Path 04 · Module 03

Resonators &
RF Filters

Reject the threat. Preserve the burst. A filter selection must survive both questions—with loss, delay, terminations and evidence attached.

01 / 10

Failure: rejection passes, the burst fails

The 2.750 GHz marker looks excellent. Why does the receiver still lose the wanted burst?

Continue the illustrative engineering case: a 2.450 GHz condition-monitoring gateway must pass a pulse-shaped wanted signal while attenuating a nearby blocker. The routing network from 04.2 has established the ports. Now the review sees 49.411712 dB of ideal rejection and declares victory too early.

Think about itIf the blocker exceeds its rejection target by 9.411712 dB, is this filter acceptable?
Answer

No. The default third-order Chebyshev-I has a 0.500000 dB ripple and a separate Q = 200 passband debit of 1.596032 dB. Its worst passband loss is 2.096032 dB against 2.000000 dB allowed. The exact complex waveform proxy is 3.875175% against 3.0%, and passband delay ripple is 5.653218 ns against 5.0 ns.

Illustrative decision · p04-m03-filter-threat-v1Reject: worst passband loss and passband delay ripple and waveform distortion proxy

Chebyshev-I, n = 3, Aedge = 0.5 dB, Q = 200; 2.450 GHz / 100 MHz. Every failed axis is binding; no combined score hides it.

Numeric requirements · positive margin passes · full precision decides
AxisResultRequirementMarginDecision
Worst passband loss2.096032 dB≤ 2.0 dB-0.096032 dBreject
Blocker attenuation49.411712 dB≥ 40.0 dB9.411712 dBpass
Passband delay ripple5.653218 ns≤ 5.0 ns-0.653218 nsreject
Waveform distortion proxy3.875175 %≤ 3.0 %-0.875175 %reject

Required before hardware selection: physical realization, power, temperature, tolerance, size, measured response, hidden modes and mismatch-shaped waveform evidence. An accepted screen leaves these unresolved.

These are three distinct failures, not three names for attenuation. Magnitude tells how much each component is scaled; relative phase tells how those components recombine. A finite-Q screening rule adds an estimate of dissipative loss without pretending to predict the complete lossy waveform.

Common misconceptionA time-domain burst can be judged from magnitude response alone.

Two responses can share magnitude while differing in phase. The output waveform follows the complex transfer function. Inspect the full passband, the pulse-shaped record and the conditions at both R1 planes.

By the end, you will write a filter requirement, distinguish the Q definitions, compare consistently normalized responses, reject the binding failure and create a traceable selection row.

Bring Path 03’s two-port waves and S-parameters and matching under real terminations. Here those foundations become filter selection conditions.

02 / 10

Turn spectra into passband and stopband constraints

Start with the spectrum that must survive, then the population that must not. Give each boundary a definition and a plane before choosing a response family. A single blocker is a useful check point; a real threat specification is a set of frequencies, powers, modes and uncertainty margins.

Illustrative gateway requirement ledger · write this before selecting a family
QuantityRequired value / definitionPlane and condition
Passband2.400000…2.500000 GHz; B = 100.000 MHz; displayed fc = 2.450 GHzR1 component input/output, real Zref = 50 Ω
Wanted support60.000 Msymbol/s QPSK, RRC α = 0.35; ideal RC null-to-null width 81.000 MHzHigh-rate Path 04 variant; normalized complex envelope
LossWorst passband GT loss + local Q debit ≤ 2.000 dBMatched teaching default; do not substitute a typical center value
Blocker2.750000 GHz; +300.000 MHz offset; Pavs = −20.000 dBm; attenuation ≥ 40.000 dBSource-available at input R1; load-delivered output uses GT
Delay / waveformPassband delay ripple ≤ 5.000 ns; periodic distortion proxy ≤ 3.0%Intrinsic matched H; correction contract in Section 7
Analytic environment25 °C, small signal, source/load VSWR 1.000Power and environment qualification still missing
Hardware obligationsPower, size, temperature corners, tolerance population and reference-plane evidenceRequirements to obtain from product owner; no invented compliant limits
BNN=(1+α)Rs=1.35×60MHz=81MHzB_{\mathrm{NN}} = (1 + \alpha)R_{\mathrm{s}} = 1.35 \times 60 \mathrm{MHz} = 81 \mathrm{MHz}
Derived ideal RC support; α is dimensionless roll-off, Rs is symbol/s. This is not a measured occupied bandwidth or a regulatory mask.

The ideal wanted support leaves 9.5 MHz between each ±40.5 MHz null and the ±50 MHz passband edge. Finite truncation has spectral tails. Frequency error, manufacturing spread and environmental movement consume margin; they are not automatically covered by the 9.5 MHz arithmetic.

This high-rate variant does not replace Path 02’s 10 ksymbol/s baseline. The lesson uses no named air-interface mask. Compliance and complete receiver blocker/noise allocation belong to later paths.

Think about itDoes an 81 MHz null-to-null signal justify calling a 100 MHz filter its occupied bandwidth?
Answer

No. One number describes ideal waveform support, another names this passband span. Occupied bandwidth requires an integrated power fraction, spectral population and stated integration procedure.

Mark transition bands between the guaranteed passband and guaranteed rejection regions. A marker at +300 MHz leaves every unsampled threat—including one closer to the passband—unresolved.

03 / 10

Resonators, stored energy, loss, and coupling

A resonator exchanges electric and magnetic energy. Loss drains that exchange; coupling allows the source to feed it and the load to receive it. Recall resonance and ring-down, then separate intrinsic loss from deliberate external loading. [1]

Q=ω0WmaxPlossQ = \frac{\omega _{0} W_{\mathrm{max}}}{P_{\mathrm{loss}}}
Definition for a weakly damped resonant mode: Wmax [J] is total stored energy at a cycle phase where one reactive energy reaches its maximum, equal at ideal resonance to the sum of cycle-average electric and magnetic energies. Do not sum both separate maxima. Ploss [W] is cycle-average loss; ω0=2πf0 [rad/s].
Definition · one resonant mode with two independent external coupling channels
QEnergy leaves throughDecision consequence
Q0 · unloadedInternal conductor, dielectric and other intrinsic lossesSets dissipation before intended port loading
Qe1, Qe2 · externalEnergy extraction through each named coupling portPort coupling can broaden a response even with a high-Q resonator
QL · loadedAll internal and external loss channels togetherSets this mode’s decay and approximate half-power linewidth
1QL=1Q0+1Qe1+1Qe2\frac{1}{Q_{L}} = \frac{1}{Q_{0}} + \frac{1}{Q_{\mathrm{e1}}} + \frac{1}{Q_{\mathrm{e2}}}
Derived when independent loss powers add for the same stored-energy mode. It is not a universal multi-mode/coupling-matrix synthesis law.

With Q0 = 200 and Qe1 = Qe2 = 100, 1/QL = 0.025, so QL = 40. Near 2.450 GHz the simple isolated-mode half-power bandwidth is approximately f0/QL = 61.25 MHz. The amplitude ring-down time is 2QL/ω0 ≈ 5.197 ns; stored energy decays with half that time constant.

Think about itIf intrinsic loss vanished while both coupling ports remained at Qe = 100, would loaded Q become infinite?
Answer

No. With Q0 → ∞, QL approaches 50 because energy still leaves through the ports. External loading is not the same as heating inside the resonator.

Common misconceptionFilter order alone determines realizability.

Order counts poles in a defined prototype. Realization also needs adequate resonator Q, coupling range, parasitic control, operating power, size and stable geometry. The same order may be easy at one bandwidth and infeasible at another.

Go deeperWhy a narrow bandwidth spends more of a finite-Q budget

The same fractional dissipation matters more when a mode must store energy for many cycles before delivering it. Our later debit scales as n/(Q·FBW), capturing that screening trend. It has no coupling coefficients and therefore cannot replace a physical loss calculation.

