3. Fourier Series and Spectral Representation

Introduction and Context

Fourier analysis provides one of the most powerful ways of viewing signals and systems. In the time domain, a signal is described by how its amplitude varies with time. Fourier analysis offers a complementary frequency-domain description: a signal is represented as a collection of sinusoidal components, each having a particular frequency, amplitude, and phase. This change of representation is especially valuable because many physical systems respond differently to different frequencies.

Periodic signals are naturally studied using Fourier series. A Fourier series expresses a periodic signal as an infinite weighted sum of harmonically related sinusoids or complex exponentials. The representation is not merely a mathematical rearrangement. It reveals the spectral content of a waveform, permits convenient analysis of linear time-invariant systems, and explains phenomena such as bandwidth, filtering, resonance, waveform distortion, and harmonic interference.

The chapter develops trigonometric and exponential Fourier series, establishes the conditions under which the series converges, and interprets magnitude, phase, power, and energy spectra. Particular attention is given to symmetry, Gibbs oscillations, Parseval’s theorem, and the practical use of Fourier coefficients in engineering systems.

Learning Objectives

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

1. Explain why periodic signals can be represented through harmonically related sinusoids.

2. Derive and use the trigonometric and exponential Fourier-series forms.

3. Determine Fourier coefficients using integration and symmetry arguments.

4. Interpret line spectra, magnitude spectra, phase spectra, and power spectra.

5. State the principal convergence conditions and explain the Gibbs phenomenon.

6. Apply Parseval’s theorem to calculate average signal power.

7. Analyze the response of an LTI system to a periodic input using Fourier coefficients.

8. Recognize practical limitations, numerical issues, and engineering applications of Fourier-series representations.

Periodicity, Harmonics, and Spectral Representation

A continuous-time signal x(t) is periodic if there exists a positive number T_0 such that

x(t + T_0) = x(t)

for every t. The smallest positive value satisfying this relation is the fundamental period, and the corresponding fundamental angular frequency is

omega_0 = 2 pi / T_0.

The frequencies n omega_0, where n is an integer, are called harmonic frequencies. The zeroth harmonic is the constant or dc component. The first harmonic is the fundamental frequency, while the second, third, and subsequent harmonics occur at integer multiples of the fundamental frequency.

The importance of harmonics can be seen in a square wave. Although the waveform has abrupt transitions, it can be constructed from a fundamental sinusoid and an appropriate set of higher-frequency harmonics. A waveform’s sharpness is associated with the presence of substantial high-frequency components. Removing these components with a low-pass filter rounds the transitions and changes the time-domain shape.

A Fourier series is appropriate for periodic signals. Nonperiodic signals require the Fourier transform, which may be regarded conceptually as the limiting case in which the period becomes infinitely large and the discrete spectral lines become a continuous spectrum.

The Trigonometric Fourier Series

For a continuous-time periodic signal with period T_0, the trigonometric Fourier series is

x(t) = a_0 + sum from n=1 to infinity [a_n cos(n omega_0 t) + b_n sin(n omega_0 t)].

The coefficients are obtained over any interval of length T_0:

a_0 = (1/T_0) integral over one period of x(t) dt,

a_n = (2/T_0) integral over one period of x(t) cos(n omega_0 t) dt,

b_n = (2/T_0) integral over one period of x(t) sin(n omega_0 t) dt.

The coefficient a_0 is the average or dc value of the signal. The nth harmonic can also be written in amplitude-phase form:

A_n cos(n omega_0 t - phi_n),

where

A_n = sqrt(a_n^2 + b_n^2),

phi_n = atan2(b_n, a_n).

The two-argument arctangent is preferable because it preserves the correct quadrant of the phase. Expanding the amplitude-phase expression gives a_n = A_n cos(phi_n) and b_n = A_n sin(phi_n).

The trigonometric form is intuitive for real signals because it directly displays cosine and sine contributions. It is particularly convenient when symmetry is present. If x(t) is even, all b_n vanish. If x(t) is odd, a_0 and all a_n vanish. If a signal has half-wave symmetry, x(t + T_0/2) = -x(t), its dc component and all even harmonics vanish.

