Lesson 37 Flashcards

Click a question to reveal the answer.

1. Name the four trends of L37 and the assumption each one breaks.

Cognitive EW — that the threat library is complete and the threat behaves as catalogued. AI/ML emitter ID — that a classifier's label is a fact. NAVWAR — that you always know where you are. Space EW — that the satellites and links behind your support package are always there.

2. What are the three steps of the cognitive-EW loop?

Sense — intercept and characterize the emitter (ES). Learn — model what the emitter does, without needing a library entry. Adapt — choose a response, then score it by what the emitter does next and feed that back. The feedback is what a library lookup lacks.

3. Why does a fixed threat library fail against an agile, software-defined radar?

The library holds catalogued parametric signatures, per mode (L12). A radar whose waveform is software can present a mode nobody has catalogued, so the lookup finds no match or the wrong one — and the collect, analyze, reprogram cycle that fixes the library takes weeks, not seconds.

4. What is a cognitive radar, and why does it matter to the jammer?

The same sense–learn–adapt loop run by the adversary: a radar that senses your jamming and changes its own waveform in reply. L15's electronic-protection menu stops being a design-time choice and becomes an in-flight one.

5. What are the rows, columns and diagonal of a confusion matrix?

Rows are what each test emitter really was; columns are what the classifier said; the diagonal counts the correct calls. Every off-diagonal cell is one particular mistake, such as a tracking radar called an acquisition radar.

6. How do you read \(P_d\) and \(P_{fa}\) for one class off a confusion matrix?

\(P_d\) along that class's row: the diagonal cell over the row total (true members caught). \(P_{fa}\) down its column, off the diagonal: the non-members called that class, over all non-members. In the lesson's example, TTR gives \(P_d = 88/100 = 0.88\) and \(P_{fa} = 10/200 = 0.05\).

7. Why is one confusion matrix only one point on a ROC?

The classifier outputs a score, and a threshold decides when it names a class. Lower the threshold for "TTR" and both \(P_d\) and \(P_{fa}\) rise; sweep it and the pairs trace L8's ROC. Because a missed tracking radar costs more than a false one, that threshold is set low.

8. Why can half of a classifier's warnings be false even when \(P_{fa}\) is only 0.05?

Because the class is rare. If one intercept in twenty is a TTR, 1000 intercepts give \(0.88 \times 50 = 44\) true TTR calls and \(0.05 \times 950 \approx 48\) false ones. It is L8's many-cells lesson: a small false-alarm rate over many chances fills the display.

9. What does an ML emitter classifier do with an emitter type it never trained on?

It still answers with one of the labels it knows — its matrix has no "unknown" column — often with high confidence. Unless it is built to reject unfamiliar inputs, it turns a library's "no match" into a confident, wrong match. Adversarial emitters, shaped to be mislabeled, exploit the same weakness.

10. Why is GPS a signal below the noise, and what digs it out?

The L1 C/A signal (1575.42 MHz) arrives from about 20,000 km at roughly \(-130\) dBm, about 20 dB under the thermal noise in its 2 MHz band (\(\approx -111\) dBm). The code's processing gain, \(G_p = 10\log_{10}(1.023\times10^6/50) \approx 43\) dB, collapses the signal back out of the noise (L16).

11. What J/S can a GPS receiver tolerate, in the course's model?

Despreading gives \((\text{J/S})_\text{after} \approx (\text{J/S})_\text{in} - G_p\), and the loops need a margin \(M\) on top, so \((\text{J/S})_\text{tol} = G_p - M\). With the notional \(M = 13\) dB that is about 30 dB — the jammer can arrive a thousand times louder than the signal.

12. How do you find a GPS jammer's denial radius?

Set J/S equal to the tolerable level and solve the one-way link for range: \(20\log_{10} R_\text{deny} = \text{ERP} - S - (\text{J/S})_\text{tol} - 20\log_{10} f_\text{GHz} - 92.45\), with ERP and \(S\) in dBm and \(R\) in km (L2's FSPL). A notional 10 W jammer denies about 150 km.

13. Quadruple a GPS jammer's power. What happens to its denial radius, and why?

It doubles. The jammer's signal travels one way, so its power falls as \(1/R^2\) and the radius grows as \(\sqrt{\text{ERP}}\): 6 dB of power buys one doubling of range. A radar's echo pays \(1/R^4\) and would need sixteen times the power. Conversely, every 6 dB of receiver anti-jam halves the circle.

14. How does spoofing differ from jamming?

Jamming denies: it raises the noise until the loops lose lock, and the receiver knows. Spoofing deceives: counterfeit signals capture the loops and walk the position away, so the receiver reports a confident, wrong fix — L18's range-gate pull-off aimed at your position. Defenses: authentication, INS cross-checks, integrity monitoring.

15. How much does the INS drift on a denied leg, and why does the route planner care?

With a notional constant drift rate, \(e \approx \dot{e}\,t\) with \(t = \ell/V\) for a chord \(\ell\) through the circle. A navigation-grade INS drifts about 1.5 km/h. A clearance past a MEZ is real only if it exceeds \(e\): inflate the MEZ, spend less time denied, or plan a fix before the leg that matters.

16. Why does uplink jamming need the satellite's receive beam, while downlink jamming is local?

An uplink jammer attacks the satellite: from inside its receive beam, a relay satellite retransmits the jam to every user; from outside, the antenna barely hears it. A downlink jammer competes with the satellite's faint signal at one receiver at a time, so it denies only the terminals near it — which is why GPS denial is a circle.

17. What must the Final Project's Limitations section do with an L37 trend?

Pick one trend and say what it would change in your route — a number or a line on the map, and by how much if you can. "GPS could be jammed" is a disclaimer; comparing the INS error on a denied leg with your smallest MEZ clearance is analysis.