04 / 10

Response transformations and filter-order intuition

A normalized low-pass prototype removes units so we can examine response shape once. A frequency transformation then asks where that shape should act. Keep normalized s separate from physical s in rad/s. [2] [3]

Definition / Derived · standard prototype substitutions for orientation only
Desired responseSubstitution into normalized HLPMeaning
Low-passsLP = s/ωcKeep DC; scale the specified edge
High-passsLP = ωc/sExchange low- and high-frequency roles
Band-passsLP = (s² + ω0²)/(Bω s)Bω=ω2−ω1; ω0=√(ω1ω2); each prototype pole produces two physical poles
Band-stopsLP = Bω s/(s² + ω0²)Map prototype rejection toward the center notch

The allocator uses a deliberately simpler local mapping. Near resonance, (f²−f0²)/(Bf) ≈ 2(f−f0)/B. We use the displayed arithmetic center fc and impose B/fc ≤ 0.20. Even inside this guard, it remains a teaching approximation, not an error guarantee for a physical band-pass design.

Ω=2(ffc)BFBW=Bfc=1002450=0.04081632653\begin{aligned}\Omega &= \frac{2(f - f_{c})}{B} \\ \mathrm{FBW} &= \frac{B}{f_{c}} = \frac{100}{2450} = 0.04081632653\end{aligned}
Derived local teaching variable: f and fc in Hz; B is the full edge-to-edge passband span in Hz. Nominal edges map to Ω=±1; the actual signed blocker maps to Ω=6.

A proper physical band-pass transform of 2.400 and 2.500 GHz would use geometric center √(2.4×2.5) ≈ 2.449490 GHz. The difference is small here but real. The interaction does not silently switch between those conventions.

HΩnattenuation slope20ndBdecade|H| \propto |\Omega|^{-n} \to \text{attenuation slope} \approx \frac{20n \mathrm{dB}}{\mathrm{decade}}
Derived ideal all-pole asymptote for normalized low-pass order n: at sufficiently large |Ω|. Finite zeros, modes and feedthrough alter real responses.
Common misconceptionSteeper rejection is always better.

More order can buy rejection while spending delay flatness, loss, tolerance and size. A steeper curve that fails the wanted signal is not an improvement in this decision.

05 / 10

Butterworth, Chebyshev, elliptic, and Bessel tradeoffs

Native prototypes normalize different properties. Before comparing them, state what “one unit of bandwidth” means. The static cases below all use n = 3, Q = 200, B = 100 MHz and an equal-spec outer-edge attenuation of 0.5 dB with a unity passband peak.

Definition · native prototype versus common edge; no equal-native-bandwidth claim
FamilyNative |Ω|=1 attenuation, n=3Equal-spec ruleWhat is optimized
Butterworth3.010300 dBHcompare(jΩ)=Hnative(jαΩ); α=(10^(Aedge/10)−1)^(1/(2n))Maximally flat magnitude near DC
Chebyshev-I0.500000 dB for Rp=0.5Rp=Aedge; α=1; unity ripple peak; even n DC remains −RpEquiripple magnitude trades phase/transient behavior for transition selectivity
Delay-normalized Bessel0.902973 dB (approximately)Positive α found by bracketed bisection so the outer edge is AedgeMaximally flat DC delay; gradual transition under the same edge spec
Hn(jΩ)2=11+ϵ2Tn2(Ω)ϵ=10Rp101\begin{aligned}|H_{\mathrm{n}}(j\Omega)|^{2} &= \frac{1}{1 + \epsilon ^{2}T_{\mathrm{n}}^{2}(\Omega)} \\ \epsilon &= \sqrt{10^{\frac{R_{\mathrm{p}}}{10}} - 1}\end{aligned}
Definition: Chebyshev-I ripple Rp [dB], ε dimensionless, Tn is the nth Chebyshev polynomial; native Ω has ripple edge ±1.

The complex response is evaluated from stable left-half-plane poles, not reconstructed from this power equation. For third order, the poles are −0.313228243170137 ± j1.021927491047360 and −0.626456486340275, with numerator K = 0.715693790310798.

Go deeperInspect every pole and the reverse Bessel polynomial
β=asinh(1ϵ)nθk=(2k1)π2npk=sinh(β)sin(θk)+jcosh(β)cos(θk)H(s)=K(spk)\begin{aligned}\beta &= \frac{\operatorname{asinh}(\frac{1}{\epsilon })}{n} \\ \theta_k &= \frac{(2k-1)\pi }{2n} \\ p_{\mathrm{k}} &= -\sinh (\beta)\sin (\theta_k) + j \cosh (\beta)\cos (\theta_k) \\ H(s) &= \frac{K}{\prod (s-p_{\mathrm{k}})}\end{aligned}
Definition: k=1…n; positive normalized frequency uses s=+jΩ. K=∏(−pk) for odd n; divide K by √(1+ε²) for even n. Only roundoff imaginary residue is discarded.

Butterworth uses pk = −sin(θk) + jcos(θk) and K = ∏(−pk). All real parts are negative. For Bessel, evaluate the polynomial directly:

θn(s)=k=0n(2nk)!2nkk!(nk)!skHn(s)=θn(0)θn(s)\begin{aligned}\theta_n(s)&=\sum_{k=0}^{n}\frac{(2n-k)!}{2^{n-k}k!(n-k)!}s^k\\H_n(s)&=\frac{\theta_n(0)}{\theta_n(s)}\end{aligned}
Definition: delay-normalized reverse Bessel, order 1…9. Coefficients are generated from factorials; no copied coefficient table. DC gain and normalized DC group delay are one.

For n = 3: θ3(s) = 15 + 15s + 6s² + s³. The equal-spec scale is α = 0.750824662261… for Aedge = 0.5 dB. Bisection starts at [0,1], doubles the upper bound up to 2²⁰, then resolves α to max(10⁻¹²,10⁻¹²α). Failure to bracket is invalid, not a guessed curve.

Three families at the same outer-edge attenuationButterworth solid, Chebyshev-I dashed and Bessel dotted. The shaded interval is the common passband. This intrinsic magnitude comparison has no finite-Q debit; the following table also compares delay and waveform distortion.Intrinsic attenuation (dB, downward)04080-8.000.008.00Blocker Ω = 6Normalized signed offset Ω; fc = 2.450 GHz, B = 100 MHz
Derived · Butterworth solid; Chebyshev-I dashed; Bessel dotted. All n = 3, Aedge = 0.5 dB, unity passband peak, R1 / 50 Ω matched. Each curve uses the disclosed 4050-row grid. Curves stop visually at 80 dB; exact blocker values and native scales follow. The Q = 200 debit appears only in the table’s passband-loss column.
Derived / Simulated · fixed family candidates · each row declares Aedge, n and Q
Family / n / QAedge / αfamilyNative edge (dB)Pass loss (dB)Blocker (dB)Delay ripple (ns)Waveform (%)Screen
Butterworth / 3 / 2000.5 dB / 0.7042674013453.0103002.09603237.5540931.4854204.965292Reject: worst passband loss and blocker attenuation and waveform distortion proxy
Chebyshev-I / 3 / 2000.5 dB / 1.0000000000000.5000002.09603249.4117125.6532183.875175Reject: worst passband loss and passband delay ripple and waveform distortion proxy
Bessel / 3 / 2000.5 dB / 0.7508246622610.9029732.09603217.2584520.0016913.560812Reject: worst passband loss and blocker attenuation and waveform distortion proxy

The sample-domain proxy is not a universal family ranking. Integer-only timing correction leaves a residual fractional-sample delay, and the actual waveform spectrum weights the result. Bessel’s flatter delay alone does not promise the smallest proxy for every order and normalization.

Elliptic orientation · Informative, not computed

Finite-frequency transmission zeros can shorten a transition at the cost of passband/stopband ripple and demanding phase, tolerance and implementation behavior. A deep zero is followed by a response rise; it does not establish uniformly infinite rejection. No elliptic curve or solver is included. [2]

Think about itAt the same 0.5 dB outer edge and third order, what does Bessel give up to improve delay flatness?
Answer

Here it gives up blocker attenuation: 17.258452 dB versus Chebyshev’s 49.411712 dB at Ω = 6. Both retain the same declared outer-edge attenuation, so the comparison has a common requirement.

06 / 10

Insertion loss, return loss, rejection, and finite-Q floor

Separate what is reflected from what is dissipated. A lossless mismatched network can reject a signal while creating no internal heat. “Insertion loss” also needs a before/after reference; it is not automatically equal to every negative S21 number.

