Infinite impulse response¶
An IIR filter has an ideal impulse response with nonzero support unbounded in time, commonly realized through persistent physical state or digital recursion.
Core Idea¶
An infinite impulse response (IIR) filter is a linear time-invariant signal filter whose ideal response to a unit impulse has nonzero values over an unbounded stretch of time. For a causal filter, the tail extends arbitrarily far into the future; a noncausal filter can have unbounded support in the negative-time direction. This is a property of the complete input–output response, not simply of the circuit or computer program that produces it. An IIR response can decay and be stable, but neither is required by the definition.[^ref-678ba113b7bd]
A digital recurrence and an analog circuit can both meet this test. A first-order digital low-pass with \(y[n]=a\,y[n-1]+x[n]\) and \(0<a<1\) has \(h[n]=a^n u[n]\). An RC high-pass has an exponential response after the immediate impulse. The first stores a previous output; the second stores energy in a capacitor. Both have a nonzero ideal tail.[ref-678ba113b7bd][ref-b86ebe355100]
Scope of Application¶
Use IIR for an LTI filter when its impulse response is not confined to a finite interval. In digital signal processing, a difference equation and its uncancelled modes often reveal the tail. In analog signal processing, a circuit transfer function and stored physical state can reveal it. A low-pass or high-pass goal alone does not decide IIR versus FIR; different filter designs can pursue the same goal.[ref-678ba113b7bd][ref-b86ebe355100]
Design guides often compare IIR with FIR on memory, calculation cost, phase and possible instability. These comparisons help choose a realization for specified requirements, but no one advantage or limitation is an identity condition for every IIR filter.[^ref-b894680d90e8]
Clarity¶
The impulse response is the output after a unit impulse under declared LTI and zero-state conventions. A finite impulse response (FIR) becomes exactly zero outside some finite time interval. An IIR has nonzero support that no finite interval can contain. For a causal example, look for nonzero output at arbitrarily late positive times; for a noncausal example, inspect both time directions. This is an ideal mathematical test, so a small number rounded to zero by a machine does not necessarily make the underlying model FIR.[^ref-678ba113b7bd]
For \(y[n]=a\,y[n-1]+x[n]\), the causal response is \(a^n u[n]\). At \(a=0.9\), it persists while shrinking. At \(a=0\), it reduces to a single impulse and is FIR. At \(|a|\geq1\), it remains IIR for nonzero \(a\) but is not BIBO stable. Response duration and stability answer different questions.[^ref-678ba113b7bd]
Manages Complexity¶
A recurrence with one stored value and one coefficient can represent an ideal response containing indefinitely many nonzero samples. An RC time constant similarly summarizes a continuous exponential tail. These compact descriptions make it possible to reason about long responses without listing every moment.[ref-678ba113b7bd][ref-b86ebe355100]
The diagram must still be interpreted carefully. A recursive FIR implementation can have internal poles that cancel zeros exactly, leaving a finite overall impulse response. Compute or infer the whole filter's response before classifying it. Then check stability and design constraints separately.[ref-74c61f503d2b][ref-678ba113b7bd]
Abstract Reasoning¶
First state the time domain, LTI assumption and initial-state convention. Derive the complete impulse response from a difference equation, differential equation or transfer function, accounting for exact cancellations. Ask whether its nonzero support fits within a finite interval. If yes, call the response FIR; if no, call it IIR. After classification, check causality, stability, phase or implementation needs as separate properties.[ref-678ba113b7bd][ref-74c61f503d2b]
This order blocks three misleading shortcuts: feedback does not guarantee IIR because it may cancel; a decaying tail does not define IIR because the defining issue is duration; and a finite measurement window cannot prove an ideal response ends.[ref-74c61f503d2b][ref-678ba113b7bd]
Knowledge Transfer¶
A causal digital low-pass and a causal analog RC high-pass use unlike devices and different time variables. In each, however, an impulse probes an LTI filter and exposes a nonzero tail extending indefinitely forward. The same response-duration test transfers between sampled signals and continuous circuits, even though the implementation, frequency objective and physical state differ.[ref-678ba113b7bd][ref-b86ebe355100]
The named filter does not transfer to every ordinary process that has a lasting consequence. Its technical identity needs a signal, an LTI response, an impulse probe and an exact support test. It is a strict child of live Filter (Signal Processing) and Linear Time-Invariant System. Through those parents it inherits portable Prime Transformation and Prime Linearity roles, without becoming a Prime itself.
