RF Journey

01 / 10

Failure: sensitivity passes, the nearby transmitter wins

The gateway meets its quiet sensitivity allocation. Why does a nearby transmitter still stop reception?

Continue the fictional telemetry gateway from Sensitivity, Gain & Noise Budget. Its wanted input is −105 dBm at R2, the antenna-feed plane. The matched quiet chain has 42.5 dB gain, 2.799879362 dB NF and conditional sensitivity −118.165007875 dBm. That gives 13.165 dB above the quiet threshold. None of those numbers says how much unwanted energy its LNA, mixer or converter can tolerate.

Now admit one −10 dBm CW blocker, 20 MHz above the wanted carrier. With no extra rejection, the front-end loss makes its LNA input −11.5 dBm. The synthetic LNA input P1dB is −15 dBm: the blocker alone is 3.5 dB above the supplied compression point. The small wanted signal must traverse that same device. A later IF gain reduction cannot restore the lost linear relationship.

Think about itIf IF gain drops by 40 dB, does the LNA recover its input headroom?

Answer

No. The LNA still receives −11.5 dBm from the blocker. IF gain acts after the LNA and mixer. It can lower the converter level only if the preceding analog amplitude model remains valid. After this LNA failure, later levels are uncompressed extrapolations; neither quiet sensitivity nor a calculated downstream SNR qualifies reception.

Opening contrast · R2 input, high gain, matched synthetic gateway
Corner	First evidence	Decision
Quiet · all blockers off	NF 2.799879362 dB; sensitivity −118.165007875 dBm	Conditional weak-signal screen only.
−10 dBm at +20 MHz	RX-LNA input −11.5 dBm vs IP1dB −15 dBm	Linear domain exceeded at R1-LNA-in. Stop downstream fidelity claims.
−25 dBm burst at +1 MHz	Mixer input −6.5 dBm vs IP1dB −5 dBm	1.5 dB margin fails 6 dB reserve; this is not proof of exactly 1 dB compression.

Common misconceptionExcellent small-signal NF guarantees blocker performance.

NF measures small-signal SNR degradation under a specified noise convention. Strong signals can change gain, create products or push converter peaks across the rail. Build a separate dynamic-range argument from actual threats and stage conditions.

Begin that argument at the point where each unwanted source enters, not at the output where the failure becomes visible.

02 / 10

Catalogue every blocker and coupling path

What must you know about a blocker before adding a number to the budget?

A threat row names frequency, on-time level, waveform and occupied width, duty/timing, receiver mode, coupling path and entry plane. Antenna coupling enters at R2; a board clock coupled at the mixer input skips the front-end filter and LNA. A conducted level at a transmitter output is not a receiver input level until a coupling model or measurement connects the ports.

Four canonical gateway threats · illustrative engineering case, not conformance levels
ID / waveform	Frequency / width	Level and entry	Timing / mode / path
Quiet · no blocker	Wanted 2450 MHz; 13.5 kHz ideal signal support	Wanted −105 dBm at R2	High gain; continuous wanted.
B1 · nearby CW screen	2470 MHz; +20 MHz; zero-width tone	−10 dBm at R2	[0,2000) µs; high gain; antenna-feed coupling.
B1 + B2 · two CW tones	2451 and 2452 MHz; zero-width tones	−55 dBm EACH at R2	[0,2000) µs; high gain; two distinct source frequencies.
B1 · coexistence burst	2451 MHz; +1 MHz; gated CW screen	−25 dBm on-time at R2	[50,150) µs; 5% of 2000 µs; automatic starts high.

Blockers default to off, represented by absence, not a magic −999 dBm input. A CW burst is a plateau screen: the exact event times are modeled, but spectral splatter from its edges is not. A wideband or modulated source keeps its frequency support for overlap inspection while losing supported CW amplitude claims. Do not quietly turn Wi-Fi into this exact two-tone fixture.