Definition · loss, match and bandwidth vocabulary
TermDefinition and condition
Transducer gain GT / lossPload/Pavs; loss = −10log10(GT). Source-available input and delivered load power, same declared conditions.
Insertion loss10log10(Pload,before/Pload,after), with source, load and planes held fixed. Equals −20log10|S21| for a matched, unity-through reference.
Return loss RL−20log10|Γ| at a named plane. S11-based RL assumes port 2 matched. Γ=0 is a perfect match, not an infinite output power.
RejectionAttenuation at a specified frequency/region and reference. Here absolute −10log10(GTideal), not attenuation relative to the ripple edge.
Shape factorBandwidth at one attenuation divided by bandwidth at another, e.g. B60dB/B3dB. Both reference levels and crossings must be explicit.
3 dB bandwidthWidth between specified crossings 3 dB below a declared passband reference; not automatically this lesson’s Aedge span.
Occupied / channel / measurement bandwidthAn integrated-power fraction / assigned channel interval / instrument resolution or integration setting. Three distinct definitions.
Noise-equivalent bandwidthFor a two-sided envelope, ∫|H(ν)|² dν / |H|max², over stated bounds. Include or exclude discrete spurs explicitly; no ENBW value is inferred from one edge.
Common misconceptionReturn loss and insertion loss measure the same failure.

Return loss describes a reflected wave relative to the incident wave at a plane. Insertion loss describes changed delivered power after inserting a network. Dissipation, reflection and coupling to other ports must be separated to explain why delivery changed.

T3(6)=846ϵ=0.3493114002Ablocker=10log10(1+ϵ2×8462)=49.411712dBMargin=49.41171240=+9.411712dB\begin{aligned}T_{3}(6) &= 846\qquad \epsilon = 0.3493114002 \\ A_{\mathrm{blocker}} &= 10\log_{10}(1 + \epsilon ^{2} \times 846^{2}) = 49.411712 \mathrm{dB} \\ \text{Margin} &= 49.411712 - 40 = +9.411712 \mathrm{dB}\end{aligned}
Derived canonical Chebyshev-I, n=3, Rp=0.5 dB. T3(Ω)=4Ω³−3Ω. The ideal blocker and margin are independent of the passband-only finite-Q debit.

With −20.000 dBm source-available blocker power and matched ports, its ideal residual is −69.411712 dBm. Do not add passband loss to claim a still better stopband residual.

ILQ,dB=10ln10nQFBW=10ln103200×0.04081632653=1.596032dB\begin{aligned}\mathrm{IL}_{Q,\mathrm{dB}}&=\frac{10}{\ln10}\frac{n}{Q\,\mathrm{FBW}}\\&=\frac{10}{\ln10}\frac{3}{200\times0.04081632653}\\&=1.596032\,\mathrm{dB}\end{aligned}
Illustrative first-order loss-screening rule, model p04-m03-filter-prototype-v1. n is prototype order; Q is a requested unloaded-Q proxy; FBW=B/fc. It is a scalar passband debit, not a full physical filter-loss equation.
Lpass,total=10log10(GT,ideal)+ILQWorst=0.500000+1.596032=2.096032dBLoss margin=2.0000002.096032=0.096032dB\begin{aligned}L_{\mathrm{pass,total}}&=-10\log_{10}(G_{\mathrm{T,ideal}})+\mathrm{IL}_Q\\\text{Worst}&=0.500000+1.596032=2.096032\,\mathrm{dB}\\\text{Loss margin}&=2.000000-2.096032=-0.096032\,\mathrm{dB}\end{aligned}
Derived matched default. Worst loss uses all 2001 passband rows, including exact edges. Full precision determines rejection.

For loss alone at the fixed default response and matched terminations, Q must exceed 212.804296…; integer Q = 213 clears that screen. It leaves the delay and waveform failures intact. The “finite-Q floor” here means a passband loss burden. It does not define a universal out-of-band rejection floor.

NF=ILQ=1.596032dBF=10NF/10=1.4441198Te=(F1)T0=128.7947K\begin{aligned}\mathrm{NF}&=\mathrm{IL}_Q=1.596032\,\mathrm{dB}\\F&=10^{\mathrm{NF}/10}=1.4441198\\T_e&=(F-1)T_0=128.7947\,\mathrm K\end{aligned}
Derived separate matched attenuator-equivalent interpretation of the dissipative debit only. Tp=T0=290 K, F dimensionless, Te in K. The main analytic fixture’s 25 °C must not be substituted for 290 K.

At other physical temperatures the matched passive attenuator relation is F = 1 + (L−1)Tp/T0, where L is linear dissipative loss. In the ideal two-port below, ripple is reflected, so it is not added as thermal noise. This is not a complete filter noise model; Path 01 Noise owns the noise definitions.

Common misconceptionA lossless ideal response predicts real resonator rejection indefinitely.

Parasitic input-to-output coupling, finite loss and higher modes create paths absent from the prototype. A calculated −100 dB tail is not evidence that a package, PCB or measurement fixture achieves it.

07 / 10

Phase and group delay can damage the waveform

A constant delay translates a waveform. Delay that changes across occupied frequencies rearranges the relative timing of its components. Recovering a burst therefore needs more than removing one common phase or cable delay. [4]

τg=dϕdω\tau _{g} = -\frac{d\phi }{d\omega }
Definition under e^(+jωt). φ is unwrapped radians and ω=2πf in rad/s, so τg is seconds. For H=e^(−jωτ), the result is +τ.

Unwrap from the lowest to highest frequency, repeatedly adding/subtracting 2π so each increment lies in (−π,+π]; an exact −π tie becomes +π. The passband grid is Ωi = −1 + i/1000 for i = 0…2000. Here Δf = 50 kHz and Δω = 2π×50 kHz.

τg,iϕi+1ϕi12ΔωΔτg=maxpassτgminpassτg\begin{aligned}\tau_{g,i}&\approx-\frac{\phi_{i+1}-\phi_{i-1}}{2\Delta\omega}\\\Delta\tau_g&=\max_{\mathrm{pass}}\tau_g-\min_{\mathrm{pass}}\tau_g\end{aligned}
Derived finite differences: central in the interior; forward at the first row, backward at the last. The endpoint stencils are valid, lower-order rows. No interpolation or smoothing.

Phase/delay is suppressed where |H| < 10⁻¹². Polynomial conditioning κeval = Σ|ak||s|ᵏ / |Σak sᵏ| is inspected above 10¹⁰ and invalid above 10¹⁴. A plotted stopband floor is not evidence of physical rejection. The canonical defined delay range is 6.128549…11.781767 ns, giving 5.653218 ns ripple.

Common misconceptionGroup delay is merely cable delay or a removable constant shift.

A constant component can be aligned away; frequency-dependent delay remains. Magnitude variation and residual timing error also contribute to waveform distortion, so delay ripple alone cannot predict the exact proxy.

A fixed record, a disclosed correction

Use 256 Gray QPSK symbols from PRBS-9 x⁹+x⁵+1, restarted at 0x1FF. Emit bit 0; feed bit 0 XOR bit 4 into bit 8 after a right shift. Consume b0 as the first/MSB bit of each pair: 00, 01, 11, 10 map to (1+j), (−1+j), (−1−j), (1−j), divided by √2.

Place symbol m at index 8m of 2048 samples. Linearly convolve with a unit-energy 65-tap symmetric RRC, α=0.35 and span 8 symbols. Of the 2112 convolution samples retain indices 32…2079 inclusive, then normalize mean |x|² to one. This valid record is treated as one periodic steady-state record, not a simulated burst-start transient.

Go deeperRRC taps, signed DFT bins and alignment algorithm
h(t)=[sin(πt(1α))+4αtcos(πt(1+α))][πt(1(4αt)2)]h(t) = \frac{[\sin (\pi t(1-\alpha)) + 4\alpha t \cos (\pi t(1+\alpha))]}{[\pi t(1-(4\alpha t)^{2})]}
Definition: t=k/8 symbol periods, k=−32…32, α=0.35. Normalize taps so Σh²=1.