Example¶
Digital low-pass. Greenberg and Delgutte give the causal recurrence \(y[n]=a\,y[n-1]+x[n]\); at \(a=0.9\), its impulse response is \(0.9^n u[n]\). Mapped back: the fixed recurrence is the LTI filter; \(x[n]=\delta[n]\) is the probe; nonzero values for every \(n\geq0\) are the unbounded support; the delayed output is the state realization; and \(|a|<1\) explains stability for this example alone.[^ref-678ba113b7bd]
Analog RC high-pass. Rowell gives \(H(s)=\tau s/(1+\tau s)\), with \(\tau=RC>0\). Algebra and a causal inverse Laplace transform give \(h(t)=\delta(t)-e^{-t/\tau}u(t)/\tau\); this impulse formula is a curator derivation from the published transfer function. Mapped back: the RC transfer is the LTI filter; a Dirac impulse is the probe; its exponential part is nonzero for every finite \(t>0\), giving unbounded support; capacitor charge is the state realization; and positive \(R,C\) give a decaying tail in this case. The analog high-pass is unlike the digital low-pass while sharing the IIR test.[^ref-b86ebe355100]
Relationships to Other Abstractions¶
Current abstraction Infinite impulse response Domain-specific
Parents (2) — more general patterns this builds on
-
Infinite impulse response is a kind of Filter (Signal Processing) Domain-specific
An IIR filter is a signal-conditioning filter distinguished by unbounded impulse-response support.
-
Infinite impulse response is a kind of Linear time-invariant system Domain-specific
The IIR classification depends on an LTI operator characterized by its impulse response.
Hierarchy paths (2) — routes to 2 parentless roots
- Infinite impulse response → Filter (Signal Processing) → Transformation → Function (Mapping)
- Infinite impulse response → Linear time-invariant system → Linearity
Neighborhood in Abstraction Space¶
Infinite impulse response sits in a sparse region of the domain-specific corpus (87th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Wavelets & Function-Space Approximation (19 abstractions)
Nearest neighbors
- Linear time-invariant system — 0.81
- Polyphase matrix — 0.81
- Non-separable wavelet — 0.81
- Lulu smoothing — 0.81
- Impulse vector — 0.80
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
An FIR filter has a finite overall impulse response, even if it uses an exactly canceling recursive implementation. Feedback is common in digital IIR, yet an analog circuit can have an IIR tail without a visible output-return wire. Stability is separate: an IIR can be unstable. Causality is separate too: the nonzero support of a noncausal IIR may extend into negative time. A late response falling below numerical precision is not the exact zero needed to establish FIR.[ref-678ba113b7bd][ref-74c61f503d2b][^ref-b86ebe355100]
References¶
[^ref-678ba113b7bd]: Julie Greenberg and Bertrand Delgutte, “Biomedical Signal and Image Processing,” Harvard–MIT Division of Health Sciences and Technology course materials, MIT OpenCourseWare, Spring 2007, Chapter 2 “Digital Filters”. §§2.1.1–2.1.3 and 2.2.1–2.2.2, printed pp.2–5, 8; Fig.2.5 printed p.13; §2.3.2 printed p.14. Consulted for the LTI impulse/convolution identity, finite versus infinite response, anticausal boundary, geometric low-pass example and stability limits. https://ocw.mit.edu/courses/hst-582j-biomedical-signal-and-image-processing-spring-2007/b60ddec52ee7fc26607f8afc8a77daf5_ch2_dfilt.pdf
[^ref-b86ebe355100]: Derek Rowell, “Introduction to Time-Domain Digital Signal Processing,” Lecture 13 of 2.161 Signal Processing: Continuous and Discrete, Massachusetts Institute of Technology, Fall 2008, printed pp.13-1–13-2/PDF pp.1–2. Printed p.13-1 gives the continuous RC high-pass \(H(s)=RCs/(RCs+1)\) and \(\tau=RC\); the displayed inverse-Laplace impulse formula in this entry is a curator derivation from that source, not quoted source text. https://ocw.mit.edu/courses/2-161-signal-processing-continuous-and-discrete-fall-2008/13532904b5dcaa23f42cb00bb1b79ad4_lecture_13.pdf
[^ref-74c61f503d2b]: MIT OpenCourseWare, “Network Structures for Finite Impulse Response (FIR) Systems and Parameter Quantization Effects in Digital Filter Structures,” Lecture 13 transcript in Digital Signal Processing RES.6-008, instructor Alan V. Oppenheim, Spring 2011, PDF pp.5–6. The instructor explains a recursive frequency-sampling FIR realization whose internal poles cancel zeros so the overall response remains finite; the transcript, rather than a formal journal article, is the consulted source. https://ocw.mit.edu/courses/res-6-008-digital-signal-processing-spring-2011/4260578ccde71f509549a31a9d2e3e97_AsSsGjaBbas.pdf
[^ref-b894680d90e8]: Analog Devices, “Mixed-Signal and DSP Design Techniques,” Analog Devices 2000 edition, Section 6 “Digital Filters” by Walt Kester, printed pp.6.5–6.6 and 6.30/PDF zero-index pp.6–7 and 31, especially Fig.6.38. Manufacturer design guide consulted for bounded FIR/IIR implementation, phase and stability comparisons; no such tendency is treated as an IIR identity condition. https://www.analog.com/media/en/training-seminars/design-handbooks/MixedSignal_Sect6.pdf