--- title: "Coherence" slug: "coherence" updated: 2026-05-26T14:52:06Z published: 2026-05-26T14:52:06Z canonical: "doc360.rationalacoustics.com/coherence" --- > ## Documentation Index > Fetch the complete documentation index at: https://doc360.rationalacoustics.com/llms.txt > Use this file to discover all available pages before exploring further. # Coherence ## What is Coherence? Coherence is a statistical estimation of the causality or linearity between the reference and measurement signals in a Transfer Function measurement. Coherence can not only detect contamination of the measurement Signal by unrelated signals (such as background noise or reverberation) but it is also sensitive to timing mismatches. We use it in Smaart to gauge the quality of transfer function measurement data, frequency by frequency, in real time. Since the same factors that affect coherence (noise, reverberation, etc.) also affect speech intelligibility, the coherence trace can also give you a sense of how intelligible a system is. This is, however, separate from [Smaart’s STI calculation functionality](/v1/docs/calculating-sti). In simple terms, the coherence display tells you how much reference signal you sent through the system is returning to the measurement signal, and in certain cases, whether the reference signal has been timed to energy in the measurement signal. ![](https://cdn.document360.io/a725c8f7-3a1a-4c09-9518-5d15d09a2ae2/Images/Documentation/image(273).png?sv=2026-02-06&spr=https&st=2026-11-18T17%3A23%3A56Z&se=2026-11-18T18%3A23%3A56Z&sr=b&sp=r&sig=1bRZ0IFNmaetZlphTmM%2ByAtEQl9x%2FiJeNJZanZaz8uE%3D) *Coherence shown on a scale of 0-100 in the upper half of the Magnitude plot.* The coherence calculation essentially asks the question “How confident can we be that what we are seeing in the measurement signal at this frequency was caused by the reference signal?” The answer is a value between zero and one, which Smaart displays as a percentage. A value of 100% indicates perfect correlation between the two signals and 0% means there is no discernible relationship between them. And in practical terms, coherence works by comparing the cross-Spectrum (the frequency-Domain representation of a cross-correlation) of the reference and measurement signals to the product of their averaged power spectra. That means it must be calculated across multiple readings of the two signals in order to be meaningful. As such, the feature turns itself off when averaging is not in use. ## The Coherence Display The coherence trace in Smaart is plotted in the top half of the Magnitude plot (or the top quarter if ¼ coherence is selected in [*Transfer Function Graph Settings*](/v1/docs/transfer-function-graph-settings#coherence)). Its x axis, like the magnitude trace below it, indicates frequency and its y axis indicates its coherence value as a percentage value between zero and one hundred. Coherence is **always** calculated for all transfer function measurements that use averaging, but only the coherence for the top magnitude trace in the Z order is displayed. This can be a live measurement or a stored trace, so long as it is the topmost (or only) trace. ![](https://cdn.document360.io/a725c8f7-3a1a-4c09-9518-5d15d09a2ae2/Images/Documentation/image(264).png?sv=2026-02-06&spr=https&st=2026-11-18T17%3A23%3A56Z&se=2026-11-18T18%3A23%3A56Z&sr=b&sp=r&sig=%2FUO4LJZ00qoGyefJOc8Md%2FlT63YCWtwI1T%2BnyP%2FpkOA%3D)The arrowhead-shaped widget on the right edge of the Magnitude graph sets the threshold for the [coherence blanking](/v1/docs/data-protections) function. Coherence blanking removes data from magnitude and Phase traces at any frequency where coherence does not meet or exceed the specified threshold. To set or change the threshold, you can left-click and drag the widget to move it up and down. Coherence blanking works for **all** displayed traces on the Magnitude and Phase displays that use averaging (not just the front one) and it works even if the coherence trace is not displayed (hot-key [C]). ![](https://cdn.document360.io/a725c8f7-3a1a-4c09-9518-5d15d09a2ae2/Images/Documentation/image(266).png?sv=2026-02-06&spr=https&st=2026-11-18T17%3A23%3A56Z&se=2026-11-18T18%3A23%3A56Z&sr=b&sp=r&sig=bZHt4N8BQOtXSCdiEp5TyWpXQFWnsLowW8MBS9OMMfs%3D) *The Coherence section of the Transfer Function Graph Settings dialog.