Pilot tones in evacuation systems

Emergency evacuation and voice alarm systems sometimes use high-frequency pilot tones to continuously monitor the connections between the individual elements. Failure of any speaker to produce the pilot tone can thus be detected very quickly and an appropriate error message can then be generated. These pilot tones have a frequency beyond the range of human hearing at around 20 kHz. An example of such a system is the Bosch Praesideo.
In rare cases, the pilot tones have been found to interfere with acoustic measurements.

A typical magnitude spectrum measured in a hall equipped with an emergency system using pilot tones can be seen in the image below:

Pilot tone spectrum

Note that the pilot tone frequency is slightly higher than 20 kHz. There seems to be quite some variance in the pilot tone frequencies, but most fall within the range of 19 - 21 kHz.

The pilot tone can potentially pose several problems for acoustic measurements because although it is not audible, it does fall well within the 16 kHz octave band and the 20 kHz third-octave band.
The first indication that a pilot tone is present will be that the LAeq measurements will show a much higher level than would be expected from the audible ambient sound level. Obviously, these elevated signal levels may also appear in some calculated parameters.

In most cases, for safety reasons, the pilot tone cannot be turned off for the duration of the measurements. So how does the pilot tone affect the measurements? For many types of measurements, the pilot tone will have no influence on the calculated parameters whatsoever. The reverberation time for instance can be measured without problems. The same holds for many other parameters such as the energy ratio and spaciousness parameters. Frequency weighted parameters such as LAeq and LCeq will, and should, be affected of course.

Evaluating the background noise levels for speech intelligibility measurements according to ISO 3382-3 may result in values that are too high, because one might look at the LAeq which includes the 16 kHz octave band. One could argue that the pilot tone is effectively part of the background noise, but being inaudible it clearly doesn't influence the speech intelligibility, and the calculated A-weighted background noise value will therefore be invalid in the context of speech intelligibility. Dirac therefor calculates the Lp,A,B as the noise level over the 125 Hz thru 8 kHz octave bands, thus avoiding any influence of the pilot tone.

The pilot tone may also cause the trigger option of the AutoMeasure function to be activated surreptitiously. A careful adjustment of the trigger level to a level higher than that of the pilot tone may then be required.

If the measurement microphone is very close to the source of the pilot tone, and the normal stimulus signal has a relatively low level, then it may be impossible to correctly identify and deconvolve an Echo signal.
As one might expect, Dirac has a solution for the rare cases in which this may happen. This is to record the raw responses to the stimulus and then filter the raw response before using the import function to retrieve the impulse response. Recording the raw response can easily be achieved by selecting 'None' as the stimulus in the measurement window when using an external stimulus such as the Echo. When using an internal stimulus, the 'Skip deconvolution' option on the 'Process options' tab could be checked.

Skip deconvolution

The most effective filter for removing the pilot tone will be a notch filter centered at the pilot tone's frequency with a narrow bandwidth. Depending on the strength of the signal the filter may need to be applied more than once.

Notch filter

As you can see, when using Dirac pilots tones need not be feared, but you do need to be aware of them. Whether a future update of Dirac needs to pre-filter Echo measurements automatically to remove the pilot tone before processing, depends in part on how often the problem is encountered. Therefore, let us know if pilot tones interfere(d) with your Echo measurements.