Use h(0)=1+α(4/π−1). At t=±1/(4α), use (α/√2)[(1+2/π)sin(π/(4α))+(1−2/π)cos(π/(4α))]. The analytic limits prevent a numerical 0/0.

X[k]=nx[n]ej2πkn/NY[k]=Hcompare(jΩk)X[k]y[n]=1NkY[k]ej2πkn/N\begin{aligned}X[k]&=\sum_n x[n]e^{-j2\pi kn/N}\\Y[k]&=H_{\mathrm{compare}}(j\Omega_k)X[k]\\y[n]&=\frac1N\sum_kY[k]e^{j2\pi kn/N}\end{aligned}
Definition: N=2048; Fs=480 Msample/s; q=k for k<1024, otherwise q=k−2048; νk=qFs/N; Ωk=2νk/B. Forward DFT has negative exponent, inverse positive exponent and 1/N.

Test every circular integer lag ℓ=−1024…1023, using yℓ[n]=y[(n+ℓ) mod 2048]. Maximize |Σyℓ conj(x)|; ties take the smallest |ℓ|, then negative lag. Use all 2048 pairs with no additional transient exclusion, fractional delay, timing recovery, adaptive equalizer or DPD.

c=ny[n]conj(x[n])nx[n]2Proxy=100ny[n]cx[n]2ncx[n]2%\begin{aligned}c&=\frac{\sum_n y_\ell[n]\operatorname{conj}(x[n])}{\sum_n|x[n]|^2}\\\text{Proxy}&=100\sqrt{\frac{\sum_n|y_\ell[n]-cx[n]|^2}{\sum_n|cx[n]|^2}}\,\%\end{aligned}
Definition of the local sample-domain distortion proxy. One fitted complex gain removes common amplitude/phase after the integer circular lag. This is not standards-defined symbol EVM. If Σ|cx|²≤10⁻¹²Σ|x|², report an unavailable denominator.
Simulated pinned result · p04-qpsk-rrc-periodic-v1 · matched Chebyshev-I n=3 / Rp=0.5
QuantityDefault result
Best circular lag+3 samples = +6.250000 ns
Fitted gain c0.968095425927111 − j0.001498446673922
Distortion proxy3.875175% > 3.0% local criterion
Waveform frequency domain−240…+240 MHz, upper endpoint excluded; every fc+ν must be positive
Separate blocker+300 MHz is outside this waveform Nyquist interval; analytic check only
Common misconceptionA synthetic EVM proxy proves standards compliance.

Reference normalization, symbol selection, timing correction and equalization determine EVM interpretation. This metric uses waveform samples and one fixed correction. No named standard or compliance limit is implemented. [5]

The allocator now combines these definitions. Its entire default result is also available without JavaScript and in print.

Class 1 · one decision model

Filter Threat Allocator

Choose a response that meets the wanted-signal and blocker requirements together. You have now seen the loss and complex-response definitions; use them to explain each decision.

  1. Predict whether third-order Chebyshev passes 40 dB at +300 MHz; compare the result with its Q = 200 loss.
  2. Change only order, then only Q. Name the constraints that moved.
  3. Compare Bessel and Chebyshev at equal order and Aedge. Check delay and the periodic burst.
  4. Introduce independent source/load phase. Choose the next technology and evidence request.
01 · Wanted signal and threat at R1
0.110 GHz; step 0.001; default 2.45.
11000 MHz; step 1; default 100. Require span/center ≤ 0.20.
06 dB; step 0.1; default 2.
-55 GHz; step 0.001; default 0.3. Keep its sign; blocker must be outside the band at positive RF frequency.
10100 dB; step 0.5; default 40.
-12030 dBm; step 0.5; default -20.
0100 ns; step 0.1; default 5.
020 %; step 0.1; default 3.
02 · Equal-spec prototype and loss region
Three closed choices; default Chebyshev-I. Elliptic behavior is qualitative only.
19 integer; step 1; default 3.
0.11 dB; step 0.1; default 0.5. Equal-spec edge for every family; also ripple for Chebyshev-I.
205000 integer; step 1; default 200.
03 · Source and load conditions
13 :1; step 0.01; default 1.
13 :1; step 0.01; default 1.
-180180 °; step 1; default 0. Not applicable at VSWR 1; stored phase retained.
-180180 °; step 1; default 0. Not applicable at VSWR 1; stored phase retained.

ρ = (VSWR−1)/(VSWR+1), Γ = ρ exp(jφ). Independent phases change GT of the lossless algebraic two-port. They do not reshape the intrinsic waveform in this model.

04 · Fixed waveform screen
Default canonical. 480 Msample/s, 2048 periodic samples, 81 MHz ideal RC null-to-null support. Waveform RF domain requires fc > 240 MHz. The +300 MHz analytic blocker lies beyond waveform Nyquist.
Results describe the applied inputs.

Keyboard-selectable markers; no dragging required. Outside-passband points have no derivative row.
Illustrative decision · p04-m03-filter-threat-v1Reject: worst passband loss and passband delay ripple and waveform distortion proxy

Chebyshev-I, n = 3, Aedge = 0.5 dB, Q = 200; 2.450 GHz / 100 MHz. Every failed axis is binding; no combined score hides it.

Numeric requirements · positive margin passes · full precision decides
AxisResultRequirementMarginDecision
Worst passband loss2.096032 dB≤ 2.0 dB-0.096032 dBreject
Blocker attenuation49.411712 dB≥ 40.0 dB9.411712 dBpass
Passband delay ripple5.653218 ns≤ 5.0 ns-0.653218 nsreject
Waveform distortion proxy3.875175 %≤ 3.0 %-0.875175 %reject

Required before hardware selection: physical realization, power, temperature, tolerance, size, measured response, hidden modes and mismatch-shaped waveform evidence. An accepted screen leaves these unresolved.

Ideal complex-response magnitude and terminated transducer attenuationSolid intrinsic matched H and dashed current-termination GT. Shading marks the passband; dotted lines mark edges, the blocker and its target. Finite-Q debit is not applied to these curves. Exact values follow in the key table.Attenuation (dB, downward)03060-8.000.008.00Blocker target ≥ 40.0 dBNormalized signed offset Ω = 2(f − fc)/B
Derived · solid: intrinsic H; dashed: current ΓS/ΓL; shaded: passband. Frequency = 2.450 GHz + Ω × 50.000 MHz. 4050 sorted samples; exact blocker 2.750000 GHz. Magnitude floor −240 dB; deep-null phase is suppressed. No finite-Q stopband curve.
Unwrapped intrinsic passband phasePhase from the stable left-half-plane prototype. Positive normalized frequency uses s=+jΩ. Current selected passband point is marked. The blocker is outside this derivative region.Unwrapped phase (°)-1800180-1.000.001.00Passband Ω; edges −1 and +1
Derived · 2001 inclusive passband rows. Phase is unwrapped low to high, with increments in (−π,+π]. Positive slope is not silently converted to positive delay.
Intrinsic passband group delayMinus the derivative of unwrapped phase in radians per radian per second. All passband samples and both valid one-sided endpoints are used. Dotted bound equals minimum delay plus the allowed ripple, not a maximum absolute cable delay.Group delay (ns)0.006.3612.72-1.000.001.00Passband Ω; physical spacing B/2000
Derived · range 6.12854911.781767 ns. Ripple 5.653218 ns; limit 5.0 ns. Central differences, first-order endpoints, Δf = 0.050000 MHz; no smoothing or interpolation. Dotted bound, when in view: minimum + allowance.
Derived key frequencies · equal-spec H · R1 / 50 Ω · Q debit only inside passband
Exact rowRF (GHz) / ΩH real / imag|H| (dB)Phase (°)Delay (ns)GT loss (dB)Pass loss + Q (dB)Algebraic RL (dB)
Lower edge2.400000 / -1.000-0.668997142 / 0.666103417-0.500000135.124184 unwrapped11.781767 · forward endpoint0.5000002.0960329.635745
Center2.450000 / 0.0001.000000000 / 0.0000000000.0000000.000000 unwrapped6.826553 · central0.0000001.596032perfect algebraic match
Upper edge2.500000 / 1.000-0.668997142 / -0.666103417-0.500000-135.124184 unwrapped11.781767 · backward endpoint0.5000002.0960329.635745
Actual signed blocker2.750000 / 6.000-0.000710197 / 0.003308509-49.411712102.115126 wrappedoutside passband derivative grid49.411712not defined here0.000050
Worst passband loss2.475000 / 0.5000.501188918 / -0.800037878-0.500000-57.934647 unwrapped6.207742 · central0.5000002.0960329.635745
Derived / Illustrative loss bookkeeping · separate quantities and conditions
QuantityValueCondition
Ideal matched blocker49.411712 dBIntrinsic prototype |H|²; no Q debit
Current-termination blocker49.411712 dBGTideal of the algebraic two-port at current ΓS/ΓL
Load-delivered blocker-69.411712 dBm-20.0 dBm source-available at R1; no Q debit
Finite-Q passband debit1.596032 dBLocal screening rule; does not define a complex lossy filter
Dissipative NF1.596032 dBSeparate matched attenuator-equivalent debit, Tp=T0=290 K
Linear F / Te1.4441198 / 128.7947 Klog10F=0.159603; ideal ripple reflects, not heat
Minimum Q for loss alone212.804; choose integer ≥ 213Fixed family/order/edge and terminations; other requirements still apply

