4. Transform-Domain Signal Analysis

Introduction and Context

Discrete-time signals are often difficult to interpret directly from their sample sequences. A sequence may represent a measured waveform, an audio signal, a control variable, or the output of a digital sensor, yet its most useful characteristics may be hidden in the time-domain description. Transform-domain methods provide an alternative description in which frequency content, system dynamics, stability, causality, and convolution can be studied more systematically.

The discrete-time Fourier transform (DTFT) describes the frequency composition of a sequence. The z-transform extends this idea by introducing a complex variable and thereby providing a unified framework for transient behavior, pole-zero analysis, causality, stability, and inverse transformation. The region of convergence (ROC) is essential: a rational expression alone does not uniquely specify a sequence unless its ROC is also known.

This chapter develops the DTFT and z-transform as analytical tools for signals and linear time-invariant (LTI) systems. Particular attention is given to interpretation, inverse transforms, system functions, and the relationship between pole locations and system behavior. These ideas support later work on digital filter design, frequency response, spectral analysis, and implementation.

Learning Objectives

After studying this chapter, the reader should be able to:

1. Define and interpret the DTFT and z-transform of a discrete-time sequence.

2. Explain the significance of periodicity in the DTFT and the role of the unit circle in the z-plane.

3. Determine and interpret regions of convergence for common signal classes.

4. Obtain inverse z-transforms using partial fractions, power-series expansion, and contour concepts.

5. Derive system functions from difference equations and impulse responses.

6. Use poles, zeros, ROC, causality, and stability to characterize LTI systems.

7. Apply transform-domain methods to practical signal-processing problems.

8. Identify common analytical errors and make technically justified decisions about system behavior.

From Time Sequences to Complex Frequency

For a discrete-time sequence x[n], the DTFT is defined by

X(e^{jω}) = Σ from n = −∞ to ∞ x[n]e^{−jωn}.

When the sum converges, X(e^{jω}) describes the complex amplitude associated with each normalized angular frequency ω, measured in radians per sample. The inverse DTFT is

x[n] = (1/2π) ∫ from −π to π X(e^{jω})e^{jωn} dω.

The frequency variable is periodic with period 2π:

X(e^{j(ω+2π)}) = X(e^{jω}).

This periodicity follows from e^{−j(ω+2π)n} = e^{−jωn} for integer n. Consequently, one interval of length 2π, commonly −π ≤ ω ≤ π, contains all distinct frequency information.

The DTFT converts convolution into multiplication. If y[n] = x[n] * h[n], then

Y(e^{jω}) = X(e^{jω})H(e^{jω}).

This property is central to digital filtering. The magnitude |H(e^{jω})| indicates how sinusoidal components are amplified or attenuated, while the phase ∠H(e^{jω}) describes phase displacement. For a stable LTI system, the frequency response is the DTFT of h[n], provided h[n] is absolutely summable.

The DTFT is powerful but does not always provide a convenient description of initial conditions, exponential growth or decay, or unilateral system behavior. The z-transform addresses these issues by replacing the unit-circle variable e^{jω} with a general complex variable z.

The Bilateral z-Transform

The bilateral z-transform of x[n] is

X(z) = Σ from n = −∞ to ∞ x[n]z^{−n},

where z is complex and may be written as z = re^{jω}. Substitution gives

X(re^{jω}) = Σ x[n]r^{−n}e^{−jωn}.

Thus, the z-transform is a collection of DTFTs of exponentially weighted versions of x[n]. The radius r controls the weighting, while ω controls oscillation. The DTFT is obtained by evaluating the z-transform on the unit circle, z = e^{jω}, when the unit circle belongs to the ROC.

The z-transform is especially useful because it converts time-domain convolution into algebraic multiplication:

Y(z) = X(z)H(z).

For an LTI system with impulse response h[n], the system function is

H(z) = Σ h[n]z^{−n}.

If the input transform exists and the relevant ROC conditions are satisfied, the output is described by Y(z) = H(z)X(z). This relationship makes the z-transform a natural language for system analysis.

Important terminology

A pole is a value of z at which a transform becomes unbounded after cancellation of common factors has been performed. A zero is a value at which the transform is zero. The pole-zero diagram gives a geometric representation of a rational system function. Poles indicate natural modes and strongly influence magnitude, decay rate, resonance, and stability. Zeros produce attenuation or cancellation at particular locations.

A rational transform is commonly written as

X(z) = B(z)/A(z) = b0 Π(1 − zk z^{−1}) / [a0 Π(1 − pk z^{−1})].

