Harmonics


HARMONICS, BATS AND ANABAT

I've recently read a number of papers, or drafts of papers, which have quite seriously misrepresented aspects of Anabat and what it can do. To some extent, I can't blame the authors for this, since there isn't much published on what Anabat really can do. On the other hand, it gets tiresome to see the same old nonsense repeated ad nauseum as if it were fact, by people who clearly don't have a good understanding of how Anabat works or of how it can be used.

One subject which comes up again and again, is the claim that Anabat cannot deal with harmonics. To quote a recent example, in the paper "Constant-frequency and frequency-modulated components in the echolocation calls of three species of small bats (Emballonuridae, Thyropteridae, and Vespertilionidae)" by Fenton, Rydell, Vonhof, Eklof and Lancaster, Can J. Zool. 77: 1891-1900 (1999), the authors state (page 1895) "Variation in harmonic composition is a further difficulty, and the calls of R. naso make it obvious that a bat-detection system like Anabat ........., which does not display information about harmonics, may not be reliable for identifying some species ..... such as R. naso." This statement is extremely misleading, so I'd like to put the record straight here, and explain what this means and why it's wrong.

SPECTRAL ANALYSIS AND ZCA

Firstly, let me explain what spectral analysis and ZCA are, and why there is a general perception that ZCA can't deal with harmonics.

In spectral analysis, as it is generally employed, a bat call is sampled at a very high rate and the samples are passed through a software algorithm called Fast Fourier Transform (FFT) which can be used to separate out the various frequency components which make up the signal being investigated. Output can take several forms, such as a power spectrum, which is typically presented as a summation of the various components in the whole bat call, or as a sonogram, which is effectively derived from many short term power spectra, and which shows graphically how the power spectra vary in time. Using spectral methods, many harmonics of the same signal can be displayed at once.

In ZCA, dots are drawn on the screen to show the average frequency of the loudest part of the signal, since the last dot. The time between dots is inversely proportional to the frequency being measured, but also depends on the frequency division ratio being used. The result, for a bat call, is a single curve which shows the way in which the frequency of the bat call varied in time. At any one time, ZCA can only respond to the loudest (dominant) component of the signal being investigated, so if more than one harmonic is present in the original signal, only one of these can be displayed and it will always be the loudest (but see below what that means).

So let me be really clear about this. If you want to investigate the harmonic structure of bat calls, then spectral analysis is the better way to do that, because it will show, for each bat call studied, the full range of component harmonics present in the signal (within certain limitations - if a component is too faint, even spectral analysis won't show it). If you wanted to quantify the relative intensities of the different harmonics, then spectral analysis is certainly the way to go. But for many purposes, full harmonic detail isn't required. Anabat, using ZCA, is designed specifically to present data useful for species identification and it is optimized for real-time display of frequency-time characteristics of bat calls, and for convenient storage of recordings of those calls. While it can't match the harmonic detail possible with spectral analysis, it can still be used to reveal a great deal about the harmonic structure of bat calls, and in particular, it can reveal the details of most interest for species identification.

HARMONICS

Any real sound consists of many frequency components. To have only a single component, a sound would have to consist of a perfect sinewave of infinite duration. As soon as we modulate a sound (turn it on and off, for example), we increase the spectral complexity of the sound (we increase the number of frequency components). Many real sounds have multiple sources. If we strike a bell, then that bell has several independent modes of vibration, which cause it to simultaneously emit several different signals, each with a different frequency composition, decay time, etc. Each of these modes appears as a separate source of sound.

Real sounds also aren't perfect sinewaves, and any distortion of a sinewave results in the production of a harmonic series. A harmonic series is just a collection of harmonics, where the frequency of each harmonic is an exact, whole number multiple of the fundamental frequency. Harmonics are, by definition, EXACT whole number multiples of the fundamental. We can refer to the fundamental as H1, the second harmonic as H2, the third as H3 etc. Note that the component which has a frequency exactly twice the frequency of the fundamental is called the SECOND harmonic, not the first as some people seem to think.

In some systems, different modes of vibration can have frequencies which are very close to, though not exactly, whole number multiples of the primary mode. If we pluck a violin string, it will vibrate with several modes, one of which will have a frequency very close to twice the frequency of the primary mode. But this second mode is not a harmonic, as the frequency is not exactly twice that of the primary mode, and it is produced by a different mechanism. Yet it is often called a harmonic. A better word would be overtone.

For practical purposes, we can think of a bat call as produced by a single mode of vibration and consisting of a series of harmonics. The fundamental (H1) is the lowest harmonic present in the series. The harmonic structure (the proportions of energy in the different harmonics) of the sound leaving the face of the bat may not closely resemble that of the signal emanating from the vocal chords. The different harmonics will be amplified or attenuated by various resonances inside the bat, or even outside it, in such a way that certain harmonics may be greatly emphasised in the signal leaving the bat, while others may be profoundly suppressed. It isn't necessary that the fundamental will be dominant at the face of the bat, and it also isn't necessary that the same harmonic will be dominant throughout a whole bat call.

