Filter Shape Factor and Selectivity

Filter shape factor compares the bandwidths of a bandpass filter at two specified attenuation levels, while selectivity describes how effectively the filter separates wanted signals from nearby unwanted signals.

This article explains their relationship to bandwidth, Q factor, filter order and receiver architecture, with practical selection guidance for RF design engineers and component engineers.

The discussion concerns RF and signal-processing filters, particularly bandpass filters. It develops the original explanation by Peter Matthews of Knowles Precision Devices [1], with additional distinctions needed when comparing datasheets and measured responses. Power-line and SMPS filters require additional consideration of operating current, source and load impedance, damping and safety.

Key Takeaways

Shape factor is useful for comparison only when its measurement conditions are defined. The following points connect the headline figure to practical filter performance.

• A common bandpass shape factor is the 60 dB bandwidth divided by the 3 dB bandwidth. A value closer to unity indicates a narrower combined transition region for those specified levels.
• Always state the attenuation reference: levels relative to the pass band peak are not the same as absolute insertion loss measured from a through reference.
• Bandwidth Q describes how narrow the pass band is relative to its centre frequency. It does not, by itself, specify normalised skirt shape, insertion loss or far-out rejection.
• Resonator Q, response family, coupling and filter order jointly influence the achievable response. Higher order can improve rejection, but practical loss, size and sensitivity to tolerances must also be considered.
• Check rejection on both sides of the pass band, pass band ripple, group delay and the complete stop band. A single shape-factor value cannot replace the full filter specification.

An Ideal Filter

An ideal unity-gain filter has a magnitude response of one, or 0 dB, throughout its pass band and zero transmission in its stop band. Its magnitude response changes discontinuously at the band edges: it has no transition band. This brick-wall response is a mathematical reference rather than a physically realisable causal filter [2].

To preserve the waveform of a signal within the pass band, amplitude response alone is not sufficient. The filter must also have a suitable phase response, ideally linear phase over the occupied signal bandwidth, corresponding to constant group delay [2, 3].

Real passive filters introduce insertion loss and may have pass band ripple. Stop band attenuation is finite, and some response families also exhibit stop band ripple. Between the pass band and each stop band is a transition region, often called the filter skirt. Figures 1 and 2 contrast the idealised and practical cases [1, 2].

The transition region can determine whether a filter is suitable for a particular application. Shape factor describes its relative width between specified response levels; selectivity expresses the broader ability to pass the wanted band while rejecting frequencies outside it [1].

Filter Shape Factor

Filter shape factor is a dimensionless ratio of bandwidths measured at two specified attenuation levels. For the convention used here, the bandwidth at the greater attenuation is the numerator, so a value closer to unity represents a sharper transition for otherwise comparable specifications [1].

In Eq. (1), SF is the shape factor, BW_{60 dB} is the frequency difference between the upper and lower 60 dB crossings surrounding the wanted pass band, and BW_{3 dB} is the corresponding difference between the 3 dB crossings. Both bandwidths must use the same frequency units and the same attenuation reference.

The original example uses 3 dB to identify the pass band bandwidth and 60 dB for the wider rejection bandwidth. These levels are a convention, not a universal requirement. A datasheet may use another pair of attenuation levels, so the levels must accompany the quoted shape factor [1].

Define the attenuation reference

For a normalised response, attenuation is measured relative to a stated pass band reference, commonly the maximum transmission. A 3 dB bandwidth then means the response has fallen by 3 dB from that reference. Absolute insertion loss instead compares transmission with a through reference. If the filter already has pass band insertion loss, these definitions produce different crossing levels and potentially different bandwidths [2, 3].

Record whether the reference is the pass band peak, the centre-frequency response or an absolute insertion-loss baseline. Do not combine bandwidths taken from differently normalised plots. The usable pass band may also be defined by a ripple or insertion-loss limit rather than by the 3 dB points [2].