The factorization is useful, but the expression must always be accompanied by an ROC when a unique sequence is required.

Regions of Convergence: The Missing Part of a Transform

The ROC is the set of complex values z for which the defining summation converges. It is not merely a mathematical technicality; it identifies the time orientation and growth properties of the sequence.

Consider the right-sided exponential

x[n] = a^n u[n].

Its z-transform is

X(z) = Σ from n = 0 to ∞ a^n z^{−n} = 1/(1 − az^{−1}),

with convergence condition |az^{−1}| < 1, or |z| > |a|. The sequence is therefore associated with an exterior ROC.

For the left-sided sequence

x[n] = −a^n u[−n−1],

the same algebraic expression is obtained,

X(z) = 1/(1 − az^{−1}),

but the ROC is |z| < |a|. The same rational formula represents two different sequences because the ROCs are different.

For a finite-duration sequence, the ROC is generally the entire z-plane except possibly z = 0 or z = ∞. For a right-sided rational sequence, the ROC lies outside the outermost pole. For a left-sided rational sequence, it lies inside the innermost pole. A two-sided sequence has an annular ROC between poles. If poles occur on both sides of the origin, the ROC must be an annulus that contains no poles.

The ROC cannot contain any pole. For a stable LTI system, the unit circle must lie in the ROC. For a causal rational LTI system, the ROC is outside the outermost pole. Therefore, a causal rational system is stable if and only if all poles lie strictly inside the unit circle.

These statements must be applied together. Pole locations alone do not establish causality, and an algebraic expression alone does not establish the time sequence.

Transform Properties and Their Interpretations

Linearity states that the transform of ax1[n] + bx2[n] is aX1(z) + bX2(z), with an ROC determined by the intersection of the individual ROCs, except where cancellation changes the final result.

A delay by n0 samples gives

x[n − n0] ↔ z^{−n0}X(z).

An advance gives z^{n0}X(z), subject to ROC considerations. Multiplication by a^n gives

a^n x[n] ↔ X(z/a),

with the ROC radially scaled by |a|. Time reversal gives x[−n] ↔ X(z^{−1}), and reverses the radial interpretation of the ROC.

The convolution property is

x[n] * h[n] ↔ X(z)H(z).

The multiplication of sequences in time corresponds to convolution in an appropriate transform domain, although such operations require care because convergence conditions may become restrictive.

The initial-value theorem, when applicable, is

x[0] = lim as z→∞ X(z)

for right-sided sequences with suitable convergence. The final-value theorem is more conditional and must not be applied mechanically. In discrete time, if the appropriate poles of (1 − z^{−1})X(z) lie inside the unit circle, then

lim as n→∞ x[n] = lim as z→1 (1 − z^{−1})X(z).

A failure to verify pole conditions can produce an incorrect final value.

Inverse z-Transforms

Inverse transformation is the process of recovering x[n] from X(z) and its ROC. Three approaches are particularly useful.

Power-series expansion

If the ROC is outside the outermost pole, expand in negative powers of z. For example,

X(z) = 1/(1 − 0.5z^{−1}), |z| > 0.5,

has the expansion

X(z) = 1 + 0.5z^{−1} + 0.5^2z^{−2} + ⋯,

so x[n] = (0.5)^n u[n]. If the ROC is |z| < 0.5, expand in positive powers of z instead, producing the left-sided sequence −(0.5)^n u[−n−1].

Partial-fraction expansion

For rational transforms, partial fractions reduce the problem to standard pairs. Suppose

X(z) = 1/[(1 − 0.5z^{−1})(1 − 0.8z^{−1})].

A decomposition of the form

X(z) = A/(1 − 0.5z^{−1}) + B/(1 − 0.8z^{−1})

can be found algebraically. The ROC then determines whether each term is right-sided or left-sided. A common ROC outside 0.8 gives a causal sequence; an ROC inside 0.5 gives a left-sided sequence; the annular ROC 0.5 < |z| < 0.8 gives a two-sided sequence in which one component is right-sided and the other left-sided.

Repeated poles lead to terms such as 1/(1 − pz^{−1})^k. Their inverse transforms contain polynomial factors in n, for example n p^n u[n] for a second-order repeated pole. Such factors explain why repeated poles can produce slower decay or larger transient responses even when |p| is less than one.

Contour interpretation

The inverse z-transform can be expressed through a contour integral,

x[n] = (1/2πj) ∮ X(z)z^{n−1} dz,

where the contour is a closed path within the ROC enclosing the appropriate singularities. Residue theory provides a rigorous basis for inversion and is useful in advanced system analysis, although partial fractions are usually more efficient for standard engineering problems.

