Coherence Mathematics

March 22, 2019

Coherence indicates how closely a pair of signals (x and y) are statistically related. It is an indication of how closely x coheres or “sticks to” y.

In other words, coherence is how much influence events at x and events at y have on one another. The magnitude of coherence will always be between zero (no influence) and 1 (direct influence).

Mathematical literature defines two coherence functions.

Complex coherence:

(1)   \begin{equation*} C_{xy}(f)=\frac{\hat{S}_{xy}(f)}{\sqrt{\hat{S}_{xx}(f)\hat{S}_{yy}(f)}} \end{equation*}

Ordinary coherence:

(2)   \begin{equation*} \gamma_{xy}^2(f)=\frac{|\hat{S}_{xy}(f)|^2}{\hat{S}_{xx}(f)\hat{S}_{yy}(f)} \end{equation*}

where:

Of the two definitions, ordinary coherence is more common in literature. However, ordinary coherence is the magnitude squared of complex coherence (γxy2(f)= |Cxy(f)|2). Thus, ordinary coherence is a special case of the more general complex coherence.

Applications of Coherence

In practical use, coherence has two main uses.

First, it establishes a measure of confidence that a peak observed in a transfer function is a resonant frequency of the device under test and not a spike due to measurement noise. For example, suppose that a peak at frequency f in a transfer function is observed. Then:

Ordinary Coherence

Complex Coherence

Description

γxy2(f) ≥ 0.70

   or

|Cxy| ≥ 0.85

A high degree of confidence that the peak is a resonant frequency.

γxy2(f) ≤ 0.50

   or

|Cxy| ≤ 0.707

Not likely that the peak is a resonant frequency.

0.50 < γxy2(f) < 0.70

   or

0.707 < |Cxy| < 0.85

Somewhat likely that the peak is a resonant frequency.

Coherence also serves as a measure of a system’s linearity at frequency f. An ordinary coherence value close to 1 indicates that the system is linear time-invariant at frequency f. An ordinary coherence value close to zero indicates that the system is either non-linear, statistically changing with time, or both.

Note, if coherence drops within a resonance, the surrounding coherence values should be evaluated before making a judgment.

Example Walk-Through

  1. Start with a vibration test setup where the accelerometers can measure acceleration in the x and y directions across a range of frequencies. Generate the graphs shown as images in Figures 2.1, 2.2, and 2.3.
  2. To determine the spectral relationship between the x and y accelerations, use the transfer function estimators Ĥ1 and Ĥ2.
  3. Around 435Hz, both Ĥ1 and Ĥ2 exhibit remarkable gains. Initially, it may be assumed that 435Hz is a resonant frequency for Hxy.
  4. However, complex coherence equals 0.007, indicating that 435Hz is probably not a resonant peak.
  5. Conversely, the peak at 5330Hz does appear to indicate a resonant frequency, as there is a high coherence value in that vicinity.
Ĥ1 transfer function graph

Figure 2.1. Ĥ1 transfer function graph.

Ĥ2 transfer function graph

Figure 2.2. Ĥ2 transfer function graph.

Cxy complex coherence graph

Figure 2.3. Cxy complex coherence graph.

Mathematical Details

Coherence as a Measure of Linearity

In many cases, the signal x may represent an input to a system (e.g., a device under test such as a mechanical device or the earth). The signal y may represent the output.

Suppose that an impulse to this system results in a response h[n] (the “impulse response”) and the discrete-time Fourier transform (DTFT) of this response is H(ω). If the system is linear time-invariant (LTI), then:

(3)   \begin{equation*} S_{xy}(\omega)=H(\omega)S_{xx}(\omega), S_{yy}(\omega)=|H(\omega)|^2S_{xx}(\omega) \end{equation*}

The coherence will be:

(4)   \begin{equation*} |C_{xy}(\omega)|\triangleq\left|\frac{S_{xy}(\omega)}{\sqrt{S_{xx}(\omega)S_{yy}(\omega)}}\right|\text{ by definition of coherence }C_{xy}(\omega) \end{equation*}

(5)   \begin{equation*} =\left|\frac{H(\omega)S_{xx}(\omega)}{\sqrt{S_{xx}(\omega)S_{yy}(\omega)}}\right| \end{equation*}

(6)   \begin{equation*} =\frac{|H(\omega)|S_{xx}(\omega)}{\sqrt{S_{xx}(\omega)S_{yy}(\omega)}} \end{equation*}

(7)   \begin{equation*} =\frac{|H(\omega)|S_{xx}(\omega)}{\sqrt{S_{xx}(\omega)|H(\omega)|^2S_{xx}(\omega)}} \end{equation*}

(8)   \begin{equation*} =1 \end{equation*}

For a system that is linear time-invariant, the ordinary coherence γxy2(f) will be 1 for all frequencies. Conversely, if the coherence is less than 1, then it is not LTI. And, if γxy2(f) is 1 only at certain frequencies, then the system is LTI at those frequencies.

Coherence as a measure of confidence in transfer function H1  

If a system is linear time-invariant with frequency response (Sxy(ω) = H(ω) Sxx(ω) ), then H(ω) = Sxy(ω)/Sxx(ω). This is the transfer function H1. To measure how “good” an estimator H1 is likely to be, check the ordinary coherence function γxy2(f). If γxy2(f) is 1 for all frequencies or close to it, then Hxy1(f) is likely a good Hxy1 estimator. If γxy2 is close to zero, then one should be much less confident about the quality of the  Hxy1 estimator.

Coherence as a measure of distance between H1 and H2

The ordinary-coherence function γxy2(f) can also be viewed as a measure of the “distance” between Hxy1 and Hxy2. This is because:

(9)   \begin{equation*} \gamma_{xy}^2=\frac{|S_{xy}|^2}{[S_{xx}S_{yy}]}=\frac{S_{xy}S_{xy}}{S_{xx}S_{yy}}=\frac{[\frac{S_{xy}(f)}{S_{xx}}]}{[\frac{S_{yy}}{S_{xy}}]}=\frac{H_{1}}{H_{2}} \end{equation*}

(10)   \begin{equation*} \text{As 0} \leq \gamma_{xy}^2(f) \leq 1, 0 \leq |\frac{H_{1}}{H_{2}}| \leq 1, \text{then } |H_{1}(f)| \leq |{H_{2}(f)}| \end{equation*}

By using the ordinary-coherence function, the distance between |H1(435)|=19 and |H2(435)|=607 is very great. This is reflected in the following equation:

(11)   \begin{equation*} \gamma_{xy}^2(435)=|C_{xy}(435)|^2=(0.007)^2=(0.0098)^2=\left(\frac{19}{607}\right)^2=\left(\frac{|H_{1}|}{|H_{2}|}\right)^2 \end{equation*}