The Exponential Fourier Series

Using Euler’s identities,

e^{j theta} = cos(theta) + j sin(theta),

cos(theta) = (e^{j theta} + e^{-j theta})/2,

sin(theta) = (e^{j theta} - e^{-j theta})/(2j),

the trigonometric series can be written as

x(t) = sum from n=-infinity to infinity of C_n e^{j n omega_0 t},

where

C_n = (1/T_0) integral over one period of x(t)e^{-j n omega_0 t} dt.

For a real-valued signal,

C_{-n} = C_n^*.

Consequently, the magnitude spectrum is even and the phase spectrum is odd, apart from phase ambiguities at zero-valued coefficients. The connection between the two forms is

C_0 = a_0,

C_n = (a_n - j b_n)/2 for n > 0,

C_{-n} = (a_n + j b_n)/2.

The exponential form is often superior in system analysis. Differentiation, time shifting, convolution with an LTI system, and modulation can be expressed compactly through complex exponentials.

For an LTI system with frequency response H(j omega), if the input is

x(t) = sum C_n e^{j n omega_0 t},

then the steady-state output is

y(t) = sum C_n H(j n omega_0)e^{j n omega_0 t}.

Thus, the system independently scales and phase-shifts each harmonic. This property is the central reason Fourier series is so important in signal and system analysis.

Calculating Coefficients: A Structured Method

A reliable coefficient calculation begins by identifying the period and selecting a convenient integration interval. The signal should then be examined for even, odd, quarter-wave, or half-wave symmetry. Symmetry can eliminate many integrals before any algebra is performed.

For example, consider the periodic square wave of amplitude A defined over one period by x(t) = A for 0 < t < T_0/2 and x(t) = -A for -T_0/2 < t < 0. The signal is odd and has half-wave symmetry. Therefore, its dc term and cosine coefficients are zero. Its sine coefficients are

b_n = (2A/(n pi))[1 - (-1)^n].

Hence b_n = 4A/(n pi) for odd n and b_n = 0 for even n. The series is therefore

x(t) = (4A/pi)[sin(omega_0 t) + (1/3)sin(3 omega_0 t) + (1/5)sin(5 omega_0 t) + ...].

The expression demonstrates that a square wave contains only odd harmonics and that the magnitude of the nth nonzero harmonic decreases as 1/n. The slow decay is consistent with the waveform’s discontinuities.

As another example, a periodic triangular wave has continuous but sharply changing slopes. Its Fourier coefficients decay more rapidly, typically as 1/n^2. A smoother waveform therefore requires less high-frequency content to reproduce it accurately. In general, greater time-domain smoothness is associated with faster spectral decay.

Convergence and the Gibbs Phenomenon

A Fourier series does not necessarily converge pointwise to the signal at every time. The classical Dirichlet conditions provide sufficient conditions for convergence over a period. A signal should be absolutely integrable over one period and have a finite number of maxima, minima, and discontinuities. Under these conditions, the Fourier series converges at a point t to

[x(t^+) + x(t^-)]/2.

At a point of continuity, the limiting value is x(t). At a jump discontinuity, the series converges to the midpoint of the jump rather than to either one-sided value.

Near a discontinuity, partial sums exhibit overshoot and oscillation. This is the Gibbs phenomenon. As more harmonics are included, the oscillatory region becomes narrower, but the maximum overshoot does not disappear; for a jump of height D, its limiting magnitude is approximately 0.08949D on each side. The phenomenon is not caused by an error in the coefficient calculation. It results from representing a discontinuous function with globally smooth sinusoids.

Gibbs oscillations matter in digital signal processing, image reconstruction, control, and communications. Truncating a Fourier representation near abrupt edges can generate ringing. Windowing, spectral shaping, or alternative expansions can reduce visible ringing, although these methods generally broaden transitions or alter other characteristics.

Spectral Descriptions of Periodic Signals

The exponential coefficients C_n form a discrete line spectrum located at frequencies n omega_0. The magnitude spectrum plots |C_n| against frequency, while the phase spectrum plots angle C_n. For a real signal, positive- and negative-frequency components occur in conjugate pairs. The two-sided spectrum is mathematically natural; the one-sided spectrum commonly used for real engineering signals combines the positive and negative contributions and requires appropriate amplitude scaling.