What the burst sees

3.875175% sample-domain distortion proxy · limit 3.0%. Intrinsic matched H only; no Q debit and no mismatch shaping.

I waveform comparison after alignment and complex-gain removalFirst 128 of 2048 periodic samples: solid input, dashed aligned output divided by fitted complex gain. This view has normalized amplitude; it does not portray actual loss or mismatch-shaped modulation.I (normalized amplitude)-1.60.01.60.00132.29264.58Time in periodic record (ns), first 128 samplesQ waveform comparison after alignment and complex-gain removalFirst 128 of 2048 periodic samples: solid input, dashed aligned output divided by fitted complex gain. This view has normalized amplitude; it does not portray actual loss or mismatch-shaped modulation.Q (normalized amplitude)-1.60.01.60.00132.29264.58Time in periodic record (ns), first 128 samples
Simulated · solid input, dashed aligned output / c. First 128 samples shown; all 2048 pairs determine the metric. Lag +3 samples; c = 0.968095425927 j0.001498446674. Plot normalization removes common gain and phase; it must not be read as delivered power.
Simulated waveform table · first four of all 2048 aligned pairs · normalized I/Q
Sample / nsInput I / QAligned output I / QCorrected output I / Q
0 / 0.000000-0.736386 / -0.739278-0.423050 / -0.380180-0.436384 / -0.393384
1 / 2.083333-0.772998 / -0.785499-0.605165 / -0.621916-0.624113 / -0.643378
2 / 4.166667-0.788233 / -0.807086-0.762037 / -0.786601-0.785892 / -0.813740
3 / 6.250000-0.787204 / -0.806759-0.851284 / -0.855638-0.877969 / -0.885195
Technology region to investigate

Explore LC or LTCC only with measured use-frequency Q and loss. Compare ceramic/acoustic and cavity options against the same waveform.

Q is a requested screening parameter, not a measured technology capability. Increasing Q changes this loss debit and its noise consequence; it does not repair intrinsic delay or waveform error.

Inspect model, grids and waveform contract

κeval > 10¹⁰: inspect the affected required metric; > 10¹⁴ or denominator ≤ 10⁻¹²: invalid. A plot-only deep-null floor does not reject a candidate. Required metric unavailability is explicit. Valid endpoint derivatives do not trigger inspect.

Filter selection / evidence row

Copy the applied case with its result, model, planes and unresolved verification points. Print preserves the decision and readable tables.

