Loaded Q and Bandwidth of a Matching Network

Fractional bandwidth equals 1/Q_L for a single resonant circuit, so Q_L = 10 at 1 GHz gives 100 MHz. An L-match has Q = sqrt(R_high/R_low − 1), so matching 5 to 500 ohm forces Q = 9.95 and only about 20% bandwidth.

Engineer’s note: A matching network's bandwidth follows its loaded Q, but its loss depends on the unloaded Q of the chip inductors you actually mount. Don't trust a datasheet Q quoted at 100 MHz. Check the measured curve at your real frequency, 2 GHz or 5 GHz, and make sure the inductor's Q hasn't collapsed there, or insertion loss follows.

For practitioners

A lossless matching network transforms one resistance into another at a single frequency. Away from that frequency the match degrades. How fast it degrades depends on the Q of the network, and the impedance ratio you ask for sets that Q. This page shows how the ratio sets the Q and how the Q sets the bandwidth.

Loaded Q from the 3 dB bandwidth

Loaded Q is defined from the measured 3 dB bandwidth of the response:

QL=f0ΔfQ_L = \frac{f_0}{\Delta f}

The fractional bandwidth is Δf/f0. For a single resonant circuit, that makes it the inverse of the loaded Q:

Δff0=1QL\frac{\Delta f}{f_0} = \frac{1}{Q_L}

At 1 GHz, a circuit with Q_L = 10 has a 3 dB bandwidth of 100 MHz, or 10%. Double the Q and the bandwidth halves. A matching network at its design frequency behaves like a resonator. Its series and shunt reactances cancel, and the circuit stores energy in them. The more energy it stores per cycle relative to the power delivered, the higher the Q and the narrower the response.

Unloaded, external and loaded Q

Two mechanisms remove energy from the resonator. Component losses do, and their effect is described by the unloaded Q, Q_U. The source and load terminations also do, and their effect is described by the external Q, Q_E. The loaded Q combines them:

1QL=1QU+1QE\frac{1}{Q_L} = \frac{1}{Q_U} + \frac{1}{Q_E}

In a matching network the terminations do the useful work, so Q_E is the design quantity. Q_U is a parasitic. A finite Q_U always pulls Q_L below Q_E. That widens the response, but only because power is being dissipated in the parts. With good components Q_U is much larger than Q_E and Q_L ≈ Q_E. The analysis below assumes lossless parts.

The L-network: Q fixed by the resistance ratio

An L-network has two elements: a series reactance X_s on the low-resistance side and a shunt reactance X_p on the high-resistance side. Each branch has its own Q:

Qs=XsRlow,Qp=RhighXpQ_s = \frac{X_s}{R_{\text{low}}}, \quad Q_p = \frac{R_{\text{high}}}{X_p}

At the match the two are equal, and the shared node Q depends only on the resistance ratio:

Q=RhighRlow−1Q = \sqrt{\frac{R_{\text{high}}}{R_{\text{low}}} - 1}

You get no free design parameter here. Once you choose the two resistances, the Q is fixed, and so is the bandwidth. Matching 50 to 200 ohm (ratio 4) gives Q = √3 = 1.732. Matching 5 to 500 ohm (ratio 100) gives Q = √99 = 9.95.

The reason a higher Q means a narrower match is simple. Q is the reactance divided by the resistance. With a high Q, the reactances are large compared with the resistances they transform. A given fractional change in frequency changes each reactance by the same fraction, but that change is a large error compared with the small resistance. The cancellation that produced the match is lost quickly.

When the network is driven from a matched source, the source resistance loads the resonator as much as the load does. The loaded Q is therefore about half the node Q, and the 3 dB fractional bandwidth is about:

Δff0≈2Q\frac{\Delta f}{f_0} \approx \frac{2}{Q}

Worked example: 5 to 500 ohm at 1 GHz

Take a lowpass L-match from a 5 ohm source to a 500 ohm load at f0 = 1 GHz.

  1. Node Q: Q = √(500/5 − 1) = √99 = 9.95.
  2. Series reactance on the 5 ohm side: X_s = Q × R_low = 9.95 × 5 = 49.75 Ω. As an inductor at 1 GHz, L = 49.75 / (2π × 10⁹) ≈ 7.92 nH.
  3. Shunt reactance on the 500 ohm side: X_p = R_high/Q = 500/9.95 = 50.25 Ω. As a capacitor, C = 1 / (2π × 10⁹ × 50.25) ≈ 3.17 pF.
  4. Loaded Q with a matched source: Q_L ≈ 9.95/2 ≈ 5.
  5. Fractional bandwidth: about 2/Q ≈ 0.20, so roughly 200 MHz, from about 0.9 to 1.1 GHz.

Ask for the same transformation at Q = 1.732 (the 50 to 200 ohm case) and the reactances are only 86.6 Ω in series and 115.5 Ω in shunt. Both are comparable to the resistances around them, so frequency changes do far less damage.

Where the 2/Q estimate stops working

The figure compares the two lowpass L-matches. Both reach their best return loss at f/f0 = 1.

0.50.751.01.251.50−10−20−30−40Normalized frequency f/f0|S11| (dB)−3 dB5 to 500 ohm, Q = 9.9550 to 200 ohm, Q = 1.73minimum at f/f0 = 1
Return loss of lowpass L-matches from 50 to 200 ohm with Q = 1.73 and from 5 to 500 ohm with Q = 9.95, both tuned to f0. Computed from the formula.