Worked example and limitations

As an illustrative calculation, consider a filter with BW_{3 dB} = 20 MHz and BW_{60 dB} = 30 MHz, measured relative to the same pass band reference. Substitution in Eq. (1) gives SF_60/3 = 30/20 = 1.5. The two transition regions together occupy 10 MHz between these response levels. Only if the response is symmetric on the plotted linear frequency axis can this be assigned as 5 MHz on each side.

The ideal brick-wall limit has a shape factor of unity. In practical comparisons, seek a value closer to unity only after matching the pass band requirement, attenuation levels and reference convention. There is no single shape factor that characterises all LC, SAW or BAW filters: response family, order, bandwidth, coupling and implementation also matter [1–3].

A total bandwidth ratio can conceal unequal lower and upper skirts. It also cannot show whether rejection deteriorates again farther from the pass band. Specify attenuation at the actual interfering frequencies and inspect the complete required stop band rather than relying on the ratio alone [2, 3].

Selectivity

For an RF filter, selectivity is its ability to distinguish between frequencies that should pass and frequencies that should be rejected. Shape factor is one figure of merit for this behaviour, but a complete selectivity requirement usually states attenuation at specified frequencies or offsets from the wanted band [1, 3].

Narrower transition bands can reduce the frequency spacing needed between wanted and unwanted signals. However, a filter alone does not determine the required guard band: transmitter emissions, modulation bandwidth and receiver performance also constrain the system. Compare rejection at the relevant lower and upper frequency offsets, together with insertion loss and ripple across the entire occupied signal bandwidth [1, 3].

The response depends on more than a single Q value. Resonator loss, filter order, coupling and the placement of transmission zeros all influence the achievable skirts and stop band. A transmission zero is a frequency at which the ideal transfer function becomes zero; placing zeros near an unwanted band can provide strong local rejection [2, 4].

Response shape and signal fidelity

At comparable order, a Butterworth response offers a maximally flat pass band, while a Chebyshev response trades pass band ripple for a narrower transition. An elliptic response uses transmission zeros to obtain a sharper transition, with ripple in both pass band and stop band. A Bessel response prioritises group-delay behaviour rather than the sharpest amplitude discrimination [2, 3].

These are design trade-offs, not a ranking from universally best to worst. For modulated RF signals and pulses, group-delay variation across the occupied band can distort the waveform even when the amplitude specification is satisfied. Selectivity must therefore be evaluated alongside the required phase and time-domain performance [2, 3].

Q Factor and Selectivity

Q factor has several meanings in filter engineering. Bandwidth Q describes the relative width of a bandpass response, whereas resonator Q describes energy storage and loss. These quantities are related through the design, but they are not interchangeable [2, 4].

In Eq. (2), Q_BW is the bandpass bandwidth figure of merit, f_c is the stated centre frequency, and Δf is the 3 dB bandwidth. The lower and upper 3 dB frequencies are f_L and f_H, respectively, so Δf = f_H − f_L. Use the same frequency units throughout.

For a conventional second-order bandpass resonator, this bandwidth relationship describes loaded Q, and its resonant frequency is the geometric mean of the two half-power frequencies. For a multi-resonator filter, Eq. (2) is a useful overall bandwidth descriptor but does not identify the Q of each resonator. Use the centre-frequency convention specified by the supplier, particularly for wideband or asymmetric responses [2].

At fixed centre frequency, reducing bandwidth increases Q_BW. This makes the accepted band narrower, but it does not necessarily improve the shape factor or rejection at an offset expressed as a multiple of bandwidth. Those properties also depend on response family, order and coupling [2, 4].

Loaded and unloaded resonator Q

Unloaded resonator Q describes internal losses in the resonator. Loaded resonator Q also includes energy coupled into external circuits. Component Q refers to the loss behaviour of individual reactive components, while pole Q describes a pole pair in the transfer function. A statement such as “high Q” is incomplete unless its context and operating frequency are clear [3, 4].

