Fact-checked by Grok 2 weeks ago

Negative frequency

In signal processing and Fourier analysis, negative frequency refers to a frequency component with a value less than zero in the frequency-domain representation of a signal, arising from the use of complex exponentials in the Fourier transform, where e^{-j\omega t} (with \omega > 0) describes a component rotating clockwise in the complex plane, in contrast to the counterclockwise rotation of positive frequencies e^{j\omega t}.^[1] For real-valued signals, such as sinusoids, the frequency spectrum is Hermitian symmetric (or conjugate symmetric), meaning the magnitude at negative frequency -\omega equals the magnitude at +\omega, while the phase at -\omega is the negative of the phase at +\omega, ensuring that the negative frequency components are the complex conjugates of their positive counterparts and contain no additional information beyond this redundancy.^[2] This symmetry stems from the Euler formula decomposition of real sinusoids, for example, \cos(\omega t) = \frac{e^{j\omega t} + e^{-j\omega t}}{2}, where the e^{-j\omega t} term explicitly contributes the negative frequency -\omega.^[1] The concept is fundamental to understanding the full two-sided spectrum in tools like the fast Fourier transform (FFT), where the output array includes both positive and negative frequencies for real signals, with the negative half being redundant and often discarded or doubled in power calculations to form a single-sided spectrum for practical analysis.^[3] Negative frequencies play a key role in applications such as analytic signal generation, where filtering out negative frequency components produces a complex-valued signal containing only positive frequencies, which is essential for tasks like amplitude modulation demodulation, Hilbert transformation, and envelope detection in communications and audio processing.^[4] Beyond signal processing, negative frequencies appear in wave physics and quantum mechanics, such as in the decomposition of electromagnetic waves or quantum fields into positive and negative frequency parts associated with creation and annihilation operators, though their interpretation remains tied to mathematical convenience rather than physical rotation direction.^[5]

Mathematical Representation

Complex Exponentials

The complex exponential function forms the foundational mathematical representation for oscillatory signals, particularly through Euler's formula, which states that e^{i \omega t} = \cos(\omega t) + i \sin(\omega t).^[6] This equation decomposes the exponential into its real part, a cosine wave, and imaginary part, a sine wave, both with angular frequency \omega. In the complex plane, where the horizontal axis represents the real part and the vertical axis the imaginary part, the term e^{i \omega t} traces a circular trajectory of unit radius centered at the origin. For positive \omega > 0, this rotation proceeds counterclockwise as time t increases, reflecting the standard convention for forward progression in phase.^[7] Conversely, for negative frequency \omega < 0, the rotation is clockwise, inverting the directional sense of the oscillation.^[7] Negative frequency is formally defined in the context of the complex exponential e^{i \omega t} when \omega < 0, distinguishing it from positive frequencies by the reversal of rotational direction in the complex plane. To illustrate, consider \omega = 1 rad/s: the trajectory of e^{i t} starts at (1, 0) at t = 0 and rotates counterclockwise, completing one full cycle every $2\pi seconds, with the point at time t given by (\cos t, \sin t).^[1] For \omega = -1 rad/s, e^{-i t} = \cos(-t) + i \sin(-t) = \cos t - i \sin t, the trajectory mirrors the positive case but rotates clockwise, passing through (1, 0) at t = 0 and reaching (\cos t, -\sin t) at time t, effectively reflecting over the real axis.^[6] This clockwise motion highlights the interpretive role of negative frequencies in capturing reversed phase progression, essential for symmetric signal representations. A key property arises from the conjugation of complex exponentials: for real t and positive \omega, e^{-i \omega t} is the complex conjugate of e^{i \omega t}, since \cos(\omega t) is even and \sin(\omega t) is odd, yielding e^{-i \omega t} = \cos(\omega t) - i \sin(\omega t).^[1] This relationship underscores the mirror symmetry between positive and negative frequency components, where the negative counterpart inverts the imaginary part while preserving the real part's magnitude. In general, complex sinusoids extend this form to A e^{i (\omega t + \phi)}, where A is the complex amplitude (incorporating magnitude and initial phase) and \phi is the phase offset.^[6] When \omega < 0, the phase progression \omega t + \phi decreases with increasing t, effectively inverting the direction of phase advancement compared to positive \omega. Real sinusoids can be expressed as sums of such positive and negative frequency components, a decomposition explored further in subsequent sections.^[1]