The dc line occurs at zero frequency and has magnitude |C_0|. A missing harmonic corresponds to a zero coefficient. Spectral nulls may result from symmetry, cancellation, or the transfer function of a system. Phase is meaningful only where the magnitude is nonzero, and phase should normally be plotted with a stated wrapping convention.

For the trigonometric representation, the amplitude of the nth sinusoidal harmonic is A_n = 2|C_n| for n > 0. Confusing |C_n| with the amplitude of the corresponding real sinusoid is a common error. In a two-sided exponential spectrum, half of a real sinusoid’s amplitude appears at positive frequency and half at negative frequency.

Power and Energy in Periodic Signals

A nonzero periodic signal generally has infinite total energy because it continues indefinitely. Its useful measure is average power:

P_x = (1/T_0) integral over one period of |x(t)|^2 dt.

For the exponential Fourier series, Parseval’s relation is

P_x = sum from n=-infinity to infinity of |C_n|^2.

For a real signal in trigonometric form,

P_x = a_0^2 + (1/2) sum from n=1 to infinity of (a_n^2 + b_n^2).

This result has both theoretical and practical significance. It states that average power calculated in the time domain equals the sum of the powers associated with all spectral components. It can be used to verify coefficient calculations and to estimate how many harmonics are required to capture a specified fraction of signal power.

For the square wave of amplitude A, x^2(t) = A^2 except at transition instants, so its average power is A^2. The Fourier coefficients must satisfy the same result. This yields the classical relation involving the sum of reciprocals of odd squares, demonstrating how a physical signal property leads to a mathematical identity.

The terms energy spectrum and power spectrum must be distinguished. A finite-energy aperiodic signal has an energy spectrum, while an indefinitely repeated periodic signal is characterized by spectral lines and average power. Applying finite-energy formulas indiscriminately to periodic signals can produce incorrect conclusions.

Fourier Series Through the Lens of LTI Systems

Suppose a periodic input contains a finite or infinite set of harmonics. An LTI system with frequency response H(j omega) produces an output coefficient

D_n = C_n H(j n omega_0).

The output spectrum is therefore obtained by multiplying each input line by the system response at that line. A low-pass system suppresses higher harmonics and smooths sharp features. A band-pass system may select only a narrow group of harmonics, producing a waveform with a different periodic shape. A resonant system can greatly amplify selected harmonics, subject to stability and physical constraints.

This frequency-by-frequency interpretation is useful in power electronics, where switching waveforms generate harmonics; in communication systems, where filters select spectral bands; in vibration analysis, where rotating machinery generates harmonic lines; and in instrumentation, where unwanted periodic interference can be identified and attenuated.

A system may also introduce phase distortion. Two systems with similar magnitude responses but different phase responses can produce different time-domain waveforms. For periodic signals, phase changes alter the alignment of harmonics and therefore the shape and timing of peaks, edges, and zero crossings.

Applied Case Study: Harmonic Analysis of an Inverter Output

Background

Consider an illustrative single-phase power inverter intended to produce an approximately sinusoidal voltage for an isolated electrical load. The switching device generates a periodic bipolar waveform with fundamental frequency 50 Hz. The idealized switching waveform has amplitude ±V and initially resembles a square wave.

Situation and Problem

The load includes sensitive instrumentation that responds poorly to voltage distortion. Although the fundamental component has the required frequency, the square-like waveform contains odd harmonics. The engineering team must estimate the dominant harmonic content and decide how filtering will affect the delivered voltage.

Analysis

Because the waveform has odd symmetry and half-wave symmetry, its dc term and even harmonics are zero. Its Fourier series is

v(t) = (4V/pi)[sin(omega_0 t) + (1/3)sin(3 omega_0 t) + (1/5)sin(5 omega_0 t) + ...].

The fundamental peak amplitude is 4V/pi, while the third and fifth harmonic peak amplitudes are one-third and one-fifth of the fundamental, respectively. The total average power of the ideal waveform is V^2, but this power is distributed among the fundamental and harmonic components.