Reference-plane contract · semantic portfolio names and local suffixes
Plane	What crosses it	Local convention
R2 → R1-LNA-in	Antenna feed → RX-FE → LNA input	50 Ω matched available power; RX-FE at physical 290 K.
R1-LNA-out	LNA output = mixer RF input	50 Ω; real LO drive and image response require evidence.
A1-IF → A1-ADC-in	Mixer 100 MHz IF → driver → converter	100 Ω differential teaching power at A1; actual ADC impedance is not asserted resistive.
R0 / R3	RF demodulator boundary / detector decision boundary	R3 is not an antenna connector. A0 codes and D3 80 kcomplex-sample/s remain separate.

Common misconceptionThe strongest blocker is always the dominant failure.

A weaker pair can place a cubic product exactly on the wanted channel. A closer blocker may sample a much noisier part of the LO mask. Rank paths and mechanisms, not only a list of dBm values.

Exact technology test levels belong to Path 07, coupling distributions to Path 06, and repeatable receiver tests to Path 08. Here, preserve a source’s identity as it moves through the chain.

03 / 10

Propagate levels and compression headroom

Which local input margin fails before you ever reach the ADC?

Carry each wanted, blocker and product frequency through its own transfer. The nominal stage gains are −1.5/+20/−6/+30 dB and NFs 1.5/1/6/4 dB. We retain p05-m03-receiver-ledger-v1, including its effective wanted-channel mixer NF. We add synthetic input P1dB/IIP3 conditions in p05-m04-blocker-agc-v1; these do not replace the different bilateral device fixture in 04.4.

Pinned local CW conditions · nominal bias, active devices 25 °C, matched interfaces
Stage	Gain / NF · dB	Input P1dB / IIP3 · dBm	Condition
RX-FE	−1.5 / 1.5	not modeled / not modeled	Linear passive teaching network; no real power-handling claim.
RX-LNA	+20 / 1	−15 / −5	CW compression; two equal CW tones near 2.450 GHz, 1 MHz spacing.
RX-MIX	−6 / 6	−5 / +10	LO 2.350 GHz; adequate fixed drive assumed; effective wanted-band product convention.
RX-DRV	+30, +10, −10 / 4	−10 / +5	100 MHz IF. Fixed NF and input intercepts across gain states are synthetic assumptions.

The screening law pins flat parameters over RF 2.430–2.470 GHz and IF 80–120 MHz. Outside that window the frequency map remains useful, but amplitude evidence stops. RF/IF band edges, input P1dB and the independent detector range are different limits.

\begin{aligned}P_{\mathrm{total,in}}&=10\log_{10}\left(\sum_i10^{P_{i,\mathrm{in}}/10}\right)\\M_{\mathrm{stage}}&=\mathrm{IP1dB}-P_{\mathrm{total,in}}\\\mathrm{OP1dB}&=\mathrm{IP1dB}+G_{\mathrm{small}}-1\,\mathrm{dB}\end{aligned}

Pi in dBm; mW conversion is 10^(Pi/10). M is conditional margin to an INPUT compression point, not a peak or absolute-damage rating.

For the LNA, OP1dB = −15 + 20 − 1 = +4 dBm. Omitting the final −1 dB mistakes the uncompressed output extrapolation for the compressed output point. Negative input margin leaves the small-signal domain. Margin below 6 dB fails this lesson’s local reserve; a positive margin alone is not measured fidelity. Intercept estimates need the stricter 10 dB guard.

Common misconceptionP1dB at the output is enough without stage input levels.

Use the local input total and its waveform against IP1dB. A small final output can hide a compressed upstream stage. Absolute maximum voltage, current, temperature and input-power ratings are unknown here; P1dB is not a damage limit.

Once a predecessor compresses, retain the first invalid stage and label every downstream level as an uncompressed extrapolation. Do not carry a fabricated nonlinear gain correction into an apparently precise SNR.

04 / 10

Intermodulation products and intercept approximations

Can two harmless-looking frequencies create an unwanted signal right on the carrier?

Think about itFor f1 = 2451 MHz and f2 = 2452 MHz, where does 2f1−f2 land?

Answer

At 2450 MHz, exactly on the wanted channel. The other close-in permutation is 2f2−f1 = 2453 MHz. A channel filter that preserves wanted energy at 2450 MHz cannot separate an already coincident IM line by frequency alone.

