Types of Bursting with a Two-Block Spiking Primitive

arXiv:2611.08045 · eess.SY, cs.SY · Submitted 2026-11-05 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: We're looking at the paper: "Types of Bursting with a Two-Block Spiking Primitive".

Dev: Here's the gist of the paper. Excitable spikers are attractive systems for sparse, event-driven computation This paper asks which bursting classes that primitive can produce with two elementary selectors: a first-order high-pass filter, and a second-order resonant band-pass filter The nonlinearity decides whether the system fires, and where the rheobase sits The selector decides whether rest and repetitive firing can coexist, and therefore which bursting class is available.

Rosa: Primitive Definition The two-block setup is the positive feedback interconnection of a static nonlinearity and a linear frequency selector The system dynamics are defined by the equation y = ϕk(v), v ≜ u+w, w = H(s)y.

Dev: The nonlinearity is described as being monotone increasing, bounded, and odd, with a slope maximal at the origin so that ϕ′k(0) = k.

Rosa: The frequency selector H(s) is assumed to block the constant component of its input with H(0) = 0.

Title and authors: Rosa: We're now looking at the title and authors of "Types of Bursting with a Two-Block Spiking Primitive" again, just to set the stage for what we’ve been talking about.

Dev: It’s by Rosa and Dev, but it pulls in several other authors too, which shows this is a collaborative piece exploring how these basic circuits can be used as design elements.

Rosa: The title itself points us toward the core idea: figuring out the different types of bursting that can be produced when you use this two-block spiking primitive.

Dev: It’s about using that primitive to generate specific firing patterns, and the authors are using these selectors to control those outcomes.

Taro: I wonder how this relates to the other papers we've heard about, especially since we're dealing with autonomous systems and real-world failure modes.

Rosa: It connects because both papers look at how simple components can be used for specific tasks, like our paper looking at event-driven computation versus some of these others looking at safe navigation in cluttered environments.

Dev: Our paper focuses on the internal dynamics of a spiking system, while the other ones deal more with external control or environmental sensing.

Taro: That makes sense because if you can reliably generate a specific bursting class, it might be useful for designing robust circuits that can handle unexpected inputs gracefully.

Rosa: Exactly. If we know which bursting classes are reachable, we know what kind of behavior to expect when the world misbehaves and how the system should recover.

Dev: It gives us a better set of tools for characterizing the limitations and capabilities of these minimal excitable systems before we go into building more complex things.

The paper's summary: Rosa: So, what’s the paper actually summarizing? It boils down to using this two-block setup to classify all the bursting classes that can come out from it.

Dev: Basically, it shows that the nonlinearity determines firing and where the threshold sits, and then the selector determines if rest and repetitive firing can coexist.

Rosa: The main finding is that for any selector blocking constant inputs, you get a unique resting point set by the input alone, which rules out squarewave bursting entirely.

Taro: So that means we don't have to worry about the system just flipping between two fixed states in this specific architecture.

Dev: And without bistability, as with the high-pass filter, you can't get any bursting at all using a single linear slow feedback branch with one state. You need a second slow oscillator for parabolic bursting instead.

Rosa: That’s because the system structure demands it when you can't get that rest and firing coexistence easily.

Taro: It seems like the authors are mapping out exactly what kind of behavior we can expect from this minimal primitive under different conditions.

Dev: They are using frequency-domain tools to derive the onset of firing, rest and firing bistability, and then figure out which bursting classes are supported by each selector.

Rosa: It’s a detailed look at how the selection mechanism is the key decision maker in this system.

The paper's improvements: Rosa: Now let’s talk about what improvements the authors suggest based on their findings, because they don't just stop at classification.

Dev: They suggest that for a few cases, we can actually build systems that produce parabolic bursting by using a piecewise linear nonlinearity and then adding a second copy of the primitive.

Taro: So they are suggesting combining components to get the desired results rather than relying on just one component doing everything perfectly.

Rosa: That’s right. They show that with a piecewise linear nonlinearity, it gives type one excitability, and a second copy of the primitive can then produce parabolic bursting.

Dev: And for elliptic bursting, they suggest employing the second-order resonant band-pass filter and adding a single additional slow linear feedback branch to achieve it.

Taro: So that means the second selector is powerful enough on its own if you just add one more simple loop to get those more complex dynamics.

Rosa: It seems that for elliptic bursting, the paper suggests that a single additional slow linear feedback then gives elliptic bursting, which requires using the resonant band-pass selector and meeting a certain gain condition.

Dev: It really shows how you can engineer these specific firing patterns by combining the elements in a way that exploits those bistability windows they found.

Taro: It’s about taking the classification and turning it into actionable design rules for achieving what we want in hardware, not just theoretical possibilities.