System Functions and Difference Equations

Consider a linear constant-coefficient difference equation

y[n] + a1y[n−1] + ⋯ + aNy[n−N]

= b0x[n] + b1x[n−1] + ⋯ + bMx[n−M].

Under initial-rest conditions, taking z-transforms gives

Y(z)[1 + a1z^{−1} + ⋯ + aNz^{−N}]

= X(z)[b0 + b1z^{−1} + ⋯ + bMz^{−M}].

Therefore,

H(z) = Y(z)/X(z) = [b0 + b1z^{−1} + ⋯ + bMz^{−M}] /

[1 + a1z^{−1} + ⋯ + aNz^{−N}].

The numerator determines zeros and feedforward behavior; the denominator determines poles and feedback dynamics. In a causal implementation, the denominator coefficients govern recursive memory. If poles approach the unit circle, the system may exhibit long transients and sharp frequency selectivity. If a pole lies outside the unit circle, a causal system is unstable.

The impulse response is the inverse z-transform of H(z). The frequency response is H(e^{jω}), obtained by evaluating the system function on the unit circle when the unit circle lies in the ROC. A system can possess a formal expression H(z) but lack a valid frequency response if the unit circle is outside its ROC.

Worked Analytical Example: Stability and Realization

Consider

H(z) = (1 − 0.4z^{−1}) / [(1 − 0.7z^{−1})(1 − 1.1z^{−1})].

The poles are 0.7 and 1.1, while the zero is 0.4. If the system is causal, its ROC is |z| > 1.1. Because the unit circle is not in this ROC, the causal system is unstable. The pole at 1.1 produces a growing mode proportional to (1.1)^n.

Could a stable realization be associated with the same algebraic expression? A stable ROC would need to include |z| = 1. Since the poles are at radii 0.7 and 1.1, the annular ROC 0.7 < |z| < 1.1 contains the unit circle. This corresponds to a two-sided impulse response, not a causal one. Thus, the same pole-zero expression can describe a stable noncausal system but not a stable causal system.

This example illustrates an important engineering trade-off: causality and stability are not independent labels attached to a rational expression. They are jointly determined by pole locations and the selected ROC.

Application Case: Vibration Monitoring in a Rotating Machine

Background

An industrial motor is monitored using a discrete accelerometer sampled at a fixed rate. Engineers observe a periodic vibration component and need to distinguish a harmless rotational harmonic from a developing bearing fault.

Situation and problem

The raw time record contains operating noise and slowly varying amplitude. A direct inspection of samples does not clearly reveal whether the suspected component is stationary. The team models a digital analysis filter using a second-order system function with complex-conjugate poles near the unit circle.

Analysis

The pole radius controls the persistence of the filter response, while the pole angle identifies the normalized frequency at which the response is concentrated. Moving the poles closer to the unit circle increases selectivity but also lengthens the transient and makes coefficient quantization more critical. The engineers verify that the poles remain inside the unit circle and evaluate H(e^{jω}) to determine the passband and phase behavior.

The measured data are transformed using an FFT, which numerically samples the DTFT. The suspected harmonic appears as a stable spectral peak. The team also examines successive records and uses the filter output envelope to determine whether the component grows with load.

Approach and outcome

A stable narrowband filter is selected rather than an arbitrarily high-Q filter. Its transient duration is measured before deployment so that startup data are not misclassified. The design is implemented with a numerically robust second-order-section structure, and pole locations are checked after finite-precision coefficient conversion.

Learning points

The case demonstrates that transform-domain analysis is not limited to plotting spectra. Pole radius, pole angle, ROC, transient behavior, numerical realization, and operating conditions must be considered together. A sharp spectral response may improve detection but can also increase sensitivity to uncertainty and delay.

Practical Uses and Professional Significance

Transform-domain analysis is used in digital filter design, communications receivers, speech and audio processing, biomedical instrumentation, radar, vibration monitoring, image and video systems, and control systems. In communications, poles and zeros help characterize equalizers and resonators. In biomedical systems, frequency-domain descriptions assist in separating physiological rhythms from interference. In embedded products, z-domain analysis helps engineers choose stable architectures and estimate computational requirements.

In professional practice, analysis often proceeds from a specification to a model, then from a model to a realization. The engineer should identify the required frequency behavior, determine acceptable transient and phase distortion, select pole-zero locations, verify stability and sensitivity, and finally validate the finite-precision implementation. A mathematically correct transfer function can still be unsuitable if it creates excessive delay, overflow risk, coefficient sensitivity, or unacceptable behavior under changed sampling conditions.