Decomposition of Real Signals

Real-valued sinusoidal signals can be decomposed into sums of complex exponentials with positive and negative frequencies using Euler's formula, which states that e^{i\theta} = \cos \theta + i \sin \theta.^[8] To derive the decomposition for the cosine function, start by applying Euler's formula to both e^{i\omega t} and e^{-i\omega t}:

e^{i\omega t} = \cos(\omega t) + i \sin(\omega t),

e^{-i\omega t} = \cos(\omega t) - i \sin(\omega t).

Adding these equations eliminates the imaginary parts:

e^{i\omega t} + e^{-i\omega t} = 2 \cos(\omega t),

yielding the identity

\cos(\omega t) = \frac{e^{i\omega t} + e^{-i\omega t}}{2}.

This expresses the real cosine as an equal sum of positive-frequency and negative-frequency complex exponential components.^[1] A similar derivation applies to the sine function. Subtracting the equations for e^{-i\omega t} from e^{i\omega t} isolates the imaginary parts:

e^{i\omega t} - e^{-i\omega t} = 2i \sin(\omega t),

\sin(\omega t) = \frac{e^{i\omega t} - e^{-i\omega t}}{2i}.

Here, the imaginary unit i in the denominator accounts for the phase shift inherent in the sine relative to the cosine, ensuring the result is real-valued.^[1] For a concrete example, consider a cosine wave \cos(2\pi f t) with frequency f = 5 Hz. Substituting \omega = 2\pi f = 10\pi rad/s gives

\cos(10\pi t) = \frac{e^{i 10\pi t} + e^{-i 10\pi t}}{2},

which decomposes into two complex exponentials of equal amplitude $1/2, one at +5 Hz and the other at -5 Hz.^[1] This decomposition has a key implication for the frequency spectra of real signals: the spectrum F(\omega) must exhibit Hermitian symmetry, satisfying F(-\omega) = F^*(\omega), where F^*(\omega) denotes the complex conjugate of F(\omega). This symmetry ensures that the inverse Fourier transform reconstructs a real-valued time-domain signal.^[9]^[10]

Physical and Conceptual Interpretation

Rotational Direction in Complex Plane

In the complex plane, phasors provide a geometric interpretation of sinusoidal signals, where a positive frequency component corresponds to a vector rotating counterclockwise around the origin, tracing a circular path with increasing phase angle over time.^[11] Conversely, a negative frequency component represents a phasor rotating clockwise, with a decreasing phase angle, which manifests as motion in the opposite direction along the same circular trajectory.^[11] This directional distinction arises naturally from the complex exponential form, where the sign of the angular frequency determines the rotation sense. This rotational behavior offers an analogy to mechanical systems, such as a wheel spinning in reverse or a wave propagating backward along a medium, capturing the intuitive notion of "negative" motion without implying unphysical phenomena.^[11] For instance, consider the real part of the complex exponential \operatorname{Re}\{e^{i\omega t}\}, which equals \cos(\omega t). When \omega > 0, the phasor starts at (1, 0) and rotates counterclockwise, so the projection on the real axis decreases from 1 toward -1 as time advances. For \omega < 0, letting \omega = -|\omega|, the expression becomes \operatorname{Re}\{e^{-i|\omega| t}\} = \cos(|\omega| t), tracing the identical cosine curve but with the phasor rotating clockwise, effectively reversing the temporal progression along the path.^[11] The angular frequency \omega serves as a signed measure of rotational speed, with units of radians per second, where positive values indicate counterclockwise motion and negative values denote clockwise rotation.^[11] This concept traces back to 19th-century vector analysis, notably advanced by Charles Proteus Steinmetz, who in 1893 introduced phasors as rotating vectors to simplify alternating-current circuit analysis, building on earlier geometric interpretations of complex numbers.^[12] In real-valued signals, such rotations pair via Hermitian symmetry to ensure the overall waveform remains real, as explored in spectral decomposition.^[11]

