Filter Types Overview

A filter is a circuit that allows some frequencies to pass while blocking others. That simple description covers an enormous range of devices — from a $0.10 capacitor wired across a power rail to a precision crystal filter machined from quartz, from a two-component audio tone control to a seven-pole filter protecting a sensitive input from powerful nearby transmitters. Understanding what the different types of filters do and why they are used is the essential foundation before diving into specific filter designs.

You rely on filters constantly, even when you aren't aware of them. The low-pass filter in front of a sound card's analog-to-digital converter stops ultrasonic noise from folding back into the audio. The band-pass filter in a Wi-Fi module lets the 2.4 GHz band through and keeps out phone and broadcast signals. The crossover in a loudspeaker sends bass to the woofer and treble to the tweeter. The EMI filter on a power inlet keeps switching noise in the equipment and out of the mains. Filters are the guardians of signal quality and the gatekeepers of selectivity.

The Four Filter Types

Every filter belongs to one of four fundamental categories, defined by which part of the frequency spectrum it passes and which it blocks. These categories apply regardless of whether the filter is built from coils and capacitors, quartz crystals, digital signal processing code, or a metal cavity at microwave frequencies.

The four fundamental filter types: low-pass, high-pass, band-pass, and band-stop (notch). The shaded regions show the stopband; the flat regions show the passband.

View Larger

Low-Pass Filter (LPF)

A low-pass filter passes signals at frequencies below its cutoff frequency (f_c) and attenuates signals above f_c. Think of it as a drain strainer that blocks large particles but lets water flow freely. Low-pass filters are everywhere: in front of every analog-to-digital converter (the anti-aliasing filter), after every digital-to-analog converter and class-D amplifier (to remove switching and sampling products), and after RF oscillators and wireless transmitter chips, whose output contains harmonics at 2×, 3× and higher multiples of the wanted frequency. A 433 MHz transmitter module, for example, also produces energy at 866 MHz and 1.3 GHz; radio regulations limit how much, and a low-pass filter with its cutoff just above 433 MHz passes the fundamental while rolling off the harmonics.

A practical low-pass filter for a 7 MHz oscillator would have its cutoff at roughly 10 MHz. Any signal below 10 MHz passes through with minimal loss (typically under 0.2 dB). At 14 MHz (the second harmonic) the filter provides perhaps 30–40 dB of attenuation depending on the design and number of poles. The rolloff slope beyond f_c is determined by the filter order: a first-order filter rolls off at 20 dB per decade (6 dB per octave); a third-order filter at 60 dB per decade; a seventh-order filter at 140 dB per decade.

High-Pass Filter (HPF)

A high-pass filter is the mirror image of the low-pass: it passes signals above its cutoff frequency and attenuates those below. High-pass filters block DC and low-frequency noise: the rumble filter in a phono preamplifier, the high-pass section of a loudspeaker crossover feeding the tweeter, and the filter at the input of a VHF or UHF receiver module that keeps strong AM and FM broadcast signals from overloading it. A device located near a broadcast transmitter may see signals of +30 dBm or more at its input; without a high-pass filter, those signals saturate the first amplifier and leave the device unable to hear anything else.

Another common application is the AC coupling capacitor in an audio amplifier chain. A single capacitor in series with the signal path acts as a first-order high-pass filter; nothing above its cutoff frequency is blocked, but DC bias is rejected and the corner frequency can be set to filter out low-frequency hum or rumble.

Band-Pass Filter (BPF)

A band-pass filter passes a defined band of frequencies and attenuates everything above and below that band. The passband is centered on a center frequency f₀ and has a defined bandwidth, usually measured at the −3 dB points. Band-pass filters are the most important filter type in any system that has to pick one signal out of many. Every superheterodyne receiver has at least one — the IF filter — and wireless modules use band-pass filters at their inputs to reject signals outside their band.