WHICH HARMONIC IS DOMINANT?

From the point of view of an Anabat detector, which harmonic is dominant will depend on many factors, some extrinsic to the detector and others internal to it.

Firstly, the bat itself will determine the mix emanating from it. Secondly, the mix will be different from different directions relative to the bat. Typically, higher frequencies will be more focused by the bat than lower frequencies, so the proportions of energy in the different harmonics will depend to some extent on which way the bat is facing. We should expect that higher frequencies will be favoured in a beam out the front of the bat, while lower frequencies would be relatively emphasised to the side or rear. However, this relationship is going to be very complex, and will vary greatly between species.

Thirdly, the atmosphere attenuates high frequencies more than lower frequencies, so that will also have an effect. What this will mean, is that at a distance, lower harmonics will be favoured over higher harmonics. Even if H3 was dominant close to the bat, H1 might be still be dominant from a few metres away.

Then there are the frequency-dependent factors intrinsic to the bat detector. The microphone of the Anabat is much more directional to higher frequencies than lower frequencies. The microphone also does not have a flat frequency response, and the preamplifiers inside are designed to roll off in sensitivity to lower frequencies, so that peak sensitivity should occur around about 40 to 50 kHz, with much less sensitivity down at 20 kHz or below.

When you consider all these factors, it should be apparent that which harmonic is dominant won't be easy to predict. However, we can illustrate how it works with a few examples.

Consider firstly a bat producing constant frequency calls of H1 = 20 kHz and H2 = 40 kHz with equal energy into both harmonics. The Anabat will be much more sensitive to H2, while the atmosphere will transmit H1 better. So at close range, such a bat would always be detected on H2, while from a long distance, it would always be detected on H1. If such a bat emitted low intensity calls, it might always be detected on H2, since it couldn't be detected from far enough away for the frequency response of the atmosphere to become an issue. An interesting consequence of the CF call would be that at some distance, neither H1 nor H2 will really dominate the other, and in that situation (which will rarely happen in real life) the Anabat would detect the bat as at some intermediate frequency which isn't present in either of the harmonics. Such an intermediate frequency signal would probably not look at all like a bat call, as it would fluctuate wildly even within a single call. Since the distance between bat and detector is continually changing in nearly all circumstances, the probability is extremely low that H1 and H2 could remain equally dominant for any length of time. In practice, I have never seen this happen.

Now consider another example. Suppose a bat produces an FM sweep which on H1 sweeps from 30 down to 10 kHz and on H2 sweeps from 60 down to 20 kHz, again placing the same energy into both harmonics throughout the call. From a long distance, such a bat would again be detected only on H1, while at close range, H2 would always dominate. At some intermediate range, however, H1 would be dominant in the early part of the call, and H2 would be dominant in the later part of the call. This would happen because of the interplay between the frequency responses of the Anabat and the atmosphere. At such a distance, early in the call, H1 will dominate because it is less attenuated by the atmosphere. But as the frequency sweeps downwards, it will reach a point where the frequency response of the detector overwhelms the response of the atmosphere, and H2 will then become dominant. The user would see a pattern in which the harmonic switched midway through the call from H1 to H2. A typical manifestation might be a call which appears to sweep down from 30 to 20 kHz, then jumps up to 40 and continues to sweep down to 20. Just where in the call the switch from H1 to H2 happens, will depend on the distance to the bat, so as distance is increased, the switch between harmonics will occur progressively later until at some distance it doesn't occur at all.

Thirdly, consider another bat which produces a sweep down from 40 to 20 kHz on H1 and a sweep from 80 to 40 kHz on H2. This bat differs from the preceding bats in that H1 is produced with much more intensity than H2 for most of the call, but just as the H1 is reaching 25 kHz, the bat suddenly switches its output so that H2 is now much louder than H1. The pattern of harmonic switching shown by such a bat would depend on the extent to which the harmonic produced with most intensity overwhelms the other. It may be that H1 would still always be detected at distance and H2 at close range. But whether or not this is the case, there will be a range of intermediate distances at which the call will always appear to switch from H1 to H2 at the same point in the call. So at these distances, the call will be seen by Anabat as sweeping down from 40 to 25 kHz, then jumping up to 50 kHz and sweeping down to 40 kHz.

Harmonic Switching, where Anabat detects different harmonics at different times, is often encountered in Anabat, and it can tell us a lot about the harmonic structure of the call emitted by the bat.

