Two-sided Laplace transform
The two-sided Laplace transform, also known as the bilateral Laplace transform, is an integral transform that converts a time-domain function f(t) defined for all real t into a complex frequency-domain representation F(s), given by the formula F(s) = \int_{-\infty}^{\infty} f(t) e^{-st} \, dt, where s = \sigma + j\omega is a complex variable and the integral must converge absolutely in a specified region.[1] This transform generalizes the more common one-sided (unilateral) Laplace transform, which integrates only from t = 0 to \infty and assumes causality (i.e., f(t) = 0 for t < 0), by accommodating signals and systems that extend to negative time, such as non-causal or two-sided functions in signal processing.[2] The two-sided Laplace transform emerged in the early 20th century as an extension of Pierre-Simon Laplace's original unilateral transform from 1774–1785, with significant development attributed to the independent efforts of mathematicians like Balth van der Pol and H. Bremmer in their 1950 work Operational Calculus Based on the Two-Sided Laplace Integral Transform, which unified it with the Fourier transform for broader analytical applications.[3] Its convergence is determined by a region of convergence (ROC) in the complex s-plane, typically a vertical strip bounded by the real parts \sigma where the integral exists; for right-sided signals (decaying for t > 0), the ROC is a right half-plane (\sigma > \sigma_0); for left-sided signals, a left half-plane (\sigma < \sigma_0); and for two-sided signals, an annular strip (\sigma_1 < \sigma < \sigma_2).[2] If the ROC includes the imaginary axis (\sigma = 0), the transform reduces to the Fourier transform, highlighting its role as a bridge between time-domain analysis and frequency-domain methods.[4] Key properties mirror those of the unilateral transform but account for the full time domain, including linearity (\mathcal{L}\{a f(t) + b g(t)\} = a F(s) + b G(s), with ROC at least the intersection of individual ROCs); time shifting (\mathcal{L}\{f(t - t_0)\} = e^{-s t_0} F(s), ROC unchanged); differentiation in time (\mathcal{L}\{\frac{df}{dt}\} = s F(s), assuming boundary terms vanish, ROC including the original); and convolution (\mathcal{L}\{f * g\} = F(s) G(s), ROC intersection).[2] The inverse transform is given by the Bromwich integral: f(t) = \frac{1}{2\pi j} \int_{\sigma - j\infty}^{\sigma + j\infty} F(s) e^{st} \, ds, where \sigma lies in the ROC, often computed via residue theorem or partial fractions for rational functions.[5] In applications, the two-sided transform is essential in signals and systems analysis for modeling linear time-invariant (LTI) systems, including non-causal filters and anti-causal signals, where the system function H(s) determines stability (ROC includes the j\omega-axis for BIBO stability) and causality (ROC to the right of the rightmost pole).[2] It facilitates solving differential equations algebraically, as in control theory and circuit analysis, and extends to advanced areas like fractional-order systems and biomedical signal processing for stability assessment via pole-zero plots.[4][3]Fundamentals
Definition
The two-sided Laplace transform, also known as the bilateral Laplace transform, is an integral transform that maps a time-domain function f(t) defined for all real t to a complex frequency-domain function F(s). It is defined by the integral F(s) = \int_{-\infty}^{\infty} f(t) e^{-st} \, dt, where s = \sigma + j\omega is a complex variable, with \sigma denoting the real part (often interpreted as a damping factor) and \omega the imaginary part (representing angular frequency).[6][7] This formulation integrates over the entire real line from -\infty to \infty, distinguishing it from the one-sided (unilateral) Laplace transform, which integrates only from 0 to \infty.[8] The standard notation for the transform is \mathcal{L}\{f(t)\} = F(s), where the uppercase F(s) represents the transform and the lowercase f(t) the original function (with the inverse transform often denoted \mathcal{L}^{-1}\{F(s)\} = f(t)./13:_Laplace_Transform/13.02:_Laplace_transform) For the integral to exist, it must converge, which requires absolute integrability: \int_{-\infty}^{\infty} |f(t) e^{-\sigma t}| \, dt < \infty for values of \sigma in a certain strip in the complex plane known as the region of convergence.[8] This region comprises the set of s values for which the transform is well-defined.[6]Region of Convergence
