But Can You Trust It? Systematic uncertainties in bioacoustic recording devices

Experimental design in bioacoustics is determined by several factors. One of these factors that is often overlooked is the uncertainty of the recording device and the information recorded. From the clock speed to the transducer efficiency, the spatial positioning to the frequency response of the device’s ADC, there are many different sources of uncertainty to be contended with.

Methods of quantifying the uncertainties of different instruments need to be rather diverse. Some examples include the frequency response of the recording device and how to characterize it, the temperature-induced drift in recording clock frequency, and the reference voltage shift in the recorder Analog to Digital Converter (ADC).

Another main point to be addressed is the traceability of measurements. A measurement of SPL, for example, is completely useless without knowing its relation to a known standard, such as 1 μPa or 20 μPa. Furthermore, those standards must themselves be traceable to some form of measurement standard, so that disparate measurements made in different environments by different people and equipment might be reasonably compared to one another.

In the case of the frequency response of a recorder, a handy example would be that of a recorder which has as its input stage a series of filters, especially Sallen-Key filters. As these filters rely on capacitors of set values to create a filter, and on variable resistors to tune the corner frequency of the filter, any uncertainty in the values of the resistor and capacitor will translate into the uncertainty of the frequency response of the filter. Now add to this the fact that both capacitors and resistors change in their values with temperature, and the problem only gets worse when a recording device is placed outside, where most bioacoustic recordings are done.

The temperature-induced drift in clock frequency is of particular concern to those doing any kind of localization work. Over the course of several months of recording, an induced error of several minutes to several hours can occur, and depending on the speed of sound within the recording medium, this can account for a localization error drift on the order of km to even Mm. Luckily, this effect is often reversible, but care must be taken to characterize the drift as a function of temperature, and the temperature of the recorder must be logged.

Finally, the reference voltage drift in an ADC can be induced by either the temperature of the recorder changing, the source voltage in the recorder’s power source changing, or both. The effect of either of these causes is that the dynamic range of the recorder may be increased or decreased, thus changing the actual values of SPL which correlate to the output of the recorder data. This, too, is correctable, but only by re-characterization (and post-processing re-calibration) of the signal chain from transducer through ADC and into memory.

In summary, there are many sources of uncertainty and error in bioacoustic recording devices, and in the protocols used to acquire data from them. The only way to begin to address this problem is to admit that it exists, then learn as much as possible about it.