Question 1

How do I increase coherence in my transfer function measurement?

Accepted Answer

The most effective strategies in order: increase the number of averages (from 4 to 16 or 32 — this is the easiest improvement). Increase the source level relative to the ambient noise. Use a larger FFT size at low frequencies for better frequency resolution. Move the microphone away from reflective surfaces that create interference. Reduce background noise (turn off HVAC, close doors, measure during quiet periods). Use a longer measurement time (more data improves statistical reliability). Ensure the reference signal is clean (direct console feed, not a microphone near the source). Check that the system is time-invariant during measurement (no changing conditions, no fading signals). SonaVyx shows real-time coherence so you can see the effect of each improvement as you make changes.

Question 2

Does increasing the source volume improve coherence?

Accepted Answer

Yes, if the coherence loss is due to ambient noise. Increasing the source level raises the signal-to-noise ratio, which directly improves coherence. The relationship is approximately: doubling the SNR (adding 3 dB) improves coherence from 0.5 to about 0.67, from 0.7 to about 0.82, and from 0.9 to about 0.95. However, if coherence is low due to reflections or time-varying conditions rather than noise, increasing the volume will not help significantly. A diagnostic test: if coherence improves uniformly across all frequencies when you increase volume, noise was the problem. If only some bands improve, those specific bands were noise-limited while other bands have different issues. Be careful not to overdrive the speakers into distortion, which itself reduces coherence.

Question 3

Why does moving the microphone sometimes improve coherence?

Accepted Answer

Microphone position affects coherence because of the room interference pattern. At certain positions, the direct sound from the speaker and a strong reflection from a nearby surface partially cancel at specific frequencies, creating a null. At the null, the signal level is very low while the noise remains constant, reducing SNR and coherence at that frequency. Moving the microphone even 30-50 cm can shift away from the null, improving coherence at those frequencies (but potentially reducing it at others). Additionally, moving away from walls reduces the strength of nearby reflections, generally improving broadband coherence. The optimal position is typically 1 m or more from any surface, on the direct axis of the speaker, at a distance where the direct sound is well above the ambient noise.

Question 4

How does FFT size affect coherence?

Accepted Answer

Larger FFT sizes improve coherence at low frequencies by providing finer frequency resolution. Room modes create very narrow interference features at low frequencies. With a coarse FFT (2048 points, 23.4 Hz bins), a mode at 87 Hz is averaged with energy at 64-110 Hz, including nulls and peaks, which reduces coherence. With a fine FFT (16384, 2.9 Hz), the same mode is measured more precisely, and the bins around it have better SNR, improving coherence. The trade-off: larger FFT means slower updates and longer measurement time to accumulate the same number of averages. For system tuning, start with 4096 or 8192 and increase to 16384 if low-frequency coherence is insufficient. SonaVyx lets you change FFT size during measurement and see the coherence response in real time.

Question 5

Can I improve coherence by using a different test signal?

Accepted Answer

The test signal matters when measuring transfer function in real time (not using a sweep). Pink noise gives the best broadband coherence because it excites all frequencies simultaneously with consistent spectral density. White noise over-represents high frequencies. Music or speech may not excite all frequencies, leaving gaps where coherence is low. For loopback measurement with SonaVyx, the sweep signal gives excellent results because deconvolution concentrates the energy into a clean impulse response. For real-time transfer function using the H1 estimator, pink noise is preferred. If you must use program material, choose material with broad spectral content: full-band music works better than solo instruments or speech. Monitor the coherence to verify sufficient excitation across the frequency range.

Question 6

Why is my coherence worse when I measure the full system versus a single speaker?

Accepted Answer

When measuring multiple speakers simultaneously, interference between them creates position-dependent comb filtering. At the microphone position, the signals from different speakers arrive at slightly different times, creating constructive interference at some frequencies and destructive at others. At the null frequencies, the combined signal is weak while noise remains constant, reducing coherence. This is normal and expected — it does not mean the system is malfunctioning. To get clean coherence, measure one speaker at a time (mute the others). When measuring the combined system, accept that coherence will be lower in the interference regions. Use more averages to compensate. SonaVyx system alignment workflow recommends measuring individual speakers first, then the combined response.

Question 7

Does using exponential averaging instead of linear averaging affect coherence?

Accepted Answer

The averaging type affects the coherence estimate differently. Linear (Welch) averaging gives the most statistically accurate coherence because each FFT frame contributes equally, and the estimate converges to the true value as the number of averages increases. Exponential averaging weights recent frames more heavily, which is useful for tracking slow changes but can bias the coherence if conditions changed during the measurement. Peak hold does not apply to coherence (it is used for magnitude only). For coherence assessment, use linear averaging with at least 16 frames. For live monitoring where you want to see coherence update as conditions change, exponential averaging with a time constant of 2-4 seconds provides a responsive but reasonably stable display. SonaVyx defaults to linear averaging for transfer function mode.

Question 8

Is it possible to achieve coherence above 0.95 across the entire audio range?

Accepted Answer

Achieving coherence above 0.95 from 20 Hz to 20 kHz is very difficult in a real room. Practical expectations: above 200 Hz, coherence of 0.95+ is achievable with a good signal-to-noise ratio (30+ dB), 16 or more averages, and the microphone in the direct field of the speaker. Between 50-200 Hz, 0.85-0.95 is more realistic due to room modes and background noise. Below 50 Hz, even 0.80 is good. In an anechoic chamber or free-field outdoor measurement, 0.99+ is achievable across the full range. The key factors for high broadband coherence are: high SNR, stable conditions, no nearby reflections, and sufficient averaging. For system tuning purposes, coherence of 0.85 across the 100 Hz to 10 kHz range is excellent and provides reliable data for EQ decisions.

Improving Measurement Coherence: Practical Tips for Better Transfer Function Data

Frequently Asked Questions