Read or manually copy the complete record
{
  "axes": [
    {
      "limit": "2.000000",
      "margin": "-0.096032",
      "name": "Worst passband loss",
      "outcome": "reject",
      "unit": "dB",
      "value": "2.096032"
    },
    {
      "limit": "40.000000",
      "margin": "9.411712",
      "name": "Blocker attenuation",
      "outcome": "pass",
      "unit": "dB",
      "value": "49.411712"
    },
    {
      "limit": "5.000000",
      "margin": "-0.653218",
      "name": "Passband delay ripple",
      "outcome": "reject",
      "unit": "ns",
      "value": "5.653218"
    },
    {
      "limit": "3.000000",
      "margin": "-0.875175",
      "name": "Waveform distortion proxy",
      "outcome": "reject",
      "unit": "%",
      "value": "3.875175"
    }
  ],
  "blockerResidualDbm": "-69.411712",
  "decision": "Reject: worst passband loss and passband delay ripple and waveform distortion proxy",
  "delayRangeNs": [
    "6.128549",
    "11.781767"
  ],
  "errors": {},
  "fixture": {
    "alignment": "all integer circular lags -1024…1023; max |Σ y[n+lag]conj(x[n])|; ties smallest |lag| then negative",
    "convolutionLength": "2112.000000",
    "correction": "c=Σy_lag conj(x)/Σ|x|²; all 2048 pairs; no fractional delay, receiver matched filter, timing recovery, equalizer or DPD",
    "denominatorRelativeFloor": "0.000000",
    "grid": "q=k if k<1024, else k-2048; ν=q*480e6/2048; [-240,+240) MHz",
    "id": "p04-qpsk-rrc-periodic-v1",
    "mapping": "00,01,11,10 → (1+j),(-1+j),(-1-j),(1-j), divided by sqrt(2)",
    "metric": "100 sqrt(Σ|y_lag-cx|²/Σ|cx|²); sample-domain distortion proxy, no standard limit",
    "normalization": "RRC tap energy 1; valid record mean |x|²=1",
    "nullToNullHz": "81000000.000000",
    "prbs": "x^9+x^5+1; emit bit0; feedback bit0 XOR bit4; shift right into bit8; b0 is MSB",
    "record": "periodic steady state; no additional transient exclusion",
    "retainedIndices": "32…2079 inclusive",
    "rollOff": "0.350000",
    "sampleRateHz": "480000000.000000",
    "samples": "2048.000000",
    "seed": "0x1FF",
    "spanSymbols": "8.000000",
    "sps": "8.000000",
    "symbolRateHz": "60000000.000000",
    "symbols": "256.000000",
    "taps": "65.000000",
    "transform": "X=DFT(x), Y=Hcompare(j2ν/B)X, y=IDFT(Y); circular, no padding"
  },
  "ilQDb": "1.596032",
  "inputs": {
    "blockerDbm": "-20.000000",
    "centerGHz": "2.450000",
    "delayNs": "5.000000",
    "distortionPercent": "3.000000",
    "edgeDb": "0.500000",
    "family": "Chebyshev-I",
    "loadPhase": "0.000000",
    "loadVswr": "1.000000",
    "lossDb": "2.000000",
    "offsetGHz": "0.300000",
    "order": "3.000000",
    "q": "200.000000",
    "rejectionDb": "40.000000",
    "sourcePhase": "0.000000",
    "sourceVswr": "1.000000",
    "spanMHz": "100.000000",
    "waveform": "canonical"
  },
  "keyRows": [
    {
      "name": "Lower edge",
      "row": {
        "actualAttenuationDb": "0.500000",
        "condition": "5.940502",
        "delayNs": "11.781767",
        "derivative": "forward endpoint",
        "floor": false,
        "gt": "0.891251",
        "h": {
          "im": "0.666103416985",
          "re": "-0.668997142008"
        },
        "hz": "2400000000.000000",
        "idealAttenuationDb": "0.500000",
        "magnitudeDb": "-0.500000",
        "omega": "-1.000000",
        "phaseDeg": "135.124184",
        "returnLossDb": "9.635745",
        "sigma": "1.000000",
        "totalPassLossDb": "2.096032"
      }
    },
    {
      "name": "Center",
      "row": {
        "actualAttenuationDb": "0.000000",
        "condition": "1.000000",
        "delayNs": "6.826553",
        "derivative": "central",
        "floor": false,
        "gt": "1.000000",
        "h": {
          "im": "0.000000000000",
          "re": "1.000000000000"
        },
        "hz": "2450000000.000000",
        "idealAttenuationDb": "0.000000",
        "magnitudeDb": "0.000000",
        "omega": "0.000000",
        "phaseDeg": "0.000000",
        "returnLossDb": null,
        "sigma": "1.000000",
        "totalPassLossDb": "1.596032"
      }
    },
    {
      "name": "Upper edge",
      "row": {
        "actualAttenuationDb": "0.500000",
        "condition": "5.940502",
        "delayNs": "11.781767",
        "derivative": "backward endpoint",
        "floor": false,
        "gt": "0.891251",
        "h": {
          "im": "-0.666103416985",
          "re": "-0.668997142008"
        },
        "hz": "2500000000.000000",
        "idealAttenuationDb": "0.500000",
        "magnitudeDb": "-0.500000",
        "omega": "1.000000",
        "phaseDeg": "-135.124184",
        "returnLossDb": "9.635745",
        "sigma": "1.000000",
        "totalPassLossDb": "2.096032"
      }
    },
    {
      "name": "Actual signed blocker",
      "row": {
        "actualAttenuationDb": "49.411712",
        "condition": "1.281458",
        "delayNs": null,
        "derivative": "outside passband derivative grid",
        "floor": false,
        "gt": "0.000011",
        "h": {
          "im": "0.003308509479",
          "re": "-0.000710196600"
        },
        "hz": "2750000000.000000",
        "idealAttenuationDb": "49.411712",
        "magnitudeDb": "-49.411712",
        "omega": "6.000000",
        "phaseDeg": "102.115126",
        "returnLossDb": "0.000050",
        "sigma": "1.000000",
        "totalPassLossDb": null
      }
    },
    {
      "name": "Worst passband loss",
      "row": {
        "actualAttenuationDb": "0.500000",
        "condition": "2.534450",
        "delayNs": "6.207742",
        "derivative": "central",
        "floor": false,
        "gt": "0.891251",
        "h": {
          "im": "-0.800037878171",
          "re": "0.501188918098"
        },
        "hz": "2475000000.000000",
        "idealAttenuationDb": "0.500000",
        "magnitudeDb": "-0.500000",
        "omega": "0.500000",
        "phaseDeg": "-57.934647",
        "returnLossDb": "9.635745",
        "sigma": "1.000000",
        "totalPassLossDb": "2.096032"
      }
    }
  ],
  "match": "Algebraic two-port only; intrinsic matched waveform",
  "minimumQForLoss": "212.804296",
  "model": {
    "bessel": "θn(s)=Σk (2n-k)!/[2^(n-k)k!(n-k)!] s^k; α bracket [0,1], double to 2^20, bisect abs/rel 1e-12",
    "conditions": "25 °C, small signal; separate dissipative attenuator-equivalent NF at Tp=T0=290 K",
    "differentiation": "unwrap low→high increments (-π,+π], -π tie→+π; central interior, one-sided first-order endpoints; Δω=2πB/2000; no interpolation/smoothing",
    "evidence": "Illustrative engineering case; Derived analytic prototype; Simulated periodic waveform",
    "id": "p04-m03-filter-threat-v1",
    "loss": "ILQ=(10/ln10)n/(Q*FBW), scalar passband screening debit only; Lpass=-10log10(GTideal)+ILQ",
    "mismatch": "lossless reciprocal algebraic S11=a, S21=S12=H, S22=-a H/conj(H); a=sqrt(1-|H|²). No hardware S11 prediction",
    "native": "Butterworth -3.010300 dB edge; Chebyshev-I -Rp edge; Bessel θn(0)/θn(s), DC delay=1",
    "noise": "NFdB=ILQ, log10F=NFdB/10; F/Te tagged beyond numeric range when log10F>300",
    "normalization": "Equal-spec unity passband peak; outer |Ω|=1 attenuation=Aedge. Chebyshev even n retains -Rp DC",
    "numerics": {
      "denominatorFloor": "0.000000",
      "floor": "0.000000",
      "inspectCondition": "10000000000.000000",
      "invalidCondition": "100000000000000.000000",
      "passivityTolerance": "0.000000"
    },
    "passbandGrid": "Ωi=-1+i/1000, i=0…2000; fi=fc+BΩi/2; exact center and edges",
    "phasor": "e^(+jωt), H(+jΩ); positive-real DC gain; positive envelope offset raises RF frequency",
    "planes": "R1 input/output; real Zref=50 Ω; no fixture, board, or access lines",
    "plotGrid": "union passband + 1025 log magnitudes 1…max(8,2|offset|/B), signed mirrors, exact blocker; deduplicate; omit nonpositive physical plot-only rows",
    "prototype": "p04-m03-filter-prototype-v1",
    "rounding": "p04-m03-decimal-v1: fixed 6 metric decimals, 12 complex/α decimals; normalize displayed -0; JSON LF, stable field order",
    "transform": "Ω=2(f-fc)/B; local narrowband mapping, B/fc<=0.20; not physical bandpass synthesis",
    "waveformPath": "intrinsic matched equal-spec H; no finite-Q debit or mismatch-shaped modulation"
  },
  "noise": {
    "factor": "1.444120",
    "linearStatus": "defined",
    "log10F": "0.159603",
    "nfDb": "1.596032",
    "teK": "128.794742"
  },
  "power": "not qualified",
  "prototype": {
    "alpha": "1.000000000000",
    "coefficients": [
      "0.715694",
      "1.534895",
      "1.252913",
      "1.000000"
    ],
    "edgeDb": "0.500000",
    "family": "Chebyshev-I",
    "normalization": "Equal-spec: unity passband peak; Aedge=0.5 dB at |Ω|=1; α=1; Chebyshev even-order DC remains -Rp",
    "numerator": "0.715694",
    "order": "3.000000",
    "poles": [
      {
        "im": "1.021927491047",
        "re": "-0.313228243170"
      },
      {
        "im": "0.000000000000",
        "re": "-0.626456486340"
      },
      {
        "im": "-1.021927491047",
        "re": "-0.313228243170"
      }
    ]
  },
  "size": "allocation required",
  "status": "reject",
  "technologyRegion": "Explore LC or LTCC only with measured use-frequency Q and loss. Compare ceramic/acoustic and cavity options against the same waveform.",
  "temperature": "25 °C analytic; corners unverified",
  "verificationPoints": [
    "Physical resonator/coupling realization and use-frequency Q/loss model",
    "Condition-matched measured complex response, S11/S22 and mismatch-shaped waveform",
    "Passband/stopband populations, tolerance and temperature corners",
    "Average/peak/burst power, acoustic or contact nonlinearity, heating and lifetime",
    "Package/board/fixture planes, size allocation, parasitic feedthrough and hidden modes"
  ],
  "warnings": [],
  "waveform": {
    "c": {
      "im": "-0.001498446674",
      "re": "0.968095425927"
    },
    "lag": "3.000000",
    "percent": "3.875175",
    "reason": "Simulated intrinsic matched H; periodic sample-domain proxy"
  }
}
08 / 10

Choose an RF filter technology

Choose a feasible technology region before requesting a product. Frequency alone is insufficient: bandwidth, realizable Q, loss, tuning, power and physical volume must fit the same operating conditions. Analog Devices’ block overview provides context for where filters sit in an RF chain. [6]

Informative technology orientation · attributes are not interchangeable data-sheet limits
TechnologyUseful region / tradeEvidence needed before selection
Lumped LCCompact and tunable at electrically small dimensions; component Q, SRF, pads and tolerance constrain narrow GHz responsesUse-frequency component models, assembly parasitics, voltage/current peaks, tuning range and temperature coefficients
Distributed / cavityGeometry stores EM energy; attractive Q and power regions can require more space; re-entrant/higher modes matterConnector/board planes, material loss, coupling, dimensions, tuning and wide frequency scans at power/temperature corners
Ceramic / LTCCCeramic resonators or integrated multilayer LC structures; compact integration does not imply the same Q or mode behaviorIdentify actual topology; package land pattern, dielectric temperature behavior, fixture treatment and resonator versus LTCC mechanism
SAW / BAWAcoustic resonances enable compact selective responses; material/cut/mode affect Q, thermal drift and nonlinear/power behaviorExact acoustic technology, matching network, temperature, average/peak waveform, harmonics and application-board evidence
Active analogUseful at suitable baseband/IF with gain and tunability; amplifier bandwidth, noise, headroom and stability limit performanceBias, source/load, dynamic range, noise, distortion, supply and amplifier excess phase
DigitalPrecisely adjustable response after conversion; computation and latency buy flexibilitySample rate, alias protection, converter overload and arithmetic limits. It cannot undo clipping before the ADC

The following exact examples teach how to preserve conditions. They are not purchasing recommendations, and their generic attributes are not used as coefficients in the allocator.