Symmetry in Real-Valued Signals

In the frequency-domain representation of real-valued signals, negative frequencies impose a fundamental constraint known as Hermitian symmetry to ensure the time-domain signal remains real. For a real signal x(t), its continuous-time Fourier transform X(\omega) satisfies X(-\omega) = X^*(\omega), where X^*(\omega) denotes the complex conjugate of X(\omega). This property arises directly from the definition of the Fourier transform:

X(\omega) = \int_{-\infty}^{\infty} x(t) e^{-j \omega t} \, dt.

Taking the complex conjugate yields

X^*(\omega) = \int_{-\infty}^{\infty} x(t) e^{j \omega t} \, dt,

since x(t) is real. Substituting -\omega into the original transform gives

X(-\omega) = \int_{-\infty}^{\infty} x(t) e^{j \omega t} \, dt,

which matches X^*(\omega). Thus, the real part of X(\omega) is even, while the imaginary part is odd, mirroring the signal's reality across the frequency axis.^[13] A key consequence of this symmetry is observed in the power spectral density (PSD), defined as S(\omega) = |X(\omega)|^2. Since |X(-\omega)| = |X(\omega)| from the Hermitian condition, the PSD is even: S(\omega) = S(-\omega). This mirroring implies that the energy distribution is symmetric about zero frequency, with negative frequencies contributing an identical amount of power to their positive counterparts. In practice, this redundancy allows the total power to be computed by integrating over positive frequencies and doubling the result, excluding the DC component.^[13] Consider the example of a real cosine pulse, x(t) = \cos(\omega_0 t) \rect(t/T), where \rect(\cdot) is the rectangular window of duration T. The Fourier transform consists of sinc-like lobes centered at +\omega_0 and -\omega_0, with identical magnitudes but conjugate phases: the phase at -\omega_0 is the negative of that at +\omega_0. This ensures reconstruction yields a real signal, as the negative-frequency component provides the necessary phase opposition to cancel imaginary parts in the inverse transform.^[14] The presence of negative frequencies has significant implications for spectral representations. In a two-sided spectrum, the apparent bandwidth doubles due to the symmetric replication across zero, but this full view is essential for preserving phase information critical in applications like filtering or modulation. Conversely, one-sided spectra, common in power analysis, discard negative frequencies and scale the positive side by a factor of two to account for the mirrored energy, simplifying visualization without losing total power but at the cost of phase details. Negative frequencies thus underpin phase accuracy in two-sided formulations, preventing distortions in signal recovery.^[3]

Applications in Analysis

Fourier Transform Formulation

The two-sided Fourier transform provides a mathematical framework for decomposing signals into their frequency components, explicitly incorporating negative frequencies. It is defined as

X(\omega) = \int_{-\infty}^{\infty} x(t) \, e^{-i \omega t} \, dt,

where the integration over \omega from -\infty to \infty includes both positive and negative angular frequencies, with the complex exponential basis functions e^{-i \omega t} representing oscillations in opposite directions for \omega > 0 and \omega < 0.^[13] This inclusion of negative frequencies streamlines the formulation by enabling a uniform treatment of all spectral components through complex exponentials, as opposed to separate handling of sines and cosines in real-valued representations, which simplifies derivations in areas such as signal reconstruction and spectral analysis.^[13] For real-valued signals x(t), the transform exhibits Hermitian symmetry X(-\omega) = X^*(\omega), ensuring that negative frequency components are the complex conjugates of their positive counterparts and thus carry redundant but essential phase information.^[13] The inverse Fourier transform recovers the original signal via

x(t) = \frac{1}{2\pi} \int_{-\infty}^{\infty} X(\omega) \, e^{i \omega t} \, d\omega.

