Depending on the number of drivers in your speaker, will determine how far away you do the measurement. A single driver would be in the 1"-3", so quite close.

But, if you have 2 drivers, you would need to determine where a proper mix/integration was realistic. Certainly not more than 12". Best to move the speaker as far away as possible from any boundaries.

The mic is pointed directly at the speaker (horizontal axis) and at 0 degrees. You can play with taking measures of horizontal off axis (30 deg.) too, to see how the off-axis distribution looks.

REW remains the same for taking the measurement, but once taken, then you can work with the impulse response to set an appropriate gate time depending on the frequencies you're measuring. Simply open the IR Windows pop-up and select a gate that you want and watch the impulse response to ensure it's appropriate (with respect to reflections), then select Apply Windows and your response graph will adjust to that gate time response plot. I've used a 1msec left window and a 5 msec right window quite successfully for 500-20000Hz measures. The popup will show the frequency resolution for you, so you don't need to do the math. You're trying to exclude reflections from any surfaces nearby. It could be as simple as the mic stand, but predominantly hard surfaces such as the floor.

So, a gate time of 5 msec would give you a reflection free time of about 0.86 meters. Remember the sound has to hit the boundary and bounce back. So, distance=(time*speed)/2 = (5msec*344m/sec)/2. Remember that a short gate time limits the lowest frequency you can get useful response information for. So 5 msec limits your lowest frequency to about 200Hz (freq=1/period).

So, for reflection free measurements where your closest boundary is 1 meter away, you could use a 5 or 6 msec gate and check your response fairly accurately without smoothing for a measure from maybe 500Hz-20000Hz. Then for the bandwidth below 500Hz, you could measure from 0-500Hz and extend your gate out.

If you wanted an overall full spectrum measure, you could use a wide gate and use smoothing. A bit less accurate, but suitable to take a wide general look.

brucek