Z-transform
The Z-transform is a mathematical technique in signal processing and control theory that converts a discrete-time signal, represented as a sequence of values x where n is an integer, into a complex-valued function X(z) of a complex variable z, enabling analysis in the frequency domain similar to how the Laplace transform operates on continuous-time signals.[1] Formally defined as X(z) = \sum_{n=-\infty}^{\infty} x z^{-n}, it generalizes the discrete-time Fourier transform by allowing evaluation anywhere in the complex z-plane rather than restricting to the unit circle.[1] The Z-transform emerged in the early 1940s when M.F. Gardner and J.L. Barnes applied it to solve linear constant-coefficient difference equations in sampled-data systems.[2] In 1947, W. Hurewicz extended its use to transform sampled signals or sequences in control contexts.[2] The term "Z-transform" was coined in 1952 by a sampled-data control research group at Columbia University, led by John R. Ragazzini, marking its formal introduction and popularization in engineering literature.[2] Key properties of the Z-transform include linearity, time-shifting, convolution (which becomes multiplication in the z-domain), and scaling, facilitating the simplification of difference equations into algebraic forms for system analysis.[1] The inverse Z-transform recovers the original time-domain signal, often via contour integration or partial fraction expansion, though practical computation uses tables or numerical methods.[1] In the z-plane, system stability is determined by pole locations—all poles must lie inside the unit circle for bounded-input bounded-output stability.[1] Applications of the Z-transform are central to digital signal processing (DSP), where it models linear time-invariant discrete systems through transfer functions H(z) = Y(z)/X(z), aiding in filter design such as infinite impulse response (IIR) filters via techniques like the bilinear transform.[1] It is also essential in control systems for analyzing sampled-data feedback loops and in communications for processing discrete signals like those in digital modulation schemes.[1]Fundamentals
Definition
The Z-transform is the discrete-time counterpart of the continuous-time Laplace transform, converting a sequence representing a discrete-time signal into a function of the complex variable [z](/page/Z), thereby enabling algebraic manipulation of difference equations in the z-domain for applications such as digital signal processing and control systems.[3] This transformation is particularly valuable for analyzing sampled data systems, where signals are sequences of values indexed by integers n.[3] The bilateral Z-transform applies to general two-sided sequences x defined for all integers n, and is formulated as X(z) = \sum_{n=-\infty}^{\infty} x z^{-n}, provided the infinite sum converges in absolute value. Convergence occurs within an annular region in the complex z-plane, referred to as the region of convergence (ROC), which depends on the signal's characteristics and ensures the transform's uniqueness when specified alongside the ROC.[3] For one-sided or causal sequences where x = 0 for n < 0, the unilateral Z-transform is used, defined by X(z) = \sum_{n=0}^{\infty} x z^{-n}. This form facilitates the analysis of causal systems by naturally incorporating initial conditions when solving linear constant-coefficient difference equations, a common requirement in digital signal processing.[4] In standard notation, the factor z^{-1} represents a unit delay in the discrete-time domain, aligning the transform with shift properties essential for system modeling.[3]Historical Background
The Z-transform has its conceptual roots in the generating functions developed in the 18th century for solving recurrence relations and difference equations, with early contributions from Abraham de Moivre in 1718 who applied such functions to the Fibonacci sequence, and later expansions by Leonhard Euler in the 1730s and 1750s for broader discrete problems.[5] These ordinary generating functions provided a formal power series representation of sequences, laying the groundwork for transforming discrete-time problems into algebraic forms amenable to analysis. Although not explicitly termed the Z-transform, this approach anticipated its structure by associating sequences with series expansions, influencing 19th-century mathematicians in probability and combinatorics who refined methods for linear difference equations. The modern Z-transform emerged in the context of post-World War II engineering, particularly for analyzing sampled-data control systems, where continuous signals are discretized. It was reintroduced in 1947 by Witold Hurewicz in his work on pulsed data filters and servomechanisms, drawing analogies to the Laplace transform but adapted for discrete sampling.