This can be derived by considering the orthogonality of the complex exponentials over the frequency domain, where the factor $1/(2\pi) arises from the normalization in the angular frequency convention; the contributions from negative frequencies ensure the imaginary parts cancel, yielding a real-valued x(t) through conjugation symmetry.^[13] Historically, Joseph Fourier's 1822 treatise Théorie analytique de la chaleur implicitly utilized negative frequencies in the Fourier series expansion of periodic functions, treating them as part of the sinusoidal decomposition without explicit distinction. This concept was formalized for the continuous Fourier integral by Augustin-Louis Cauchy around 1827, who proved the integral theorem enabling representation of non-periodic functions over the full frequency axis.^[15]^[15]

Convolution and Linear Systems

In the analysis of linear time-invariant (LTI) systems, the convolution theorem states that the Fourier transform of the convolution of an input signal x(t) and the system's impulse response h(t) is equivalent to the product of their individual Fourier transforms: \mathcal{F}\{x(t) * h(t)\} = X(\omega) H(\omega).^[16] This relationship allows efficient computation of the output y(t) = x(t) * h(t) by performing multiplication in the frequency domain followed by an inverse transform. Negative frequencies play a critical role here, as they encode essential phase information; without accounting for components at negative \omega, the phase shifts introduced by the filter would be incomplete, leading to distortions in the reconstructed time-domain signal.^[17] Consider a low-pass filter example, where the transfer function H(\omega) passes frequencies below a cutoff \omega_c while attenuating higher ones. For real-valued signals, H(\omega) exhibits Hermitian symmetry, with symmetric attenuation applied to both positive and negative frequency components (e.g., H(-\omega) = H^*(\omega)), ensuring the filtered output remains real and free of imaginary artifacts.^[18] This symmetry preserves the signal's physical reality, as demonstrated in applications like image smoothing, where multiplying the signal's spectrum by a Gaussian-shaped H(\omega) in the frequency domain yields a convolved result that attenuates high-frequency noise without phase distortion.^[16] The transfer function H(\omega) of an LTI system, obtained as the Fourier transform of h(t), is defined across the entire frequency axis, including negative \omega, to fully characterize system behavior. For causal systems, the response at negative frequencies often reflects non-causal idealizations or bidirectional wave propagation in models like acoustic or electromagnetic systems, enabling accurate prediction of transient responses.^[18] This comprehensive frequency coverage is vital for operations sensitive to phase, such as the Hilbert transform, where suppressing negative frequencies isolates the analytic signal, avoiding errors that would arise from folding only positive frequencies into the analysis.^[19]

Implications in Discrete Processing

Sampling and Nyquist Criterion

In the context of sampling continuous-time signals, the Nyquist-Shannon sampling theorem specifies that a bandlimited signal with frequency components up to a maximum frequency f_{\max} (encompassing both positive frequencies up to +f_{\max} and negative frequencies down to -f_{\max}) can be perfectly reconstructed from its samples if the sampling rate f_s satisfies f_s > 2 f_{\max}.^[20]^[21] This minimum rate, known as the Nyquist rate, ensures that the full spectral content is captured without overlap in the frequency domain.^[22] The bandwidth of such a signal is defined as the total span from -f_{\max} to +f_{\max}, yielding a width of $2 f_{\max}; the inclusion of negative frequencies doubles the effective range relative to the highest positive frequency alone, directly dictating the factor of 2 in the sampling requirement.^[22]^[21] For real-valued signals, the spectrum exhibits conjugate symmetry, meaning negative frequency components mirror their positive counterparts, but they still contribute to the overall bandwidth that must be sampled adequately.^[22] Consider a bandlimited signal whose spectrum extends from -10 kHz to +10 kHz, so f_{\max} = 10 kHz. To avoid distortion, the sampling rate must exceed 20 kHz; if undersampled at, say, 15 kHz, the spectral replicas would overlap, causing the higher frequencies (including negative ones) to fold into the lower baseband, preventing accurate reconstruction.^[21] In discrete-time processing, this maps to the discrete-time Fourier transform (DTFT), where the normalized frequency \omega ranges from -\pi to \pi, corresponding to the full principal period that includes negative frequencies wrapping around from the sampling process.^[23] This range ensures that the sampled signal's spectrum aliases correctly within the Nyquist interval without loss of the original negative frequency information.^[23]

