Question 1

What does a coherence value of 0.9 versus 0.5 mean?

Accepted Answer

Coherence of 0.9 means 90% of the output signal energy at that frequency is linearly related to the input signal. The transfer function measurement at that frequency is highly reliable. Coherence of 0.5 means only 50% of the energy is related to the input — the other 50% comes from noise, reflections, or nonlinear distortion. The transfer function at that frequency has significant uncertainty and should be interpreted cautiously. Magnitude accuracy at a given coherence: at coherence 0.9, the magnitude uncertainty is about ±1 dB. At coherence 0.5, uncertainty is ±3-6 dB. At coherence below 0.3, the magnitude is essentially unreliable. SonaVyx color-codes coherence: green (above 0.85), amber (0.5-0.85), red (below 0.5), helping you immediately identify which parts of the frequency response are trustworthy.

Question 2

Why is coherence always high at some frequencies and low at others?

Accepted Answer

Coherence is frequency-dependent because different factors affect different frequency ranges. High coherence frequencies are where: the signal-to-noise ratio is good, the direct sound dominates over reflections, and the system behavior is linear. Low coherence frequencies are typically at: very low frequencies (below 100-200 Hz) where room modes create position-dependent interference patterns and background noise is strongest, frequencies where a deep cancellation occurs (two paths cancel, leaving mostly noise), frequencies not well excited by the source (speaker rolloff at low end, or gaps in program material), and high frequencies far from the speaker where air absorption and distance reduce the signal. SonaVyx shows coherence alongside the transfer function so you can correlate poor coherence with specific measurement conditions.

Question 3

Can coherence be greater than 1.0?

Accepted Answer

Theoretically, coherence (gamma-squared) is bounded between 0 and 1 by definition. In practice, measurement artifacts can cause values slightly above 1.0 due to: insufficient averaging (coherence estimates are biased high with few averages), spectral leakage from windowing, numerical precision issues, or the reference and measurement signals being processed through the same digital path (introducing artificial correlation). SonaVyx clips coherence display at 1.0 and requires a minimum number of averages for reliable coherence estimates. If you consistently see coherence above 0.99 across all frequencies, check that the reference signal is not accidentally feeding directly into the measurement channel, which would give perfect but meaningless coherence.

Question 4

What coherence threshold should I use to trust a transfer function measurement?

Accepted Answer

For system tuning decisions, use a threshold of 0.85 (some practitioners use 0.80). Above this threshold, the transfer function magnitude is accurate to within ±1-2 dB, which is precise enough for EQ decisions. Between 0.5 and 0.85, the data is usable for identifying broad trends but not for precise EQ setting. Below 0.5, do not make EQ decisions based on the measurement at those frequencies. For phase and time alignment, higher coherence (above 0.90) is needed because phase is more sensitive to noise than magnitude. For research and precision measurement, require 0.95 or above. The threshold you need depends on the decision: coarse tonal adjustment tolerates lower coherence than fine-grained parametric EQ. SonaVyx shows all coherence values but highlights reliable regions in green.

Question 5

What is the relationship between coherence and the number of FFT averages?

Accepted Answer

With few averages, the coherence estimate is biased upward — even uncorrelated signals show non-zero coherence. With one average, coherence is always 1.0 (mathematically trivial). With 4 averages, the expected coherence for uncorrelated signals is about 0.25. With 16 averages, it drops to about 0.06. More averages give more accurate coherence estimates and lower the noise floor of the coherence function. A rule of thumb: you need at least 10-16 averages for a meaningful coherence measurement. The coherence uncertainty decreases as 1/sqrt(K), where K is the number of averages. SonaVyx uses Welch averaging with 50% overlap, so 16 averages correspond to about 8.5 blocks of data at whatever FFT size you have selected.

Question 6

How does coherence differ from signal-to-noise ratio?

Accepted Answer

Coherence and SNR are related but not identical. Coherence measures the linear correlation between input and output at each frequency, regardless of the absolute levels. SNR measures the ratio of signal power to noise power. In a purely additive noise scenario (no reflections, no distortion), coherence = 1/(1 + 1/SNR), so coherence of 0.9 corresponds to SNR of 9 (about 9.5 dB), and coherence of 0.5 corresponds to SNR of 1 (0 dB). However, coherence also decreases due to nonlinear distortion and time-varying conditions, even when SNR is high. A system with 40 dB SNR but strong early reflections can show low coherence due to the interference pattern, not noise. SonaVyx provides both coherence (from transfer function mode) and SNR indication (from SPL comparison) for comprehensive signal quality assessment.

Question 7

What does it mean when coherence is high but the magnitude response looks wrong?

Accepted Answer

High coherence with an unexpected magnitude response usually means the measurement is capturing a real system characteristic that you did not expect. Possible explanations: a filter in the signal chain that you forgot about (HPF, crossover, EQ), a cable or connector problem causing frequency-dependent loss, the speaker actually has a resonance or dip at that frequency, or a nearby reflection is creating constructive/destructive interference that is consistent enough to maintain coherence. High coherence confirms the measurement is reliable — the system really does behave that way at the measurement position. Before blaming the measurement, check the signal chain from source to microphone for any processing or hardware issues. SonaVyx system health check helps verify the signal chain is clean.

Question 8

How do I display and interpret coherence alongside the transfer function?

Accepted Answer

SonaVyx displays coherence as a horizontal bar below the magnitude plot, color-coded from green (high, reliable) through amber (moderate, cautious) to red (low, unreliable). When making EQ decisions, focus your attention on frequency ranges with green coherence. Amber ranges can inform broad decisions but should not drive narrow-band corrections. Red ranges should be investigated (why is coherence low?) rather than EQed. A useful technique is to overlay the coherence on the magnitude plot by shading regions: dim the magnitude trace where coherence is low and brighten it where coherence is high. This visually directs your attention to the reliable data. SonaVyx magnitude plot dims unreliable frequency ranges based on the coherence threshold setting.

Coherence Interpretation: What the Numbers Mean for Measurement Reliability

Frequently Asked Questions