[6] This formulation addressed the need to solve linear constant-coefficient difference equations arising in radar and feedback control, building on earlier 1940s efforts by researchers like M.F. Gardner and J.L. Barnes who applied similar transforms to sampled systems.[2] The term "Z-transform" was coined in 1952 by John R. Ragazzini and Lotfi A. Zadeh in their seminal paper on sampled-data systems analysis, which unified input-output relations and established it as a standard tool in control theory.[7] Eliahu I. Jury further advanced the Z-transform in 1958 through his book Sampled-Data Control Systems, which systematized its application to stability analysis and design of discrete-time systems, extending earlier work on difference equations.[8] Post-WWII adoption accelerated in control theory and signal processing, driven by the rise of digital computers and sampled systems in aerospace and communications. By the 1960s, amid the emergence of digital signal processing (DSP), the Z-transform was formalized for digital filter design, as seen in techniques like the bilinear transformation for converting analog filters to discrete equivalents, enabling efficient implementation of IIR and FIR filters.[9] Jury's later 1973 text, Theory and Application of the Z-Transform Method, solidified its role in both unilateral (causal) and bilateral variants for broader engineering applications.[10]Region of Convergence
Properties
The region of convergence (ROC) of the Z-transform is defined as the set of all complex values z in the complex plane for which the infinite series \sum_{n=-\infty}^{\infty} x z^{-n} converges absolutely. This region is fundamentally an annular area centered at the origin, expressed as r_1 < |z| < r_2, where r_1 and r_2 are non-negative real numbers with $0 \leq r_1 < r_2 \leq \infty; special cases include a disk when r_1 = 0 (interior of a circle) or an exterior region when r_2 = \infty (outside a circle).[11][12] Several key properties characterize the ROC. It is always an annulus, disk, or exterior region and cannot contain any poles of the Z-transform X(z), as poles represent points of divergence. For finite-duration signals—those nonzero only over a finite range of n—the ROC encompasses the entire complex plane except possibly the origin z = 0 for right-sided sequences (nonzero for n \geq 0), and it includes the point at infinity |z| = \infty. Conversely, right-sided signals generally exclude z = 0 from their ROC due to the form of the series involving negative powers of z.[13][3][11] The type of discrete-time signal significantly influences the ROC's form. For causal or right-sided signals (nonzero only for n \geq 0), the ROC is the exterior of a circle, |z| > r, extending to infinity. Anti-causal or left-sided signals (nonzero only for n \leq 0) yield an ROC that is the interior of a circle, |z| < r, including the origin. Two-sided signals, nonzero over both positive and negative n, result in a strip or annulus, r_1 < |z| < r_2, between two circles. These distinctions arise from the bilateral Z-transform definition, which applies to general signals, while the unilateral variant assumes causality and adjusts the summation limits accordingly.[12][13][3] A fundamental uniqueness theorem states that a given Z-transform X(z) paired with its specific ROC uniquely determines the underlying time-domain signal x; the same rational X(z) can correspond to different signals if associated with distinct ROCs. For operations on signals, the ROC of the sum (or linear combination) of two signals is at least the intersection of their individual ROCs, ensuring convergence where both transforms are defined. Similarly, for the Z-transform of the convolution of two signals—which is the product of their individual Z-transforms—the ROC is at least the intersection of the two ROCs, though it may extend further depending on pole-zero interactions.[14][15][15]Examples
To illustrate the region of convergence (ROC) for the Z-transform, consider specific signal types that highlight different convergence behaviors in the complex z-plane.[12] For a finite-duration signal, such as a rectangular window x = 1 for $0 \leq n \leq N-1 and x = 0 otherwise, the Z-transform is a finite polynomial sum X(z) = \sum_{n=0}^{N-1} z^{-n}, which converges for all z \neq 0. The ROC is thus the entire z-plane except at z = 0, where a pole may exist if the signal starts at n=0. This example demonstrates that finite-duration signals have the broadest possible ROC, excluding only potential singularities at the origin or infinity, allowing the transform to exist almost everywhere. A causal (right-sided) exponential signal x = a^n u, where u is the unit step function and |a| is a constant, has Z-transform X(z) = \frac{1}{1 - a z^{-1}}. The series converges for |z| > |a|, forming an exterior region outside a circle of radius |a| centered at the origin. This ROC shape reflects the right-sided nature of the signal, where convergence requires |z| to dominate the exponential growth for positive n.