* Display options for the coherence trace are found in the Coherence section of *Transfer Function Graph Settings:* - ***Show Coherence*** – displays the coherence function by frequency for the trace at the top of the z-axis on the transfer function Magnitude display. The coherence trace is normally shown in the upper half of the Magnitude display between 0 dB and the top of the graph. Please note that since coherence is calculated by comparing averaged and un-averaged transfer function data, the coherence trace is not displayed when averaging is set to None. - ***1/4 Height Coherence*** – compresses the coherence display into the topmost 1/4 of the transfer function Magnitude graph. When this box is un-checked, the upper half of the magnitude graph is used. Select this option if you want more unobstructed graph area for magnitude traces when displaying coherence. - ***Squared Coh*** – displays coherence data as magnitude squared coherence when selected, which makes small ripples in the coherence trace more pronounced. - ***Coh Blanking Threshold %*** – sets the minimum allowable coherence value for transfer function Magnitude and Phase displays. Frequency data points with coherence values that fall below the value specified here will not be displayed. This setting can also be changed by moving the arrowhead-shaped widget that appears on the upper right edge of transfer function Magnitude graphs in the main window. - ***Line Thickness Follows Foreground*** – when enabled, causes the coherence trace's line thickness to match the Foreground Line Thickness setting. When disabled, the option below will become available and allow you to choose a separate line thickness value for the coherence trace. ## **Common Causes of No Coherence** As previously mentioned, zero coherence (or at the very least extremely low coherence) is caused when there is no correlation between your reference and measurement signals. When there is no coherence, the [delay finder/tracker](/v1/docs/activate-delay-finder) will flash between several different values and fail to lock onto any value in particular. **This can be caused by:** - **Using random (or pseudorandom) noise from two different sources** (i.e. your reference channel is Smaart-generated Pink Noise, but the System Under Test is receiving pink noise from a different external source (mixing console oscillator, pink noise injector, etc). While they may sound alike, these are two independently generated signals and cannot be correlated. **To Resolve:** Ensure that the same noise source is being used for both the reference signal and output to the SUT. - **Using two different channels of “Stereo” or “Multi”-output Noise.** Selecting “Stereo” from within [*Signal Generator Settings*](/v1/docs/signal-generator-measurement-settings) will send two different seeds of noise to odd- and even-numbered outputs. When using the Left output for a Loopback/reference channel and sending the Right output out to the System Under Test, these signals cannot be correlated as they are being generated separately. Similarly, the “Multi” setting generates a unique noise seed for *every* output. **To Resolve:** Select “Mono” in *Signal Generator Settings* or ensure that the same output is being used for both the reference signal and output to the SUT, or make sure that the reference input is receiving the same noise output as the SUT. ## Causes of Poor Coherence There are three common causes for a coherence loss: - A problem with the measurement system - Environmental noise causing measurement signal contamination - Reverberation ### Problems with the Measurement System The most common measurement system issue affecting coherence is a timing mismatch between reference and measurement signals. Loss of coherence from a timing mismatch will show up in the higher frequencies first, although the mismatch needs to be a significant fraction of the measurement Time Window to become obvious. As such, small timing issues may not be visibly apparent when using large FFT sizes. The MTW transfer function uses extremely small time windows in the upper octaves and is therefore much more sensitive to small timing mismatches than large single-FFT-size measurements. If you don’t see coherence falling off more on the high end than at lower frequencies in an MTW measurement, then timing probably isn’t the problem. Other potential factors that could impact coherence on the measurement system end include excessive electronic noise, distortion, nonlinear processes (such as compression and limiting), Crosstalk, or other mixing of signals in the measurement system signal path, but these are less common. ![](https://cdn.document360.io/a725c8f7-3a1a-4c09-9518-5d15d09a2ae2/Images/Documentation/image(274).png?sv=2026-02-06&spr=https&st=2026-11-18T17%3A23%3A56Z&se=2026-11-18T18%3A23%3A56Z&sr=b&sp=r&sig=moApkYaG%2BAvu6FCc%2FvYMszYFpQG%2Bt0IDAuEMsARhTfo%3D) *HF coherence loss in an MTW Transfer Function due to delay mismatch.