An intercept is where extrapolated fundamental and intermodulation lines would meet. It is not a safe input power. Use the approximation only while contributing parents remain at least 10 dB below each pinned CW input P1dB, under compatible frequency, spacing, bias and match. Unknown IIP2 permits frequency enumeration but no numeric IM2 level.

\begin{aligned}P_{2f_a-f_b,\mathrm{out}}&\approx2P_a+P_b+G-2\mathrm{IIP3}\\\text{Equal tones:}\quad P_{\mathrm{IM3,out}}&\approx3P+G-2\mathrm{IIP3}\end{aligned}

Third-order cubic normalization at one stage. Pa and Pb are individual INPUT CW tone powers, G is small-signal gain, and IIP3 is input-referred, all in dB/dBm as labelled. Unequal-tone normalization retains the doubled parent.

Independent −55 dBm EACH R2 fixture · zero extra rejection, high gain
Stage output	Per-tone power	Generated in-band IM3 propagated to A1	Origin calculation
RX-FE → R1-LNA-in	−56.5 dBm	not modeled	Passive transfer only.
RX-LNA → R1-LNA-out	−36.5 dBm	−115.5 dBm	3(−56.5)+20−2(−5)−6+30
RX-MIX → A1-IF	−42.5 dBm	−105.5 dBm	3(−36.5)−6−2(10)+30
RX-DRV → A1-ADC-in	−12.5 dBm	−107.5 dBm	3(−42.5)+30−2(5)

Increasing both parents by 1 dB raises each cubic product by 3 dB while the model remains valid. Increasing only the doubled parent gives 2 dB; increasing only the other parent gives 1 dB. Intercept estimates near compression display “inspect—intercept extrapolation unreliable.” ADC clipping invalidates the digital result but does not erase a valid analog contribution generated before the ADC.

\begin{aligned}P_{\mathrm{bound}} &= (\sum _{i} \sqrt{P_{\mathrm{i}}})^{2} \\ \text{Bound for the three A1 contributions} &= -99.012062399 \mathrm{dBm}\end{aligned}

Common-plane powers in mW. Coincident products from different stages have unknown relative phase. This is a constructive bound, not a measured level or random-noise sum.

Go deeperCascade orientation, phase and the separate IM2 fixture

In a separate flat-gain, aligned-product cascade, constructive cubic coefficient addition gives 1/IIP3total = Σ(Gpreceding/IIP3i), with gains and intercept powers in linear units. The corresponding aligned second-order normalization is 1/√IIP2total = Σ√(Gpreceding/IIP2i). Gains include only preceding stages. Frequency-selective gains, mixing conversions and unknown phases prevent treating either shortcut as a general receiver truth. ADI’s selectivity derivation explicitly uses an in-phase worst case and different parent/product filtering.

A separate one-stage example supplies IIP2 = +30 dBm, G = +10 dB, and two −40 dBm tones: PIM2,out = Pa + Pb + G − IIP2 = −100 dBm. Raising both tones by 1 dB gives −98 dBm, a 2:1 slope. At 101 and 102 MHz, the sum is 203 MHz and difference 1 MHz; neither is the wanted 100 MHz IF. These values do not invent IIP2 for the gateway.

Common misconceptionIP3 predicts all modulated blocker behavior.

Memory, bias dynamics, cross-modulation, waveform peaks and frequency-dependent coefficients need their own evidence. The pairwise table omits three-distinct-tone cubic terms, wanted/blocker mixing, self harmonics and higher orders. No complete spur-free or modulation-quality claim follows.

IM is one mechanism. Next distinguish it from gain change and noise contamination so the proposed remedy addresses the cause.

05 / 10

Cross-modulation, compression, and NF degradation

What exactly changed when the receiver became less sensitive?

“Desense” describes a degraded receive threshold; it does not identify its cause. A strong signal can change the incremental gain experienced by a weak wanted signal. If the blocker’s envelope changes, the weak signal’s gain can vary with it: cross-modulation. That requires a waveform-dependent nonlinear model, not a single interpolated IP3 line.