[16] In contrast, an anti-causal (left-sided) exponential signal x = -a^n u[-n-1] yields the same algebraic Z-transform expression X(z) = \frac{1}{1 - a z^{-1}}, but with ROC |z| < |a|, an interior disk inside the circle of radius |a|. Here, convergence occurs for smaller |z| to counteract the exponential behavior for negative n. This illustrates how the same rational function can represent different signals based on the ROC, emphasizing left-sided convergence.[3] For a two-sided signal such as x = a^{|n|}, the Z-transform becomes X(z) = \frac{1 - a^2}{1 - 2a z^{-1} + a^2 z^{-2}}, but the ROC is an annular strip |a| < |z| < 1/|a| (assuming |a| < 1). This ring-shaped region arises from combining right-sided and left-sided components, where the outer boundary is set by the causal part and the inner by the anti-causal part. Such ROCs are typical for signals extending infinitely in both directions.[12] The ROC uniquely determines key signal attributes: an exterior ROC (|z| > r) indicates a causal signal, an interior ROC (|z| < r) an anti-causal one, and an annulus a two-sided signal; moreover, inclusion of the unit circle |z| = 1 within the ROC ensures the signal is stable, as it allows the discrete-time Fourier transform to exist.Inverse Z-transform
Contour Integration
The inverse Z-transform can be obtained through complex contour integration, providing an analytical method to recover the time-domain sequence from its Z-transform representation. The formula is given by x = \frac{1}{2\pi j} \oint_C X(z) z^{n-1} \, dz, where C is a counterclockwise closed contour lying within the region of convergence (ROC) and encircling the origin.[17] This integral arises from the orthogonality of the complex exponentials z^n on the unit circle, generalized to arbitrary contours in the ROC. To evaluate the integral, Cauchy's residue theorem is applied, which states that the value of the contour integral is $2\pi j times the sum of the residues of the integrand at its poles enclosed by C. For the integrand X(z) z^{n-1}, since z^{n-1} is analytic everywhere (an entire function), the poles are solely those of X(z). Thus, the inverse Z-transform simplifies to x = \sum_k \operatorname{Res} \left\{ X(z) z^{n-1}, z = p_k \right\}, where the sum is over all poles p_k of X(z) inside C, and \operatorname{Res}\{\cdot, z = p_k\} denotes the residue at pole p_k.[17] For rational X(z) = P(z)/Q(z) with simple poles, the residue at a pole p_k (where Q(p_k) = 0 and Q'(p_k) \neq 0) is P(p_k) / Q'(p_k) \cdot p_k^{n-1}. Higher-order poles require the general residue formula involving derivatives. The contour C must satisfy specific requirements to ensure the integral correctly captures the sequence: it must be closed and traversed counterclockwise, lie entirely within the ROC to guarantee convergence of X(z), and enclose all relevant poles contributing to x. For right-sided (causal) sequences, where the ROC is exterior to the outermost pole (|z| > r_{\max}), a common choice is a circular contour |z| = r with r > r_{\max}. For left-sided sequences, the ROC is interior to the innermost pole, and the contour is chosen accordingly inside that region. Failure to select an appropriate contour, such as one outside the ROC, results in divergence or incorrect recovery of the sequence.[17] This method excels in symbolic computation, allowing exact closed-form expressions by isolating pole contributions, which aids in theoretical analysis and verification of Z-transform pairs. It is particularly valuable for functions with known pole structures, enabling insight into how individual poles influence the time-domain behavior, such as growth rates determined by pole magnitudes. However, practical limitations arise in numerical implementation: evaluating residues requires precise pole locations and can be sensitive to computational precision, while parameterizing and integrating over arbitrary contours is inefficient for digital tools compared to algebraic alternatives. Thus, contour integration is primarily used for analytical purposes rather than routine computation. A simple example illustrates the approach. Consider X(z) = \frac{z}{z - a} with ROC |z| > |a|, corresponding to a causal geometric sequence. The integrand is X(z) z^{n-1} = \frac{z^n}{z - a}, with a simple pole at z = a. The residue at z = a is \lim_{z \to a} (z - a) \cdot \frac{z^n}{z - a} = a^n. Since this is the only pole inside a contour such as |z| = r > |a|, x = a^n for n \geq 0, and x = 0 for n < 0. This confirms the standard pair for the unit step modulated by a^n.[17]Power Series Expansion
