BAT DETECTORS

Introduction

The bat detector is one of the most fundamental tools of the researcher into ultrasonic bio-acoustics since it enables direct observation of the acoustic behaviour of an animal whether in the laboratory or in the field.

The basic requirement of a bat detector is that it should give some indication that an ultrasonic sound has occurred. The indication that it gives can be either audible or visual. In either case the common feature to all bat detectors is that they must have a microphone and headstage amplifier for detecting the ultrasonic sounds. Ideally this should have a frequency response extending from 15 to over 200 kHz, with good rejection of audio frequency signals.

The simplest form of bat detector is to monitor the waveform from the microphone on an oscilloscope. This technique is useful for monitoring general activity levels, and with a suitable type of camera can be used to record the signals on film (Griffin 1958).

It is generally more useful if information about frequency, rather than amplitude, can be displayed. The easiest way of achieving this is with a zero-crossing period meter. The amplified signal from the microphone is converted into a square-wave, using a Schmitt trigger, and each cycle or half cycle is integrated. At the end of the cycle or half cycle the output of the integrator is sampled in a sample and hold circuit, and displayed on a scope for the duration of the sample period.

The squaring circuit only extracts information about the dominant harmonic in the signal. If the period of each half cycle is displayed and the waveform is asymmetric then a double trace may be produced indicating the presence of other frequency components, but interpretation of such a display must be done with great caution.

The period meter output is the reciprocal of the normal sonagraph type of display. Simmons et. al. (1979) developed an instrument using a period to frequency conversion circuit, which provides an output which is linearly proportional to frequency over a wide range.

I designed a less complex alternative to this which uses an exponential law as an approximation to a reciprocal law (see Fig.). The circuit is basically the same as the period meter, except that the integrator is a simple RC combination providing the exponential function, with the capacitor connected to the supply rail rather than ground so as to provide the signal inversion. The circuit gives an adequate approximation to a linear function of frequency over a 10:1 frequency range, and alternative ranges can be selected by switching in different values of capacitor in the integrator.

Fig BD1 Comparison of an exponential (dotted) and a reciprocal (solid) function

Later development of the ambiguity diagram computer suggests that the period function of the signal may be more significant to the echolocation characteristics of the signal. The period may therefore be found more suitable for the study of bats.

The main disadvantages of these visual bat detectors is that they need to be looked at and it is not possible to observe the subject at the same time. Also they require the use of an oscilloscope, which tends to be very expensive, especially if sufficiently portable for field use. Visual bat detectors can be useful when recording signals on high speed tape, since it is then necessary to monitor the signals on an oscilloscope anyway (Silver and Halls 1980).

The basic form of audio bat detector is the simple envelope detector (McCue and Bertolini 1964, Simmons et al 1979). The high frequency signal is rectified and smoothed and the envelope is used to drive a loudspeaker. This instrument is inherently a broadband device so that any ultrasonic sound present will be detected, but the output contains no information about the frequency content of the signal. It also suffers from a problem with long constant amplitude signals where the envelope is a single cycle of very low frequency, and is therefore very poorly audible.

The simple envelope detector may be improved in two ways.

The use of band limiting filters in the headstage amplifier allows some measure of the frequencies present in the signal (McCue and Bertolini 1964), although unless large numbers of filters are available this tends to be a fairly coarse measurement. Secondly it is possible to incorporate a tone generator which is modulated by the envelope of the ultrasonic signal, which enables even very long constant amplitude signals to be detected (Fenton pers. comm.).

A refinement of this technique would be to use the control voltage of a phase-locked loop, locked onto the hf signal, to control the frequency of the tone generator. The output would then contain audible information on the dominant frequency in the signal, as well as information about the amplitude. The technique is not ideal since a PLL requires several cycles of signal before it can lock, and in the case of the shorter bat signals there would not be enough cycles in the original signal. Since tone discrimination in such short sounds is poor however, this would not be a severe disadvantage, although it would be advisable to use a design of PLL in which the out-of-lock frequency is in the centre of the frequency band covered.

A simple alternative to the envelope detector is to use a super-heterodyne receiver (Noyes and Pierce 1938, Pierce 1948, Kay and Pickvance 1963, Pye and Flinn 1964, Pye 1968, Hooper 1969). Although the circuitry is more complex it is a simple matter to convert a vlf radio receiver by replacing the aerial with a microphone and headstage (Pye and Flinn 1964).

Alternatively it is possible to convert a portable long-wave radio to cover the 15-160 kHz band, and this is the technique used in the QMC instruments Ltd. Mini bat detector. In this case a BFO is included to overcome the problem with long constant amplitude envelopes. Since the super-het. is a tuned instrument with a bandwidth of about 5 kHz, it is practicable to use a microphone with a non-flat frequency response. An audio electret hearing aid microphone (such as the Knowles type BT1759) with a response which is flat to 15 kHz and then decays at 12 dB / 8e, provides a signal to noise ratio, and therefore a sensitivity, which is reasonable constant from 20 to 110 kHz.