Mechanism ledger · keep each failure and evidence request distinct
Mechanism	Causal path	Discriminating next evidence
Direct leakage	Unwanted spectrum enters the wanted passband	Measure coupling and the actual channel response with the declared waveform.
Compression / cross-modulation	Strong envelope changes gain at an upstream device	Wanted-plus-blocker AM/AM, modulation transfer and recovery at named bias and temperature.
Intermodulation	Two or more parents generate a new in-band line	Change parent levels/spacing; preserve origin and receiver/instrument residual checks.
Reciprocal noise	Blocker translates LO noise into wanted band	Blocker offset/power and LO SSB density over exact integration bounds.
Gain-state noise / converter noise	Reduced gain changes input-referred later contributions or converter allocation	State-specific gain/NF and actual in-band converter noise.
AGC blanking / clipping	Finite response or rails remove samples	Time-aligned source, detector, gain command and valid-code capture; packet/preamble overlap.

Common misconceptionOut-of-band means irrelevant.

The label depends on the reference filter and plane. A blocker outside the digital channel may still reach the LNA, mixer and ADC. Filtering after those blocks cannot prevent the distortion or clipping they already created.

Small-signal NF at one bias is not a large-signal NF-degradation curve. Once compression invalidates the incremental linear model, the ledger suppresses current downstream noise claims. It retains the quiet NF as an explicitly separate reference, so you can still see what the design was trying to preserve.

Before changing gain, inspect the oscillator path: it can move blocker energy into the wanted band without the same two-tone mechanism.

06 / 10

LO phase noise causes reciprocal mixing

How can an offset blocker raise the wanted-band noise even before strong compression?

A real LO has phase-noise sidebands. Mixing a CW blocker with that LO translates a slice of those sidebands into the wanted channel. Keep the LO at 2.350 GHz, wanted RF at 2.450 GHz and IF at 100 MHz. In the anchor, the blocker is +1 MHz from wanted and −35 dBm at the mixer RF input, not at R2.

\begin{aligned}N_{\mathrm{rm,mixer-in}} &= P_{\mathrm{blocker}} \times \\ &\int |H(f)|^{2} 10^{\frac{L(|\Delta f-f|)}{10}} d f\end{aligned}

CW narrowband teaching proxy. Pblocker is linear power at mixer RF input. L is LO SSB density in dBc/Hz at positive offset |Δf−f|. H is normalized amplitude response. Integrate signed wanted f = −10…+10 kHz.

Think about itThe wanted integral runs from −10 to +10 kHz. Should a factor of two be added for SSB phase noise?

Answer

No. One blocker translates one relevant LO side into that full 20 kHz wanted interval. The signed integral already covers both edges of the wanted band. Doubling it again would add an unjustified 3.0103 dB. A second independent symmetric blocker supplies a second contribution; that is a different physical source.

Independent one-side reciprocal anchor · H = 1 in ±10 kHz; L = −120 dBc/Hz over 990–1010 kHz offset
Quantity / plane	Calculation	Result
Mixer RF input equivalent	−35 −120 +10log10(20000)	−111.989700043 dBm
R2 input equivalent	Mixer result −18.5 dB preceding wanted gain	−130.489700043 dBm
R2 baseline noise	Same quiet receiver, same 20 kHz ENBW	−128.165007875 dBm
Noise-limited degradation	10log10(1 + 10^[(Nrm−Nbaseline)/10])	2.001676806 dB

Common-plane power addition is essential. The −111.99 dBm mixer figure cannot be added directly to −128.17 dBm at R2. Two equal, independent reciprocal contributions add 3.010299957 dB to the reciprocal noise, not automatically 3.0103 dB to total receiver degradation.

Common misconceptionReciprocal mixing is additive thermal noise.

Its source is the blocker and oscillator phase-noise process, not a new resistor temperature. This lesson adds its proxy to thermal/receiver noise only under an explicit independence assumption at the same plane and bandwidth. The resulting SNR degradation is not PER or measured Wi-Fi desense.

Go deeperBounded masks, deterministic integration and spurs

The anchor is flat only from 990 to 1010 kHz. Shifting the blocker so any required LO offset leaves that mask makes the integral unknown. The optional synthetic mask has knots (10 kHz, −90), (100 kHz, −105), (1 MHz, −120), (10 MHz, −140), (30 MHz, −145), in dBc/Hz. It interpolates in log-frequency/dB and uses 1024 equal signed-frequency Simpson panels across ±10 kHz; both continuous integration limits must lie inside the mask. Discrete spurs are excluded and require separate lines. Wideband convolution and phase correlations are deferred.