High unloaded resonator Q is valuable when a narrow, selective response must be achieved with limited insertion loss. It provides a lower-loss starting point for the required coupling and response synthesis; it does not automatically produce a particular bandwidth or shape factor. Compare measured insertion loss, ripple and rejection rather than selecting a filter from a Q claim alone [3, 4].

Filter Order and Selectivity

Filter order is formally the degree of the denominator of the reduced transfer function, or equivalently its pole count. RF filter descriptions also commonly use resonator count or low-pass prototype order. These conventions must be distinguished when comparing designs [2].

For a conventional low-pass-to-bandpass transformation, a prototype of order n produces a bandpass transfer function of order 2n. In the corresponding ideal n-resonator bandpass model, each resonator contributes a pair of poles. Eq. (3) retains the original article’s relationship with this explicit limitation: N_BP is the full bandpass transfer-function order, and n is the prototype order or corresponding resonator count [2].

Do not interpret Eq. (3) as a universal rule for every filter topology or every supplier’s order label. Ask whether a quoted order means resonators, prototype poles or full bandpass transfer-function poles. Parasitic modes and practical implementations may also require a more detailed model [2].

Within a comparable response family, increasing order generally permits a sharper transition or greater rejection. In a practical passive implementation, additional resonators and couplings can increase insertion loss, size and sensitivity to manufacturing tolerances. Higher order is therefore a trade-off rather than a free improvement [1–3].

Technology selection changes the available compromise. LC, distributed, ceramic-resonator, cavity, SAW, BAW and LTCC filters offer different combinations of resonator Q, bandwidth, size, power handling and integration. Acoustic filters are not confined to operation below 2 GHz; suitability must be established from the specific device’s frequency range and guaranteed response rather than from a generic technology boundary [3].

For receivers that must tune over a wide frequency range, switched filter banks and digitally tunable RF filters provide alternatives to a single fixed filter. Their practical benefit is adjustable preselection, but insertion loss, rejection and bandwidth must be checked across the required tuning states [3].

Selectivity of Receivers

At system level, selectivity is a receiver’s ability to recover a wanted signal in the presence of unwanted signals at other frequencies. It depends on the complete signal chain, not only on the small-signal transmission response of one filter [1, 3].

Adjacent channel rejection, adjacent channel selectivity, in-band blocking and out-of-band blocking describe different interference situations. Their definitions depend on the applicable system specification, including wanted and unwanted signal levels, frequency separation, modulation and the required receiver performance. A filter shape factor cannot be directly substituted for any of these system tests [1, 3].

A strong unwanted signal can compress an amplifier or mixer, or create intermodulation products that fall inside the wanted band. Filtering before the vulnerable nonlinear stage can reduce that risk. A highly selective filter placed only later in the chain cannot reverse distortion already generated upstream [3].

Receiver architecture determines where selectivity is most useful. RF preselection can suppress blockers before amplification or mixing. In a superheterodyne receiver, frequency conversion allows additional channel selection at an intermediate frequency, but unwanted image responses must also be rejected. Depending on the architecture, image rejection may be provided by preselection, dedicated filters, image-reject mixing or a combination [3].

Strong selectivity close to the antenna can simplify downstream requirements, but a lossy filter before the low-noise amplifier also penalises receiver sensitivity. Balance blocker protection against insertion loss and noise performance. Direct-sampling architectures still require suitable analogue filtering to manage unwanted signals and aliasing at the converter input [3].

Design and Selection Checklist

Start with the wanted signal and interference environment, then compare complete filter responses under matching conditions. The checklist below turns shape factor and selectivity into a practical purchasing and validation specification [2, 3].

