SIGNAL COLLECTION AND ANALYSIS
Introduction
Bat detectors are invaluable for making field observations, but they cannot entirely substitute for detailed signal analysis carried out in the laboratory. For this it is necessary to record the signals on a high speed instrumentation recorder, taking great care to ensure that the signal is not distorted in any way during the recording process. Various forms of signal analysis may then be carried out at leisure, so as to provide a complete description, quantification and estimate of the functional properties of the signal.
Since most commercially available equipment for signal analysis is either designed to be used on audio frequency signals, or on radio frequency signals, it is necessary to replay the signals for analysis at reduced speed, thus lowering the frequency and increasing the duration of the sounds. Instrumentation tape recorders are so designed that it is possible to replay signals at any speed without distorting the signal.
Recording Techniques
The majority of field recordings were made with a Pemco 110 instrumentation recorder with a top speed of 30 i.p.s. (72 cm/s) and a frequency response extending to 150 kHz. This machine was especially modified to enable the speeds to be electronically switched to 3 ¾ i.p.s. (9 cm/s).
Laboratory recordings were made either on the Pemco or on a PI 6100 recorder, with a top speed of 37.5 i.p.s. (90 cm/s), and a frequency response which is flat to 100 kHz. This instrument was also used for replay of recordings from all machines for analysis.
Two SE-84 recorders, manufactured by SE Labs Ltd, were also available for use in the laboratory. These were equipped with a 30 i.p.s. speed option, and the record and replay circuits were modified to give a frequency response flat to 150 kHz.
In all cases the record and replay heads were obtained from Pemco ( later Pemtek) to ensure compatibility between the different machines.
All machines were set to give an output of 1V rms for an input signal of the maximum acceptable level. When making recordings the output channel of the recorder was always monitored on an oscilloscope to ensure that the signal did not exceed this level. In practice a small proportion of signals were allowed to exceed the nominal signal level so that the majority of signals were at the correct level. This is especially important because of the limited dynamic range of the instrumentation recorders (typically 38dB). Overloaded signals are rejected for analysis when they are replayed. For field use a Philips PM6010 portable oscilloscope was used to monitor recording levels. Since this is a dual beam oscilloscope it was possible to use the second channel to monitor activity when the tape recorder was not running, or for examining the signals with a period or frequency meter.
Analysis
The most basic form of analysis is to photograph the signal waveform. This can be done in one of two ways. The signal may be copied at slow speed onto the drum of a sonagraph, where it can be replayed continually into an oscilloscope, and a microswitch, activated by a pin attached to the sonagraph drum, can be used to trigger the oscilloscope. The oscilloscope trace can then be photographed over one or more sweeps across the screen. Alternatively a special camera can be used in which the film is moved continuously past the lens, at a selected rate. In this case the oscilloscope trace is halted in the centre of the screen, and the tape may be replayed at full or reduced speed. In this way long sections of a recording may be illustrated on a single trace permitting detailed analysis of pulse repetition rates and durations.
A period or frequency meter can be used with either of the above techniques, either alone or at the same time as photographing the signal waveform. Calibration signals should be recorded from a signal generator in conjunction with an accurate timer counter, on each film recorded in this way.
A spectrum analyser (manufacturer and type) was used to measure mean frequency patterns over periods of several minutes. In this case the spectrum of the signal would be measured and the spectra continuously averaged and displayed. When the spectrum had been collected it was then written out to an x-y plotter. Internal calibration signals were used to indicate the frequency scale. Linear or logarithmic amplitude scales were available.
The most common form of analysis is the sound spectrograph or sonagraph. Two models were used in this study, a Kay Missilyzer type 675 and a Kay Sonagraph type 6029. The principle is the same in both instruments although the frequency ranges covered and the filter bandwidths are different.
The signal to be analysed is first copied onto a rotating magnetic drum, from which it can then be replayed continually. A sheet of electrostatic paper is wrapped around the upper part of the drum and a stylus then traverses slowly upwards on a screw thread, writing a raster on the paper. The signal is used to modulate a carrier, the frequency of which is controlled by the vertical position of the stylus. The modulated signal is then filtered in a fixed frequency filter, with a switch selectable bandwidth, and the output is used to drive the stylus. The resulting sonagram is a plot of the spectrum of the signal as a function of time, with the amplitude indicated by the darkness of the trace.
Certain precautions must be taken when making sonagrams. The signal to be analysed must first be carefully monitored on an oscilloscope to ensure that it was not overloaded when the signal was recorded. It must then be monitored again with an oscilloscope connected across the VU meter of the sonagraph to ensure that it is distorted as it is re-recorded on the sonagraph drum. The VU meter is not a reliable indicator of signal strength for short duration signals because of its slow response time. Finally the signal level must be monitored as it is replayed from the sonagraph drum to ensure that the analysis circuits of the sonagraph are not overloaded.
If these precautions are taken then reliance can be placed on the frequency content of the signal as displayed in the sonagram. Caution must still be exercised however in the interpretation of the fine details of the sonagram. The effective bandwidth of the filter will have a marked effect on the display. If a wide bandwidth filter is used then the sonagram will have good time resolution, but will appear to be smeared vertically, in the frequency plane. Conversely a narrow band filter will give good frequency resolution but will smear horizontally along the time axis. The effect of this is most marked with a signal with a very rapid train of pulses. With the wideband filter this will appear as separate wideband pulses, but with the narrowband filter they will appear as a series of harmonically related frequencies, with a spacing equal to the rate of occurrence of the pulses. Both these descriptions of the signal are correct but it is important to be able to recognise the identity of such signals.
