I think what you are seeing there is essentially the truth. As you move beyond 90°, diffraction effects become more dominant. The transition from smooth to erratic happens a bit more abruptly there than I might have expected, although the speaker you're measuring is very square/boxy. So yeh, I suspect this is not a measurement error but simply the reality of the sound field the speakers produce.
To illustrate, here are the horizontal polars of the Genelec 8351A (measured by Princeton's
3D3A lab). The 8351 has better edge rounding/smoothness than the speaker you're measuring, which pushes the erratic behaviour out a bit beyond 90°, but the trend should still be fairly obvious:
View attachment 93094
View attachment 93095
This is part of it, but it's also a specific quirk or REW measurements. See below! A good FYI for anyone else looking to measure their own speakers.
Same time window, same smoothing, yet very different looking results.
View attachment 93091
View attachment 93092
Ah yes! I used to have the same problem past 90 degrees
@TimVG . It was especially prevalent with box-shaped speakers.
What's happening is that by default, REW sets the reference point/ t=0 on the impulse response based on the
loudest sound -- when the impulse response peaks.
However, once you go past 90 degrees, it's often the case that the
reflection off the nearest wall is louder than the first sound hitting the microphone. So REW gets confused. It thinks the IR peak is at the reflection, rather than when the sine sweep begins.
To illustrate this more clearly, here is the difference between the H100 and H110 measurements:
This can't be explained by simple diffraction. Note how the treble is ~5dB
higher at H110 than at H100.
Taking a look at the impulse response, we see the problem. Here it is for H100:
There are some reflections that would ideally be windowed out, but this is scaled to show the worst of it. It's still roughly usable as is, though there's a notable reflection at 9ms.
But now look at H110:
Yikes! This is not usable.
But now look what happens when we "go back in time" to when the sweep actually starts
It looks just like the earlier measurement.
So the actual result, once adjusted for timing, should look like this:
Tada! The expected result. So we'd just need to go back in time for all those measurements, sighs
Luckily, there's an easy solution for future measurements. In the REW measurement window, set the timing reference to begin at when the impulse response starts, not when it peaks.
If you have a dual channel mic and can do a loopback measurement for accurate timing even better, but the Umik-1 can't do that. Setting t=0 at IR start has sufficed for all the spins I've posted here. As a general rule, it's good to check the impulse response when something seems amiss in a few of the measurements.
EDIT:
@hardisj beat me to it! Glad we both spotted it though!