Informative manufacturer examples · specifications retain their original frequency and test conditions
Example / technologyInspected definitionWhy it is not an automatic candidate
Mini-Circuits SLP-1000+ · lumped LC low-pass [7]DC–900 MHz, loss <1 dB; nominal 990 MHz 3 dB cutoff; separate >20/>40 dB regions; 50 Ω SMA interfacesDifferent frequency and response type. The 0 dBm characterization does not establish behavior at every power.
Mini-Circuits BFCN-2450+ · ceramic LTCC band-pass [8]2400–2550 MHz passband; separate lower/upper stop regions; 50 Ω package, board conditions matterA ceramic package is not a generic unloaded-Q promise; do not transfer typical group delay into a full-temperature guarantee.
Mini-Circuits ZVBP-2450-S+ · cavity band-pass [9]2400–2500 MHz; max passband loss 1.3 dB; minimum 40 dB at 2635–2780 MHz, at 25 °CConditions cover our passband and 2.75 GHz marker; modulated waveform, power corners and physical allocation still need verification.
Qorvo 885136 · BAW band-pass [10]Rev. F, May 2017; channel-specific 2.4 GHz passbands and adjacent rejection bands; actual matching/board conditions applyIts selected-channel spans do not equal our 2400–2500 MHz requirement. Stopband attenuation within that interval can conflict with the wanted band.
Common misconceptionA SAW/BAW/LC data sheet stays valid at any impedance, power, temperature and board plane.

Its limits belong to specified conditions. Moving the plane or changing the matching network may alter loss and phase; larger drive can create heating or nonlinear response. Preserve the document’s test conditions, not just its product title.

A cavity example may provide a useful evidence starting point for the gateway; an acoustic or LTCC option may better fit a small node. Neither decision follows from the prototype family name alone.

09 / 10

Mismatch, tolerance, temperature, power, and hidden modes

A scalar H cannot determine an arbitrary filter’s return loss. To expose termination dependence without inventing measured S11, this lesson uses a disclosed lossless reciprocal algebraic two-port. It is a teaching embedding, not a physical synthesis.

S11=aS21=S12=tS22=atconj(t)\begin{aligned}S_{11} &= a\qquad S_{21} = S_{12} = t \\ S_{22} &= -\frac{a t}{\operatorname{conj}(t)}\end{aligned}
Definition of this embedding only: t=H, a=√max(0,1−|t|²), real 50 Ω normalization. SᴴS=I in exact arithmetic; at t=0 choose S11=+1, S22=−1, S21=S12=0.

Ideal ripple/rejection returns energy as reflection. The separate Q debit is kept outside this matrix. Each evaluated row checks σmax ≤ 1+10⁻¹². A roundoff-only negative 1−|t|² sets a to zero without renormalizing H.

D=(1S11ΓS)(1S22ΓL)S12S21ΓSΓLGT=(1ΓS2)S212(1ΓL2)D2Pload=PavsGT\begin{aligned}D &= (1-S_{11}\Gamma _{\mathrm{S}})(1-S_{22}\Gamma _{\mathrm{L}}) - S_{12}S_{21}\Gamma _{\mathrm{S}}\Gamma _{\mathrm{L}} \\ G_{\mathrm{T}} &= \frac{(1-|\Gamma _{\mathrm{S}}|^{2})|S_{21}|^{2}(1-|\Gamma _{\mathrm{L}}|^{2})}{|D|^{2}} \\ P_{\mathrm{load}} &= P_{\mathrm{avs}} G_{\mathrm{T}}\end{aligned}
Derived from two-port wave equations at current source/load R1 planes. ΓS and ΓL are independent complex reflections, not interchangeable VSWR magnitudes. Singular |D|≤10⁻¹² is invalid.

For an ideal through H=1 with ΓS=ΓL=+0.5, GT=1. Rotate only ΓL to −0.5 and GT=0.36, or 4.436975 dB loss. Both terminations still have VSWR 3. Phase changes delivered power even when the scalar VSWR labels do not move. The allocator keeps its waveform metric intrinsic and matched; it does not claim to simulate the mismatch-shaped burst.

Informative operating-region failure map · these effects are outside the synthetic prototype
MechanismObservable consequenceDiscriminating evidence
Tolerance / correlationResonances and couplings move together or apart; ripple, match and skirts changeBounded corners or declared population/distribution/correlation; actual lot and assembly data
TemperatureResonance, loss and dimensions drift; margins migrate across the bandHot/cold complex response at stable DUT temperature, power and mounting state
Power / heatingInternal field/current peaks exceed small-signal conditions; compression or mixing may appearAverage, peak, duty, pulse length, mismatch and thermal conditions; acoustic/contact nonlinear characterization
Parasitic couplingA path around resonators limits observed rejectionBoard/package/fixture separation and response at multiple planes
Higher / hidden modesOut-of-band transmission reappears beyond the first rejection regionWideband response through relevant blocker/harmonic populations; probe mode and environment sensitivity
Go deeperA first-order frequency sensitivity check

For an isolated LC mode, f0=1/(2π√LC) gives Δf0/f0≈−½(ΔL/L+ΔC/C). If both change +1%, the first-order shift is −1%, or −24.5 MHz at 2.450 GHz. The exact common-scale result is f0/1.01, a −24.257426 MHz shift. Couplings and parasitics make a real multi-resonator response more complicated.

Common misconceptionA deterministic tolerance sweep is measured production yield.

A sweep explores a specified set of values. Yield needs a defensible population, distributions, correlation, process and measurement uncertainty. No synthetic tolerance population or production-yield metric is claimed here.

Common misconceptionOne stopband marker proves blocker immunity.

A receiver can fail from other frequencies, intermodulation, noise, compression, mixing, or input clipping. This marker answers one linear transfer question. System immunity is a later allocation and verification task.

10 / 10

Select the node coexistence filter

Select a bounded numerical candidate, then specify what would falsify it. Do not soften the requirement until a preferred response appears to pass. The default Chebyshev and third-order Bessel are both rejected under the canonical requirements.

Derived / Simulated · fixed family candidates · each row declares Aedge, n and Q
Family / n / QAedge / αfamilyNative edge (dB)Pass loss (dB)Blocker (dB)Delay ripple (ns)Waveform (%)Screen
Chebyshev-I / 3 / 2000.5 dB / 1.0000000000000.5000002.09603249.4117125.6532183.875175Reject: worst passband loss and passband delay ripple and waveform distortion proxy
Bessel / 3 / 2000.5 dB / 0.7508246622610.9029732.09603217.2584520.0016913.560812Reject: worst passband loss and blocker attenuation and waveform distortion proxy
Butterworth / 4 / 5000.4 dB / 0.7465409702723.0103001.25121752.0964192.4673151.866213Accept for this bounded prototype screen only

The fourth-order Butterworth candidate changes two response choices explicitly: n=4 and Aedge=0.4 dB, with Q=500. It retains the same 100 MHz passband, 2 dB loss, 40 dB rejection, 5 ns delay-ripple and 3% waveform thresholds. It passes this screen with 1.251217 dB passband loss, 52.096419 dB blocker attenuation, 2.467315 ns delay ripple and 1.866213% proxy.

Think about itDoes this passing Butterworth row qualify a 2.45 GHz hardware filter?
Answer

No. It justifies requesting a physical realization and discriminating data. The algebraic return loss, constant Q debit and periodic proxy do not qualify power, environment, tolerance, size, hidden modes or a particular manufacturer part.