Band-pass filters are just as common outside radio. A tone decoder or a vibration-monitoring sensor uses a band-pass filter to pick out one frequency; a telephone line uses 300–3,400 Hz for speech; and a graphic equalizer is simply a bank of band-pass filters, each controlling one slice of the audio spectrum.

Band-Stop Filter (BSF) / Notch Filter

A band-stop filter, also called a notch filter or band-reject filter, attenuates a specific band of frequencies while passing everything above and below. It is the complement of the band-pass filter: the passband is everywhere except in the notch. Notch filters eliminate specific interferers — 50 or 60 Hz mains hum in a measurement or ECG amplifier, a switching supply's noise at a known frequency, or acoustic feedback at one frequency in a PA system. Many audio processors include an automatic notch that tracks and removes a persistent tone.

The notch must be deep — often 40 dB or more of attenuation in a narrow bandwidth — to remove the interference without disturbing the wanted signal on either side. A twin-T notch filter tuned to 60 Hz is a classic example: it removes mains hum from a sensor signal while leaving the rest of the measurement untouched.

Key Filter Parameters

When selecting or designing a filter, several parameters define its performance. Understanding these parameters lets you compare filters intelligently and match the filter to the application.

Cutoff Frequency (f_c)

The cutoff frequency is the boundary between the passband and the transition to the stopband. By convention, it is defined as the frequency at which the filter's insertion loss reaches 3 dB — i.e., where the output power drops to half the input power (or output voltage to 0.707 of input). For a low-pass filter cleaning up a 7.0–7.3 MHz oscillator, f_c might be set at 9 MHz, comfortably above the highest wanted frequency (7.3 MHz) but well below the second harmonic at 14.6 MHz.

For band-pass filters, the relevant frequency is the center frequency f₀. For a crystal filter at 9.0 MHz with 2.4 kHz bandwidth, f₀ = 9.000 MHz, and the 3 dB bandwidth runs from approximately 8.9988 to 9.0012 MHz.

Insertion Loss

Insertion loss is the attenuation introduced by the filter in its passband — the signal level reduction from input to output caused by resistive losses in the filter components themselves, not by deliberate signal rejection. A perfect filter would have zero insertion loss in the passband and infinite attenuation in the stopband. Real filters fall short of this ideal because real inductors and capacitors have resistance (the inductor Q is finite, never infinite).

For a high-Q LC filter — such as a 7-pole low-pass filter wound on toroidal cores — insertion loss might be as low as 0.1 to 0.3 dB. For a crystal filter, insertion loss is typically 1 to 3 dB. For an active filter using an op-amp, loss in the passband can be zero (the op-amp provides gain), but the filter cannot handle large power levels or very high frequencies.

Passband Ripple

An ideal filter would have a perfectly flat response in the passband — the same gain at every frequency within the pass region. In practice, some filter designs intentionally allow the passband response to ripple up and down by a specified amount in order to achieve a sharper transition to the stopband. Butterworth filters have zero passband ripple (maximally flat). Chebyshev filters allow ripple, typically specified as 0.1 dB, 0.5 dB, or 1 dB. Elliptic filters have ripple in both the passband and the stopband. For audio applications, 0.5 dB of ripple is typically inaudible. For precision measurement, where the filter's gain affects the reading, even 0.1 dB of ripple (about 1%) may matter.

Stopband Attenuation

This is how much the filter suppresses signals in the stopband — the region beyond the transition band that the filter is supposed to block. It is measured in dB and must meet the system requirement. For an anti-aliasing filter in front of a 12-bit ADC, for example, any signal above half the sample rate must be reduced below the converter's resolution — about 72 dB. For a filter cleaning up an oscillator's harmonics, 40 to 60 dB of stopband attenuation is a typical target, with a comfortable safety margin.

Shape Factor

Shape factor describes how quickly a filter transitions from passband to stopband. It is defined as the ratio of the −60 dB bandwidth to the −6 dB bandwidth:

Shape factor matters most where wanted and unwanted signals sit close together. A low shape factor means the filter transitions sharply from full output to full attenuation, allowing signals very close to the wanted channel to be rejected without distorting the wanted signal. Crystal, ceramic and SAW filters achieve the best shape factors; LC filters are generally inferior, which is why they are used for broad tasks such as harmonic suppression rather than narrow selectivity.