The region of convergence (ROC) for the two-sided Laplace transform of a function f(t) is defined as the set of complex values s = \sigma + j\omega for which the integral \int_{-\infty}^{\infty} |f(t) e^{-st}| \, dt < \infty.[9] This condition ensures absolute convergence of the transform, distinguishing it from pointwise convergence.[10] Geometrically, the ROC corresponds to a vertical strip in the complex s-plane, bounded by vertical lines \operatorname{Re}(s) = \sigma_1 and \operatorname{Re}(s) = \sigma_2 (with \sigma_1 < \sigma < \sigma_2) where the integral converges.[11] For right-sided signals where f(t) = 0 for t < 0, the ROC is a right half-plane \operatorname{Re}(s) > \sigma_{\max}, determined by the rightmost pole.[9] In contrast, for left-sided signals where f(t) = 0 for t > 0, the ROC is a left half-plane \operatorname{Re}(s) < \sigma_{\min}, bounded by the leftmost pole.[10] Two-sided signals, nonzero on both sides of t=0, yield a strip ROC as the intersection of the individual half-planes from their components.[11] Within the ROC, the transform F(s) is holomorphic, meaning it is analytic and free of singularities.[9] The ROC excludes all poles of F(s), which are the singularities arising from the denominator in rational representations, though the boundaries of the ROC may contain poles.[10] Zeros of F(s) do not directly influence the ROC boundaries but affect the overall transform behavior inside it.[11] For example, the signal f(t) = e^{-at} u(t) (right-sided exponential decay with a > 0) has ROC \operatorname{Re}(s) > -a, a right half-plane excluding the pole at s = -a.[9] Similarly, f(t) = -e^{bt} u(-t) (left-sided with b > 0) yields ROC \operatorname{Re}(s) < b, a left half-plane.[10] A two-sided example, f(t) = e^{-2t} u(t) - e^{3t} u(-t), combines these to form a strip ROC -2 < \operatorname{Re}(s) < 3, bounded by poles at s = -2 and s = 3.[11]Properties
Basic Operations
The two-sided Laplace transform exhibits several fundamental properties that facilitate algebraic manipulations in the s-domain, analogous to those of the one-sided version but applicable to signals defined over the entire real line. These basic operations include linearity, time and frequency shifts, time scaling, differentiation with respect to time and the complex frequency s, and initial and final value theorems, each derived directly from the defining integral F(s) = \int_{-\infty}^{\infty} f(t) e^{-st} \, dt, where the region of convergence (ROC) is the vertical strip in the s-plane ensuring absolute integrability. These properties are essential for analyzing noncausal signals and systems.[12] Linearity is the most straightforward property, stating that for constants a and b, and functions f(t) and g(t) with transforms F(s) and G(s),\mathcal{L} \left\{ a f(t) + b g(t) \right\} = a F(s) + b G(s),
with the ROC being at least the intersection of the individual ROCs (potentially larger if pole-zero cancellations occur). To derive this, substitute into the integral:
\mathcal{L} \left\{ a f(t) + b g(t) \right\} = \int_{-\infty}^{\infty} \left[ a f(t) + b g(t) \right] e^{-st} \, dt = a \int_{-\infty}^{\infty} f(t) e^{-st} \, dt + b \int_{-\infty}^{\infty} g(t) e^{-st} \, dt = a F(s) + b G(s),
leveraging the linearity of the integral operator, provided s lies in the common ROC.[12][13] The time-shifting property addresses delays or advances in the time domain: for a real constant t_0,
\mathcal{L} \left\{ f(t - t_0) \right\} = e^{-s t_0} F(s),
with the ROC unchanged. The derivation follows by substitution: let u = t - t_0, so dt = du and the limits remain -\infty to \infty; then
\int_{-\infty}^{\infty} f(t - t_0) e^{-st} \, dt = \int_{-\infty}^{\infty} f(u) e^{-s(u + t_0)} \, du = e^{-s t_0} \int_{-\infty}^{\infty} f(u) e^{-su} \, du = e^{-s t_0} F(s).
Convergence is preserved because the exponential factor e^{-s t_0} is independent of the integration variable and finite for s in the original ROC.[12][13] Frequency shifting modulates the time-domain signal: for a complex constant a,
\mathcal{L} \left\{ f(t) e^{a t} \right\} = F(s - a),
with the ROC shifted horizontally by \operatorname{Re}(a), i.e., the new strip is \{\sigma_1 + \operatorname{Re}(a) < \operatorname{Re}(s) < \sigma_2 + \operatorname{Re}(a)\} if the original was \sigma_1 < \operatorname{Re}(s) < \sigma_2. Substituting into the integral yields
\int_{-\infty}^{\infty} f(t) e^{a t} e^{-s t} \, dt = \int_{-\infty}^{\infty} f(t) e^{-(s - a) t} \, dt = F(s - a),
where the shift in s adjusts the real part of the exponent to maintain convergence in the translated region.[12][13] Time scaling compresses or expands the signal: for a nonzero real constant a,
\mathcal{L} \left\{ f(a t) \right\} = \frac{1}{|a|} F\left( \frac{s}{a} \right),
with the ROC scaled by |a|, meaning the boundaries become \sigma_1 / |a| < \operatorname{Re}(s) < \sigma_2 / |a| (adjusted for the sign of a, which introduces no additional phase factor in the bilateral case). The derivation uses the substitution u = a t, so dt = du / |a| (accounting for the absolute value to preserve orientation):
\int_{-\infty}^{\infty} f(a t) e^{-s t} \, dt = \int_{-\infty}^{\infty} f(u) e^{-(s / a) u} \frac{du}{|a|} = \frac{1}{|a|} F\left( \frac{s}{a} \right).
