Filter Specifications: Five Key Parameters Explained

The five key filter specifications are center frequency, bandwidth, insertion loss, out-of-band rejection and selectivity. This article explains how design, purchasing and component engineers can interpret these parameters for RF and small-signal bandpass filters, compare datasheets and verify a selected part under defined operating conditions.

How to Measure Filter Performance: Insertion Loss, VNA Fixtures and CM/DM Separation

Passive Filter Technologies Compared: LC, Ferrite, Feedthrough, Ceramic, Cavity, Microstrip, SAW and BAW

Knowles A-Series RF Attenuators Target Compact 20 GHz Designs

Key Takeaways

A useful filter specification describes the complete required response and its measurement conditions, not just a center frequency and bandwidth.

State whether center frequency is a nominal catalog value, an arithmetic midpoint or a geometric mean. These values can differ substantially for a wide passband.
Specify the response level used to measure bandwidth. A 3 dB bandwidth is not necessarily the frequency range over which all passband requirements are met.
Insertion loss depends on source and load conditions. Compare typical curves and specified limits separately, including their temperature and mounting conditions.
Specify rejection at defined frequencies or over defined ranges. Selectivity describes the transition between the wanted passband and nearby rejection requirements.
Validate transmission, matching and rejection with calibrated measurements and the intended PCB implementation. RF bandpass specifications do not replace the current, thermal, damping or safety assessment required for power and mains filters.

Understanding Filter Specifications

Filter selection begins by defining the wanted signal band, allowable signal loss and unwanted frequencies that must be attenuated. The five parameters below describe important parts of that requirement, but they must be read together rather than optimized independently. The original explanation and figures were published by Knowles Precision Devices. [1] The underlying low-pass, high-pass, bandpass and bandstop responses are introduced in Basic Filter Circuits Explained. Here, the main examples concern RF bandpass transmission; power-supply and mains EMI filters require additional application-specific assessment. Figure 1 identifies the five parameters on a typical bandpass response. Such a response plot should be read with its vertical-axis definition, reference impedance and measurement conditions: transmission in decibels is not an absolute output-power measurement.

Center Frequency

Center frequency identifies a reference frequency within the passband, shown as f₀ in Figure 2. Let f_L and f_H denote the lower and upper band-edge frequencies measured at the same stated response level. Two common conventions are the arithmetic midpoint, f_0,a, and geometric mean, f_0,g, given in Eq. (1). [1]

f_{0,a}=\frac{f_{L}+f_{H}}{2}, \qquad f_{0,g}=\sqrt{f_{L}f_{H}} \tag{1}

Bandpass response showing lower band edge f L, upper band edge f H and center frequency f zero — Figure 2. Lower and upper band edges, f_L and f_H, and center frequency f₀. The center-frequency convention and the response level defining the edges must be stated. Source: Knowles Precision Devices.

The geometric mean is the center frequency of the canonical second-order resonant bandpass response when the edges are its half-power frequencies. The arithmetic midpoint is also used to describe a frequency interval. For narrow bands the two values are close, but they are not interchangeable for wide bands. A manufacturer may instead identify a nominal catalog center frequency; use the datasheet definition rather than assuming that the product name specifies an exact calculated midpoint. [1, 2]

Bandwidth

Bandwidth, BW, is the difference between the upper and lower frequencies at a specified response level, as expressed in Eq. (2). In a resonant bandpass measurement, the usual half-power bandwidth is approximately 3 dB below peak transmission, not automatically 3 dB below an ideal lossless reference. Record the reference level and measurement method when comparing results. [2, 3]

\mathrm{BW}=f_{H}-f_{L} \tag{2}

The usable passband can be narrower than the measured 3 dB bandwidth because it must also satisfy insertion-loss, ripple and matching requirements. A datasheet frequency range associated with a specified insertion-loss limit should not be relabeled as a 3 dB bandwidth unless the manufacturer defines it that way.

Fractional bandwidth, FBW, expresses bandwidth relative to the selected center frequency. Equation (3) gives its dimensionless and percentage forms. Use the same frequency units in the numerator and denominator, and state which center-frequency convention is used. [1]

\mathrm{FBW}=\frac{\mathrm{BW}}{f_{0}}, \qquad \mathrm{FBW}_{\%}=100\frac{\mathrm{BW}}{f_{0}} \tag{3}

Figure 3 provides an illustrative comparison of filter technologies by frequency and fractional bandwidth. It is an overview rather than a current product-availability map: the suitability of a particular part still depends on its complete electrical response and operating conditions. [1]

Illustrative comparison of fractional bandwidth and operating frequency for different filter technologies — Figure 3. Illustrative frequency and fractional-bandwidth ranges for different filter technologies from the original 2022 source. The ranges are not procurement limits or guarantees for individual products. Source: Knowles Precision Devices.