Build on 04.7’s LO plane and mask discipline. Its different 2.550 GHz observation-receiver handoff and real device drive examples are not transplanted into this 2.350 GHz synthetic system fixture.

Reciprocal noise is born at the mixer. Filtering parents before that point and filtering already translated noise afterward are different interventions.

07 / 10

Place filtering, linearity, and gain together

Where can a filter stop the unwanted cause before it creates an in-band effect?

The zero-extra-rejection corner is a deliberately broad-front-end screen. It keeps 05.3 wanted gain/NF but does not claim to implement 05.2’s narrow IF selection. The interaction adds ideal filters at pre-LNA and post-mixer planes: each passes ±10 kHz with its declared wanted insertion loss and applies extra rejection outside it. These are transparent allocation models, not realizable sharp-edge component responses.

Filter-placement counterexample · −10 dBm, +20 MHz blocker; zero added wanted loss
Extra rejection	LNA input	Consequences
20 dB before LNA	−31.5 dBm	16.5 dB LNA margin; parent also reaches mixer 20 dB lower. ADC can still be limiting in high gain.
20 dB after mixer	−11.5 dBm	LNA remains above P1dB; a low downstream extrapolated level cannot qualify the chain.
20 dB after mixer, −55 dBm pair	Parents into driver drop 20 dB each	Driver-generated cubic product drops 60 dB; existing in-band LNA/mixer IM sees only wanted insertion loss.

For the −55 dBm pair, post-mixer rejection of 20 dB leaves the LNA/mixer A1 in-band terms at −115.5 and −105.5 dBm when wanted insertion loss is zero; it changes the driver-generated term from −107.5 to −167.5 dBm. The parent frequencies and the product frequency must follow different filter transfers.

Common misconceptionFilters can be placed after the overloaded stage with the same result.

They can remove the parent after the damage has happened. In-band products or cross-modulation may remain. Put the necessary rejection before the vulnerable stage, then check the noise cost and what the next stage now sees.

An extra 1 dB matched loss before the LNA at 290 K increases nominal receiver NF by 1 dB and reduces wanted gain by 1 dB. Loss after substantial gain usually costs less input-referred noise, but protects fewer upstream blocks. The noise calculation inserts each extra passive loss in its physical position; it does not subtract blocker rejection from wanted gain indiscriminately.

Go deeperLinearity, noise and stability are a joint allocation

Moving gain forward suppresses later noise but raises later blocker inputs. Changing bias may improve intercept at an energy cost and change NF. A real filter’s impedance can change amplifier gain, stability and noise match. This scalar matched model cannot approve that termination; request the S/noise parameters and stability review from Path 04. ADI’s selectivity example supports tracking off-channel parent attenuation separately from the coincident product, under its stated phase assumptions.

With the placement choices explicit, add time. Gain changes do not happen the instant a blocker appears.

08 / 10

AGC needs states, thresholds, hysteresis, and time

What reaches the converter while the AGC is still deciding to act?

The three driver states are +30/+10/−10 dB, giving total gains 42.5/22.5/2.5 dB. This diagnostic observer sits at A1-ADC-in and reports composite average dBFS relative to the full-scale sine. It sums tracked wanted/blocker powers over 80–120 MHz, omits noise/error and remains ideal through +30 dBm even if ADC codes clip. It never reconstructs unclipped power from clipped codes.

Down qualifies at detector ≥−12 dBFS continuously for 20 µs; up qualifies at ≤−36 dBFS for 1000 µs. Between thresholds, hold the state and reset both timers. A broken condition resets its timer. Each event moves one adjacent state, then ignores and resets timers during [transition, transition+5) µs. Qualification restarts when the new state settles. Bound states hold.

Think about itThe burst starts at 50 µs. Does automatic gain control protect its first sample or immediately reach low gain?

Answer

Neither. High→medium occurs at 70 µs, settles at 75; a new attack hold then reaches medium→low at 95 and settles at 100. Earlier clipped samples and blanked intervals stay lost or exposed. The gain switch cannot improve the mixer’s upstream 1.5 dB reserve.