Illustrative filter selection row · Path 04 component evidence pack
FieldProposed entryVerification / rejection evidence
Selected screenButterworth n=4; Aedge=0.4 dB; Q proxy=500; local Ω mappingPrototype p04-m03-filter-prototype-v1; interaction p04-m03-filter-threat-v1
Passband / loss2400…2500 MHz; ≤2 dB total screening lossMeasured full-band loss/match at required source/load and temperature corners
Stopband / threatAt least 40 dB at 2750 MHz; −20 dBm Pavs at R1Expand to actual blocker/harmonic population; no single-marker immunity claim
Delay / waveform≤5 ns delay ripple; ≤3% local periodic proxyComplex response and modulated test; preserve declared correction/normalization
Match / planesR1 input/output, real 50 Ω; default matched source/loadObtain physical S11/S22 and fixture/de-embedding record; test actual Γ region
Power / size / temperatureSmall-signal analytic screen at 25 °C; hardware allocations unresolvedSpecify average/peak/duty, permitted dimensions, operating temperatures and thermal conditions before hardware selection
Rejected alternative AChebyshev-I n=3 / Aedge=0.5 / Q=200Rejection passes; pass loss, delay ripple and waveform fail
Rejected alternative BBessel n=3 / Aedge=0.5 / Q=200Delay flatness improves; blocker rejection, pass loss and waveform still fail
Next evidenceLC/LTCC, ceramic/acoustic or cavity physical realization studyFinite-Q physical model; corner data; power/temperature qualification; board/fixture modes

A filter is a conditional complex mapping. Its useful response must meet wanted-signal, threat, loss, delay, power, environment and evidence requirements together. Hand the conditional filter loss and terminations to 04.4: Low-Noise Amplifiers, where active gain, noise and stability enter.

Ungraded review

Check your understanding

Answer each question in your own words, then reveal the model answer.

  1. 01Why does the default candidate fail even with 49.411712 dB blocker attenuation?
    Model answer

    Its 2.096032 dB worst passband loss exceeds 2 dB, its 5.653218 ns delay ripple exceeds 5 ns, and its 3.875175% periodic waveform proxy exceeds 3%. The blocker check passes but cannot erase the other failures.

  2. 02What does increasing Q fix in this lesson, and what does it leave unresolved?
    Model answer

    It reduces only the scalar passband loss debit and its separate matched 290 K noise consequence. Intrinsic H, phase, delay and waveform remain unchanged. It creates no physical realizability or power evidence.

  3. 03Why are native Butterworth, Chebyshev and Bessel edges not an equal-bandwidth comparison?
    Model answer

    At normalized frequency one they optimize different definitions: Butterworth is -3.010300 dB, Chebyshev is -Rp and delay-normalized third-order Bessel is about -0.90297 dB. The equal-spec view rescales each to a unity passband peak and the common outer-edge attenuation. Even-order Chebyshev retains its -Rp DC droop.

  4. 04What sign and units must be used to compute group delay?
    Model answer

    With e^(+jωt), τg=-dφ/dω. Unwrap phase in radians and differentiate against angular frequency in rad/s. The result is seconds. Central interior and one-sided endpoint stencils use the physical grid; deep-null rows are suppressed.

  5. 05Why is 3.875175% not a standards-compliance EVM result?
    Model answer

    It is a sample-domain proxy over all 2048 periodic pairs after a best integer circular lag and one fitted complex gain. It has no fractional timing recovery, adaptive equalizer or standard-defined symbol population. The 3% threshold is a local illustrative decision criterion.

  6. 06What must accompany the accepted fourth-order Butterworth screen in the evidence pack?
    Model answer

    The exact family/order/edge/Q, passband and blocker population, delay/waveform conditions, source/load and planes, physical realization, power/size/temperature allocations, tolerance and hidden-mode evidence, and rejected alternatives. Accept here means the bounded numerical screen only.

Sources and further study

Accessed 6 September 2026. Theory, independent derivations and manufacturer examples are separated. All curves are drawn from the disclosed local models. No vendor response table supplies the allocator.

  1. MW-1 · D. M. Pozar, Microwave Engineering, 4th ed., Wiley, 2012. Publisher edition and contents; chapters 4, 6, 8 and 10 orient networks, resonators, filters and noise. Publisher record consulted; full text was not available for fresh page-by-page verification. The shown equations are independently derived and numerically checked.
  2. H. Zumbahlen, Basic Linear Design, Analog Devices, 2006, chapter 8. Analog Filters, standard responses and transformations. Canonical design-handbook source consulted; native normalization is explicitly replaced by the disclosed common-edge rule for comparisons. No coefficient or physical synthesis tables are copied.
  3. SciPy official reference, v1.18.0 web documentation. Bessel prototype normalization and Low-pass to band-pass transformation. Used to cross-check the natural DC-delay convention and geometric band-pass mapping. SciPy is not an application dependency.
  4. H. Zumbahlen, Analog Devices. Phase Relations in Active Filters, Analog Dialogue 43, 2009. Informative phase/order and active implementation limits. The lesson’s exact derivative and waveform algorithm are local, versioned derivations.
  5. SIG-2 · J. G. Proakis and M. Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. Publisher edition record, chapters 2, 4 and 5 for signal representation/pulse shaping; full text not newly accessible. SYS-3 · Rohde & Schwarz, Understanding EVM, 3683.8038.52, version 01.00, consulted for reference and normalization interpretation; no standard limit is imported.
  6. CIR-2 · Anton Patyuchenko, Analog Devices. RF Signal-Chain Discourse, Part 2: Essential Building Blocks, 2021. Informative signal-chain roles and filter technology context; not a component rating.
  7. Mini-Circuits SLP-1000+. Official sheet Rev. A, stamp 061024, inspected via official indexed text; current product page identifies lumped LC. 50 Ω SMA, FF99 package; DC–900 MHz <1 dB, nominal 990 MHz at 3 dB; >20 dB at 1340–1750 MHz, >40 dB at 1750–2000 MHz. Curves use 0 dBm; operating −55…+100 °C, 0.5 W maximum. Exact calibration plane and characterization temperature need confirmation. Different band: orientation only.
  8. Mini-Circuits BFCN-2450+. Official sheet Rev. OR, M143261, stamp 190725, official indexed text inspected. Ceramic LTCC band-pass, FV1206-4, 50 Ω, 2400–2550 MHz. Operating −40…+85 °C; passband input maximum 2 W at 25 °C, derated to 0.5 W at 85 °C. Preserve its separate stop regions and device/board port treatment; fixture removal and exact characterization drive were not established in the inspected excerpt. Do not infer resonator Q from the package.
  9. Mini-Circuits ZVBP-2450-S+. Official sheet Rev. A, ECO-000859, stamp 191202, official indexed text inspected. Cavity, 50 Ω SMA-female interfaces, QT2302. At 25 °C: 2400–2500 MHz loss ≤1.3 dB, VSWR ≤1.5; attenuation ≥40 dB at 2635–2780 MHz. Operating −55…+100 °C, 15 W maximum input. Typical delay data are not a corner guarantee; exact calibration plane and drive require confirmation. No power or waveform qualification is transferred to the synthetic candidate.
  10. Qorvo 885136. Official product page and manufacturer datasheet Rev. F, May 2017, distributor-hosted copy, pp. 2–5, 7 inspected. BAW, 1.1×0.9×0.50 mm; optimum 50 Ω single-ended ports and external ±3% matching, characterized on 885136-EVB. Electrical tables cover separately stated temperature regions; typical values use 25 °C. Loss integrates linear S-parameters over 19 MHz; attenuation over 5 MHz. These are not spot values across our whole band. Lifetime/power conditions specify waveform, temperature and input pin; a +39 dBm CW rating is limited to 20 ms at 25 °C. Matching may need adjustment for the actual PCB; the exact calibration/de-embedding plane remains to confirm. Technology illustration only.
Deliberate next steps

Physical prototype tables, coupling matrices, EM geometry, tuning and tolerance distributions remain outside this lesson. LNA/PA/mixer/PLL design is in 04.4–04.7; receiver allocation in Path 05; named masks in Paths 07/09; measurement and group-delay uncertainty in Path 08; manufacturing/layout in Path 10. Model p04-m03-filter-prototype-v1 remains a Learn teaching interaction.