Fractional bandwidth can exceed 100%. In the original example, a filter spans 2 GHz to 18 GHz. Its bandwidth is 16 GHz and its arithmetic midpoint is 10 GHz, giving a fractional bandwidth of 160%. These calculations are correct when the arithmetic-center convention is used. [1]

For the same edges, the geometric center from Eq. (1) is 6 GHz, giving a fractional bandwidth of approximately 266.7%. The frequency interval has not changed; only the normalization has changed. Consequently, a fractional-bandwidth value without its center-frequency convention can be misleading for a very wide passband.

Insertion Loss

Insertion loss, IL, describes the reduction in signal delivered to a load when a filter is inserted into an otherwise defined test configuration. Let P₁ be the load power without the filter and P₂ the load power with it. Let V₁ and V₂ be the corresponding load-voltage magnitudes. Equation (4) gives the power definition and its equivalent voltage form when the load impedance is unchanged. The voltage form transcribes the equation in the original article. [1]

\mathrm{IL}_{\mathrm{dB}}=10\log_{10}\left(\frac{P_{1}}{P_{2}}\right)=20\log_{10}\left(\frac{|V_{1}|}{|V_{2}|}\right) \tag{4}

The reference and inserted-filter measurements must use the same source settings, load and defined connection planes. V₁ is not simply the voltage at the filter input: comparing input and output voltages without accounting for the surrounding circuit does not necessarily measure insertion loss.

For a two-port filter measured with matched source and load at the stated reference impedance, forward transmission is described by the scattering parameter S₂₁. If S_21,dB denotes its logarithmic magnitude, Eq. (5) relates the negative transmission trace to positive insertion loss. [3]

S_{21,\mathrm{dB}}=20\log_{10}|S_{21}|, \qquad \mathrm{IL}_{\mathrm{dB}}=-S_{21,\mathrm{dB}} \tag{5}

Measured insertion loss is not necessarily all dissipated as heat in the filter: mismatch and reflection can also reduce delivered power. Compare insertion loss with return loss, and do not assume that a matched-system measurement predicts the same delivered-power reduction with arbitrary source and load impedances.

On the transmit side, insertion loss consumes the available RF power budget. On the receive side, loss affects the receiver noise budget, making filter placement and its interaction with the following receiver stages important. Compare the maximum loss across the required signal band rather than relying only on a typical center-frequency value. [1]

Out-of-Band Rejection

Out-of-band rejection describes how strongly a filter attenuates unwanted frequencies outside its passband. Specify the required attenuation at named frequencies or over defined frequency ranges on both sides of a bandpass filter. State whether the requirement is relative to the through reference or relative to passband transmission. [1, 4]

Unwanted signals may lie immediately beside the passband or much farther away. In a sampled signal chain, insufficient analog filtering can allow unwanted input frequencies to alias into the wanted band after sampling. The required rejection must therefore follow the complete signal-chain frequency plan, not just the appearance of the filter skirts. [2]

Filter rejection terminology must be distinguished from regulatory emissions terminology. ITU-R SM.328-12 defines an out-of-band emission as an emission immediately outside the necessary bandwidth that results from modulation, excluding spurious emissions. This definition describes transmitter emissions; it does not restrict a filter rejection specification to that regulatory domain. [5]

A filter can therefore have rejection requirements for nearby interferers, transmitter harmonics or other unwanted frequencies, but these are not all regulatory out-of-band emissions. Passing a component-level rejection test also does not, by itself, demonstrate compliance of the complete transmitter.

Selectivity

Selectivity describes how effectively a filter separates the wanted passband from nearby unwanted frequencies. One practical specification is the transition interval between a defined passband edge and the frequency at which a required rejection level is reached. Express that interval in frequency units or normalize it to the stated center frequency. [1]

A more selective response can satisfy a given rejection target closer to the passband, reducing the transition bandwidth allocated to the filter. This can help a system use closely spaced channels, although the final guard-band requirement also depends on the transmitted spectrum and the receiver architecture.