Assume the output filter has a low-pass response that passes 50 Hz with little attenuation but strongly attenuates 150 Hz and higher frequencies. The output then becomes substantially more sinusoidal. However, the filter must be designed with attention to resonance, component ratings, transient response, and the actual inverter’s switching pattern. A filter that attenuates the third harmonic but resonates near the fifth harmonic could create an unexpected distortion peak.

Approach and Solution

The team first computes the Fourier coefficients of the actual switching waveform, rather than relying solely on the ideal square-wave model. It then evaluates the filter response at each harmonic frequency using D_n = C_nH(jn omega_0). Total harmonic distortion is estimated from the rms harmonic components relative to the rms fundamental component. Finally, measurements are compared with the model under representative load conditions.

Outcome

The Fourier model identifies which harmonics dominate before hardware testing begins. The filter is tuned to attenuate the principal low-order harmonics without excessive reactive current or unstable resonance. Measurements confirm that the filtered waveform has lower distortion, while the design review identifies switching dead time and load variation as sources of residual harmonics.

Learning Points

The case illustrates that Fourier series is not limited to waveform plotting. It supports component sizing, filter design, compliance assessment, fault diagnosis, and communication between modeling and experimental teams. It also shows why the spectrum of the actual operating waveform—not merely an idealized waveform—should guide engineering decisions.

Practical Computation and Numerical Considerations

In practical work, the infinite series is truncated. The partial sum

x_N(t) = sum from n=-N to N of C_n e^{j n omega_0 t}

approximates the signal using 2N + 1 spectral terms. The required value of N depends on the desired accuracy, the signal’s smoothness, and whether accuracy is needed globally or near transitions. A small number of terms may reproduce the general shape but fail to represent narrow features.

Numerical coefficient estimation may use sampled data rather than analytic integration. The discrete Fourier transform and fast Fourier transform provide efficient computational methods, but their interpretation requires care. Sampling frequency, record length, windowing, leakage, aliasing, and synchronization influence the measured spectrum. For a record containing an integer number of periods, spectral leakage is minimized. If the record does not contain an integer number of periods, windowing can reduce sidelobes but changes amplitude calibration and resolution.

Finite precision can also affect very small coefficients and phase estimates. A sound engineering workflow includes checking periodicity, scaling conventions, units, conjugate symmetry, and Parseval consistency.

Common Errors and Limitations

Several mistakes recur in Fourier-series analysis:

• Using an incorrect fundamental period, which places every spectral line at the wrong frequency.
• Omitting the factor 2 in the trigonometric coefficients for n greater than zero.
• Confusing the coefficient C_n with the amplitude of the corresponding real sinusoid.
• Ignoring symmetry and consequently performing unnecessary or error-prone integrations.
• Treating the value at a discontinuity as the one-sided value rather than the midpoint limit.
• Interpreting Gibbs overshoot as a coefficient-calculation failure.
• Mixing peak and rms amplitudes in power calculations.
• Applying one-sided spectrum conventions without adjusting amplitudes.
• Assuming that a magnitude spectrum alone determines the time-domain waveform; phase is also required.

Fourier series also has limitations. It describes periodic steady-state behavior and does not by itself characterize transients. Discontinuities converge slowly, and a large number of harmonics may be needed near sharp edges. The choice of time origin affects phase, although it does not alter the physical content of the signal. Finally, ideal Fourier coefficients do not account automatically for measurement noise, nonstationarity, nonlinear components, or changes in operating conditions.

Professional and Contemporary Applications

Fourier-series methods remain central in power-system harmonic studies, electromagnetic compatibility, audio engineering, vibration monitoring, biomedical instrumentation, radar, communication, and control. In power electronics, harmonic spectra support filter and converter design. In rotating machinery, spectral lines reveal periodic mechanical faults. In audio, the relative magnitude and phase of harmonics influence timbre. In biomedical systems, periodic components can represent useful physiological rhythms or interference requiring removal.