In use it is not always convenient to have to decide between using a tuned and broadband instrument. When searching for bats or rodents of unknown species it is desirable to have a broad band instrument so that any sounds will be detected. Having established the presence of ultrasound it is then useful to be able to establish the frequency or frequencies being used, and for this a tuned instrument is necessary. Pye (Sales and Pye 1974) achieved this by using a tuned direct conversion receiver in which the output is the difference frequency between the input signal and a tuned sinewave local oscillator (a technique which also obviates the need for a BFO). In this case the broad band mode can be implemented by using a comb spectrum with lines every 10 kHz as the local oscillator. Since the audio bandwidth is about 5 kHz, and no signal can be more than 5 kHz away from a line in the spectrum, then any signal in the range can be detected. By arranging that the comb spectrum can be fine tuned to alter the line spacing slightly it is possible to avoid nulls due to a signal occurring at exactly the same frequency as one of the spectral lines, and also allows tuning to get an optimally pleasant tone to listen to. Switching the local oscillator between the comb spectrum and the tuneable sine wave oscillator gives instant switching between tuned and broad-band modes of operation.

This instrument is very versatile and pleasant to use, but does suffer from some technical drawbacks. If the local oscillator is not a pure sinewave, then an output will be produced when its harmonics occurs near the frequency of an external signal. If this happens it appears that there is a weak frequency component at harmonics of the actual frequencies present. Since the sinewave oscillator has to be tuneable over a range of greater than 10:1 it must therefore be very carefully designed and set up.

An alternative approach to bat detector design was developed by Andersen and Miller (1977). The approach in this case is to extract the dominant frequency component of the signal and then to use digital techniques to reduce this to an audible frequency. The signal is squared with a Schmitt trigger, and the resulting wavetrain divided in frequency by a factor of ten. The resulting low frequency squarewave could then be recorded on a simple audio cassette recorder or listened to directly.

Since this is a broadband instrument it is necessary to compensate for the declining frequency response of the electret hearing aid microphone used. This effectively reduces the sensitivity across the band to the sensitivity at the highest frequency. In addition the user is listening to a full amplitude output signal all the time and it is therefore necessary to set the threshold of the input Schmitt trigger above the background noise level, with a consequent loss in performance. The measured average sensitivity of this instrument was 64 dB SPL (Maries pers. Comm..)which compares unfavourably with the sensitivity of the QMC Instruments Mini bat detector, with its detachable horn, of 15 dB SPL at 40 kHz.

Improved Designs

Pye (pers comm) tried a number of ways of incorporating a broad-band facility into a tuned super-het bat detector. Amplitude modulation of a carrier at the intermediate frequency, by a suitable rectangular waveform will produce a comb spectrum with the desired line spacing and covering the necessary range of frequencies, but was found to be impractical due to the high mark-space ratio which is necessary to ensure a flat spectrum.

I proposed two techniques for generating a comb spectrum local oscillator. The requirements of the comb spectrum generator are that it should provide a spectrum with discrete lines 10 kHz apart between 465 kHz and 665 kHz, with less than a 6 dB variation in amplitude. To avoid problems with signals occurring at exactly the frequency of one of the lines of the spectrum it should be possible to trim the line positions by an amount equal to the bandwidth of the i.f. filter. This may be done by either shifting the entire spectrum or by altering the line spacing.

The preferred design comprises a square wave generator which is swept linearly in frequency between the frequency limits that the line spectrum is to cover. The precise spectrum of a frequency modulated signal such as this is dependent upon the ratio of the centre frequency to the modulation range. In this case it is possible to achieve a spectrum which is largely flat within the range swept, but which falls off steeply outside those limits. The line spacing within the spectrum is equal to the frequency of the sawtooth modulating waveform, and the trim facility can therefore be implemented by adjusting the frequency of the sawtooth waveform.

The original design used a 555 tone generator as the sawtooth oscillator, with external frequency adjustment.

The voltage controlled oscillator was a CMOS circuit, the MC4046, which has the advantage of very low power consumption. When this principle was incorporated in the S100 bat detector however the design was altered to use a unijunction oscillator to generate the sawtooth waveform, and a modified transistor astable multivibrator as the VCO. The reasons for these changes were the voltage and power requirements of the 555 oscillator which were incompatible with the supplies available in the instrument, and the fact that to cover the required frequency range at the supply voltage available it would be necessary to run the 4046 VCO outside its standard specification, which was considered to be unacceptable for a commercial instrument.

