Measurement Considerations For Successful MSO Use

A Word of Warning

There are many features of MSO that have yet to be discussed here. All of these features and the effort to learn them can be rendered moot by using measurements that might appear to be perfectly fine if you're only familiar with traditional low-frequency equalization techniques. In reality, failure to take certain important yet subtle precautions when measuring can cause MSO's response predictions to have substantial errors. We'll spell out these potential problems here, so that when you're working to get the best out of MSO, your efforts won't be compromised by avoidable errors.

MSO projects that make use of problematic measurements might appear to be perfectly normal. MSO won't "know" that there's a problem. It might proceed to create solutions with great-looking graphs and low calculated errors. But when you export filter information from MSO, import the filters into your DSP device and measure the final result using REW, you might find that the final result differs substantially from MSO's predictions.

On the other hand, careful adherence to the measurement guidelines can yield surprisingly accurate results. That's why understanding and following the guidelines is so important.

If you've followed the guidelines and are still experiencing data correlation problems, the FAQ lists a number of possible causes that have been identified by users over the years.

The Two-Sub Timing Reference Test

To understand how such data correlation errors can sneak in, a simple experiment can be performed by measuring two subs using REW. Before performing the test, set REW to use an acoustic timing reference for the measurements. For this test to work correctly, the acoustic timing reference must be used in both of the individual sub measurements below, with the same (non-sub) speaker used for the timing reference of both measurements.

• Measure the frequency response of a single sub at the MLP.
• Repeat the measurement using a second sub.
• Using REW's "A+B" trace arithmetic feature, compute the sum of the responses.
• Measure the two subs energized together at the MLP.
• Compare the result with the sum that's computed from the trace arithmetic.
  • These results should be within tenths of a dB of one another.
  • Do not attempt to hand-wave errors larger than this away.

If the acoustic timing reference is working properly, you'll find that the computed sum using the "A+B" trace arithmetic matches the response of the two subs energized together almost perfectly. Before making your final measurements for MSO, do this test to make sure the timing reference is working as it should.

To illustrate what can go wrong, perform the steps above without using an acoustic timing reference. Unless a strange coincidence occurs, the summation computed by REW's "A+B" trace arithmetic won't match the measurement of the two subs combined very well at all. That's because the REW "A+B" trace arithmetic function takes phase into account. When you measure each sub individually without an acoustic timing reference, the correct phase relationship between the responses of the two subs you measured is not preserved, and the summation, taking phase into account, will have substantial errors.

This simple example illustrates that any measurement or optimization technique that needs to make use of summations that take phase into account requires that the measurements you provide to it be taken with an acoustic timing reference.

How MSO Differs From Conventional Equalization

Let's look at a comparison of conventional subwoofer equalization with the technique used by MSO. This comparison will show that an algorithm's need for summation that takes phase into account leads to the requirement of an acoustic timing reference for the measurements you provide to it.

Conventional Subwoofer Equalization

To simplify the discussion, only a simplified method of conventional subwoofer equalization will be summarized here. It might consist of the following steps.

• Regardless of how many subs and listening positions are used, only one REW measurement is performed.
  • All subs are measured playing together at a single listening position.
• REW is used to compute an equalizer that flattens the resulting response.
• The filter information is exported from REW and imported into the DSP device.
• A verification measurement is performed afterward.

This technique only makes use of the magnitude response of the measurement from the first step. The equalization filter has dips where the measured response has peaks and vice versa. The end result is a flattened magnitude response without the need to take phase information into account at all.

MSO Optimization Approach

Let's assume we have 4 subs and 3 listening positions. Optimization of this setup requires 12 measurements. Each sub must be measured at each listening position. Then the following steps are performed.

• For each of the 12 measurements, the measurement is modified (both in magnitude and phase) by the response of the DSP channel to which the sub used for that measurement is connected.
• For each of the 3 listening positions, the sub measurements taken at that position and modified by the DSP channel responses as above are summed, taking phase into account.
• These three summations represent the equalized response of the system at that listening position.

In the second step above, summations taking phase into account must be performed. Just as with our simple example above of two subs measured at one position, the measurements provided must be taken with an acoustic timing reference. If they are not, all the computed responses will be incorrect and MSO's predictions won't match the actual measured result.

Some Problems in Forum Discussions

When people are having problems with the predicted MSO results not matching the final results measured with REW, they'll often post forum questions containing many graphs. There is a reasonable expectation that some person more experienced than they will be able to look at the graphs and determine what happened to cause the problem.

Unfortunately, it doesn't often work that way. As mentioned earlier, MSO can't "know" that there's a problem with the timing reference, nor can a human observer (unless the timing reference was not used at all, in which case the imported text data will show that). MSO will make the best of the data it has, sometimes producing what appears to be a very high-quality result, even in the presence of faulty data. One cannot know from looking at such graphs whether the data upon which they are based has a problem with the timing reference or not.

Almost all of the FAQ entries on this topic are based on the experiences of ordinary users who performed various experiments to determine what went wrong, sharing with others what they found.

All of these characteristics of the problem make it very difficult to troubleshoot in a forum environment. To track down what's going on, it's best to first perform the simplest experiments possible, such as the two-sub experiment described above using REW "A+B" arithmetic. Then look through the FAQ for possible solutions. The measurement details documentation on the next page may also lead to a solution.