Aliasing from Negative Frequencies

In discrete-time signal processing, aliasing arises when the sampling interval T is such that frequencies satisfying |\omega| > \pi/T are present in the continuous-time signal, causing them to fold into the principal frequency range [-\pi/T, \pi/T] upon sampling. This folding mechanism implies that high-magnitude positive frequencies map to lower positive or negative aliases within the baseband, while negative frequencies similarly alias to positive or negative equivalents, leading to spectral overlap and distortion in the reconstructed signal.^[24] The aliased angular frequency is determined by the relation \omega_{alias} = \omega - \frac{2\pi k}{T} for an integer k selected to ensure \omega_{alias} lies within [-\pi/T, \pi/T]. For a negative frequency component -\omega where \omega > 0, choosing an appropriate k (often positive) shifts it to a positive alias, such as \omega_{alias} = -\omega + \frac{2\pi k}{T} > 0, thereby contaminating the positive frequency spectrum with unintended contributions from the negative side. This bidirectional mapping exacerbates artifacts in real-valued signals, where the spectrum is inherently conjugate symmetric. A representative example illustrates this effect: consider a signal sampled at rate f_s (so T = 1/f_s) containing a negative frequency component at -0.7 f_s. Applying the aliasing formula with k=1 yields \omega_{alias}/(2\pi) = -0.7 f_s + f_s = 0.3 f_s, mapping the negative component to a positive frequency of $0.3 f_s within the baseband [-0.5 f_s, 0.5 f_s], which distorts the perceived spectrum by adding spurious energy to the positive band. Aliasing from negative frequencies is mitigated through anti-aliasing filters, typically low-pass filters that attenuate signal components beyond the Nyquist frequency \pi/T, with symmetric application to ensure both positive and negative high-frequency sides are suppressed prior to sampling. While traditional analog designs suffice for basic prevention, post-2000 digital filter advancements, such as those leveraging multirate structures and sigma-delta modulation, enable more precise control in oversampled architectures, addressing limitations in earlier approaches by reducing transition band widths and improving stopband attenuation.^[25]