The power series expansion method provides an algebraic approach to inverting the Z-transform by representing X(z) as a Laurent series within its region of convergence (ROC), where the coefficients directly yield the time-domain sequence x. Specifically, the bilateral Z-transform is given by X(z) = \sum_{n=-\infty}^{\infty} x z^{-n}, and the inverse transform is obtained by identifying the coefficient of z^{-n} in the Laurent series expansion of X(z) valid in the ROC. This expansion is unique in the annular ROC, ensuring the series converges to X(z) pointwise in that region.[18] For causal signals, where x = 0 for n < 0, the ROC is exterior to a circle |z| > r, and X(z) expands as a one-sided power series in negative powers of z: X(z) = \sum_{n=0}^{\infty} x z^{-n}. To compute this, one can perform long division on the rational function expressed in powers of z^{-1}, or recognize known series expansions such as the geometric series. Equivalently, dividing by z yields \frac{X(z)}{z} = \sum_{n=0}^{\infty} x z^{-n-1}, from which coefficients are extracted term by term. For instance, consider X(z) = \frac{1}{1 - a z^{-1}} with ROC |z| > |a|; this expands via the geometric series as X(z) = \sum_{n=0}^{\infty} a^n z^{-n}, so x = a^n for n \geq 0 and x = 0 otherwise.[18][19] Two-sided signals, with nonzero values for both positive and negative n, have an annular ROC r < |z| < R, and the Laurent series includes both positive and negative powers of z. The terms with negative exponents (z^{-n} for n > 0) capture the causal (right-sided) component, expanded using series in z^{-1} valid for |z| > r. Conversely, the positive exponent terms (z^{|n|} for n < 0) represent the anti-causal (left-sided) component, obtained by expanding in powers of z valid for |z| < R, often after manipulating X(z) to isolate these parts. This separation allows sequential computation of each side's coefficients via appropriate series methods, such as long division or binomial expansions, ensuring convergence within the ROC.[19][20] The power series approach to the Z-transform is intrinsically linked to generating functions, as both represent sequences via formal power series; the Z-transform generalizes the ordinary generating function by incorporating negative indices through Laurent series in z^{-1}, facilitating analysis of bilateral sequences in signal processing and probability.[21][22]Partial Fraction Expansion
The partial fraction expansion provides a practical method for computing the inverse Z-transform of a rational function X(z) = \frac{P(z)}{Q(z)}, where the degree of P(z) is strictly less than the degree of Q(z), ensuring a proper rational form.[23] This approach decomposes X(z) into a sum of simpler terms whose inverse Z-transforms are known from standard tables of Z-transform pairs.[24] It is particularly effective for causal sequences, assuming the region of convergence (ROC) includes the exterior of the circle defined by the outermost pole and extends to infinity.[25] For stable discrete-time systems, the ROC must include the unit circle |z| = 1.[26] The procedure begins by forming \frac{X(z)}{z}, which facilitates the partial fraction decomposition into terms centered at the poles of X(z).[27] For distinct simple poles p_k of Q(z), the decomposition is \frac{X(z)}{z} = \sum_k \frac{A_k}{z - p_k}, where the coefficients A_k, known as residues, are computed as A_k = \lim_{z \to p_k} (z - p_k) \frac{X(z)}{z}. Multiplying through by z yields X(z) = \sum_k \frac{A_k z}{z - p_k}. The inverse Z-transform then follows directly from the standard pair \mathcal{Z}^{-1}\left\{ \frac{z}{z - p_k} \right\} = p_k^n u, giving the causal sequence x = \sum_k A_k p_k^n, \quad n \geq 0, where u is the unit step function.[28] For repeated poles of multiplicity m_k > 1 at p_k, the expansion of \frac{X(z)}{z} includes higher-order terms: \frac{X(z)}{z} = \sum_k \sum_{j=1}^{m_k} \frac{A_{k,j}}{(z - p_k)^j}, with residues A_{k,j} found via the general formula A_{k,j} = \frac{1}{(m_k - j)!} \lim_{z \to p_k} \frac{d^{m_k - j}}{dz^{m_k - j}} \left[ (z - p_k)^{m_k} \frac{X(z)}{z} \right]. [29] The corresponding inverse involves generalized pairs, such as \mathcal{Z}^{-1}\left\{ \frac{z}{(z - p_k)^2} \right\} = n p_k^{n-1} u for a double pole.[30] As an illustrative example, consider inverting X(z) = \frac{z(z + 1)}{z^2 - 2z + 1} = \frac{z(z + 1)}{(z - 1)^2}, which has a repeated pole at p = 1 with multiplicity 2 and ROC |z| > 1.[23] First, form \frac{X(z)}{z} = \frac{z + 1}{(z - 1)^2}. The partial fraction expansion for the double pole is \frac{z + 1}{(z - 1)^2} = \frac{A}{z - 1} + \frac{B}{(z - 1)^2}. Solving yields A = 1 and B = 2, so X(z) = \frac{z}{z - 1} + \frac{2z}{(z - 1)^2}. Applying the known inverses gives x = 1^n u + 2 n (1)^{n-1} u = (1 + 2n) u, \quad n \geq 0. This sequence represents a linear combination of a unit step and a discrete ramp, common in responses of systems with integrators or accumulators.[24]Properties
Algebraic Properties