The scaling of the ROC arises because the effective decay rate in the exponent is altered by the factor $1/a.[12][13] Differentiation in the time domain corresponds to multiplication by s in the s-domain:
\mathcal{L} \left\{ \frac{df}{dt} \right\} = s F(s),
with the ROC at least that of F(s) (possibly expanded). This is derived by integration by parts:
\int_{-\infty}^{\infty} \frac{df}{dt} e^{-s t} \, dt = \left[ f(t) e^{-s t} \right]_{-\infty}^{\infty} + s \int_{-\infty}^{\infty} f(t) e^{-s t} \, dt,
where the boundary term vanishes in the ROC due to absolute convergence. Higher-order derivatives follow: \mathcal{L} \left\{ \frac{d^n f}{dt^n} \right\} = s^n F(s).[12] Differentiation in the s-domain corresponds to multiplication by time in the t-domain. For the first order,
\mathcal{L} \left\{ -t f(t) \right\} = \frac{d}{ds} F(s),
with the ROC unchanged (or possibly expanded if poles are removed). Higher-order derivatives follow by iteration: the nth derivative \frac{d^n}{ds^n} F(s) = \mathcal{L} \left\{ (-t)^n f(t) \right\}. This is derived by differentiating under the integral sign, assuming uniform convergence allows interchange:
\frac{d}{ds} F(s) = \frac{d}{ds} \int_{-\infty}^{\infty} f(t) e^{-s t} \, dt = \int_{-\infty}^{\infty} f(t) \frac{\partial}{\partial s} \left( e^{-s t} \right) \, dt = \int_{-\infty}^{\infty} f(t) (-t) e^{-s t} \, dt = \mathcal{L} \left\{ -t f(t) \right\}.
The ROC remains at least the original, as differentiation may introduce poles but does not shrink the strip.[12] The initial value theorem relates the behavior at t = 0^+ to the s-domain: if \lim_{t \to 0^+} f(t) exists, then
\lim_{t \to 0^+} f(t) = \lim_{s \to \infty} s F(s),
provided the limit in s is taken within the ROC (typically along a path where \operatorname{Re}(s) \to +\infty). Similarly, the final value theorem states that if \lim_{t \to +\infty} f(t) exists, then
\lim_{t \to +\infty} f(t) = \lim_{s \to 0} s F(s),
with the limit in s within the ROC (all poles of s F(s) in the open left half-plane). These theorems are derived by considering the transforms of step functions or using L'Hôpital's rule on appropriate contours, but apply to two-sided transforms with the noted caveats for noncausal components.[12][13]
Advanced Theorems
The convolution theorem for the two-sided Laplace transform states that the transform of the convolution of two functions f(t) and g(t) is the product of their individual transforms: \mathcal{L}\left\{ \int_{-\infty}^{\infty} f(\tau) g(t - \tau) \, d\tau \right\}(s) = F(s) G(s), provided the regions of convergence (ROCs) of F(s) and G(s) overlap sufficiently to include a common vertical strip in the complex plane where both integrals converge absolutely.[5] This result holds under the assumption that the functions are such that the transforms exist in that common ROC, ensuring the convolution integral is well-defined.[14] A proof of the convolution theorem can be obtained by applying Fubini's theorem to interchange the order of integration in the double integral representation of the transform of the convolution. Specifically, substituting the convolution into the Laplace integral yields \mathcal{L}\{f * g\}(s) = \int_{-\infty}^{\infty} \left( \int_{-\infty}^{\infty} f(\tau) g(t - \tau) \, d\tau \right) e^{-st} \, dt = \int_{-\infty}^{\infty} \int_{-\infty}^{\infty} f(\tau) g(t - \tau) e^{-st} \, d\tau \, dt, and changing variables or recognizing the inner integral as the transform of g shifted by \tau leads to F(s) G(s) after justification of the interchange via absolute convergence in the common ROC.[5] The uniqueness theorem asserts that if two functions f(t) and g(t) have the same two-sided Laplace transform F(s) over a common region of convergence that includes a vertical strip of positive width, then f(t) = g(t) almost everywhere on \mathbb{R}.[5] This follows from the analyticity of the Laplace transform in its ROC: if F(s) coincides on an open set (the common strip), analytic continuation implies equality throughout the connected ROC, and the inversion formula then recovers identical time-domain functions almost everywhere.[14] The result relies on Titchmarsh's theorem for the uniqueness of Fourier transforms, adapted via the relationship between the two-sided Laplace transform along the imaginary axis and the Fourier transform when the ROC includes that axis.