Independent timed fixture · blocker −25 dBm at R2, +1 MHz, interval [50,150) µs
Boundary / interval	State / event	Causal interpretation
0…49 µs	High, quiet	No blocker source; detector below lower, already at high bound.
50…69 µs	High, attack qualifying	Uncompressed A1 blocker +17.5 dBm; rail screen fails before AGC reacts.
70 → 75 µs	High→medium; blank [70,75)	At 75, restart qualification from settled medium.
95 → 100 µs	Medium→low; blank [95,100)	No skipping a state; low output follows settling.
150 µs	Blocker removed before detector evaluation	Start low-state release hold; stop endpoint is exclusive.
1150 → 1155 µs	Low→medium; blank [1150,1155)	Next release starts at 1155; high would occur at 2155, outside the horizon.

At the burst plateau, LNA input is −26.5 dBm, mixer input −6.5 dBm and driver input −12.5 dBm. LNA margin 11.5 dB clears the screens; mixer/driver margins 1.5/2.5 dB fail the 6 dB reserve in every gain state. This conditional timing exercise is allowed while inputs remain below P1dB. If a predecessor leaves the linear domain, quantitative AGC timing becomes unavailable from that boundary.

Common misconceptionAGC prevents front-end overload.

This AGC changes IF driver output gain. It cannot reduce LNA/mixer input or undo prior clipping. Real receivers may distribute attenuation and detectors elsewhere, which changes the noise trade and timing contract.

Go deeperHysteresis, chatter and a real implementation evidence request

The 24 dB default threshold separation exceeds the 20 dB state step. This avoids the simple deterministic toggle in which a down-step immediately demands an up-step. Custom separation ≤20 dB warns about chatter; threshold ordering is still mandatory. Hysteresis does not remove detector noise or delay in real hardware.

ADRV9001-family UG-1828 Rev. A, Receiver Gain Control, distinguishes detector functions, gain modes and gain-block timing. Use those categories to request detector bandwidth/range, gain table, gain-dependent noise, settling and source/packet timing. Its implementation does not validate the three-state synthetic timers here.

At every time boundary the model applies source events, evaluates the settled detector, then tests holds and transitions. This ordering makes the event trace reproducible without a browser clock. Now compare those states with the converter’s instantaneous rail.

09 / 10

Protect converter full scale and digital headroom

Does an average level below full scale leave enough room for the largest peak?

The ADC boundary is a 2 Vpp differential sine across a teaching 100 Ω load. Sine RMS is Vpp/(2√2), so PFS = 6.989700043 dBm. The rail-equivalent instantaneous power is PFS+3.010299957 = 10 dBm. Average dBFS uses the first number; a peak-to-rail comparison uses the second.

\begin{aligned}P_{\mathrm{FS}} &= 10 \log_{10}[\frac{V_{\mathrm{pp}}^{2}}{8R_{\mathrm{load}} \times 1 \mathrm{mW}}] \\ P_{\mathrm{peak}} &= P_{\mathrm{average}} + \mathrm{CF} \\ M_{\mathrm{rail}} &= P_{\mathrm{FS}} + 3.010299957 - P_{\mathrm{peak}}\end{aligned}

Differential voltage across 100 Ω. CF is a supplied composite peak/average power ratio; reserve R is additional distance below rail. Generic CF = 6 dB; two equal CW tones have exact worst-envelope CF = 6.020599913 dB.

Independent pair-only peak anchors · excludes the much weaker wanted signal; A1 high gain, two equal CW tones
R2 each	A1 each / total average	Worst peak	Rail reserve / decision
-55 dBm	-12.500000000 / -9.489700043 dBm	-3.469100130 dBm	13.469100130 dB · clears 3 dB reserve
-45 dBm	-2.500000000 / 0.510299957 dBm	6.530899870 dBm	3.469100130 dB · clears 3 dB reserve
-43 dBm	-0.500000000 / 2.510299957 dBm	8.530899870 dBm	1.469100130 dB · reserve fails; no clipping

At −45 dBm each, reserve is 3.469100130 dB: it clears the 3 dB allocation by 0.469100130 dB. At −43 dBm each, reserve is 1.469100130 dB: the reserve fails, but the 10 dBm rail has not been crossed. These exact anchors concern the blocker pair alone. The live composite average also sums wanted power, so its final digits differ. Applying the exact-pair CF to any additional waveform is a declared proxy, not an exact envelope reconstruction.