The Z-transform exhibits linearity, meaning that for sequences x and y with Z-transforms X(z) and Y(z), respectively, and constants a and b, the transform of a linear combination is the corresponding combination of the transforms: Z\{ a x + b y \} = a X(z) + b Y(z), where the region of convergence (ROC) is at least the intersection of the ROCs of X(z) and Y(z).[31] A key algebraic property is the convolution theorem, which states that the Z-transform of the convolution of two sequences x and y is the product of their individual Z-transforms: Z\{ x * y \} = X(z) Y(z), with the ROC being at least the intersection of the ROCs of X(z) and Y(z), potentially larger if pole-zero cancellations occur.[32][31] Differentiation in the Z-domain corresponds to multiplication by the time index in the time domain: for a sequence x with Z-transform X(z), Z\{ n x \} = -z \frac{d}{dz} X(z), with the same ROC as X(z).[33] Unlike the continuous-time Laplace transform, the Z-transform has no direct analog for arbitrary time scaling due to the discrete nature of the time index n. However, for exponential scaling, Z\{ a^n x \} = X(z/a), with ROC scaled by |a|. For integer decimation (downsampling by factor M), an exact formula exists involving a sum of scaled and aliased versions of X(z), but it lacks the simplicity of continuous-time scaling and is often approximated in practice for analysis.[34][35] The initial value theorem provides the value at n=0 directly from the Z-transform: if x is causal (x=0 for n<0), then x{{grok:render&&&type=render_inline_citation&&&citation_id=0&&&citation_type=wikipedia}} = \lim_{z \to \infty} X(z). This holds provided the limit exists.[36] The final value theorem gives the steady-state value as n \to \infty: \lim_{n \to \infty} x = \lim_{z \to 1} (1 - z^{-1}) X(z), assuming the limit exists and all poles of (1 - z^{-1}) X(z) lie inside the unit circle except possibly a simple pole at z=1.[36]Time-Domain Properties
The time-domain properties of the Z-transform elucidate the effects of common signal operations—such as shifting, reversal, accumulation, and multiplication by the time index—on the transform representation. These properties facilitate the analysis of discrete-time signals by translating temporal manipulations into algebraic operations in the z-domain, aiding in the solution of difference equations and system responses. While some properties apply to both unilateral and bilateral Z-transforms, others, particularly those involving causality, are tailored to the unilateral form, which assumes signals are zero for n < 0. The region of convergence (ROC) often adjusts with these operations, reflecting changes in signal duration or causality. The time shift property describes how delaying or advancing a sequence impacts its Z-transform. For the unilateral Z-transform, a right shift by n_0 ≥ 0 (delay) yields Z\{x[n - n_0]\} = z^{-n_0} X(z) + \sum_{k=0}^{n_0 - 1} x[k - n_0] z^{-(n_0 - k)}, where the summation accounts for the influence of initial conditions before the shift, as the unilateral transform incorporates only n ≥ 0. The ROC remains the same as that of X(z), though it may shrink if the added polynomial terms introduce new poles. For a left shift (advance) by n_0 > 0, the formula is Z{x[n + n_0]} = z^{n_0} \left( X(z) - \sum_{k=0}^{n_0 - 1} x z^{-k} \right), again preserving the ROC of X(z) except possibly at z = 0 due to the positive powers of z.[4] Time reversal flips the sequence about n = 0, primarily relevant for the bilateral Z-transform to handle anti-causal components. The property states that Z\{x[-n]\} = X(1/z), with the ROC inverted: if the original ROC is r < |z| < R, the new ROC becomes 1/R < |z| < 1/r. This inversion arises because reversal maps the signal's temporal extent, potentially exchanging inner and outer convergence boundaries; for causal signals, the property may not apply directly without extending the sequence. The ROC adjustment ensures convergence where the original did, scaled reciprocally.[36] Accumulation, or the running sum s = \sum_{k=0}^n x for unilateral transforms (assuming causality), represents discrete-time integration. Its Z-transform is Z\left\{ \sum_{k=0}^n x \right\} = \frac{X(z)}{1 - z^{-1}}, valid for |z| > 1, with the overall ROC being the intersection of the ROC of X(z) and the exterior of the unit disk (excluding |z| ≤ 1) due to the pole at z = 1 introduced by the denominator. This property excludes regions inside or on the unit circle to maintain convergence, particularly for stable accumulators where the sum grows without bound unless x decays sufficiently. For bilateral transforms, the summation from -∞ to n adjusts similarly but requires careful ROC handling for two-sided signals.[37] Multiplication by the time index n, akin to time-domain differentiation, weights the sequence by its index. The corresponding property is Z\{n x\} = -z \frac{d X(z)}{dz}, applicable to both unilateral and bilateral Z-transforms, with the ROC unchanged from that of X(z) provided the derivative exists within it. This operation generates a new rational function if X(z) is rational, and it connects to algebraic manipulations like differentiation in the z-domain, though its primary utility lies in time-domain emphasis for moment-like computations or system stability analysis. Repeated application yields higher-order terms, such as Z{n^2 x} = -z \frac{d}{dz} \left( z \frac{d X(z)}{dz} \right), but the ROC remains invariant.[37]Common Z-Transform Pairs
