One of the problems with trying to measure the response of any subwoofer is that if you conduct the test in a room, the effects of the room tend to completely dominate and mask the very effects you are trying to measure. Reflections from the walls, floor and ceiling, room gain and modal resonances completely dominate the results. Even if you perform all your measurements in the same room, it doesn't really help because of the physical differences between subs, each one energises the room slightly differently. The configuration of the sub in terms of whether it is forward-firing/downward firing, whether it is ported, the orientation of the port(s) all influence how the sub couples with the air in the room and therefore how it excites the room's modes.

Testing in an anechoic chamber would remove these problems, but the size and cost of a chamber that would be truly anechoic right down to below 20 Hz would be prohibitive! It would have to be huge and we certainly don't have the money to do it! The alternative is to perform the measurements out of doors, on a large, hard, flat surface - so-called ground-plane measurement. This is what we decided to do for our tests, albeit taking a risk with the notorious British weather. We placed the microphone on the ground, exactly 2m from the sub which was placed well away from any buildings that might cause reflections. Using a large flat paved area like a school playground proved to be a pretty important choice.

Of course, being a hard surface, there are reflections from the ground itself but because the sub and microphone are placed directly on the ground the difference in path lengths is minimised so the phase effect is small. In fact the microphone 'hears' two versions of the sub, the real one and its virtual image reflected in the ground which is why a ground plane measurement made at 2m is 6dB higher than it would be if you measured at 2m with the sub somehow floating in free space.

In order to keep measurement files down to reasonable sizes and to improve calculation speeds, we set WinMLS to use a sampling rate of 8kHz which is the lowest the Lynx 2B can manage. This is still way above the Nyquist frequency for the frequency range we are testing here. The sweep tone was a logarithmic sine wave sweep of 2.5 seconds duration between 10Hz and the maximum for the sampling rate (4kHz). To this we applied a second order Butterworth pre-emphasis filter to limit the test signal supplied to each sub to below 315Hz.

We set the subwoofer gain by first setting WinMLS to deliver a drive signal at -40dbFS (i.e., 40dB below full scale) and then set the subwoofer gain control to about mid-way. We then sent a series of sweeps to the sub, adjusting the sub's gain control so that over the band 30Hz to 80Hz we got an average SPL of 80dB or as close as we could manage. We then fine-tuned the drive signal from WinMLS to generate a sweep where the average in the range 30Hz to 80Hz was 80dB +/- 0.1dB. This was then our reference starting point.

Because we had no idea of the relative noise-floor we were working with, we conducted a series of test sweep measurements for each sub: 3 at 80dB, 3 at 85dB and 1 each at 90dB, 95dB, 100dB, 105dB and 110dB for those that lasted that long! We had the intention of averaging the 3 measurements at the 2 lowest SPLs. After examining the results we decided to discard the 80dB sweeps because they were too near the noise-floor. We also selected the 85dB measure that had the least noise contamination each time. Each of the measurements consisted itself of multiple 2.5 second sweeps averaged (to reduce the effect of random background noise). The number of sweeps was determined by the duration depending on the SPL: 40 seconds for 80 and 85dB, 20 seconds for 90 and 95dB, and 10 seconds for 100 and 105dB and 5 seconds for 110dB.

We pondered for some time how to plot the +/-3dB band for the frequency responses. In the end, w decided to use the most generous interpretation by measuring the maximum SPL on the 85dB FR plot and treating that as the +3dB point, then subtracting 6dB from that value and using that as the -3dB point. It's swings and roundabouts but it does tend to be flattering to subs that have a very flat response anyway. The alternative is to use an average measure and plot lines at +3dB and -3dB from the average. The trouble is that several of the subs don't keep their response in any zone that you could reasonably apply uniformly across the group.

Update

In response to some comments from previous rounds, we made a few changes to the testing method, while retaining the basic comparability between these results and the earlier rounds. The most significant change was that we used reverse sine sweeps. This gives the subs a slightly easier time of it because any power compression will tend to occur towards the end of the sweep (at the bottom end), allowing the true performance further up the range to be measured first. In practice, the differences between this method and the forward sweep are fairly subtle.

In response to some comments from previous rounds, we made a few changes to the testing method, while retaining the basic comparability between these results and the earlier rounds. The most significant change was that we used reverse sine sweeps. This gives the subs a slightly easier time of it because any power compression will tend to occur towards the end of the sweep (at the bottom end), allowing the true performance further up the range to be measured first. In practice, the differences between this method and the forward sweep are fairly subtle.

In order for WinMLS to extract the distortion products from the impulse response with sufficient resolution in the reverse-sweep mode, it proved necessary to increase the sweep time to 14 seconds. This resulted in rather larger measurement files and caused the plotting of the charts to be much slower. Although, the longer sweep does have the advantage of allowing a better signal to noise ratio to be achieved.

Once again, we performed sweeps starting at the 85dB level as before. For the 85dB measurement we allowed WinMLS to take the average of 4 x 14 secs sweeps (again improving the SN ratio), and chose the best of 3 measurements to be the baseline as before. For the 90 and 95dB levels, WinMLS averaged 2 x 14 secs sweeps and for the 100, 105, 110 and 115 levels we used a single sweep for each measurement.

I had wanted to plot sweeps at 2.5dB level increments to allow a more precise measurement of maximum output, but with the longer plotting times produced by the longer sweeps, there simply wasn't time.

There has been much discussion elsewhere over the supposed shortcomings of ground plane measurement techniques which I won't bore you with. But one of the things cited as making a difference to any measurements taken is the distance from the microphone to the ground.

To test this, we took a comparative measurement of the DD-18 at 100dB sweep level with the microphone placed directly on the ground without its windshield for protection. The difference caused by the ground proximity should be greatest at the highest frequencies. We noticed from the results that this is certainly less than 0.1dB and substantially less than the difference further down the scale, which is due to the natural difference in the background noise levels between any two measurements, and consistent with the fact that one of the measurements was carried out without the windshield in place.