How to Use Coherence to Validate Measurements

TL;DR

Coherence is the most underused and most important diagnostic in transfer function measurement. It tells you how much of the output is linearly related to the input — in other words, how trustworthy your magnitude and phase data are at each frequency. A beautiful-looking frequency response means nothing if coherence is below 0.5. This guide explains how to read and improve coherence.

What Is Coherence?

Magnitude-squared coherence gamma-squared(f) ranges from 0 to 1. A value of 1.0 means the output at that frequency is entirely caused by the input — the measurement is perfectly reliable. A value of 0.0 means there is no linear relationship — the data at that frequency is pure noise. SonaVyx displays coherence as a color-coded bar: green (>0.8), amber (0.5-0.8), red (<0.5).

Interpreting Coherence

High coherence (>0.8): The transfer function magnitude and phase are reliable. You can trust the data for EQ decisions, alignment, and system characterization.
Medium coherence (0.5-0.8): The data is partially reliable. There is some contamination from noise or nonlinearities. Treat the data as directional (trends are meaningful, exact dB values less so).
Low coherence (<0.5): The data is unreliable. The magnitude and phase at these frequencies should not be used for EQ or alignment decisions. Diagnose the cause before proceeding.

Common Causes of Low Coherence

Cause	Frequency Pattern	Fix
Insufficient signal level	Broadband low coherence	Increase source level
Ambient noise	Low coherence where noise is loud	Reduce noise or increase signal
Strong reflections	Notches at specific frequencies	More averaging, closer mic position
Speaker distortion	Low coherence at high SPL	Reduce level, check speaker
System not time-aligned	Broadband degradation	Use delay finder
Digital processing delay	Broadband, especially HF	Compensate delay in SonaVyx

Step-by-Step: Improving Coherence

Check signal level. If coherence is low across all frequencies, your signal-to-noise ratio is insufficient. Increase the pink noise level until coherence improves above 0.8 from 100 Hz to 10 kHz.
Increase averaging. More averages improve coherence by reducing random noise. Switch from 4 averages to 16 or 32. The trade-off is slower update rate.
Move the microphone closer. In reverberant spaces, the direct-to-reverberant ratio drops with distance. Moving from 10 m to 5 m can dramatically improve coherence.
Window the measurement. If a strong reflection is causing coherence dips, use a shorter FFT window or time windowing (if available) to exclude late reflections.
Check time alignment. In multi-speaker systems, if the delay compensation is wrong, the reference and measurement are mismatched, destroying coherence. Use the delay finder to set the correct compensation.
Reduce ambient noise. If coherence is low at specific bands where ambient noise is high (e.g., HVAC rumble at 125 Hz), either reduce the noise source or increase the test signal level at those frequencies.

Common Mistakes

Ignoring coherence entirely. Many engineers look only at the magnitude trace. Without coherence, you cannot distinguish real system response from noise artifacts.
EQing at low-coherence frequencies. If coherence is 0.3 at 160 Hz, the +6 dB peak you see might be noise, not the system. Boosting or cutting at unreliable frequencies wastes time and can make things worse.
Assuming coherence = 1.0 means perfect system. High coherence means the measurement is reliable, not that the system sounds good. A system with terrible frequency response can still have perfect coherence.

Tool Bridge

Open SonaVyx Transfer Function mode to see real-time coherence alongside magnitude and phase. The color-coded coherence bar makes it easy to spot unreliable frequency ranges at a glance.