The issue here is the low end limit and resolution due to the early reflections. There are ways to get around this, but each has its proponents and detractors.
The driver may be measured outdoors, possibly the best solution, but not practical in many cases, certainly not in mine. Then there's the issue of raising the drivers and mic high enough off of the ground, since ground reflections will still be present.
A ground-plane measurement can be made if the conditions allow it, but if done indoors there will still be a low frequency limit. Again, it's not practical for me.
A close-mic measurement (approximately 1/4") can be spliced to a quasi-anechoic measurement with adjustment of the close-mic measurement to account for the driver surface area and referenced to the distance used for the quasi-anechoic portion. This has room for error in area measurement as well as mic positioning. Small changes in mic positioning when this close can easily increase or decrease the SPL measurement by 1db.
A techique that I don't see used, but one that I rely upon, is a quasi-anechoic measurement imported into CAD (CALSOD in this case) with the driver/box T/S parameters then used to predict the highpass. T/S parameters are reasonably accurate and provide acceptible results in typical CAD programs, so I use this method. For woofers I set the sensitivity to be the 4-pi below-step level plus 2-4 db to account for some floor bounce reinforcement at the low end. Midranges have the full 6db of baffle step already in them due to the relatively narrow baffles and step involved, so no baffle step alteration is made.
Consider the curves in the graph below. It shows the differences that can occur in the model. The curves are:
|: Original quasi-anechoic measurement, shown down to 500Hz|
|: Close-Mic measurement adjusted for driver area and distance to mic (best estimate)|
|: Splice of quasi-anechoic and close-mic for quasi-anechoic down to 300Hz|
|: Quasi-anechoic coupled with CAD extension using T/S parameters, estimated 4-pi sensitivity|
It appeared to me that the spliced response was not a good choice. The limit of the quasi-anechoic response to 300Hz is problematic. The low limit SPL can vary by several db one way or the other due to measurement conditions, signal levels used and windowing. Then there are the issues with the close-mic measurement mentioned above.
Note that at the low limit the curves are converging, if not to the same point, at least to the same slope. The splice ( smoothed) shows 3db difference at the low end, due to the adjustment made to match its SPL at the splice point. Above 90Hz, the variation isn't bad. But below 90Hz it is significantly different from the T/S model.
The T/S model and the unadjusted close-mic measurement converge at 60Hz and overlay below that. This makes me a bit more confident of the technique, since the model is set to be the 4-pi response as the T/S section starts below the step and the convergence is in the area of the measured response low frequency area.
The question arises as to why the close-mic response shows an increasing response above 60Hz vs. the other two methods shown. In fact, typical of close-mic measurements, this one shows the turnover to a decreasing response above some point, 300Hz in this case. This is due to delays between the short distance to the mic at the center of the driver and the much longer distances from the outer areas of the driver to the mic. At some point there will be very dramatic comb-filtering seen. Close-mic measurements are very limited at higher frequencies. This is the reason I tried a splice at 300Hz as shown, since I had quasi-anechoic response to that point. Had I spliced at the close-mic upper limit, it would have had a rather significant upper bass bump. For a midrange, this is more important than the reponse below, say, 100Hz.
One final point. I haven't checked into the reason for the slight overlay mismatch of the spliced response in the upper frequencies. I think that this is due to the fact that the sample points used varies with different software and may be an interpolation artifact from importation and differing interpolation of the same file.