Basic Pairs
The basic pairs of the Z-transform refer to the transforms of fundamental discrete-time signals, typically assuming causality where the signals are zero for negative time indices. These pairs form the foundation for analyzing more complex signals through properties like linearity and time-shifting.[38] The Z-transform of the unit impulse \delta, defined as \delta{{grok:render&&&type=render_inline_citation&&&citation_id=0&&&citation_type=wikipedia}} = 1 and \delta = 0 for n \neq 0, is derived directly from the definition: X(z) = \sum_{n=-\infty}^{\infty} \delta z^{-n} = 1. The region of convergence (ROC) is the entire z-plane since the signal has finite support.[39] For the unit step u, which is 1 for n \geq 0 and 0 otherwise, the Z-transform is X(z) = \sum_{n=0}^{\infty} z^{-n}. This sum is a geometric series with ratio r = z^{-1}, converging to \frac{1}{1 - z^{-1}} = \frac{z}{z-1} for |z^{-1}| < 1, or equivalently |z| > 1.[40] The exponential signal a^n u has Z-transform X(z) = \sum_{n=0}^{\infty} (a z^{-1})^n = \frac{1}{1 - a z^{-1}} = \frac{z}{z - a}, again using the geometric series formula, with ROC |z| > |a| to ensure convergence.[41] The unit ramp n u is obtained via the differentiation property: if X(z) is the transform of u, then the transform of n u is -z \frac{d}{dz} X(z). Differentiating \frac{z}{z-1} yields \frac{z}{(z-1)^2}, with ROC |z| > 1.[40]| Signal x | Z-transform X(z) | ROC |
|---|---|---|
| \delta | $1 | All z |
| u | \frac{z}{z-1} | \|z\| > 1 |
| a^n u | \frac{z}{z-a} | \|z\| > \|a\| |
| n u | \frac{z}{(z-1)^2} | \|z\| > 1 |
Derived Pairs
Derived pairs for the Z-transform are constructed by applying algebraic and time-domain properties to the basic pairs, enabling the analysis of more sophisticated discrete-time signals such as sinusoids, damped oscillations, and finite-duration pulses without resorting to direct evaluation of the transform sum. These derivations leverage linearity to express signals as combinations of exponentials, scaling for damping effects, time-shifting for windowing, and convolution for composite waveforms like triangular pulses. The region of convergence (ROC) for these pairs typically extends outside the farthest pole, ensuring stability for causal signals. A fundamental derived pair is that of the unit-amplitude sinusoid, obtained by expressing \sin(\omega n) as the imaginary part of complex exponentials and using the basic geometric pair for a^n u. The Z-transform of \sin(\omega n) u is given by X(z) = \frac{z \sin \omega}{z^2 - 2 z \cos \omega + 1}, with ROC |z| > 1. [42] Similarly, the cosine pair follows from the real part, yielding \cos(\omega n) u \leftrightarrow \frac{z (z - \cos \omega)}{z^2 - 2 z \cos \omega + 1}, also with ROC |z| > 1. [43] Damped sinusoidal signals incorporate exponential modulation, derived via the scaling property: if x \leftrightarrow X(z), then a^n x \leftrightarrow X(z/a). Applying this to the undamped cosine pair gives the Z-transform of the damped cosine a^n \cos(\omega n) u as X(z) = \frac{z (z - a \cos \omega)}{z^2 - 2 a \cos \omega \, z + a^2}, with ROC |z| > |a| for |a| < 1 ensuring decay. [40] The corresponding sine pair is a^n \sin(\omega n) u \leftrightarrow \frac{a z \sin \omega}{z^2 - 2 a z \cos \omega + a^2}, with the same ROC. [40] Finite-duration signals like the rectangular pulse are derived using time-shifting and summation properties on the unit impulse or step. For a causal rectangular pulse of length N, defined as x = 1 for $0 \leq n \leq N-1 and 0 otherwise, the Z-transform is the finite geometric series X(z) = \sum_{n=0}^{N-1} z^{-n} = \frac{1 - z^{-N}}{1 - z^{-1}}, valid for all z \neq 0 (no ROC restriction beyond the origin due to finite support). [44] More complex shapes, such as the triangular pulse, are obtained via the convolution property: the Z-transform of the convolution of two signals is the product of their individual transforms. For instance, convolving two rectangular pulses of lengths M and L yields a triangular signal whose Z-transform is the product of the respective geometric series expressions. To summarize key derived pairs and illustrate property applications, the following table lists selected examples, emphasizing how shifting creates windows and scaling/modulation builds oscillatory signals:| Time-domain signal x | Z-domain X(z) | ROC | Derivation notes |
|---|---|---|---|
| \sin(\omega n) u | \frac{z \sin \omega}{z^2 - 2 z \cos \omega + 1} | $ | z |
| \cos(\omega n) u | \frac{z (z - \cos \omega)}{z^2 - 2 z \cos \omega + 1} | $ | z |
| a^n \sin(\omega n) u | \frac{a z \sin \omega}{z^2 - 2 a z \cos \omega + a^2} | $ | z |
| a^n \cos(\omega n) u | \frac{z (z - a \cos \omega)}{z^2 - 2 a \cos \omega \, z + a^2} | $ | z |