Modern software packages calculate transforms, pole-zero diagrams, frequency responses, and inverse transforms automatically. Such tools improve productivity but do not replace reasoning about ROC, assumptions, initial conditions, or numerical accuracy. FFT output is also not identical to the continuous DTFT: it is a set of samples of the DTFT, affected by record length, windowing, leakage, and sampling frequency.

Common Errors and Limitations

A frequent error is reporting X(z) without its ROC. This loses information about sidedness and may lead to contradictory conclusions about stability. Another error is assuming that every rational expression represents a causal sequence. Causality requires a right-sided ROC for the relevant system.

Students also commonly infer stability merely from a pole being less than one without specifying the ROC. A pole-zero cancellation may conceal internal modes, especially in practical realizations; therefore, cancellation in an external transfer function does not always eliminate implementation sensitivity. The final-value theorem is often used without checking its pole conditions. Similarly, evaluating H(e^{jω}) when the unit circle is outside the ROC is invalid.

Numerical limitations become important for poles near the unit circle. Finite word length can move poles, alter notch frequencies, or cause limit cycles in recursive systems. High-order direct-form implementations may be more sensitive than cascaded second-order sections. In safety-critical applications, independent verification, bounded arithmetic analysis, and testing across component tolerances are appropriate professional practices.

Emerging Directions

Transform-domain analysis is increasingly integrated with adaptive filtering, time-frequency methods, model-based fault diagnosis, and automated design optimization. Data-driven systems may estimate spectral features or pole locations from large datasets, but transform-domain constraints remain valuable for interpretability and stability guarantees. In embedded machine learning, spectral preprocessing is often combined with learned classifiers; engineers must still account for sampling, aliasing, latency, and robustness.

A growing emphasis is placed on reproducible computational analysis. Design records should specify sampling rate, transform conventions, windowing choices, normalization, coefficient precision, and stability tests. These practices make results auditable and reduce errors when algorithms move from simulation to hardware.

Best-Practice Workflow

A reliable analysis generally follows this sequence:

1. State the signal or system assumptions, including causality, initial rest, and sampling convention.

2. Write the transform definition and identify convergence conditions.

3. Factor rational expressions to locate poles and zeros.

4. Determine the possible ROCs from the signal class or physical realization.

5. Use the ROC with pole locations to decide sidedness, causality, and stability.

6. Perform inversion using partial fractions or series expansion.

7. Evaluate the unit-circle response only when the unit circle belongs to the ROC.

8. Check results through limiting cases, direct substitution, simulation, and numerical realization analysis.

Chapter Summary

The DTFT represents the frequency content of a discrete-time signal and is periodic in normalized frequency. The z-transform generalizes the DTFT by introducing radial weighting through the complex variable z. Its central advantages are the treatment of exponential behavior, convolution, difference equations, poles and zeros, and system stability.

A transform expression is incomplete without its ROC. Exterior, interior, and annular ROCs correspond respectively to right-sided, left-sided, and two-sided sequences in common rational cases. Causal rational systems have an exterior ROC, while stability requires the unit circle to lie within the ROC. Consequently, a causal rational system is stable precisely when all its poles lie strictly inside the unit circle.

Inverse z-transforms can be obtained through power-series expansion, partial fractions, or contour integration. Difference equations lead directly to system functions, whose poles determine natural modes and whose zeros shape spectral attenuation. Practical analysis must also consider transients, phase, numerical precision, implementation structure, and the difference between an ideal transform model and a finite sampled computation.

Key Terms

DTFT: The discrete-time Fourier transform that represents a sequence as a periodic function of normalized frequency.

z-transform: A complex-frequency transform defined by the sum Σx[n]z^{-n}.

ROC: The region of the z-plane in which a z-transform summation converges.

Unit circle: The locus |z| = 1, on which the frequency response is obtained from a valid z-transform.

System function: The transform ratio H(z) = Y(z)/X(z) under initial-rest conditions.

Pole: A value of z at which a reduced rational transform becomes unbounded.

Zero: A value of z at which a transform becomes zero.

Right-sided sequence: A sequence that is zero sufficiently far to the left, commonly associated with an exterior ROC.

Left-sided sequence: A sequence that is zero sufficiently far to the right, commonly associated with an interior ROC.

Two-sided sequence: A sequence extending in both time directions, commonly associated with an annular ROC.

Causality: The property that the output at a time depends only on present and past inputs.

BIBO stability: Bounded-input, bounded-output stability; for an LTI system it requires an absolutely summable impulse response.

