Comparison of detector technologies

Comparison of typical acoustic sampling technologies

 

The last decades show development of various bat detection techniques. From the basic forms of transforming ultrasound to the audible frequency range of humans a couple of high speed digital storage solutions exist. The different techniques all have an - sometimes huge - impact on how reliable bats are detected and species identified. Only with time-expansion systems as well as with real time recording systems all necessary sound information is available for an objective analysis as well as independent control. Furthermore only these systems are suitable for automated monitoring, if reliable species identification is necessary.

Presence of bats can easily be detected by means of acoustical methods. This works very well due to the fact that bats echolocate between roughly 2 to 20 times a second and thus advertise their presence. Most of these are in the ultrasonic frequency range and can't be heard by humans. Thus, a technical device for call detection has to be utilized. Today many different types and models of these so-called bat detectors exist. They either transform ultrasound to human hearing range and/or store the sound waves for later analysis. Technical differences in these devices have an influence on the reliability of detecting and identifying different bat species. Not all device types are therefore equally well suited for different tasks. We're going to introduce you to the common types of detectors, how they can be used and compare them regarding automated monitoring as well as species identification. The following text is intended as a rough overview. The complexity of bat detection in the field for scientific or consulting work is more complex than shown here.

Detector techniques / usage 101

Techniques for bat detection

  • Heterodyne detector: these detectors are rather simple devices that transform ultrasonic frequencies to our hearing range. This is achieved by mixing the microphone signal with a selectable frequency. This results in an addition as well as a subtraction of the microphone and the selected frequency. The frequency resulting from subtraction are audible for humans. Are the calls of a common pipistrelle for example at 45 kHz and the detector is set to 40 kHz, you'll be able to hear that call at 5 kHz. Most heterodyne systems have a window around the selected frequency of ±5 kHz thus all other input is lost. Thus heterodynes usually only allow to sample a small spectrum of bats, those whose calls are falling within the selected frequency band. This also allows a more accurate determination the calling frequency. So you can use rhythm as well as frequency for species identification.
    Left image shows the heterodyne principle while right image the frequency division method
  • Frequency division: The principle of this technique is to divide the microphone signal by factor 10 (some have other factors like 8, 14 or 20) to transform ultrasound to a lower frequency range (10 times lower for a device with factor 10). A common pipistrelle call at 45 kHz will be audible at 4.5 kHz. In contrast to the heterodyne systems which only make narrow frequency audible, this system is a broadband detector. While this at first sounds like the perfect solution, it has its drawback. For example amplitudes may be misinterpreted as well as short calls don't get very good sampled. The Anabat's basic unit is a frequency division detector.
  • Anabat SD1/2: the Anabat SDX system consists of a frequency division detector as well as a ZCAIM modul. This modul calculates zero crossings after the division stores these highly reduced information digitally on CF card. The sound is not stored.
  • Time-expansion: detectors based on that principle store the sound on a digital ring memory. This memory holds between 1 to 3 seconds of sound. To increase the time that can be stored most such detectors only have a limited sampling rate and amplitude (usually around 300 kHz and 8 bit). The stored microphone signal can be played slowed down by factor 10 (sometimes 20) and thus be recorded with regular sound hardware (22 to 48 kHz sample rate) to be analyzed on a computer for example. Since time-expansion detectors only output sound if you stop the recording, they are not useful to manually detect bats. Thus, they are always coupled to either a heterodyne, a frequency division or both types of detectors. For along time these were the high-end recording devices due to the lack of alternatives. The multiple conversions of sound between analogue and digital (mic to ring memory to audio jack to field recorder to computer...) have a negative influence on signal quality.
  • Real-time-recording: these full spectrum systems are recorders, but not real detectors. They allow to digitally store the microphone signal at a high sampling rate, similar to the time expansion systems. They do not rely on low capacity ring memory, but store directly to a high capacity medium (CF card, SD card or even hard disk). Usually 500,000 samples are stored per second with an amplitude resolution of 16 bit. If used as a common detector they have to be coupled with heterodyne or frequency division detectors, just as the time-expansion systems. One advantage is that the chain of AD/DA conversions is not necessary with these devices. Their primary usage is in automated long term monitoring (duration of at least one night). The batcorder is an example for a very elaborate high-end real-time monitoring system.
Heterodyne-/
Frequency division
Anabat SD1Time-expansionRealtimebatcorder
Technical properties
TypAnalogueDigitalDigital
Amplitude-resolution---8 bit16 bit16 bit
Samplingrate---300-350 kHz500 kHz500 kHz
Soundsignalchanged, reducedonly zerocrossingsslightly changedoriginaloriginal
Recording quality---+++++
Storageextern, analogueinternextern, analogueintern, digitalintern, digital
Robustness+, not weatherproof+, not weatherproof-, not weatherproof-/+, not weatherproof++, weatherproof
Usage
Typ. runtime< 24 hrs.< 12 hrs.6 hrs. - 1 weekca. 1.5 weeks
Calibrationlimitedyes
Detection distancehighmediummedium
Omnidirectionality--/- ++
Usagesimplemoderatemoderate - complicatedsimple
Recordings triggernone / amplitudeAmplitude + simple parameterscall recognition
Passive monitoring-+-+(+)++
Price
price (inkl. VAT.)< 400 EURca. 1.700 EUR900-1.500 EUR1.700-5.000 EUR2.850 EUR

