Measurement / Averaging questions

[TekniskeBureau]

Hi there,

I'm observing a rather strange behavior, that I was hoping someone could enlighten me on. My best guess is that it has something to do with how REW averages.

My background is that I'm a Mechanical Engineer with specialty in vibrations. I'm trying to get a better understanding of acoustics, and I'm reading up on the acoustic literature. I'm also an active musician and have a rehearsal space / mini-studio, where I'm working on improving the acoustics. I'd like to do measurements at each step of the process (installing bass traps, diffusers, new floor etc.). I know that some of these things might be a bit out of scope, since most of you are probably using REW for measuring speakers in a room, whereas I'm trying to get measures for just the room itself, not a related to a specific set of speakers (the main focus is band rehearsal and recording).

I'm following the guidelines in the (brand new) EU standard for acoustic spaces for rehearsal rooms ISO23591:2021 and the associated standards for measuring acoustic room parameters (ISO3382-1/2). It seems entirely possible for me to do measurements of the the room. I've mapped out two speaker positions in combination with 6 measurement points, giving a total of 12 measurements. Typically, you average these to get an overall estimate frequency response, decay time, clarity etc. for the whole room. I have a calibrated measurement microphone, a very clean interface / preamps, an SPL-meter and a decent flat-response studio speaker.

The thing is that the mic is omnidirectional, but the speaker is not. However, following the findings of a research article by Nikolaos M. Papadakis, it seems reasonable to substitute a dodecahedron omnidirectional speaker with a normal studio grade speaker, and manually average several responses. I've mounted a single speaker on a chair that can turn and angled in a 45 degree position, in order to avoid too much direct response (the room is not that big). See the picture below:

So my idea was this: Instead of doing 8 separate measurements in different rotations, I could do a single measurement but with 8 repeated sweeps, changing the rotation between each sweep.

Below, you can see 8 separate measurements done with a single sweep per measurement, as well as an averaged response:

The extra 0-deg ref body, is to check if me lying below the speaker (as I will have to in the next test) makes a difference for the results. It seems to have very little effect on the results.

Below is the the results of both the above tests, the average of those curves and a single measurement doing 8 sweeps and making the rotation of the chair in-between each sweep. Purple is the average of the previous test and Red is the 1 Measurement / 8-Sweep 'turning method'.

Here's just a picture of the Average vs. the Turning Method, which is a bit more simple to compare.

I've done the measurements several time to confirm the trend. I was expecting that these seemingly different methods should give close to exact results, but they don't. I've also compared measurements for a single sweep vs. several sweeps in the same rotation, and the results are here also almost identical.

How does the averaging in REW work when several sweeps are used vs. several measurements? I suppose both are done in the frequency domain?

Sorry for the length post, hope that someone can help me out

Thanks! All the best, Emil

 

[sm52]

The chair on wheels can be easily moved from the operation of the large driver in the low-frequency region. And in any direction and on any axis. Also, the seat of the chair can move from vibrations. I would not trust such measurements.

 

[TekniskeBureau]

The chair on wheels can be easily moved from the operation of the large driver in the low-frequency region. And in any direction and on any axis. Also, the seat of the chair can move from vibrations. I would not trust such measurements.

Hey SM52. I've repeated the measurements, and (within a reasonable tolerance), the results are very consistent. The uncertainty/error arising from the chair setup is at least an order of magnitude smaller than the difference I see using the two different measurement strategies.

Edit: I've added en example below to show the robustness and precision of the setup.

These measurements are made using the following method
- 4 different mic positions were recorded.
- A single sweep was used per measurement
- For each mic position, 3 different speaker rotations were used (0, 120, 240 degree).
- SPL was calibrated with an external SPL-meter before each measurement.
- An average response was created for each mic position (averaging the 3 speaker positions).
- A total average was made for all 4 mic positions (averaging the room response).

The whole process was repeated, yielding the two average room responses above.

Hope it is clear that the problem is not the repeatability of the measurements, calibration or setup. It also rules out the possibility of it being a 'physical' phenomena. It must have to do with how REW processes repeated sweeps.

All the best, Emil

 

[sm52]

At the core of your question, I can't clarify anything. Wait a bit, maybe someone else will read your message and express their opinion.

 

[John Mulcahy]

That's a rather creative use of multiple sweeps , if not really advisable. Multiple sweeps are synchronously averaged at time of capture so if you do anything that alters the arrival time at the mic you will get phase cancellations at high frequency. Here is the relevant section of the help:

REW allows multiple sweeps to be averaged, although best results are generally obtained by using single, longer sweeps rather than multiple, shorter sweeps. Multiple sweeps are offered when not using a Timing reference or when using a loopback as a timing reference. Do not use multiple sweeps if the input and output are on different devices (for example, if the input is a USB mic). If Repetitions is more than 1 REW uses synchronous pre-averaging, capturing the selected number of sweeps per measurement and averaging the results to reduce the effects of noise and interference. The pre-averaging can improve S/N by almost 3 dB for each doubling of the number of sweeps. Averaging can be useful if the measurements are contaminated by interference tones, whether electrical or acoustic, as they typically will not add coherently in the averaging and hence will be suppressed by the process.

Warning: some interfaces do not maintain sample synchronisation between the successive sweeps which produces a corrupt measurement that has multiple, closely-spaced peaks of approximately the same level in its impulse response, 1 peak for each sweep. This can also happen if the input and output are on separate devices. If the frequency response with multiple sweeps is significantly different from the response with a single sweep, stick with single sweeps


It would be easier to use the option to make multiple measurements using the control just below the sample rate selector:

Normally a single measurement is made, but a series of measurements can be made without further intervention by setting the total number required in the Measurements count. That may be useful when making a series of polar measurements, using the Delay control to specify how long REW will wait before starting each sweep to allow time to reposition a speaker turntable, for example. The Delay control can also be used for single measurements to provide time to move away before the measurement starts. When making multiple measurements a panel is shown with the number of measurements remaining, the name that will be assigned to the next measurement and a pause button that will stop the sequence when the current measurement completes and continue it when unpaused.

 

[TekniskeBureau]

Dear John,

Thank you so much for your detailed response. That makes perfect sense, and yes I probably was a little bit too creative in the use of multiple sweeps. It makes total sense with the phase lag in higher frequencies, and correlate well with what I see.

The repeated measurement option seems exactly like what I want. I think that I might have an older version of REW running on the computer I've been using to do the measurements. I don't remember seeing that option before. I'm running the newest version now and I see the option for multiple measurements now. That provides a much easier (and less error prone) option than doing separate measurements and manual averaging. Exactly what I wanted

I'll have to check if it's necessary for me to do SPL calibration between each measurement (each angle). I don't suppose that is possible with the repeated measurement option? (y guess is that the error will be negligible..

Thanks again, really appreciate it!

Emil

 

[John Mulcahy]

SPL calibration is usually only done once, it remains valid unless the gain is changed along the input path. If your measurements have differing levels there is an Align SPL option in the controls of the All SPL graph to manage that.

 

[TekniskeBureau]

I must have misunderstood that. Kind of piecing these things together without a good main reference still I'll go for the multiple measurement strategy, just calibrating once per setup-position. Thanks again for all your advice John!