Modern engineering software automates coefficient calculation, spectral estimation, and visualization, but automation does not eliminate the need for conceptual judgment. Engineers must decide whether the signal is genuinely periodic, select appropriate sampling and windowing strategies, distinguish physical harmonics from measurement artifacts, and interpret phase consistently. Data-driven and machine-learning systems may classify spectral patterns, but reliable decisions still depend on physically meaningful features and properly validated measurements.

Professional practice also requires transparent reporting. The period, amplitude convention, time reference, sampling conditions, frequency resolution, window, and normalization should be stated. In regulated or safety-critical applications, undocumented spectral assumptions can lead to incorrect compliance claims or inadequate design margins.

Best-Practice Workflow

A robust Fourier-series analysis can follow this sequence:

1. Establish the fundamental period and angular frequency.

2. Define the signal precisely over one period, including amplitude and time origin.

3. Examine symmetry before calculating integrals.

4. Select the trigonometric or exponential form according to the application.

5. Compute coefficients and verify dimensions and conjugate symmetry.

6. Plot magnitude and phase using clearly stated one-sided or two-sided conventions.

7. Use Parseval’s theorem to check power calculations.

8. Evaluate convergence and the effect of truncation.

9. If an LTI system is involved, multiply each coefficient by the system response at the corresponding harmonic.

10. Validate the model against measurements or a direct numerical reconstruction.

Chapter Summary

Fourier series represents a periodic signal as a sum of harmonically related sinusoids or complex exponentials. The trigonometric coefficients describe cosine and sine content, while exponential coefficients provide a compact representation especially suited to LTI-system analysis. Symmetry can eliminate coefficients and reveal missing harmonics. The magnitude and phase of the coefficients form a discrete line spectrum, and Parseval’s theorem equates time-domain average power with the sum of spectral powers.

Under standard regularity conditions, Fourier series converges to the signal at continuous points and to the midpoint of a jump at discontinuities. Gibbs oscillations occur near jumps and persist in amplitude even as the transition region narrows. Fourier-series analysis supports harmonic estimation, filtering, inverter design, vibration diagnosis, communications, and many other engineering tasks. Correct interpretation requires attention to normalization, phase, sampling, truncation, and the distinction between energy and average power.

Key Terms

Periodic signal: A signal satisfying x(t + T_0) = x(t) for a positive fundamental period T_0.

Fundamental period: The smallest positive period of a periodic signal.

Fundamental frequency: The frequency corresponding to the fundamental period, with angular frequency omega_0 = 2 pi/T_0.

Harmonic: A sinusoidal component whose frequency is an integer multiple of the fundamental frequency.

Fourier series: An expansion of a periodic signal in harmonically related sinusoids or complex exponentials.

Trigonometric Fourier series: A Fourier representation using cosine and sine terms.

Exponential Fourier series: A Fourier representation using complex exponentials e^{jn omega_0t}.

Fourier coefficient: A scalar specifying the magnitude and phase contribution of a particular harmonic.

DC component: The zero-frequency or average value of a signal.

Magnitude spectrum: A plot of the absolute values of spectral coefficients versus frequency.

Phase spectrum: A plot of coefficient phases versus frequency.

Line spectrum: A discrete spectrum consisting of spectral lines at harmonic frequencies.

Conjugate symmetry: The relation C_{-n} = C_n^* for a real-valued signal.

Gibbs phenomenon: Persistent oscillatory overshoot near a discontinuity in a truncated Fourier series.

Parseval’s theorem: A relation equating time-domain average power with the sum of squared Fourier-coefficient magnitudes.

Average power: Mean squared magnitude of a periodic signal over one period.

Half-wave symmetry: The property x(t + T_0/2) = -x(t), which eliminates dc and even harmonics.

Spectral leakage: Spreading of spectral energy caused by finite or non-integer-period observation records.

Total harmonic distortion: A measure comparing the rms value of harmonic components with the rms value of the fundamental.