Group Delay and Phase Response

Group delay is the time it takes for different frequency components to travel through the filter. An ideal filter has constant group delay across its passband (linear phase) so that all frequency components of a signal are delayed by the same amount and their timing relationships are preserved. Filters with non-constant group delay introduce phase distortion that can smear transient signals.

For speech and music, some phase distortion is acceptable because the ear is relatively insensitive to phase. For digital data and pulses, phase distortion smears each symbol into its neighbors (intersymbol interference) and raises the error rate, and it distorts the shape of square waves and pulses on an oscilloscope. Bessel filters have maximally flat group delay (the most constant group delay of any classical filter family) at the expense of gentle rolloff. Butterworth and Chebyshev filters have worse group delay near the cutoff frequency. Elliptic filters have the most uneven group delay. When filtering data or pulses, group delay should be considered alongside stopband attenuation.

The Three Classical Filter Families

Filter theory has produced many different mathematical optimization criteria, each representing a different choice about what to sacrifice and what to preserve. Three families dominate practical filter design, plus a fourth for pulse and data work, and understanding their trade-offs is essential before diving into component values and design tables.

Comparing the three classical filter families at the same filter order. The elliptical filter achieves the sharpest transition but at the cost of ripple in both passband and stopband.

View Larger

Butterworth — Maximally Flat Magnitude

The Butterworth filter, described by British engineer Stephen Butterworth in 1930, optimizes for a perfectly flat passband response. At no frequency within the passband does the response ripple; it decreases monotonically as frequency increases, reaching −3 dB at f_c and then rolling off into the stopband. The rolloff slope in the stopband is −20n dB per decade, where n is the filter order. A third-order Butterworth rolls off at −60 dB/decade; a fifth-order at −100 dB/decade.

The Butterworth filter's great strength — its perfectly flat passband — is also its weakness. By insisting on zero passband ripple, it must use a gentle transition from passband to stopband. For a given order n, it achieves less stopband attenuation at any given frequency ratio than a Chebyshev or elliptic filter. Where simplicity and a clean passband matter more than a sharp cutoff, the Butterworth is the natural choice: audio filters, sensor signal conditioning, and harmonic filters where the harmonics are well above the cutoff.

Chebyshev — Equiripple Passband

The Chebyshev filter (named after the Russian mathematician Pafnuty Chebyshev) accepts a specified amount of equal-amplitude ripple in the passband in exchange for a significantly sharper transition to the stopband. The passband response oscillates between two bounds — (0 dB and −R_dB) where R is the ripple specification — but outside those bounds, the stopband attenuation is substantially greater than a Butterworth of the same order.

As a concrete example, consider a 5th-order low-pass filter with f_c = 35 MHz cleaning up a 28–30 MHz oscillator. At 70 MHz the normalized frequency ratio is 2:1. A 5th-order Butterworth provides about 30 dB of attenuation at that point. A 5th-order Chebyshev with 0.5 dB passband ripple provides about 42 dB. If you need 43 dB of harmonic suppression and are using a Butterworth, you may need to step up to a 6th or 7th order; the Chebyshev achieves the same result with fewer poles and fewer components.

The practical cost of Chebyshev ripple is modest for most applications. In a harmonic filter, 0.5 dB of passband ripple means the filter slightly favors some frequencies over others within the passband — an effect that is negligible for a single-frequency signal. For a precision test signal path or a wideband data link where amplitude flatness matters, Chebyshev ripple may be unacceptable, but for protective harmonic filtering it is an excellent trade.