Common misconceptionADC average level alone proves clipping margin.

Two tones sum to each+3.0103 dB in average power; their coherent peak needs the separate 6.0206 dB crest factor. An unknown coherent peak cannot be obtained by RSS. A supplied worst-envelope bound or a declared waveform/CCDF is needed.

For an exact two-tone envelope, crossing the deterministic worst-peak screen establishes possible rail crossing in the stated waveform, not a packet-error rate. For a modulated signal, clipping probability requires a CCDF with sample population, oversampling, filtering and observation time. The current model reports a peak-bound screen only; its clipping probability is unknown. A generic 6 dB crest factor cannot supply a probability.

Reducing driver gain by 20 dB reduces analog output signal and noise by 20 dB in this fixed-NF fixture. It leaves this receiver’s input-referred analog NF unchanged, but makes fixed converter input noise more significant. Reuse 05.3’s converter-noise allocation rather than claiming gain reduction always improves the full receiver.

Clipping prevention also does not prove alias clearance. At A0’s 80 MS/s teaching rate, the 100 MHz IF folds to 20 MHz. A 140 MHz IF tone also folds to 20 MHz; such an input is outside the current 80–120 MHz amplitude fixture and still geometrically dangerous. D3’s 80 ksample/s is a different boundary. Carry anti-alias response, clock noise and actual converter spectra into 05.6.

10 / 10

Build wanted-plus-blocker operating corners

Can you defend a gain/filter decision for every threat without hiding an invalid assumption?

Load Quiet. Reproduce 05.3 gain, NF and sensitivity with all blockers explicitly absent.
Load Strong. Find R1-LNA-in before looking at the converter. Compare 20 dB rejection pre-LNA versus post-mixer.
Load Two-tone. Identify 2f1−f2 and its three origins; compare −55, −45 and −43 dBm each without confusing reserve and clipping.
Load Reciprocal anchor. Verify mixer→R2 referral and the one-side integral; switch the LO mask to unknown.
Load Timed. Use Next event through 50, 70, 75, 95, 100, 150, 1150 and 1155 µs. Name the earlier samples and upstream margins that AGC cannot repair.
Record the stage, plane, time, invalidated claim, filter/linearity requirement and next waveform-specific test for each corner.

Interactive · Class 1 · p05-m04-blocker-agc-v1

Wanted/Blocker Headroom Map

Predict the first failure. Load a corner, compare filter placement and gain, then step through the burst. Apply commits a complete valid input; unsupported physics remains inspect.

Corner preset · restores all inputs and conditions

Wanted and gain mode

Wanted at R2 · dBm-130…-20 dBm; step 0.5. Reset restores the quiet fixture.

Gain control modeReset restores the quiet fixture. Changes need Apply corner.

Manual driver stateReset restores the quiet fixture. Changes need Apply corner.

Driver edits change the high-state gain; medium/low remain exactly 20/40 dB below it. Automatic mode always starts high.

Blocker catalogue · three stable source slots

B1 · independent source

Source stateReset restores the quiet fixture. Changes need Apply corner.

Coupling / entry planeReset restores the quiet fixture. Changes need Apply corner.

Waveform evidenceReset restores the quiet fixture. Changes need Apply corner.

On-time average level at entry · dBm-130…0 dBm; step 0.5. Reset restores the quiet fixture.

Offset from wanted at entry · MHz-200…200 MHz; step 0.001. Reset restores the quiet fixture.

Declared occupied-width proxy · MHz0…40 MHz; step 0.001. Reset restores the quiet fixture.

On boundary (inclusive) · µs0…2000 µs; step 1. Reset restores the quiet fixture.

Off boundary (exclusive) · µs0…2000 µs; step 1. Reset restores the quiet fixture.

Offset is relative to 2450 MHz at RF entries or 100 MHz at A1-IF. On-time level and [start, stop) define duty within the 2000 µs horizon. Entry is before any named extra filter at that stage. Frequencies outside ±20 MHz retain geometry only.

B2 · independent source