The alternative design of comb spectrum generator was to use a pseudo-random bit sequence (PRBS) generator. This is a device which produces a digital output in which the state of the output at each clock cycle is unrelated to the preceding sequence of output states. However the total sequence of output states is repeated continuously which is why the sequence is described as pseudo-random. A pulse train of this type has a comb spectrum with lines spaced at the repetition rate of the total sequence, with the amplitude of the lines decaying with frequency according to (sin(x)/x)^{^2}. There will thus be a null at the clock frequency (f_c), and the amplitude of the spectrum will have fallen by 6 dB at 0.45 x f_c.

The generator consists of a shift register in which the input is the exclusive OR function of the last and one or more earlier output stages of the shift register. It is important to take the feedback points from the correct stages of the shift register if a maximal length sequence is to be obtained, in which case the sequence will be of length (2ⁿ)-1 , where n is the number of stages in the shift register. It is important to ensure that the shift register does not contain all zeroes at start up, but given this condition then the maximal length sequence will always be produced.

To produce the comb spectrum required for the broad-band function of a super-het. bat detector the comb spectrum must extend to 665 kHz. This can be achieved by using a sequence with a length of 127 bits clocked at a frequency of 1.27 MHz. The -3 dB amplitude of the spectrum is at a frequency of 571 kHz, but since only a small part of the spectrum is to be used (465-665 kHz) the fall in amplitude within this range will be less than 5 dB. The spectrum then has to be filtered to remove components outside the 465-665 kHz range. Since a small change in clock frequency produces a significant change in the position of the lines in the spectrum a very stable clock is required, which should also be adjustable to provide the broad-band trim function.

A more convenient technique is to filter the comb spectrum to remove components above 200 kHz and then to use the pulse train to modulate the BFO of the bat detector. This will result in a flat comb spectrum extending 200 kHz either side of the i.f. frequency, and since the BFO is used audible beats between the comb spectrum and the BFO are prevented.

The second design of bat detector is a modification of the frequency divider detector of Miller. At the time of its development there was a need for a means of recording some information about the sounds produced by bats in the field, as part of a survey of Craseonycteris thonglongyai in Kanchanaburi Province, Thailand. An S100 and a Mini bat detector were available but in these instruments the audio output does not retain frequency information, although this can be read off the dial at the time. Weight and cost limitations precluded the use of an instrumentation recorder.

The circuit was designed to interface between the h.f. output of the S100 bat detector, and a small portable audio cassette recorder. The basic circuit is shown in fig. BD3 and improves upon the Miller design by extracting the envelope of the h.f. signal, and then using it to remodulate the frequency divided square wave. It is then possible to set the Schmitt trigger threshold well below the peak noise amplitude, thus permitting maximum possible sensitivity. There is also an advantage in using it in combination with the S100 front end, as the microphone and headstage are more sensitive than the electret microphone and have a flatter response through the ultrasonic band.

The prototype used a single CMOS analogue switch as the modulator, the l.f. square wave simply switching the envelope signal on and off. As implemented in the S200 bat detector the modulation is double sided by switching alternately between a positive envelope and a negative envelope of the original signal.

fig. BD4 shows sonagrams of echolocation pulses produced by Craseonycteris thonglongyai and Rhinolophus robinsoni. The constant frequency pulses of Rhinolophus show up reasonably clearly, but the shorter wider bandwidth signals of Craseonycteris are not at all distinct, largely due to the very short duration of the signals. Since the time bandwidth product of the sonagraph is fixed expanding the time scale simply has the effect of compressing the frequency axis, which does not significantly improve the display. It is possible to determine the area of dominant frequency from the sonagrams, but not especially accurately.

A simpler means of analysis is to examine the waveforms directly. fig. BD5 shows the waveforms of Craseonycteris thonglongyai in their roost, and waveforms of a small bat which was observed hunting in the evenings, but which was never caught for positive identification. Note that on these waveforms there is a different scale for the pulse durations and the frequency.

It can be seen from a comparison of mean frequency and pulse durations that these signals are probably from the same species of bats. Further evidence can be obtained by comparing the pulse repetition rates, which can be easily measured from the sonagrams.

Discussion

The main features required of a bat detector are that it should be portable and that it should be cheap. We have seen that these goals can be met in a very wide variety of ways, with varying trade-offs between cost and the amount of information about the detected signal that can be made available.

Visual systems can give a lot of quantitative information about the signals, but tend to be inconvenient in use and the cost of the display portion tends to be high. For use in conjunction with a high speed tape-recorder, when the display unit is usually available anyway, then they can be very useful.

Audible systems are better suited to providing qualitative information, and can use the high analytical qualities of the operators ears to do much of the work. The only quantitative information available relates to the frequencies contained in the signal, and that is only available from tuned devices. Broad band devices may preserve the information, but it is not readily available in real time.