Elliptical (Cauer) — Equiripple in Both Bands

The elliptical filter, also called the Cauer filter after the German mathematician Wilhelm Cauer, goes further still. It accepts ripple in both the passband and the stopband in exchange for the absolute minimum transition bandwidth for a given filter order. Elliptical filters achieve their sharp rolloff by placing transmission zeros — specific frequencies at which the filter output is theoretically zero — in the stopband. These zeros correspond to parallel resonant circuits or bridging elements in the filter topology that create deep notches at specific frequencies.

An elliptic filter can achieve a given attenuation level at a closer frequency ratio than either Butterworth or Chebyshev, meaning you can use fewer components to meet a given specification. For an anti-aliasing filter that must pass 20 kHz audio and reject 24 kHz by 60 dB, an elliptic response may permit a 7th-order design where a Chebyshev would need 12 or more. The penalties are complexity (extra bridging components), sensitivity to component tolerances, and non-monotonic stopband response (the attenuation has guaranteed minimum and maximum values, but is not uniformly high across the entire stopband — it is "guaranteed at least X dB everywhere beyond the transition band").

Bessel — Linear Phase

The Bessel filter (named after Friedrich Bessel) is optimized for constant group delay across the passband, meaning all frequencies within the passband are delayed by the same amount. This preserves the shape of transient signals passing through the filter — important in pulse and digital applications, and in oscilloscope and data acquisition front ends where a square wave must not overshoot. The trade-off is a very gentle rolloff, even softer than Butterworth. Bessel filters are rarely used for sharp selectivity but appear in anti-aliasing filters and measurement amplifiers where waveform shape matters.

How Filters Are Built

The same mathematical filter response can be physically realized in many different ways, each suited to a different frequency range, power level, and cost point.

LC (Inductor-Capacitor) Filters

LC filters are the workhorses of filter design from audio frequencies up through a few hundred megahertz. They consist of inductors and capacitors arranged in a ladder network. Inductors resist changes in current (they pass low frequencies easily but impede high frequencies); capacitors resist changes in voltage (they pass high frequencies but block DC and low frequencies). By arranging inductors in series and capacitors in shunt — or vice versa — you create low-pass or high-pass characteristics. Combining both produces band-pass and band-stop responses.

The primary limitation of LC filters at higher frequencies is inductor Q. A toroidal inductor might have a Q of 100 to 200 at a few megahertz. At 500 MHz, achieving that Q requires a tiny air-wound coil with only a few turns, and the self-capacitance of the coil begins to dominate at even higher frequencies. At the other end, at audio frequencies, inductors become large, heavy and expensive, which is why audio filters are usually active. Passive LC filters dominate from a few hundred kilohertz to a few hundred megahertz, and in speaker crossovers and power-line EMI filters, where they must handle real power.

Crystal Filters

Quartz crystals are mechanical resonators that exhibit very high Q — typically 20,000 to 100,000 for AT-cut crystals in a few-MHz to few-tens-of-MHz frequency range. A crystal resonator has both a series resonant frequency and a parallel resonant frequency a few hundred ppm apart, which allows both passband and rejection to be engineered precisely. Crystal filters are used where the required bandwidth is far too narrow (a few hundred hertz to a few kilohertz) to achieve with LC components, typically in receivers and frequency-selective measurement equipment working at 450 kHz to a few tens of megahertz.

Ceramic and SAW Filters

Ceramic resonators work on the same principle as quartz with lower Q and lower cost: the 10.7 MHz and 455 kHz filters in every FM and AM broadcast receiver are ceramic. Above about 100 MHz, surface acoustic wave (SAW) and bulk acoustic wave (BAW) filters take over. A SAW filter launches an acoustic wave across the surface of a piezoelectric chip through finger-shaped electrodes whose spacing sets the frequency. These tiny filters provide sharp band-pass responses in phones, Wi-Fi and Bluetooth modules, GPS receivers and key fobs, at frequencies from about 100 MHz to several gigahertz.