[5] Parseval's theorem for the two-sided Laplace transform provides a relation between inner products in the time and s-domains: \int_{-\infty}^{\infty} \overline{f(t)} g(t) \, dt = \frac{1}{2\pi i} \int_{\sigma - i\infty}^{\sigma + i\infty} \overline{F(-s)} G(s) \, ds, where \sigma lies in the intersection of the ROCs of F(s) and G(s), and the contour is along \operatorname{Re}(s) = \sigma. This holds for functions whose transforms converge absolutely in the common strip. A proof uses the inverse transform along the Bromwich contour, applying the convolution theorem to the cross-correlation, and contour integration, often relating to the Fourier case by choosing \sigma = 0 when possible. Plancherel's theorem is a special case of Parseval's theorem for energy preservation, stating that \int_{-\infty}^{\infty} |f(t)|^2 \, dt = \frac{1}{2\pi} \int_{-\infty}^{\infty} |F(j\omega)|^2 \, d\omega, provided the ROC of F(s) includes the imaginary axis (i.e., \sigma_1 < 0 < \sigma_2).[14] Here, F(j\omega) reduces to the Fourier transform of f(t), and the theorem equates the total energy in the time domain to that in the frequency domain. The proof follows from Parseval's theorem by setting g(t) = f(t) and deforming the contour to the imaginary axis, using Fourier relations and the unitarity of the Fourier transform on L^2(\mathbb{R}).[14]Relationships to Other Transforms
Fourier Transform
The two-sided Laplace transform F(s) = \int_{-\infty}^{\infty} f(t) e^{-s t} \, dt, with s = \sigma + j \omega, reduces to the Fourier transform as a special case when the region of convergence (ROC) includes the imaginary axis in the complex plane, allowing \sigma = 0. In this scenario, substituting s = j \omega yields F(j \omega) = \int_{-\infty}^{\infty} f(t) e^{-j \omega t} \, dt, which is the standard definition of the Fourier transform for functions absolutely integrable over the real line.[15][16] This equivalence holds provided the signal f(t) satisfies the conditions for Fourier convergence, such as being in L^1(\mathbb{R}). The Laplace transform serves as an analytic continuation of the Fourier transform, extending F(j \omega) from the imaginary axis into the broader s-plane. For signals where the Fourier transform diverges due to lack of absolute integrability, the Laplace transform achieves convergence by selecting \sigma \neq 0 within the ROC, effectively applying an exponential weighting e^{-\sigma t} to dampen growth or decay. For instance, the signal f(t) = e^{a t} u(t) (with u(t) the unit step function and a > 0) has no Fourier transform because it grows exponentially for t > 0, but its two-sided Laplace transform converges for \operatorname{Re}(s) > a.[16][8] This extension enables analysis of unstable or non-stationary signals common in engineering applications. The inverse transforms reflect this relationship: the inverse Fourier transform recovers f(t) = \frac{1}{2\pi} \int_{-\infty}^{\infty} F(j \omega) e^{j \omega t} \, d\omega, while the inverse Laplace transform uses the Bromwich contour integral f(t) = \frac{1}{2\pi j} \int_{\sigma - j \infty}^{\sigma + j \infty} F(s) e^{s t} \, ds, where the integration path is a vertical line with \sigma in the ROC. To derive the Fourier transform from the Laplace transform, evaluate F(j \omega) directly if the imaginary axis lies in the ROC; otherwise, take the boundary value as \sigma \to 0 from the appropriate side of the ROC. Both transforms share properties like energy preservation under Plancherel's theorem for square-integrable functions.[15][16] Historically, the Laplace transform predates the Fourier transform, with Pierre-Simon Laplace introducing related integral forms in his 1785 work on probability and differential equations, whereas Joseph Fourier developed his transform around 1807 for heat conduction, publishing it in 1822. The bilateral Laplace transform establishes a direct analytic link between them in contemporary signal processing, highlighting Laplace's generality.[8][17]One-sided Laplace Transform