| Rectangular pulse: \sum_{n=0}^{N-1} u - u[n-N] | \frac{1 - z^{-N}}{1 - z^{-1}} | All z \neq 0 | Time shifts on unit step |
Relationships to Other Transforms
Fourier Transform and Series
The Z-transform provides a powerful framework for analyzing discrete-time signals in the frequency domain through its relationship to the Discrete-Time Fourier Transform (DTFT). Specifically, if the region of convergence (ROC) of the Z-transform X(z) includes the unit circle (|z| = 1), the DTFT X(e^{j\omega}) is obtained by evaluating X(z) on this circle, given by X(e^{j\omega}) = X(z) \big|_{z = e^{j\omega}}.[20] This evaluation corresponds to setting z = re^{j\omega} and taking the limit as r \to 1 from within the ROC, ensuring convergence where the direct DTFT summation might otherwise diverge.[45] For periodic discrete-time signals, the Z-transform relates to the Discrete Fourier Series (DFS). A periodic signal with period N can be represented by its DFS coefficients, and the Z-transform of one period evaluated at the Nth roots of unity equals N times the DFS coefficients, effectively linking the two representations for frequency analysis of repeating sequences.[46] In the context of linear time-invariant (LTI) discrete systems, the Z-transform on the unit circle yields the frequency response H(e^{j\omega}), which describes the system's magnitude and phase response as a function of normalized frequency \omega. This is particularly useful for digital filters, where poles and zeros near the unit circle amplify or attenuate specific frequencies, enabling design and analysis of filtering behavior.[47] One key advantage of the Z-transform over the direct DTFT is its use of the ROC to handle signals where the DTFT does not converge in the ordinary sense, allowing frequency-domain insights via analytic continuation or limiting processes. For instance, the unit step function u has Z-transform X(z) = \frac{1}{1 - z^{-1}} for |z| > 1, and its DTFT is derived as the limit \lim_{r \to 1^+} X(r e^{j\omega}) = \frac{1}{1 - e^{-j\omega}} + \pi \sum_{k=-\infty}^{\infty} \delta(\omega + 2\pi k), capturing the DC component via Dirac deltas despite the infinite energy of the signal.[48][49]Laplace Transform
The Z-transform serves as the discrete-time analog of the Laplace transform, particularly in the context of sampled continuous-time signals, where a continuous signal x_c(t) is converted to a discrete sequence via impulse sampling x = x_c(nT) with sampling period T. This relationship arises from the Poisson summation formula, which describes the Laplace transform of the impulse-sampled signal as X^*(s) = \frac{1}{T} \sum_{k=-\infty}^{\infty} X_c(s + j k \omega_s), where \omega_s = 2\pi / T is the sampling frequency.[50] This starred Laplace transform X^*(s) accounts for aliasing effects due to sampling and directly links to the Z-transform via the substitution z = e^{sT}, yielding X(z) = X^*(s)|_{s = \frac{\ln z}{T}}.[51] A key mapping between the domains is the exponential substitution z = e^{sT}, which maps the left-half s-plane to the interior of the unit circle in the z-plane, preserving stability for causal systems. This forms the basis of the matched Z-transform method, where poles and zeros of an analog transfer function H(s) are mapped to discrete equivalents by p_d = e^{p_a T} and z_d = e^{z_a T}, ensuring the digital filter retains the transient response characteristics of the continuous prototype at sampling instants.[52] The method is particularly useful for preserving dominant pole locations in digital implementations but does not account for frequency warping. To mitigate issues like aliasing in frequency response, the bilinear transform provides a conformal mapping s = \frac{2}{T} \frac{1 - z^{-1}}{1 + z^{-1}}, which approximates the s-plane with a warped z-plane, mapping the entire jω-axis to the unit circle without aliasing but introducing nonlinear frequency compression (warping). Originally developed for analyzing linear systems via time series approximations, this transform enables direct design of digital filters from analog prototypes by substituting into the continuous transfer function.[53] In control systems, these mappings facilitate the digitization of continuous controllers, such as converting PID regulators from s-domain to z-domain for sampled-data implementations. The matched Z-transform is favored for retaining time-domain step responses, while the bilinear transform excels in preserving frequency-domain stability margins, allowing hybrid systems to approximate continuous performance with discrete hardware.Applications to Discrete Systems
Difference Equations