Active Filters

Active filters use op-amps or other amplifiers in combination with resistors and capacitors to realize filter functions. They can achieve any desired response up to the bandwidth limit of the amplifier, without using inductors (which are bulky, expensive, and lossy at audio frequencies). Active filters dominate audio and sensor work: anti-aliasing filters in front of ADCs, tone controls and equalizers, hum notches, and the band-pass filters in tone decoders. They cannot be used much above a few megahertz because op-amps run out of gain-bandwidth product, and they cannot handle significant power.

Digital / DSP Filters

Microcontrollers, DSP chips and FPGAs implement filters in software. A digital FIR or IIR filter running at 24 bits and a 100 kHz sample rate can achieve filter responses that would be physically impossible to build in hardware — a band-pass filter with a perfectly flat passband, extremely steep skirts and enormous stopband attenuation. The practical limits are the sample rate (which determines the maximum frequency the filter can handle) and the number of taps (which determines sharpness). Digital filters still need an analog anti-aliasing filter in front of the ADC, because the digital filter cannot remove frequencies that were already folded down by sampling. DSP filtering is covered in the Digital Electronics module.

Where Filters Appear in Real Equipment

Let's look inside a few everyday products and identify the filters in each:

An Audio Interface or Sound Card

At the input, a high-pass filter (often just the coupling capacitor) removes DC and subsonic rumble. Before the ADC, an anti-aliasing low-pass filter removes everything above about 20 kHz. Inside the converter chip, a digital decimation filter with a very steep response does most of the work. On the output side, a reconstruction low-pass filter after the DAC smooths the stepped output and removes sampling images.

A Wi-Fi or Bluetooth Module

Between the chip and the antenna connector sits a band-pass filter (usually a SAW or ceramic part) that passes 2.400–2.4835 GHz and rejects mobile-phone and other strong signals, and a low-pass harmonic filter that keeps the transmitter's harmonics within the legal limits. A matching network — usually an L or Pi network — matches the chip to the 50-ohm antenna line.

A Loudspeaker

The crossover is a combination of a low-pass filter feeding the woofer and a high-pass filter feeding the tweeter (and a band-pass filter for a midrange driver in a three-way system), each built from inductors and capacitors that carry the full amplifier power.

A Switching Power Supply

At the mains inlet, an EMI filter — common-mode chokes and X and Y capacitors — keeps switching noise out of the mains wiring. On the output, an LC low-pass filter removes switching ripple, and bypass capacitors throughout the circuit keep high-frequency noise off the DC supply lines.

How to Choose the Right Filter

When faced with a filtering problem, the selection process follows a logical sequence:

Step 1: Identify the type. What frequencies must pass and what must be blocked? If you need to suppress harmonics or noise above a certain frequency, you need a low-pass filter. If you need to select one band of frequencies from everything else, you need a band-pass filter. If you need to eliminate a single interfering frequency, you need a band-stop notch.

Step 2: Define the requirements. What is the passband frequency range? What is the minimum stopband attenuation at the critical interference frequency? What is the maximum acceptable insertion loss in the passband? What ripple is acceptable?

Step 3: Choose the implementation technology. What frequency range? What power level? What physical size and cost constraints? Active RC filters suit audio and sensor work up to a few hundred kilohertz. LC filters work excellently from a few hundred kilohertz to a few hundred megahertz, at power levels from milliwatts to kilowatts. Crystal and ceramic filters are required when bandwidths of a few kilohertz are needed at megahertz frequencies. SAW and BAW filters handle sharp band-pass filtering above about 100 MHz. Digital filters suit anything that is already being sampled.

Step 4: Choose the filter family. If insertion loss and passband flatness are paramount, use Butterworth. If you need sharper rolloff with acceptable passband ripple, use Chebyshev. If you need the absolute sharpest transition and can accept complex topology, use elliptic. If constant group delay matters, consider Bessel.

Step 5: Determine the required order (number of poles). The order is the minimum number that provides the required stopband attenuation at the specified frequency ratio. Filter design tables (available in textbooks and many online resources) give component values for each order and each family. The next three lessons cover this process in detail for each family.