For special purposes the sonagraph may be used in a modified way.
It is often desirable to examine only a small part of the spectrum of a signal and this can be done in two ways. The first involves a modification to the sonagraph. As the pen moves upwards it moves a slider on a pot, the voltage from which is used to control the VCO in the analyser. By altering the voltages on either end of the potentiometer it is possible to arrange that only a small portion of the spectrum is examined. On later Kay sonagraphs this facility is available as an optional extra. The problem with this mode of operation is that the resolution of the system is not increased. The sonagram is merely stretched vertically.
An alternative, and preferable, although more complicated technique is possible. In this case the original recording is replayed at high speed into a balanced mixer, where the signal is modulated onto a carrier frequency from an accurate signal generator. The output of the modulator will contain the sum and difference of the two input signals. The difference frequencies can then be analysed on the sonagraph. For example, if it is desired to look closely at the frequencies used by Rhinolophus ferrumequinum, then the oscillator would be set to 75 kHz, and the sonagraph would be set to analyse over a range of 16 kHz. The sonagram would then cover a range of frequencies from 75 to 91 kHz, with a time scale of 1.2 s in real time. If absolute frequency measurements are to be made from these sonagrams it is important that the frequency of the oscillator is accurately known, and is stable for the duration of the sonagram. It should also be borne in mind that the sonagram will also display any frequencies on the other side of the carrier frequency, 59 -75 kHz in the above example, except that for these frequencies the display will appear to be upside down.
When using this technique the increase in frequency resolution is obtained at the expense of time resolution, since the signal is no longer being slowed down first, and the duration of the sonagram is therefore longer than when using the normal technique.
The biggest disadvantage of sonagrams is the short duration that can be displayed. Although it is possible to connect several sonagrams end-to-end it is quite tricky to accurately synchronise the process of copying successive portions to the sonagraph and the process becomes impractical when the interval between sounds is greater than the duration of a single sonagram.
This has been overcome in a design by ( ) in which successive samples are analysed in a spectrum analyser and displayed vertically on an oscilloscope screen, with the intensity being used to modulate the brightness of the trace. This results in a continuous sonagram type display with a frequency bandwidth of 8 kHz.
An alternative approach is to use digital computers to perform the spectrum analysis and display of the signals. A considerable quantity of fast storage is needed to cope with signals of very long duration, but normal sonagram type displays can be carried out on quite modest computers. The limiting factor other than storage availability is the speed of the analogue to digital converter used to translate the signal into digital form. This should be capable of sampling continuously at a rate of at least three times the maximum frequency to be analysed, and the computer must be able to simultaneously store all the information collected at the same rate.
The time taken to perform the calculations necessary will vary quite considerably, depending on the type of computer used, and the efficiency of the algorithms and their implementation. A small computer may take a minute or longer to calculate each spectrum, or a large super-mini computer could be capable of displaying the results in real time (at least for audio frequency bandwidths). A digital computer is also capable of performing alternative types of analysis and display, for much of which a great deal of the calculation need only be performed once. For example it is possible to generate auto-correlation functions of the signals, or with somewhat greater effort, the sonar ambiguity function. Programs may also be provided to perform feature extraction on the resulting analyses and these may then be subjected to statistical analysis.
To date the computer displays of ultrasonic signals have tended to be of rather poor quality. They are often displayed as isometric projections of a surface, possibly with the frequency and time axes reversed with respect to a ‘normal’ sonagram, and they can be confusing and difficult to interpret. This is especially true of published examples where only a single display of a signal is available. This problem should be rapidly overcome as high resolution colour, grey-scale displays become more common.
The sonagraph type of display provides a simple description of the structure of the sound. In many cases, especially for echolocation signals, it is useful to be able to estimate the functional properties of the sound. For sonar signals it is possible to make certain broad statements about the properties of different types of signal. A long constant frequency signal will be good for measuring velocity by the doppler shift, a short wideband pulse will be ideal for measuring range. It is less obvious to tell what advantage a bat might get from using more or fewer harmonics, or by shifting the main energy in the signal from one harmonic to another.
Much of the information about the properties of a signal, assuming an ideal processing system, can be derived from an analysis of its structure. The auto-correlation function of the signal gives a measure of the effectiveness of the signal in a range measuring system. The narrower the peak of the function the greater the range resolution. If the auto-correlation function consists of a whole train of peaks of equal height and shape, then there will be multiple ambiguities in the range measurement - it is not possible to tell which peak really corresponds to the range of the target.
The sonar ambiguity diagram is a more elaborate extension of this principle and shows the effectiveness of the system at measuring the range of targets moving at different velocities (or alternatively the velocity of targets at different ranges). Derivatives of the ambiguity diagram can then be produced to show the effect of acceleration on the performance of the sonar system, or the effects of attempting to measure the angular direction of the target at the same time as other parameters. It can also be used to show whether a system has good rejection of echoes from other sources, whether the echolocation cries of other members of its own species, or jamming signals produced by its prey.
The first ambiguity diagrams of the echolocation signals of bats were produced by Cahlander ( ), using an analogue system, and computer generated diagrams of the signals of a number of species of bats have been produced by Beuter ( ).