Frequency response: The system function evaluated on the unit circle, H(e^{jω}), when valid.

Impulse response: The output of an LTI system to a unit impulse.

Pole-zero diagram: A graphical representation of poles and zeros in the z-plane.

Partial-fraction expansion: Algebraic decomposition used to invert rational transforms.

Initial-value theorem: A theorem relating x[0] to the limiting behavior of X(z) at infinity under suitable conditions.

Final-value theorem: A conditional relation between the long-term sequence value and X(z) near z = 1.

Normalized angular frequency: Discrete-time frequency measured in radians per sample and periodic modulo 2π.

FFT: A computationally efficient algorithm for evaluating samples of the discrete Fourier transform.

Review Questions

Short Answer Questions

1. Why is the DTFT periodic with period 2π?

2. Define the ROC of a z-transform.

3. What is the relationship between the DTFT and the z-transform?

4. State the ROC associated with a causal rational sequence.

5. Why must the unit circle lie in the ROC for a stable system to have a frequency response?

Long Answer Questions

1. Derive the z-transform and ROC of a^n u[n] and of −a^n u[−n−1]. Explain why the algebraic expressions are identical.

2. Explain how poles, zeros, ROC, causality, and stability are related for a rational LTI system.

3. Describe three methods for finding an inverse z-transform and identify when each is useful.

4. Derive the system function of a general constant-coefficient difference equation under initial-rest conditions.

5. Discuss the practical effects of poles near the unit circle in a digital filter implementation.

Analytical / Application Questions

1. For H(z) = 1/[(1 − 0.4z^{-1})(1 − 0.9z^{-1})], determine the possible ROCs and state which one corresponds to a causal stable system.

2. A rational system has poles at 0.6 and 1.2. Can it be both causal and stable? Justify using the ROC.

3. A vibration-monitoring filter requires a narrow passband but must operate reliably with finite-precision coefficients. Explain the trade-off involved in moving its poles closer to the unit circle and recommend validation checks.

22. Multiple Choice Questions with Answers

Question 1

For x[n] = a^n u[n], which ROC is correct?

A. |z| < |a|

B. |z| > |a|

C. |z| = |a| only

D. The entire z-plane including z = a

Correct Answer: B. |z| > |a|

Question 2

What additional information is required to uniquely identify a rational z-transform sequence?

A. Sampling frequency only

B. The phase response only

C. The region of convergence

D. The number of zeros only

Correct Answer: C. The region of convergence

Question 3

The DTFT of a discrete-time sequence is periodic in ω with period

A. π

B. 2π

C. 4π

D. Dependent on the pole locations

Correct Answer: B. 2π

Question 4

For a causal rational LTI system, the ROC is generally

A. Inside the innermost pole

B. Between the poles only

C. Outside the outermost pole

D. The unit circle only

Correct Answer: C. Outside the outermost pole

Question 5

A causal rational system is BIBO stable when

A. At least one pole is outside the unit circle

B. All poles lie strictly inside the unit circle

C. All zeros lie inside the unit circle

D. The ROC excludes the unit circle

Correct Answer: B. All poles lie strictly inside the unit circle

Question 6

If the unit circle is not contained in the ROC of H(z), then

A. H(e^{jω}) is not a valid frequency response for that realization

B. The system must be causal and stable

C. All poles must be zeros

D. The impulse response must be finite length

Correct Answer: A. H(e^{jω}) is not a valid frequency response for that realization

Question 7

For y[n] = x[n] * h[n], the z-domain relationship is

A. Y(z) = X(z) + H(z)

B. Y(z) = X(z) − H(z)

C. Y(z) = X(z)H(z)

D. Y(z) = X(z)/H(z)

Correct Answer: C. Y(z) = X(z)H(z)

Question 8

A pole at radius 0.98 in a causal system generally produces

A. A rapidly vanishing transient

B. A long-lasting transient

C. No transient response

D. An impulse response that is necessarily finite

Correct Answer: B. A long-lasting transient

Question 9

For the same rational expression, an exterior ROC and an interior ROC typically represent

A. The same sequence with different units

B. Right-sided and left-sided sequences, respectively

C. Two different sampling frequencies

D. Two identical frequency responses with no distinction

Correct Answer: B. Right-sided and left-sided sequences, respectively

Question 10

The frequency response of a system is obtained from its system function by

A. Setting z = 0

B. Setting z = ∞

C. Evaluating at z = e^{jω}, provided the unit circle is in the ROC

D. Differentiating with respect to z

Correct Answer: C. Evaluating at z = e^{jω}, provided the unit circle is in the ROC