
CHAPTER 7 OF 12 · TRANSFER THE PULSE COUNT
7. Integration gain and detection range
Once the detector model supplies the required SNR, that requirement can enter the radar equation. For a simplified monostatic radar, received power falls approximately with the fourth power of range. A change in required SNR by a linear factor L therefore changes idealised range by L^(−1/4), holding other terms fixed.
This fourth-root relation makes a useful reasonableness check. A modest integration improvement cannot create an enormous range increase. Conversely, a seemingly small correlation loss may matter near a hard probability-of-detection requirement.
Keep loss and range calculations in consistent units. Convert dB differences to linear factors before applying a power law, and do not count the same integration gain once in the detectability factor and again in the radar equation.
CHAPTER 8 OF 12 · TRANSFER THE PULSE COUNT
8. When effective pulse count is inadequate
Neff compresses a covariance structure into one number. Two correlation matrices can share the same Neff but have different eigenvalue spectra and different tail behaviour after thresholding. Non-Gaussian sea clutter, censoring, saturation and adaptive processing can make second-order correlation insufficient.
Motion can also move a target through range or Doppler cells, so pulses no longer feed the same statistic. A detector may need track-before-detect, coherent subintegrations or M-of-N logic across coherent processing intervals.
Use Neff as a diagnostic and communication tool. For certification, safety or high-cost design, validate the complete processing chain against recorded data or a traceable statistical model.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
Noncoherent integration does not “recover phase later”; the phase relationship has been discarded before combining the detected quantities. More pulses do not guarantee monotonic practical improvement if dwell time, decorrelation, cell migration or changing clutter invalidates the model.
Correlation is not always harmful. Correlated target strength can sometimes be predicted or coherently exploited if phase and dynamics are modelled. The loss described here is relative to a detector that treats samples as independent evidence when they are not.
Finally, a square-root-of-N gain is only a rough approximation in selected regimes. Authoritative detector calculations show that noncoherent gain depends on Pd, Pfa and N; a fixed exponent should not replace those curves.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Evidence and verification route
MathWorks’ authoritative technical example Introduction to Pulse Integration and Fluctuation Loss in Radar defines noncoherent integration gain through single- and multi-pulse detectability factors and demonstrates its dependence on the operating probabilities.
The related guide Introduction to Scanning and Processing Losses in Pulse Radar identifies MTI noise-correlation and batch-processing losses. For a specialist research treatment of correlated noncoherent detection, see Correlated Non-Coherent Radar Detection for Gamma-Fluctuating Targets in Compound Clutter.
Verification should include noise-only false-alarm trials, signal-plus-noise detection trials, the estimated covariance, threshold regeneration for each N and a comparison with the single-pulse baseline. Reusing one threshold across different integration lengths is not a valid audit.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Is noncoherent integration the same as averaging power?
It may sum or average detected magnitudes or powers; scaling can be absorbed into the threshold. The exact statistic and threshold distribution must be stated.
Can Neff replace a full detector simulation?
No. It is a useful approximation for variance and planning under stated correlation structures, not a complete tail-probability model.
Why not claim 10 log10 N dB?
That is the ideal coherent integration gain for phase-aligned samples. Noncoherent gain is lower and depends on the detection operating point.
What should a range budget record?
Record N, integration statistic, Pd, Pfa, target fluctuation case, noise and target correlation assumptions, processing loss and the source of every detectability factor.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Final transfer check and connected reading
Given ten pulses with an estimated equicorrelation of 0.3, calculate Neff = 10/[1 + 9(0.3)] ≈ 2.70. Then explain why this value is a warning about repeated evidence, not a substitute for the correlated detection distribution.
Change the architecture to two independent blocks of five with the same within-block correlation. Each block has Neff = 5/[1 + 4(0.3)] ≈ 2.27, giving about 4.55 across independent blocks. State which assumption made addition permissible.
Continue with the Radar root, Swerling fluctuation and detection mixtures, the Gold Standard of Education and the How X Works hub.
Top · Contents · Previous · How X Works