• Define the wanted band: state lower and upper usable frequencies, maximum insertion loss and permitted ripple across the occupied signal bandwidth. Do not assume the 3 dB bandwidth is the usable signal bandwidth.
• Define rejection on both sides: state minimum attenuation at each critical interfering frequency and across the required stop band, including any regions where spurious transmission would be unacceptable.
• State the shape-factor convention: identify both attenuation levels, their reference and whether the quoted response is typical or guaranteed.
• Check signal fidelity: include group-delay variation or other phase-response requirements where modulation or pulse shape makes them important.
• Match the electrical environment: verify reference impedance, return loss, matching requirements and the effect of the actual source and load. Include the recommended mounting and grounding arrangement.
• Compare operating limits: check the supplier’s input-power limits, temperature range and environmental ratings. Request response limits over the required operating conditions rather than relying only on a room-temperature typical plot.
• Choose catalog or custom deliberately: use a catalog part when its guaranteed response meets the complete mask. Consider custom development when rejection frequencies, bandwidth, matching, package or environmental requirements cannot be met by available standard parts.
• Define acceptance and qualification evidence: agree the response mask, measurement conditions and required environmental or reliability evidence with the supplier. RF selectivity is not evidence of mains safety suitability; power-line applications require their own ratings and safety assessment.

Measure the response, not only the headline ratio

Use a calibrated vector network analyser to measure forward transmission, S₂₁, together with input and output matching over the relevant frequency range. Establish the reference planes at the filter interfaces, or characterise the fixture so that its loss and mismatch are not mistaken for filter behaviour. Source and load conditions affect the installed response [3, 5].

Use enough frequency resolution to locate the specified crossings and enough measurement dynamic range to resolve the required rejection. Check for fixture leakage or instrument noise before accepting a deep stop band reading. If the measurement floor is reached before the required attenuation crossing, the corresponding bandwidth and shape factor have not been established.

Record the pass band reference, the lower and upper crossings at each attenuation level, the calculated bandwidths and the resulting shape factor. Then check the complete rejection mask, including both skirts and the farther stop bands. For power and EMI filters, operating current, grounding, shielding and non-standard source and load impedances can make installed attenuation differ substantially from a catalog curve [5].

Conclusion

Shape factor quantifies the relative transition width between two specified attenuation levels. Selectivity is the broader requirement to pass the wanted signal while rejecting unwanted frequencies. Both are useful only when the pass band, attenuation reference and relevant interference frequencies are clearly defined.

Bandwidth Q, resonator Q and filter order describe different aspects of the design. Use them to understand trade-offs, but select the final filter from its guaranteed insertion loss, ripple, rejection, matching and signal-fidelity performance. In a receiver, confirm that the filter’s location and loss support the required blocker tolerance and sensitivity.

FAQ

These answers summarise the definitions and practical distinctions used when comparing RF bandpass filters.

How to Measure Filter Shape Factor

Use the following procedure to calculate a bandwidth ratio from a measured bandpass response. The same method can be applied to a datasheet plot when its resolution and reference convention are sufficient.

Further Reading

The following articles provide additional background on filter specifications, Q factor and practical component implementation.

• Five Key Filter Specifications — connects centre frequency, bandwidth, insertion loss, rejection and selectivity.
• Filter Q Factor Explained — develops the distinction between loaded, unloaded, component and pole Q.
• Q Factor Modelling: How Poles and Zeros Relate to Q Factor — introduces transfer-function analysis and numerical exploration of filter response.
• RF Inductors and Filters — extends the discussion to RF inductors, parasitics and filter construction.

References

The numbered sources support the definitions, response trade-offs and receiver-design discussion in this article. Dates are included where stated in the opened source.

1. Peter Matthews, Filter Shape Factor and Selectivity, Knowles Precision Devices Blog, 2023.
2. Analog Devices, Basic Linear Design, Chapter 8: Analog Filters, Analog Devices, undated chapter PDF.
3. Brad Hall and David Mailloux, How Digitally Tunable Filters Enable Wideband Receiver Applications, Analog Dialogue, Analog Devices, June 2022.
4. Peter Matthews, Contextualizing Q Factor, Knowles Precision Devices Blog, November 2023.
5. Passive Components Blog, Insertion Loss and Performance in EMI Filtering, based on Knowles Precision Devices material, 2022.

Latest technical review and revision: 5 November 2026. Definitions of shape factor, Q factor and filter order clarified; selection, measurement and receiver-level guidance expanded.