The 5 to 500 ohm match (Q = 9.95) gives a narrow dip. Its |S11| climbs back through −3 dB near 0.9 and 1.1 f0, which is consistent with the 20% estimate. In a lossless network, |S11| = −3 dB means half the power is reflected, so these crossings are also the 3 dB transmission edges.

The 50 to 200 ohm match (Q = 1.73) gives a broad, shallow-sided dip. Applying 2/Q here predicts a fractional bandwidth of about 115%, which has no physical meaning. The resonator approximation needs a fairly high Q, roughly 5 or more. At low Q the response is too broad for it. The real circuit makes the point. At DC the series inductor is a short and the shunt capacitor is open, so 50 ohm drives 200 ohm directly. The mismatch loss is only about 1.9 dB, and |S11| is about −4.4 dB. The lowpass match never reaches its lower 3 dB edge.

Use 2/Q only when Q is about 5 or higher. Below that, simulate the network or sweep S11 and read the band edges directly.

Getting more bandwidth from a high ratio

A single L-network can't widen its bandwidth, because the ratio fixes its Q. The usual fix is to step the impedance in stages. Going from 5 to 500 ohm through an intermediate 50 ohm gives two sections with a ratio of 10 each, so each section has Q = √9 = 3 instead of 9.95. The cascade is no longer a single resonator, so 2/Q does not give its bandwidth exactly. Even so, the lower Q per section gives a broader match. Pi and T networks work the other way. They add a design Q that is at least as high as the L-network value, which is useful for selectivity but not for bandwidth. See RF Matching Networks: L, Pi and T.

In plain words · for students and beginners

Loaded Q and bandwidth in plain words

An everyday picture

Picture a swing. Push it at exactly the right rhythm and it goes high. A heavy, smooth swing only responds to the exact rhythm. Push slightly off-beat and it barely moves. A light swing that drags on the ground responds to a wider range of rhythms, but it never goes as high. Q measures how "picky" a circuit is about frequency. High Q means picky and narrow. Low Q means relaxed and wide.

1101000.50.711.42Frequency / resonant frequency|Z| / Rdots: half-power points, width 1/Q = 0.1 of f0Q = 10: the impedance is lowest at resonance
Series resonance with Q = 10: impedance against frequency, computed from the series RLC formula. Concept drawing.

The figure shows a circuit with Q = 10. It responds strongly only near its center frequency. Where the analogy breaks down: a matching network isn't trying to swing high. It's trying to pass power from one resistance to another without reflection. The pickiness comes from how hard the job is, not from how the parts are tuned.

Why a big jump is a picky job

A matching network converts one resistance into another. Resistance is how strongly a part resists current. Converting 50 ohm into 200 ohm is a small jump. Converting 5 ohm into 500 ohm is a jump of 100 times. A big jump needs large reactances. Reactance is the frequency-dependent "resistance" of a coil or capacitor. Because those values change with frequency, a big jump only lines up near one frequency. A gentle jump uses small reactances, so it keeps working over a wider band.

The worked example, step by step

  1. Divide the resistances: 500 / 5 = 100.
  2. Subtract 1 and take the square root: √99 ≈ 9.95. That is the network Q.
  3. A matched source shares the load, so the effective (loaded) Q is about half: about 5.
  4. Bandwidth as a fraction of the center frequency is about 2 / 9.95 ≈ 0.20, or 20%.
  5. At 1 GHz that is about 200 MHz of usable band.

For 50 to 200 ohm: 200 / 50 = 4, and √3 ≈ 1.73. The shortcut would give over 100%, which can't be right. The match is simply very wide, and the shortcut doesn't apply.

Mistakes beginners make

  • Using the 2/Q shortcut at low Q. It only works when Q is about 5 or more.
  • Thinking you can pick the Q of an L-network. You can't: the two resistances decide it.
  • Treating a wider dip as good news without checking loss. Lossy parts lower Q too, but they waste power doing it.
  • Forgetting that the 3 dB bandwidth means the points where half the power is lost, not where the match first looks imperfect.

Remember

  • Fractional bandwidth is about 1 / loaded Q: Q_L = 10 at 1 GHz gives 100 MHz.
  • A bigger resistance ratio forces a higher Q and a narrower match.
  • To match a big ratio over a wide band, make the jump in smaller steps.

Frequently asked questions

How is loaded Q related to the bandwidth of a matching network?

Loaded Q is defined from the 3 dB bandwidth as Q_L = f0/Δf. For a single resonant circuit the fractional bandwidth Δf/f0 is simply 1/Q_L, so Q_L = 10 at 1 GHz gives 100 MHz.

Why does a high impedance transformation ratio give a narrow match?

An L-network's Q is fixed by the ratio, Q = sqrt(R_high/R_low − 1). A larger ratio forces larger reactances relative to the resistances. Those reactances change faster with frequency, so the match falls apart sooner away from f0.

Does the bandwidth ≈ 2/Q rule work for every L-network?

No. It is a resonator approximation that holds only when Q is fairly high, roughly 5 or more. At Q = 1.732 (50 to 200 ohm) it predicts more than 100%, and the real lowpass match loses only about 1.9 dB even at DC.

How do I get more bandwidth from a high-ratio match?

Split the transformation into several smaller steps. Each section then has a lower ratio and a lower Q. Lossy components also lower the loaded Q, but they buy that bandwidth with insertion loss.

References and further reading