The Z-transform provides a powerful method for solving linear constant-coefficient difference equations, which describe the behavior of linear time-invariant discrete-time systems. A general Nth-order difference equation takes the form \sum_{k=0}^{N} a_k y[n - k] = \sum_{m=0}^{M} b_m x[n - m], where a_0 = 1, y is the output sequence, x is the input sequence, and the coefficients a_k, b_m are constants. Applying the unilateral Z-transform to both sides, leveraging the linearity and time-shifting properties, yields an algebraic equation of the form \sum_{k=0}^{N} a_k Z\{y[n-k]\} = \sum_{m=0}^{M} b_m Z\{x[n-m]\}, where Z\{y[n-k]\} = z^{-k} Y(z) + \sum_{j=1}^{k} y[-j] z^{-k+j} (assuming x=0 for n<0), resulting in Y(z) = H(z) X(z) + IC(z), with IC(z) denoting terms depending on the initial conditions y[-1], \dots, y[-N] and H(z) representing the system transfer function. Under initial rest conditions, where y = 0 for n < 0, the initial condition terms vanish, simplifying the unilateral Z-transform application to Y(z) = H(z) X(z), and transients due to prior states are ignored. This assumption is common in signal processing applications starting from rest.[54][55] To obtain the time-domain solution, the steps involve: (1) taking the Z-transform of the difference equation to express Y(z) in terms of X(z) and initial conditions; (2) solving the resulting algebraic equation for Y(z); and (3) applying the inverse Z-transform to Y(z) to recover y, typically via partial fraction decomposition into known transform pairs. The total solution decomposes into the homogeneous solution, satisfying the equation with zero input, and a particular solution addressing the specific input. The homogeneous solution arises from the characteristic equation \sum_{k=0}^{N} a_k r^{N-k} = 0, yielding y_h = \sum_i A_i r_i^n u for distinct roots r_i, where the A_i are determined by initial conditions. The particular solution y_p is found by assuming a form matching the input x or directly via the Z-transform method.[34][56] For illustration, consider the first-order equation y - a y[n-1] = x with initial rest condition y[-1] = 0. The Z-transform yields Y(z) (1 - a z^{-1}) = X(z), so Y(z) = \frac{X(z)}{1 - a z^{-1}}. The inverse Z-transform gives the convolution sum y = \sum_{k=0}^n a^{n-k} x for n \geq 0, which combines the homogeneous solution A a^n and the particular response to x.[57][58]Transfer Functions and Pole-Zero Analysis
In discrete-time systems, the transfer function H(z) is defined as the ratio of the Z-transform of the output Y(z) to the Z-transform of the input X(z), assuming zero initial conditions:H(z) = \frac{Y(z)}{X(z)}.
For linear time-invariant (LTI) systems described by linear constant-coefficient difference equations, H(z) takes a rational form
H(z) = \frac{\sum_{k=0}^{M} b_k z^{-k}}{1 + \sum_{k=1}^{N} a_k z^{-k}},
where the numerator coefficients b_k and denominator coefficients a_k are determined by the system's difference equation, and the form is typically proper with M \leq N. This representation facilitates analysis in the Z-domain, analogous to the Laplace-domain transfer function for continuous systems. The zeros of H(z) are the roots of the numerator polynomial, corresponding to values of z where H(z) = 0. These zeros influence signal cancellation and the system's frequency response by attenuating specific modes in the input signal. In contrast, the poles are the roots of the denominator polynomial, which dictate the natural modes of the system's response; repeated poles can lead to non-decaying oscillatory behaviors if located appropriately. Pole-zero plots visualize these roots in the complex Z-plane, aiding in the design and interpretation of digital filters. A key aspect of pole-zero analysis is system stability. For bounded-input bounded-output (BIBO) stability in causal LTI systems, all poles must lie strictly inside the unit circle (|z| < 1), ensuring the impulse response h is absolutely summable, and the region of convergence (ROC) must include the unit circle |z| = 1. This criterion guarantees that the system's output remains bounded for any bounded input. Marginal stability occurs when poles lie on the unit circle, leading to sustained oscillations, while poles outside the unit circle result in instability with exponentially growing responses. The output in the Z-domain is obtained by multiplication: Y(z) = H(z) X(z), which corresponds to convolution in the time domain: y = h * x, where h is the inverse Z-transform of H(z), representing the system's impulse response. This duality enables efficient computation of system responses using Z-domain tools before transforming back to the time domain. As an illustrative example, consider a second-order infinite impulse response (IIR) filter with transfer function
H(z) = \frac{1}{1 - 2r \cos(\theta) z^{-1} + r^2 z^{-2}},
where $0 < r < 1 and \theta parameterizes the pole angle. The poles are complex conjugates at z = r e^{\pm j \theta}, inside the unit circle for stability, and there is a zero at the origin. The pole-zero plot reveals a resonant peak in the frequency response near \omega = \theta, with the radial distance r controlling the bandwidth; as r approaches 1, the resonance sharpens, mimicking a narrowband filter used in applications like audio equalization.