In practice the combination of broad-band and tuned modes in a single detector, combined with the ability to record both the h.f. signal and a derived l.f. signal, provides a good compromise and this has been achieved in the QMC Instruments Ltd S200 bat detector.

There is still plenty of room for further development, but consideration must be given throughout to the cost/performance ratio.

The S200 bat detector may be improved in several ways. It would be very simple to add a visual period meter output, for display on an external oscilloscope, or if preferred at slightly more complexity a frequency meter display. The other main useful modification is slightly more complicated. When using the machine in the field is not uncommon to hear a bat when in the broad-band mode, which is only sporadically present. It is then very difficult to locate the frequency being used in the tuned mode unless the identity of the bat is known beforehand. This could be overcome by using a modification of the S100 broadband function, in which the upper and lower limits of the frequency sweep could be progressively altered, until the signal is known to lie in a narrow range of frequencies which can be rapidly searched in the tuned mode. The alternative technique of enabling both tuned and broadband modes simultaneously would be more appropriate to the S200 design, but gives the user no extra help in locating the frequency of the signal.

The possibility of doing a full frequency analysis of the signals in real time, displayed in the form of a continuous sonagram, sounds very attractive. While this may be suitable for long slow rodent type signals the short duration of bat signals gives little time to take in the amount of information being presented. A memory scope could capture a single signal, but other signals will be lost while that sample is examined, and field portable memory ‘scopes are not very common. For real time appreciation of the signals the audible bat detectors are unsurpassed. However, it would be very useful if additional information about the signal could be preserved on a low cost audio cassette recorder for more detailed analysis later.

The frequency divider type of detector is capable of doing this but suffers from the fact that information about the presence of harmonics is lost, and It is not possible to tell whether frequency changes are real or simply shifts of energy from one harmonic to another.

For harmonically structured signals it should be possible to divide down the frequency of each harmonic that is present, each modulated by its own envelope, and combine all the low frequency components into a complete low frequency replica of the original signal. Such a system could be implemented by breaking the signal up into 12 1/3 octave frequency bands from 10 to 160 kHz, and then use the existing frequency divider circuit on each channel, finally combining all the outputs. The complexity of the circuit is very high, however, and performance would still be poor on wideband non-echolocation signals, or on slow warbles from rodents or dolphins if these occurred near the overlap region of a pair of filters.

An alternative technique would be to preserve slowed down samples of the signals, so that a full analysis with no loss of information about the individual signals could be performed. In this case it would be necessary to discard the spaces between the signals, which for low repetition rates are empty anyway. The saving in recorded bandwidth is then made at the expense of loss of information about repetition rate, and loss of some signals at high repetition rates.

The most convenient method of performing this type of signal conversion would be to digitise the incoming signal at a rate of about 500,000 samples per second, and to store it temporarily in digital memory. The signal could then be copied out of the memory at low speed and converted back to analogue form for recording on a cassette recorder. One or more of several different modes of operation could be used. The incoming signal could be randomly sampled, the contents of the memory being dumped to tape whenever it had become full. The occurrence of a signal could be detected and the memory contents since the last dump could be saved as soon as the signal ended, or data could be read cyclically into the memory until a manual trigger was given, dumping the last memory full to tape.

Considerable quantities of memory would be required, but the cost of this is falling rapidly. One Megabyte of RAM would allow collection of data for two seconds at a time, with a dynamic range of 40 dB. If reduced dynamic range could be tolerated then cost could be reduced considerably. The use of more than one memory module would allow double buffering, so that further data could be sampled while the contents of a memory was being written to tape, although the overall loss of information would not be decreased. Improvement in the ratio of recorded time to lost time could be achieved by using a multi-track tape recorder. Indeed, with an eight track recorder and a speed reduction ratio of 8, then no information need be lost, although reconstitution of the signal might be complicated.

The digital memory recorder is still useful as a bat detector since the slowed down signal can be monitored as it is stored on tape, albeit the information is slightly out of date. An alternative method of recording h.f. ultrasonic signals at low cost (compared to the cost of a high-speed instrumentation recorder) is available, although it is not usable as a real time bat-detector. This technique is to use a portable video recorder, which has a very long recording time per cassette. Since the bandwidth of the video channel, while very wide, is not guaranteed to be flat, and does not come down to the ultrasonic range it is necessary to use a frequency modulation technique. The ultrasonic signal is used to modulate a carrier of about 3-4 MHz, with a deviation of about 2 MHz, and a phase-locked loop detector can be used to de-modulate the signal on replay. The disadvantages of this technique are that it is not possible to replay the signal at reduced speed (although stop-frame facilities can be used to replay the same portion of signal repeatedly) and there may be brief breaks in the signal at each revolution of the recording head.

The potential for increased complexity in bat detector development is great, but it will be hard to improve on the convenience of a simple tuned bat detector, which is small enough to fit in your pocket.