For earphones and headphones, a correct understanding of the published data on GE can get you ninety percent of the way in knowing what to expect. In case of speakers, however, there are additional variables that affect the sound - some of these arise from the characteristics of the speakers themselves, and some from the listening environment. While we cannot do much about the listening environment since everyone has a different setup, we saw the need to reinforce the data where we can in the speaker characteristics department.
When using speakers, some audiophiles like to walk around and do things while listening to music, or listen to music in different places - the problem is that this makes the frequency response graph irrelevant by itself. In fact, the frequency response graph does not accurately reflect the sound heard from anywhere other than on the axis right in between the speakers (at 0 degrees offset).
This is why GE decided to include Directivity Patterns in its measurements for speakers - this article will briefly explain how to read each graph regarding this characteristic.
ELAC BS182, Red:Duct
The above graph is the frequency response of the ELAC BS182 speakers. The uppermost line (in BLUE) is the response measured at 0 degree offset between the speakers and the listener (the microphone), and the subsequent graphs are the frequency responses measured from varying offsets in 15 degree increments (the measurements were actually made in 5 degree increments, but only the 15 degree increments were drawn for clarity).
The Directivity Pattern data is made by processing these raw frequency responses measured every 5 degrees - the data is first normalized to the response at 0 degrees (such that the intensity of a certain frequency at each location is displayed relative to the intensity of that frequency at the middle), after which the relative deviations in responses are presented in different forms (waterfall, sonogram, polar) according to need.
The rationale for normalization is that the change in sound becomes much more apparent when it is compared to the original response (down the middle). If the data is presented without normalization, the frequency response of the speakers are not accounted for, which makes it extremely difficult to determine whether the response at a particular angle is due to the location or the drivers.
Thus, the Directivity Patterns shown on GE graphs are patterns regarding directivity and directivity alone, without explicitly referencing what the frequency response of the unit looks like.
The waterfall graph is a three-dimensional graph where the horizontal axis represents frequency, the vertical axis represents intensity, and the axis into the page represents angular offset.
The graph represents the normalized, relative variance in sound with the angular offset compared to the original sound. This is apparent when looking at the 0 degree line, which manifests as a completely straight line at 0dB.
Higher frequencies have intrinsically higher directivity, and this quality along with the additional fact that twitters usually have limited angles over which they diffuse sound cause most speakers to start exhibiting weaker treble (before other frequencies) as the offset increases.
Thus, products for which the treble falls down slowly are products that have the largest sweet spots for ideal listening, and vice versa. Also, a well-engineered product will show no peaks or troughs on a waterfall graph, but will instead have a smooth, continuous surface.
The waterfall 2 graph is identical to the waterfall 1 graph except for the choice of axes.
The sonogram is a two-dimensional graph where the horizontal axis represents frequency and the vertical axis represents the angular offset.
This graph is essentially the waterfall graph viewed from the top, with the decreasing intensity represented by progressively cooler colors - a deep blue color indicates a significantly lower intensity compared to in the reference at zero offset, while a deep red indicates no change.
The sonogram is excellent for quickly determining the size of the sweet spot by glimpsing at the red area - on the other hand, more contours on the equal-intensity lines or a sonogram in every color of the rainbow is a mark of a poor directivity.
The polar graph shows the relative intensity of a given frequency as a function of the angle (in polar coordinates). It essentially consists of slices from the waterfall 2 graph that have been modified to fit on a semicircle (with the parameter of the semicircle being the reference).
For instance, the orange line representing the intensity of the 20kHz signal shows how the intensity changes over 180 degrees.
A product with a large sweet spot will have patterns that generally resemble a semicircle, with lines that lie relatively close to the perimeter. Speakers with good directivity will have smooth curves without kinks in their lines.
The contour graph is a simplified version of the sonogram, where the fill colors are precluded in favor of equal intensity lines in 3dB increments (-3dB, -6dB, -9dB, -12dB…).
The contour graph could be used to trace lines and find how large the sweet spot is for a given frequency - as an example, if -3dB is the threshold for 'listenable', then the graph shows that the midrange(2kHz) is listenable up to around 40 degrees off the center.