Conclusion: Rosa: So wrapping up this discussion on "Types of Bursting with a Two-Block Spiking Primitive." The takeaway is that the high-pass never makes rest and firing coexist under any nonlinearity, so it only supports bursters that cross the firing band like parabolic ones.

Dev: And for the resonant band-pass, it has a complex pair of poles, and its subcritical Hopf opens up that bistability window that an elliptic burster needs.

Taro: So we can get parabolic bursting by using a piecewise linear nonlinearity and adding a second copy of the primitive to do the job.

Rosa: And for elliptic bursting, you use the resonant band-pass filter and add a single additional slow linear feedback branch, which gives that specific pattern.

Dev: This whole paper is about understanding how the choice of selector dictates whether rest and repetitive firing can coexist, which directly determines which bursting class is available to produce.

Taro: It shows us the boundaries of what this minimal primitive can do before we start designing bigger systems, and that’s important context.

Rosa: That's all on this paper for now. We’ve seen how these simple frequency selectors can be used to create very specific firing patterns, whether it’s parabolic or elliptic or something else.

University of Liege

eess.SY, cs.SY

Submitted: 2026-11-05

Updated: 2026-11-05

Code: https://github.com/arthur-fyon/ECC_2027_Fyon

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 87/100

The gist: Excitable spikers are attractive systems for sparse, event-driven computation This paper asks which bursting classes that primitive can produce with two elementary selectors…

Key concepts

Primitive Definition
This is the basic circuit setup involving two blocks: a static nonlinearity and a linear frequency selector. The system's behavior depends on how these two parts are connected, specifically through the relationship y = phi_k(v) where v is an input signal modified by the selector H(s).
High-Pass Selector Results
Using a high-pass filter results in a unique resting point and prevents squarewave bursting. It does not allow for rest and firing to coexist, meaning it only supports bursting types that cross the firing band, such as parabolic bursting.
Resonant Band-Pass Selector Results
The resonant band-pass filter enables type II excitability. Above a certain gain, both resting and firing can occur simultaneously, allowing a single slow feedback loop to generate elliptic bursting.

Terminology

Summary

Excitable spikers are attractive systems for sparse, event-driven computation This paper asks which bursting classes that primitive can produce with two elementary selectors: a first-order high-pass filter, and a second-order resonant band-pass filter The nonlinearity decides whether the system fires, and where the rheobase sits The selector decides whether rest and repetitive firing can coexist, and therefore which bursting class is available. p. 0

Where the paper says it

This paper asks which bursting classes that primitive can produce with two elementary selectors: a first-order high-pass filter, and a second-order resonant band-pass filter. The nonlinearity decides whether the system fires, and where the rheobase sits.

Primitive Definition

The two-block setup is the positive feedback interconnection of a static nonlinearity and a linear frequency selector The system dynamics are defined by the equation y = ϕk(v), v ≜ u+w, w = H(s)y. The nonlinearity is described as being monotone increasing, bounded, and odd, with a slope maximal at the origin so that ϕ′k(0) = k. p. 1

Where the paper says it

(1) Here, u is the external input, v and y are the input and the output of the nonlinearity, w is the feedback signal, and H(s) the frequency selector. The nonlinearity The nonlinearity ϕk is monotone increasing, bounded, and odd, with a slope maximal at the origin, so that ϕ ′ k (0) = k, (2) Frequency selector H(s) + k u w v y Fig.

The frequency selector H(s) is assumed to block the constant component of its input with H(0) = 0. p. 0
Where the paper says it

A simple sufficient excitability motif has recently been proposed [22], [23]. A single high-gain saturating nonlinearity is closed in positive feedback around a linear frequency selector that blocks the constant component of its input and transmits its variations.

Selector Properties

The analysis shows that for any selector blocking constant inputs, the resting point is unique and set by the input alone The feedback signal w has zero mean at rest and on any periodic orbit of (1). Both selectors lose local stability at the same place, with the threshold depending only on the nonlinearity and given by ϕ′k(u) = 1. p. 2

Where the paper says it

A slow branch that senses w is blind: it reads the same zero mean whether the system rests or fires. The rheobase belongs to the nonlinearity Both selectors lose local stability at the same place.

The global stability below unit gain is guaranteed if ϕ′k(0) < 1 for either selector. p. 2
Where the paper says it

Proposition 5 (Global stability below unit gain). If ϕ ′ k (0) < 1, the resting point of (1) is globally asymptotically stable for every constant input and for either selector.

High-Pass Selector Results

With the high-pass selector, the resting point is unique and rules out squarewave bursting Without bistability, no linear slow feedback with a single state can produce bursting either, necessitating a second slow oscillator for parabolic bursting. Exogenous drive gives parabolic bursting when the external drive u(t) is periodic and slow. p. 3

Where the paper says it

A burster cannot be built by making the system remember; it has to be built by moving the system across the firing band |u| < uth. 1) Exogenous drive gives parabolic bursting: Let u(t) be periodic and slow, of period Tslow ≫ τ.