The one-sided Laplace transform, also known as the unilateral Laplace transform, is defined as \mathcal{L}\{f(t)\}(s) = F(s) = \int_{0^-}^{\infty} f(t) e^{-st} \, dt, where the integration begins just before t = 0 and extends to infinity, effectively ignoring any components of f(t) for t < 0. This formulation assumes that the signal f(t) = 0 for t < 0, making it particularly suited for causal signals that start at or after t = 0. In contrast, the two-sided Laplace transform integrates over the entire real line from -\infty to \infty, thereby capturing anti-causal components where f(t) may be nonzero for t < 0. The one-sided version simplifies analysis by focusing on future behavior relative to t = 0, but it inherently discards pre-t = 0 information, which can lead to loss of detail for non-causal signals.[18][19] A primary distinction lies in the region of convergence (ROC). For the one-sided transform of a causal signal bounded by |f(t)| \leq M e^{\alpha t} for t \geq 0, the ROC is the right half-plane \operatorname{Re}(s) > \alpha, where convergence depends solely on the exponential growth rate of the signal for positive time. The two-sided transform, however, yields an ROC that is typically a vertical strip \sigma_1 < \operatorname{Re}(s) < \sigma_2 in the complex plane, balancing convergence for both positive and negative time behaviors. For purely causal signals where f(t) = 0 for t < 0, the two-sided and one-sided transforms coincide, with their ROCs matching as the left boundary of the strip extends to -\infty. This equivalence holds because the two-sided integral reduces to the one-sided form under the causality assumption.[19][18] Properties of the transforms also differ, particularly for operations involving time shifts and convolution. In the one-sided case, the time-shift property \mathcal{L}\{f(t - t_0)\}(s) = e^{-s t_0} F(s) applies reliably only for positive shifts t_0 > 0, as negative shifts could introduce non-causal components that alter the effective ROC or violate the t \geq 0 assumption; the bilateral version handles arbitrary shifts without such restrictions, maintaining the original ROC. For convolution, the one-sided transform of the causal convolution (f * g)(t) = \int_0^t f(\tau) g(t - \tau) \, d\tau equals the product F(s) G(s), with limits restricted to ensure causality; the two-sided convolution integrates over -\infty to \infty, accommodating anti-causal interactions but requiring the intersection of ROCs for validity. These differences highlight how the one-sided transform streamlines computations for causal systems while the two-sided provides generality.[20][21][19] The choice between the two transforms depends on signal characteristics and application context. The one-sided transform is predominantly used in control systems engineering, where signals and responses are causal, enabling straightforward solution of differential equations with initial conditions. Conversely, the two-sided transform is essential for non-causal or two-sided signals, such as those modeling communication channels with pre- and post-distortion effects. For a general signal, conversion is achieved by decomposing f(t) = f_+(t) + f_-(t), where f_+(t) is the causal part (f_+(t) = 0 for t < 0) and f_-(t) is the anti-causal part (f_-(t) = 0 for t > 0); the two-sided transform then becomes the sum of the one-sided transform of f_+(t) at s and the one-sided transform of f_-(-t) at -s. This splitting allows leveraging one-sided tools for parts of the analysis while preserving full signal information.[18][14][19][22]Examples and Applications
Selected Transforms
The two-sided Laplace transform pairs for several representative signals are summarized in the table below, highlighting the role of the region of convergence (ROC) in uniquely determining the time-domain function from the transform. These examples include causal signals (ROC in the right half-plane), anti-causal signals (ROC in the left half-plane), and truly two-sided signals (ROC as a vertical strip between poles). The pairs are derived from standard analyses in signals and systems theory.[10]| f(t) | F(s) | ROC |
|---|---|---|
| δ(t) | 1 | All s |
| u(t) (unit step, causal) | 1/s | Re(s) > 0 |
| -u(-t) (unit step, anti-causal) | 1/s | Re(s) < 0 |
| e^{-2t} u(t) (causal exponential) | 1/(s + 2) | Re(s) > -2 |
| -e^{2t} u(-t) (anti-causal exponential) | 1/(s - 2) | Re(s) < 2 |
| -e^{-3t} u(-t) (anti-causal exponential) | 1/(s + 3) | Re(s) < -3 |
| e^{-3t} u(t) - e^{2t} u(-t) (two-sided) | 1/(s + 3) + 1/(s - 2) | -3 < Re(s) < 2 |
| e^{-a |t|} (even exponential decay, a > 0) | 2a / (a² - s²) | |Re(s)| < a |