Review Questions

Short Answer Questions

1. Define the fundamental frequency of a periodic signal and relate it to the fundamental period.

2. State the formulas for the trigonometric Fourier coefficients a_n and b_n.

3. What is the convergence value of a Fourier series at a jump discontinuity?

4. State the conjugate-symmetry property of exponential Fourier coefficients for a real signal.

5. Distinguish between an energy spectrum and a power spectrum.

Long Answer Questions

1. Derive the exponential Fourier-series representation from the trigonometric form using Euler’s identities.

2. Explain how even, odd, half-wave, and quarter-wave symmetries simplify Fourier-series calculations.

3. Derive Parseval’s relation for a periodic signal in exponential form and interpret its engineering meaning.

4. Explain the Gibbs phenomenon, its mathematical behavior, and its practical implications in signal reconstruction.

5. Describe how an LTI system modifies the Fourier coefficients of a periodic input and illustrate the result with a filtering example.

Analytical / Application Questions

1. A bipolar square wave has amplitude ±A and fundamental frequency f_0. Determine which harmonics are present, derive their amplitudes, and calculate the fraction of total power contained in the fundamental.

2. An inverter output contains a fundamental at 50 Hz and significant third and fifth harmonics. Explain how a low-pass filter should be assessed using its frequency response, and identify design risks that an ideal square-wave model may conceal.

3. A measured periodic waveform has a spectrum that lacks even harmonics but contains strong odd harmonics. Infer possible time-domain symmetries and propose measurements or tests to verify the inference.

22. Multiple Choice Questions with Answers

Question 1

For a real-valued periodic signal, which relation must hold for its exponential Fourier coefficients?

A. C_{-n}=C_n

B. C_{-n}=-C_n

C. C_{-n}=C_n^*

D. C_{-n}=|C_n|

Correct Answer: C. C_{-n}=C_n^*

Question 2

A signal with half-wave symmetry x(t+T_0/2)=-x(t) has which spectral property?

A. Only even harmonics are present

B. Its dc and even-harmonic components vanish

C. All sine coefficients vanish

D. Its fundamental component must vanish

Correct Answer: B. Its dc and even-harmonic components vanish

Question 3

At a jump discontinuity, the Fourier series converges, under the stated regularity conditions, to:

A. The larger one-sided value

B. The smaller one-sided value

C. The midpoint of the two one-sided values

D. Zero

Correct Answer: C. The midpoint of the two one-sided values

Question 4

For n greater than zero, what is the amplitude of the real sinusoidal harmonic associated with exponential coefficient C_n?

A. |C_n|/2

B. |C_n|

C. 2|C_n|

D. 4|C_n|

Correct Answer: C. 2|C_n|

Question 5

Which expression gives the average power of a periodic signal using exponential Fourier coefficients?

A. The sum of C_n

B. The sum of |C_n|

C. The sum of |C_n|^2

D. The square of the sum of C_n

Correct Answer: C. The sum of |C_n|^2

Question 6

The Gibbs phenomenon is best described as:

A. A frequency shift caused by changing the time origin

B. Persistent oscillatory overshoot near a discontinuity

C. Complete elimination of high-frequency harmonics

D. A failure of conjugate symmetry

Correct Answer: B. Persistent oscillatory overshoot near a discontinuity

Question 7

If an LTI system has frequency response H(jω), the output Fourier coefficient at harmonic n is:

A. D_n=C_n+H(jnω_0)

B. D_n=C_n/H(jnω_0)

C. D_n=C_nH(jnω_0)

D. D_n=H(jω_0)/C_n

Correct Answer: C. D_n=C_nH(jnω_0)

Question 8

A square wave with odd symmetry and half-wave symmetry contains sine harmonics whose nonzero magnitudes decrease approximately as:

A. 1/n

B. 1/n^2

C. n

D. Constant with n

Correct Answer: A. 1/n

Question 9

Why can a periodic signal be characterized by average power rather than finite total energy?

A. Its amplitude is always zero at discontinuities

B. It continues indefinitely, so its total energy generally diverges

C. Its Fourier coefficients are always real

D. Its fundamental frequency is finite

Correct Answer: B. It continues indefinitely, so its total energy generally diverges

Question 10

Which practical condition most directly reduces spectral leakage in a Fourier measurement?

A. Observing an integer number of signal periods

B. Using fewer samples regardless of duration

C. Removing the phase spectrum

D. Increasing the dc component

Correct Answer: A. Observing an integer number of signal periods