* ### Environmental noise Since coherence is essentially an estimation of linearity/causality, any components of the measurement signal that are uncorrelated with the reference signal will have a negative impact on coherence. These can include background noise, HVAC noise, construction noise, and people talking or shouting. Noise is easy to diagnose: If a louder test signal improves coherence, then noise is the most likely culprit. As such, solutions include either measuring louder or possibly reducing any background noise, if possible. The latter can be accomplished by shutting off HVAC systems or asking any people in the vicinity to take a break while you finish your measurement. If the noise is constant, [higher averaging](/v1/docs/averaging-over-time-temporal-averaging) won’t likely improve the overall coherence level but it *can* have a stabilizing effect. ### Reverberation If possible measurement system issues have been ruled out and coherence does not improve when measuring at higher levels, the most likely problem is reverberation. Reverberation is a non-linear phenomenon, but increases proportionally when the excitation level is increased – measuring louder will not improve the direct-to-reverberant-sound ratio. There typically isn’t much you can do about it unless you are in a position to damp reflective surfaces or re-position a loudspeaker you are measuring so as to excite the reverberant field to a lesser degree. **Application Note** When evaluating a system response measurement, one of the key skills is to be able to quickly identify causes of bad coherence, and ignore frequency response “issues” that are simply symptoms of bad coherence. A common cause of bad coherence is the reflection from the ground that appears when a measurement microphone is on a stand. Notice in the bottom example, there is a coherence drop around 250Hz. There is also a break in phase, and a clear secondary arrival 4.46ms after the direct arrival in the Live IR. An experience user will immediately recognize that the wavelength of 250Hz is roughly 4.5ms and that the measurement microphone is on a stand about 4 feet/1.5 m up. Therefore, there is content at 250Hz, there it is just getting cancelled out by the ground reflection. The actual response would be closer to what the magenta line is showing - that that 250Hz drop in magnitude is not “real” in the system. Averaging multiple measurements of multiple locations will help to reduce this effect, but being able to see this quickly is an important skill. Remember, even though we are looking at 4 graphs below, they are all created from the same fundamental data. ![](https://cdn.document360.io/a725c8f7-3a1a-4c09-9518-5d15d09a2ae2/Images/Documentation/Coherence App Note .png?sv=2026-02-06&spr=https&st=2026-11-18T17%3A23%3A56Z&se=2026-11-18T18%3A23%3A56Z&sr=b&sp=r&sig=kZgG18LWIbduS7K4XeVkDiZJUsi9Tmf5KS0Aibm%2F6f0%3D) The frequency (magnitude and phase) response of a system, function or network. The transfer function of the linear time invariant (LTI) system can be measured directly, using techniques such as dual-FFT transfer function measurements that compare the output of the system to its input signal in the frequency domain, or by taking the Fourier transform of the system’s impulse response. Strictly speaking a signal can be any set of values that depends on some other set of values. In signal processing, the independent variable, e.g., time or frequency, is said to be the domain of the signal. In audio and acoustics, the things we most commonly think of as signals are time domain signals, where voltages or numeric values representing amplitude (the dependent variable) vary over time (the independent variable). But by strict definition, most of the things we see in Smaart could also be called signals, including transfer function Magnitude and Phase and RTA displays where relative energy or phase shift are presented as a function of frequency, rather than time. The frequency content of a given signal. In signal processing, the term “domain” refers to the independent variable of a signal. By convention, when graphing a signal the independent variable is typically placed on the horizontal (x) axis of the plot with the dependent variable on the vertical (y) axis. For example in Smaart, an impulse response display, with time (the independent variable) on the x axis and amplitude (the dependent variable) on the y-axis, is referred to as a time domain display. Similarly, Spectrum and Transfer Function displays where magnitude or phase shift are plotted as a function of frequency are called frequencydomain displays. 1. A number assigned to a quantity so that it may be compared with other quantities. 2. The absolute value of amplitude. As a convention, we most often use the terms amplitude to refer to linearly scaled quantities and magnitude when discussing amplitudes cast in logarithmic units such as decibels or orders of magnitude. Phase, or Phase shift is a timing difference in a signal (relative to some reference) at one or more frequencies, typically expressed in degrees, where 360° = one full cycle at a given frequency. A random (or pseudorandom) signal in which, over a given averaging period, each Octaveband (or other logarithmically spaced interval) contains an equal amount of energy. The amount of time required for and/or represented by a measurement or other process. Often used interchangeably with Time Constant. Undesired energy in one signal (or channel) introduced from an adjacent signal or channel.