A single linear slow branch cannot close the burst, as a linear branch fails whether it senses w or y. p. 3
Where the paper says it

2) No single linear slow branch can close the burst: The next step is to replace the external drive by an internal slow branch that moves the operating point of the system. A linear branch fails, whether it senses w or y.

Resonant Band-Pass Selector Results

The resonant selector gives type II excitability with an onset frequency set by the selector alone Above a computable gain, rest and firing coexist, and a single slow negative feedback loop exploits this bistability to produce elliptic bursting. The subcritical onset above that gain creates the window where rest and firing coexist. p. 0

Where the paper says it

With a piecewise linear nonlinearity, it gives type I excitability, and a second copy of the primitive gives parabolic bursting. The resonant band-pass gives a Hopf, type II excitability, and a subcritical onset above a computable gain, hence a window where rest and firing coexist.

For k > 3/2, the bistability window can be exploited by a slow feedback loop to achieve elliptic bursting. p. 5
Where the paper says it

That stretch is the bistability window, and its width is a direct measure of how hard the onset is. Elliptic bursting For k > 3/2, the bistability window of Proposition 9 can be exploited by a slow feedback loop to achieve bursting.

Classification Summary

Table I collects the results that distinguish the two frequency selectors The high-pass never makes rest and firing coexist, whatever the nonlinearity, so it only supports bursters that cross the firing band, such as the parabolic one. The resonant band-pass carries a complex pair of poles, and its subcritical Hopf opens the bistability window that an elliptic burster needs. p. 7

Where the paper says it

The high-pass never makes rest and firing coexist, whatever the nonlinearity, so it only supports bursters that cross the firing band, such as the parabolic one. The resonant band-pass carries a complex pair of poles, and its subcritical Hopf opens the bistability window that an elliptic burster needs.

Square-wave bursting is structurally unavailable as a consequence of the uniqueness of the resting point. p. 6
Where the paper says it

Two limitations are worth stating. Square-wave bursting is structurally unavailable, as a consequence of the uniqueness of the resting point (Proposition 6).

The elliptic regime occupies a narrow region of parameter space whose width is inherited from the bistability window. p. 6
Where the paper says it

Square-wave bursting is structurally unavailable, as a consequence of the uniqueness of the resting point (Proposition 6). And the elliptic regime occupies a narrow region of parameter space, whose width is inherited from the bistability window; mapping such regions is itself an active problem in neuroscience [29].

The classification serves as a design rule as much as a taxonomy. p. 7
Where the paper says it

What that block count buys in an analog realization is still an open question. The classification is a design rule as much as a taxonomy.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improve circuit design by utilizing a minimal two-block primitive to generate specific bursting classes like parabolic and elliptic bursting based on frequency selectors such as a first-order high-pass or second-order resonant band-pass filter.

  2. Design neuromorphic circuits that can selectively produce different firing patterns by choosing the appropriate selector, where The selector decides whether rest and repetitive firing can coexist.

  3. Implement a system capable of generating parabolic bursting from the high-pass selector when paired with a piecewise linear nonlinearity, as shown by the result: With a piecewise linear nonlinearity, it gives type I excitability, and a second copy of the primitive gives parabolic bursting.

  4. Create systems that exhibit elliptic bursting by employing the second-order resonant band-pass filter and a single additional slow linear feedback branch: A single additional slow linear feedback then gives elliptic bursting.

  5. Develop event-based vibration monitoring or acoustic transient detection systems by tuning banks of resonant filters, as the paper notes: A bank of Hr filters tuned to different center frequencies detects narrow-band events directly in spikes.

Abstract

Excitable spikers are attractive systems for sparse, event-driven computation. A spiking primitive composed of a linear frequency selector closed in positive feedback around a high-gain saturating nonlinearity has recently been proposed as a minimal excitable system. This paper asks which bursting classes that primitive can produce with two elementary selectors: a first-order high-pass filter, and a second-order resonant band-pass filter. The nonlinearity decides whether the system fires, and where the rheobase sits. The selector decides whether rest and repetitive firing can coexist, and therefore which bursting class is available. We show that the high-pass never makes rest and firing coexist, whatever the nonlinearity, and that no linear slow feedback branch can then produce bursting. With a piecewise linear nonlinearity, it gives type I excitability, and a second copy of the primitive gives parabolic bursting. The resonant band-pass gives a Hopf, type II excitability, and a subcritical onset above a computable gain, hence a window where rest and firing coexist. A single additional slow linear feedback then gives elliptic bursting. We also show that the resting point is unique for both selectors, which makes square-wave bursting structurally unavailable. Closed-form thresholds, firing rates, and criticality conditions are given and validated numerically.

Sources

Related papers