Usage scenarios

  • The most simple question to ask is if bats are active or not in an area (undifferentiated bat presence/absence). Each of the simple detector types (heterodyne of frequency division) are suited equally well. Using a heterodyne the frequency has to be dialed up and down continuously to avoid overhearing bats with only certain frequencies in their calls.
  • Semi-quantitative sampling: acquisition of activity in the form of bat passes to judge bat activity. A basic differentiation between species or genera is attempted.
  • Quantitative sampling: The goal is to sample bat activity and compare it between locations or sample nights. Subjective detection methods are not really suitable for that task. A human operator usually can not work for a prolonged time with the same concentration. An objective method is much better suited to detect all bats in such a case. It is crucial that the microphone detects sound from each direction ± equally well. Data collected that way is best interpreted as a kind of relative density (seconds of activity within fixed sample period). An identification of individuals is not possible with bat calls and therefore a quantitative measurement is hard to extract.
  • Semi-qualitative sampling: a more sophisticated goal is to extract exact species identifications from acoustical data. The used technique has to allow reliable identification of bat species. While it is possible to do for many species with heterodyne or frequency division systems, it needs a lot of experience and is very subjective. It has drawbacks if many species are passing at the same time or if species encounters are rare and short. Some species are even for experienced users hard to identify. Taking high quality recordings allow more elaborate analysis of calls. While experience plays an important role as well, measurements can be fed into statistical routines and give more support for species id's. Comparison with call libraries as well as an additional control by other experts or later on with more knowledge and better tools is also an option for digitally stored bat calls.Semi-qualitative sampling uses genus or group based identification. Only simple to identify species are noted.
  • Qualitative sampling: If each contact or recording has to be identified on species level (if possible), usually only data from real-time recording systems is suitable. In some situations time-expansion recordings can be used as well. Only statistical tools achieve a reliable identification outcome.
  • Quantitative-qualitative sampling: Survey often have the goal to sample and identify species activity at different locations for comparisons. Especially if this surveys are part of a nature consulting project or job (FFH and such), the client and the public expect or need results on such a level.

Identification of species

Subjective, hearing based identification

For a long time the only possibility for distinguishing bats in the field were - apart from capture - heterodyne or frequency division detectors. The acoustic footprint, mainly call frequency and rhythm, supported by optical information (flight style, size, ...) often allow an identification on genus and species level. For many species this is rather reliable if the user is experienced. Yet, in some situations even a id on genus level is impossible. Since the user experience is heavily influencing the results and calls are not archived for cross examination, this subjective method should be rejected if reliable data is needed. It nevertheless allows rapid assessment of bat activity in the field.

Frequency division recordings

The output of a FD detector can be recorded with regular sound hardware. It allows computer based analysis. The reduction of sound data nevertheless has it drawbacks. Especially short calls of less than 10 ms are reduced to very short signals that are hard to examine in more detail. Only a tenth of the orignal sound waves are preserved. For many species, for example of genus Myotis, this is not optimal.

Anabat SD1: FD + Zero-crossing analysis

The Anabat SD1 system (and its predecessor) are primarily used in the english speaking countries. It was developed as a long term monitoring system. To reduce the amount of data that has to be stored it implements a zero crossing analysis of the frequency divided signal to only save the wavelengths. Only waves above a selectable threshold are analysed and stored. The recording quality is good enough for basic genus identification. Many species in Europe can not be determined from these signals reliably. It is inferior to time expansion or real-time systems.

Time expansion detectors

In the early 90's these systems where invented in a cooperation of Ahlen and Pettersson. The recordings generally allow a good identification of calls. Yet, the low sample rate and low amplitude resolution in most devices decreases its suitability for species identification in qualitative sampling. It also has a low usability in the field since calls stored on ring memory have to be played back ten times slower than recorded (3s recording - 30 seconds playback) to record them with common audio hardware. You also have to handle the audio recorder simultaneously. While playback is active you can not record further bat passes.

Real-times systems

A couple of years ago advances in technology allowed development of consumer devices that allow digital real-time recording of bat calls. Two of the available devices were designed specifically for automated monitoring (Pettersson D500x, ecoObs batcorder). Two other systems have heterodyne/FD detectors coupled for auditory feedback in the field (Pettersson D1000x, Avisoft UltraSoundGate, the second only works if coupled with a PC). These systems have a high recording quality. Thus, identification can be supported by statistical methods.

Yet: never simple!

Even when using the last mentioned real-time recording systems species identification can be a tough process and not always grants unambiguous results. Variability of calls, unknown calls and other influences often don't allow extraction of a species based on calls. See our document on how automatic analysis works and from what it suffers: [Download not found]