References

[1]
Positive and Negative Frequencies - Stanford CCRMA
Every real sinusoid consists of an equal contribution of positive and negative frequency components. This is true of all real signals.
[2]
EE 477 Digital Signal Processing - Montana State University
indicates that the negative frequency portion of the spectrum is the complex conjugate of the positive frequency portion. •If we know one, we can calculate ...
[3]
[PDF] The Fundamentals of FFT-Based Signal Analysis and Measurement
Thus, the negative frequency information is redundant. The two-sided results from the analysis functions include the positive half of the spectrum followed by ...
[4]
Conjugate Symmetry (Hermitian) - UCSD Music
Negative frequencies in a real signal may be ``filtered out'' to produce an analytic signal, a signal which has no negative frequency components.
[5]
[PDF] By E. C. G. Sudarshan lntroduc tion - UT Physics
The field is then broken up into positive and negative frequency parts which are respectively associated with annihilation and creation operators for positive ...
[6]
1.8: Continuous Time Complex Exponential - Engineering LibreTexts
May 22, 2022 · Eulers formula reveals that, in general, the real and imaginary parts of complex exponentials are sinusoids multiplied by real exponentials.
[7]
[PDF] Euler's Formula Where does Euler's formula eiθ = cosθ + isinθ come ...
This celebrated formula links together three numbers of totally different origins: e comes from analysis, π from geometry, and i from algebra. Here is just one ...
[8]
Symmetries for Real Signals - Stanford CCRMA
If a time-domain signal x is real, then its Fourier transform X is conjugate symmetric (Hermitian): $\displaystyle \zbox{x \in \mathbb{R}^N\Leftrightarrow X(-k
[9]
[PDF] Summary of Fourier Transform Properties - University of Toronto
Jan 17, 2017 · Thus g(t) = g∗(t) if and only if G(f) = G∗(−f), i.e., a signal is real-valued if and only if its Fourier transform exhibits Hermitian symmetry.
[10]
https://www.comm.toronto.edu/~frank/notes/fourier.pdf
[11]
https://eng.libretexts.org/Bookshelves/Electrical_Engineering/Introductory_Electrical_Engineering/A_First_Course_in_Electrical_and_Computer_Engineering_(Scharf)/03%3A_Phasors/3.02%3A_Phasor_Representation_of_Signals
[12]
[PDF] EE 261 - The Fourier Transform and its Applications
1 Fourier Series. 1. 1.1 Introduction and Choices to Make . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. 1.2 Periodic Phenomena .
[13]
Symmetry - Center for Computer Research in Music and Acoustics
The Discrete Time Fourier Transform (DTFT) is the appropriate Fourier transform for discrete-time signals of arbitrary length. It can be obtained as the ...
[14]
Highlights in the History of the Fourier Transform - IEEE Pulse
Jan 25, 2016 · Such representation of the function f, called the Fourier theorem, was first derived by the French mathematician Augustine Louis Cauchy (1789– ...
[15]
[PDF] Lecture 2: 2D Fourier transforms and applications
• Fourier transforms and spatial frequencies in 2D. • Definition and meaning. • The Convolution Theorem. • Applications to spatial filtering. • The Sampling ...
[16]
Filters and Convolution | Mathematics of the DFT - DSPRelated.com
From the convolution theorem, we can see that the phase response $ \Theta(k)$ is the phase-shift added by the filter to an input sinusoidal component at ...
[17]
Negative Frequency - an overview | ScienceDirect Topics
Negative frequency refers to a concept in signal processing where the frequency axis extends into the negative range, indicating a phase shift in the signal.Converting Analog To Digital... · Modulation Waveform And... · 3.1. 1 Analytic...<|control11|><|separator|>
[18]
Hilbert Transform - Stanford CCRMA
The Hilbert transform is an allpass filter that provides a $ 90$ degree phase shift at all negative frequencies, and a $ -90$ degree phase shift at all ...
[19]
[PDF] Communication in the Presence of Noise* - MIT Fab Lab
Shannon: Communication in the Presence of Noise ity than the message space. The type of mapping can be suggested by Fig. 3, where a line is mapped into a ...
[20]
[PDF] Discrete - Time Signals and Systems Sampling - UCSB ECE
Sampled at Nyquist rate, so, the theorem states that 2 samples are enough ... correspondingly for negative frequencies. Thus the discrete - time ...
[21]
[PDF] SAMPLING THEORY - CMU School of Computer Science
Amplitude. Freq. Nyquist Rate". Nyquist Rate". Page 33. 2/20/15. 33. Copyright ... Negative Frequencies. • Recall that FT is defined for negative as well as ...
[22]
[PDF] Discrete-Time Fourier Transform - Higher Education | Pearson
The ideal highpass filter (HPF) has its stopband centered on low frequencies, and its passband extends from | ˆω|= ˆωco out to | ˆω| = π. (The highest ...
[23]
[PDF] Lecture 6: Sampling and Aliasing
So a discrete-time sine at frequency f is also a negative sine at frequency Fs − f , and a positive sine at frequency f − Fs. Page 27. Review. Sampling.
[24]
[PDF] Sampling – I Discrete- Time frequency & Aliasing - UCSB ECE
In signal processing, “aliasing” could be defined as, “two or more discrete-time frequencies , for the same continuous frequency ”. What makes this possible ...
[25]
[PDF] Anti-Aliasing, Analog Filters for Data Acquisition Systems
Jan 18, 2002 · Digital signal processing sys- tems use filters to prevent the aliasing of out-of-band noise and interference. This application note investigates the design ...<|control11|><|separator|>