Is phase measured in degrees or radians?

Either, provided the units are stated and used consistently.

Can a magnitude spectrum determine phase?

Not generally without additional assumptions.

What does zero phase mean?

Alignment with the chosen reference convention at that frequency, not necessarily zero physical delay.

Does unwrapping prove the true delay?

No. It resolves representational jumps under assumptions and does not remove aliasing or measurement ambiguity.

AI Glossary / Time series and signal processing / Intermediate

Phase of Spectrum

Definition

The phase of a spectrum is the angle of each nonzero complex frequency component in a signal's Fourier representation. Together with magnitude, it describes how those components combine in time. Magnitude indicates component strength, while phase locates each component within its cycle relative to the chosen time origin and transform convention.

What it means

Two signals can have the same spectral magnitudes and different waveforms because their frequency components align differently. A time shift changes spectral phase even when it leaves magnitude unchanged. Phase therefore matters when timing, alignment, interference, or reconstruction is important.
A phase spectrum is not a power spectrum. Power spectral density ordinarily describes power versus frequency and does not retain the full phase information needed to reconstruct a particular waveform. Cross-spectral phase can describe relative timing between signals, but its interpretation requires additional care.

How it works

For a Fourier coefficient X(f), write X(f) = magnitude times exp(i times phase). The phase can be computed from its real and imaginary parts using a quadrant-aware angle function. It is defined modulo a full turn: adding 2 pi radians gives the same complex number.
Displayed phase often wraps into a principal interval, creating jumps near its boundaries. Unwrapping chooses a continuous-looking sequence by adding multiples of 2 pi, but it cannot recover missing information automatically. At zero magnitude, phase is undefined; near zero, noise can make it unreliable.
Under the convention using exp(-i 2 pi f t) in the forward transform, delaying a signal by tau multiplies its spectrum by exp(-i 2 pi f tau). The phase shift is therefore minus 2 pi f tau modulo a full turn. Sampling, finite windows, and transform conventions must be included when applying this relationship.

Three practical examples

An audio engineer inspects relative phase when combining microphones because alignment can strengthen or cancel frequency components.
A machine-monitoring team compares synchronized vibration channels to investigate timing relationships, while checking signal strength and consistency.
A filter designer examines phase response and group delay so a filter's effect on timing is understood alongside its attenuation.

Think of it this way

Think of several musicians playing steady notes. Their volumes resemble spectral magnitudes, while where each musician is within the repeating cycle resembles phase. The combined sound depends on both. Moving everyone later by the same time changes timing relationships with a reference even if their individual volumes remain unchanged.

Worked example

Take an original eight-sample periodic sequence x[n] = cos(2 pi n/8), for n from zero through seven. Use the unnormalized forward DFT X[k] = sum of x[n] exp(-i 2 pi k n/8). The nonzero coefficients are X[1] = 4 and X[7] = 4, so the positive-frequency component has phase zero.
Now circularly delay the sequence by two samples: y[n] = x[(n-2) modulo 8]. The DFT shift factor at bin one is exp(-i 2 pi times 1 times 2/8) = exp(-i pi/2) = -i. Thus Y[1] = -4i, with magnitude four and phase -pi/2, or -90 degrees.
At a sampling rate of eight samples per second, bin one is one hertz, and two samples represent 0.25 seconds. The delay formula gives -2 pi times 1 times 0.25 = -pi/2, agreeing with the direct calculation.
Direct summation independently verifies every DFT bin and the shift relation. This is a circular shift of a finite periodic sequence; delaying a finite nonperiodic record with zero padding requires different boundary handling. The zero-magnitude bins have no meaningful phase in this ideal example.

Illustrative case study

Fictional case: Blue Lantern Audio combines two recordings of a test tone. Their magnitude spectra match, but the combined waveform is weaker than expected. An engineer checks synchronization and relative phase before changing the gain settings.
After examining the time alignment, the team documents how the channels were combined and repeats the test across frequencies. This invented case illustrates why magnitude alone is incomplete. It does not claim that one phase adjustment solves every recording problem or that a single-frequency observation identifies a unique physical delay.

Common misconceptions and mistakes

Do not interpret a wrapped phase jump as a sudden physical timing change without checking the representation. Do not assign significance to phase where the corresponding component is absent or dominated by noise.
Do not infer a unique delay from one phase value without accounting for full-cycle ambiguity. At one frequency, several delays produce the same wrapped phase. A delay estimate across frequencies also needs a suitable signal relationship, consistent conventions, and reliable measurements.
Do not confuse phase delay with group delay, which concerns the negative slope of unwrapped phase with respect to angular frequency.

Business use

Phase analysis supports synchronization, sensor fusion, communications, audio processing, and system diagnosis. It can reveal timing behavior hidden by magnitude-only dashboards.
Record sampling rates, time references, windowing, frequency units, and the sign convention. Check synchronization and evaluate confidence in relative-phase estimates using suitable methods. When applying filters, review both magnitude and timing effects, particularly where decisions depend on event alignment. Preserve complex spectra when later reconstruction or relative-phase analysis may be required.

Check your understanding

What changes in the worked example after the delay?

Answer: The phase changes while the magnitudes stay the same.

What is the phase at bin one?

Answer: Minus 90 degrees.

Why are phases at ideal zero bins undefined?

Answer: A zero complex value has no unique angle.

What boundary convention is used?

Answer: Circular indexing over an eight-sample periodic sequence.

Related terms

Fourier transform
complex spectrum
magnitude spectrum
phase response
phase unwrapping
group delay
cross spectrum
coherence
time shift

Learn next

compare an impulse, its delayed version, and their complex spectra. Plot wrapped and unwrapped phase, then examine what happens when noise is added near frequencies with very small magnitudes.

Sources

Sources: SciPy signal-processing tutorial: https://docs.scipy.org/doc/scipy/tutorial/signal.html . SciPy Fourier-transform tutorial: https://docs.scipy.org/doc/scipy/tutorial/fft.html . SciPy group_delay documentation: https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.group_delay.html . Relevant primary documentation consulted. DFT example and fictional case are original.

Previous: Phase (Fourier) Next: Phillips - Perron test