HARMONIC SWITCHING

In the hypothetical cases above, we can see three different ways in which the dominant harmonic can be switched. In the first, the harmonic switches within a sequence but between calls. As the bat comes closer, H2 takes over from H1, and as it gets further away again, H1 will take over again. In the second case, the harmonics would again switch between calls, but they could also switch within a call, with the point at which the switch occurs being determined by the atmosphere and bat-detector characteristics and the distance between bat and detector. In the third case, the harmonics will again switch within-call, but the part of a call where the switch occurs will be controlled by the bat.

Bat-controlled Harmonic Switching seems to be very common. It often appears in species such as Tadarida brasiliensis and Eptesicus fuscus, when they are flying in high clutter or tight maneuvers, even though they never show it when flying in low clutter. It can be recognized when the switch between harmonics keeps occurring at the same place in each call, and this also often takes place at some boundary in the call (such as the "knee", where there is a sudden change of slope), producing an effect where the two (or more) harmonics seem to belong to completely different call shapes. Such an effect can look very much like the presence of two completely different bats, until a close examination of the calls shows that the higher portion always appears immediately after the lower portion.

It is my impression that in every case where Harmonic Switching is seen by Anabat, the point of switching is at least partly controlled by the bat. Even in a case like that of Townsend's Big-eared Bat Corynorhinus townsendii, where the HS can occur almost anywhere in the call, it often tends to take place at the same point for several calls in a row.

SOME REAL CASES

In Australia, the Yellow-bellied Sheathtail Bat Saccolaimus flaviventris, is a widespread species, mainly in the inland. It is one of the few clearly audible species in Australia, yet Anabat always picks it up at 20 kHz. I can't hear anything near 20 kHz, so I had always assumed that the 20 kHz being detected must be an H2, and that the audible signal was H1 at 10 kHz. But a bat researcher using spectral analysis assured me this wasn't the case, and that there was no sign of a signal at 10 kHz. Having puzzled over this for a while, I went carefully over all my recordings of the species, and eventually I found a case where there was obvious Harmonic Switching taking place within calls, between signals at 20 and 30 kHz. Clearly, these harmonics have a relationship of 2 to 3, so establishing that the 20 kHz is the second harmonic (it could be a higher number, divisible by 2, of course, but it is at least the second). In this case the H1 was presumably too weak to be detected by the spectral analysis being used (and it would probably never be detected by Anabat), yet it is clearly audible to humans. Actually, since I wrote this, I have seen many cases where H1 was detected briefly at the start of a call.

Townsend's Big-eared Bat Corynorhinus townsendii is a species whose calls could easily be confused with other species, except that it shows a very distinctive harmonic structure. In this case, Harmonic Switching occurs both within and between calls. Although the calls it gives are often faint and therefore hard to detect, even very brief sequences of their calls typically show the distinctive patterns of Harmonic Switching when Anabat is used, and there is nothing confusing or ambiguous about the way this is presented.

Several species, such as Mexican Freetail Tadarida brasiliensis and Big Brown Bat Eptesicus fuscus frequently change their harmonic mix to emphasise the H2 at the end of their calls when they are in high clutter, a fact easily observed using Anabat. Although this behavior is not in itself of diagnostic value, it is a point worth noting, since the presence of H2 in the Anabat display tells the observer that the bat is in high clutter, and can help separate such calls from those of species like Pallid Bat Antrozous pallidus, which in low clutter produce similar calls but lacking the H2.

Rhynchonycteris naso, used by Fenton et al (above) as an example of a species not reliably identified by Anabat, is in fact a good example of a species whose harmonic structure is readily apparent to a competent Anabat user (see O'Farrell and Miller, Biotropica 31(3): 507-516 (1999)).

SUMMARY

Anabat would certainly not be a good choice of system for the purposes of studying harmonic structure in detail. However, in cases where harmonic structure seems to be of significance to species identification, it is quite easy to see the main features of the harmonic structure because harmonics of roughly similar amplitude become dominant under different circumstances. Bats detected under real field conditions vary in distance from the observer throughout a sequence, so the harmonic structure becomes apparent when several calls are seen, and in many species, Harmonic Switching within a call allows something of the harmonic structure to be seen even in a single call. Contrary to what is often stated, the presence of more than one harmonic is in no way "confusing" to an Anabat, but is often helpful for species resolution. As with many aspects of Anabat, the emphasis is on variation and the recognition of patterns of variation over many calls, rather than on resolving finer detail in a smaller number of calls, which is the strength (and also a weakness) of spectral analysis. In real field situations, Anabat does a surprisingly good job of resolving harmonic structure where this is important.

Back to: Anabat Technical Notes, Anabat Contents, Home


Contact Information

I have tried to make this page as accurate as possible. If you have any comments, positive or negative, or suggestions, please Email me. If you disagree with anything I've said here, please let me know. I will always consider presenting alternative viewpoints, in the interest of fostering discussion so that we may all learn something.

Email:


Last revised: June 04, 2011.