Selectivity must be considered alongside insertion loss, ripple, phase response and transient behavior. Sharper amplitude transitions can involve trade-offs with passband behavior and ringing, depending on the response family and implementation. Neither bandwidth alone nor a single Q value completely specifies the filter skirts. [2]

Putting It All Together: A Real-World Filter Example

Figure 4 retains the original annotated S₂₁ example for the Knowles B095MB1S, described as a nominal 9.5 GHz surface-mount bandpass filter. The figure connects the five parameters to a real transmission response; use the manufacturer’s datasheet, rather than measurements estimated from this illustration, for component selection. [1, 4]

Annotated forward-transmission response for the Knowles B095MB1S bandpass filter identifying five filter specifications — Figure 4. Original annotated S₂₁ response for the nominal 9.5 GHz Knowles B095MB1S surface-mount bandpass filter. The illustration is not a substitute for the datasheet limits and test conditions. Source: Knowles Precision Devices.

The June 2025 datasheet specifies a 50 Ω characteristic impedance. For the 8.9 GHz to 10.0 GHz frequency range, its insertion-loss table lists 1.75 dB typical and 2.0 dB maximum. The low-side rejection entry covers DC to 8.0 GHz, and the high-side entry covers 11.5 GHz to 20.0 GHz; both list 40 dB minimum and 45 dB typical. These values must be read with the datasheet’s electrical-test conditions. [4]

The document states that electrical specifications are based on typical probed performance at room temperature and that insertion loss varies by ±0.5 dB over temperature. It separately presents typical de-embedded measurements on a connectorized fixture and provides a recommended PCB layout. Consequently, room-temperature typical curves, tabulated limits and mounted-board performance should not be treated as interchangeable evidence. [4]

The 8.9 GHz to 10.0 GHz interval is the frequency range associated with the listed insertion-loss requirement. It should not automatically be interpreted as the filter’s 3 dB bandwidth, nor should the nominal 9.5 GHz product designation be replaced by a calculated midpoint of those specification edges.

Measurement and Design Selection Checklist

Turn the five parameters into a response mask before choosing a catalog part or requesting a custom design. For RF filters, verify that the test setup can resolve both small passband losses and the required stopband attenuation; Keysight’s bandpass measurement procedure uses calibrated S₂₁ measurements and explicitly defined bandwidth-search settings. [3]

Define the Electrical Requirement

Record the wanted signal band, nominal center frequency, center-frequency convention and the response level used to define bandwidth.
Specify maximum insertion loss across the usable passband, allowable ripple, input and output matching, and any phase-response or transient requirements.
State rejection requirements at the actual interfering frequencies and over the required stopband ranges, including both sides of a bandpass filter.
Define source and load impedances, input power, temperature range and the PCB or fixture conditions under which compliance is required.

Compare Catalog and Custom Parts

A catalog filter is suitable when its specified response, interfaces and operating conditions satisfy the application without relying on favorable typical performance. When a custom filter is necessary, agree on the complete response mask and acceptance test rather than requesting only a center frequency and bandwidth.

Separate minimum and maximum limits from typical values and illustrative plots. Confirm the temperature and manufacturing conditions attached to each limit.
Check package, mounting, grounding, power-handling conditions and assembly requirements. The B095MB1S datasheet, for example, ties its power rating to the specified thermal mounting arrangement. [4]
Request the qualification, traceability and acceptance documentation required by the end application. Electrical performance is not a substitute for reliability qualification or a safety approval.

Verify the Mounted Response

Use an appropriate calibrated vector network analyzer setup and establish the measurement reference planes. A full two-port calibration corrects the systematic errors addressed by that calibration, but any remaining fixture sections must still be included in the result or removed using a suitable fixture treatment. The B095MB1S datasheet explicitly identifies its mounted fixture curves as de-embedded measurements. [3, 4]

Measure transmission and matching across the complete required frequency range, not only at the nominal center frequency.
Choose source power, frequency spacing and receiver bandwidth appropriate to the device and the rejection target. Confirm that the measurement system can distinguish the required stopband response from its noise floor. [3]
Record calibration, reference planes, board or fixture, temperature and bandwidth convention so that development, supplier and production results can be compared.
For power, SMPS and mains EMI applications, assess real operating impedances, current, voltage, thermal behavior, resonant peaking and applicable safety requirements separately. A matched RF response is not a complete power-filter specification.

Conclusion

Center frequency and bandwidth locate the wanted band; insertion loss describes the signal penalty; out-of-band rejection specifies suppression outside it; and selectivity describes how quickly the response reaches nearby rejection targets. Their definitions, reference levels and operating conditions determine whether two datasheets are genuinely comparable.

Select the filter against a complete response mask, distinguish typical data from specified limits, and verify the mounted implementation. Retain explicit scope limits: RF bandpass performance does not replace the thermal, current, damping or safety assessment needed for power and mains filtering.

FAQ

These answers summarize the definitions and selection checks used in this article.

What are the five key filter specifications?

The five specifications are center frequency, bandwidth, insertion loss, out-of-band rejection and selectivity. Together they describe the location and width of the wanted band, the signal penalty within it and the attenuation outside it. A usable requirement also states reference levels, source and load impedances, operating conditions and whether values are specified limits or typical data.

How is bandpass filter center frequency calculated?

Two common conventions use the arithmetic midpoint or geometric mean of the lower and upper band-edge frequencies. The geometric mean applies to the canonical second-order resonant bandpass response at its half-power edges. Wideband filters can produce substantially different results under the two conventions, so use the definition stated in the datasheet or measurement procedure.

Can fractional bandwidth exceed 100 percent?

Yes. Fractional bandwidth is bandwidth divided by a stated center frequency, so it can exceed unity. A band from 2 GHz to 18 GHz has a bandwidth of 16 GHz and an arithmetic midpoint of 10 GHz, giving 160 percent. Using its geometric center instead changes the percentage, so the center-frequency convention must accompany the result.

Is filter insertion loss the same as the S21 reading?

For a matched two-port measurement at the stated reference impedance, positive insertion loss is the negative of the S21 logarithmic-magnitude reading. A general insertion-loss measurement compares delivered load power before and after inserting the filter. Different source or load impedances can change the result, so an S21 curve does not automatically predict performance in every circuit.

What is the difference between rejection and selectivity?

Rejection states how much attenuation a filter provides at defined unwanted frequencies or over a specified frequency range. Selectivity describes how closely a required rejection level can approach the wanted passband. Compare the transition intervals, attenuation reference and passband requirements together, because a rejection value without its frequency range does not establish nearby-channel separation.

How to Specify and Verify an RF Bandpass Filter

Use this procedure to convert the five parameters into a comparable component requirement and a repeatable verification plan.

Specify an RF bandpass filter using a complete response mask and verify its mounted electrical performance.

Define the wanted band
Record the signal frequency range, center-frequency convention and response level used to define bandwidth. Specify the usable passband rather than assuming that it equals the measured 3 dB bandwidth.
Set loss and rejection limits
State maximum passband insertion loss, allowable ripple and matching requirements. Specify minimum attenuation at the unwanted frequencies and across the required stopband ranges, including the transition intervals beside the passband.
Compare parts under compatible conditions
Compare specified limits separately from typical data. Check source and load impedances, temperature, input power, mounting and qualification requirements. For a custom part, agree on the complete response mask and acceptance test.
Measure and document the mounted filter
Calibrate the network analyzer at defined reference planes and account for the fixture. Measure transmission, matching, bandwidth and rejection across the required range. Record settings and operating conditions so supplier and production results can be compared.

References

The original article, technical tutorial, measurement guide, product datasheet and ITU recommendation support the definitions and examples used here.

Peter Matthews, Knowles Precision Devices. Filter Basics Part 3: Five Key Filter Specifications to Understand. Knowles Precision Devices Blog, 2022. Original explanatory article and figures.
Analog Devices / Maxim Integrated. A Filter Primer. Analog Devices, 2002.
Keysight Technologies. Measurement Example of a Bandpass Filter. E5072A online help, undated web edition.
Knowles Precision Devices. B095MB1S: 9.5 GHz Surface Mount Bandpass Filter. Product datasheet, June 2025.
International Telecommunication Union. Recommendation ITU-R SM.328-12: Spectra and Bandwidth of Emissions. ITU, September 2025.

Latest technical review: 5 November 2026. Revised center-frequency and bandwidth conventions, clarified insertion loss and rejection terminology, checked the worked example and product datasheet, converted the insertion-loss equation to editable mathematics, and added measurement, selection and structured guidance.

Source: Knowles Precision Devices