The Calling Song of the Bladder Cicada, Cystosoma Saundersii: A Computer Analysis

Males of many insect species attract females over long distances by means of loud sounds. These are termed calling songs and often consist of relatively pure tones, especially among the families Gryllidae, Gryllotalpidae, some Tettigoniidae (Ortho-ptera) and Cicadidae (Homoptera) (reviews by Bennet-Clark, 1971 ; Michelsen & Nocke, 1974; Elsner, 1978). In the present report a detailed description is given of the calling song of the bladder cicada, Cystosoma saundersii, an insect which has been the subject of several recent studies (Young & Hill, 1977; Fletcher & Hill, 1978; Simmons & Young, 1978; Josephson & Young, 1979). Earlier descriptions of the calling song of this species were based on the limited information available from oscillograms and sonagrams (Young, 1972 a, b).

Calling songs were recorded with a Sennheiser MKH 816 directional microphone and a Nagra IVS tape recorder at a tape speed of 15 in. s⁻¹. Six undisturbed males were recorded singing in shrubs. In addition, recordings were made from four males singing while tethered in the field ; these four males were tethered as part of a study by MacNally & Young (1980) on the energetics of sound production in this species. The recorded songs were digitized for computer analysis via an analogue to digital converter at a sampling frequency of 16 kHz.

The computer displayed waveform (Fig. 1 a, c) shows the distinct sound pulses which characterize the song, with each pulse having a tripartite structure (labelled A, B, C in Fig. 1 a). The published evidence indicates that the sound is amplified by the resonant abdomen of the cicada, which is excited by a pair of tymbals (Young, 1972a; Fletcher & Hill, 1978; Simmons & Young, 1978). The two tymbals each produce two pulses per cycle but operate 25 % out of phase, thereby generating the tripartite pulses. Thus, in Fig. 1 a, pulse part A is produced by one tymbal, part C by the other and part B by the two overlapping (Simmons & Young, 1978). There are small but consistent differences between individuals in the relative timing and amplitude of parts A, B, C of each tripartite pulse. After the final excitation in each cycle, at the onset of part C, the sound waveform shows an exponential decline in amplitude (see especially Fig. 1c)

The digitized data was subjected to a linear predictive analysis (LPA), designed to detect resonances in speech (Markel & Gray, 1976). The LPA imposes a model on the data, which is analysed in terms of any specified number of resonances. Where this number is small, a very small window size may be employed without loss of accuracy. In the present case, this enables one to distinguish changes within each of the tripartite pulses (Fig. 1). The results were similar in all ten individuals. The analysis shows that there is only a single resonance in the song and that its peak frequency changes during each pulse. A measure of the sharpness of tuning (Q) is obtained by dividing the peak frequency by the bandwidth 3 dB down (Table 1). A window by window comparison of the waveform in Fig. 1 with the corresponding LPA figures in Table 1 shows that the early portion of each pulse (parts A, B, onset of C), during which the abdomen is being excited by the tymbals, is mostly characterized by a higher peak frequency, less sharply tuned. The trailing portion of each pulse (part C) showing the exponential decay, is characterized by a lower peak frequency, more sharply tuned. This is seen particularly clearly where a pulse trails away at a break in the song (Fig. 1c). This is the expected pattern if the sound is generated by the excitation and decay of a resonant system.

A Fourier analysis of the digitized data was also undertaken to obtain a power spectrum of the calling song (Fig. 2). The number of frequencies which can be distinguished by this analysis is proportional to the number of samples and hence to the window size. In the present case, a window size of 32 ms was the smallest that would allow sufficient frequency resolution and this just includes a single pulse (see inset, Fig. 2). The results were consistent both within and between individuals.

In each case a sharp peak is seen within the range 820 to 909 Hz (⁠ = 865 ; S.D. = 23·45; n = 10). At frequencies below this peak, the roll-off is steep but at higher frequencies two or three secondary peaks are found, mostly in the range 920 to 1400 Hz. This arrangement of peaks agrees well with the LPA figures (Fig. 1 ; Table 1). The main peak (around 865 Hz) is evidently generated by the trailing portion (part C) of the tripartite pulse, while the secondary peaks are generated by the early portion (parts A, B) of the pulse. A first harmonic, on average 17·5 dB below the main peak, stands out clearly at around 1700 Hz (Fig. 2) and the lesser peaks at still higher frequencies may be interpreted as harmonics of the 950–1400 Hz peaks. The peak observed below 100 Hz is probably due to the modulation envelope, i.e. the pulse repetition frequency which will be detectable, in a window of 32 ms duration. The overall shape of the log power spectrum for C. saundersii is similar to those available from other species with relatively pure tone calling songs (Bennet-Clark, 1971 ; Nocke, 1972; Counter, 1976, 1977). However the Gryllidae and Gryllotalpidae lack the secondary peaks between the main peak and the first harmonic and so may be considered to have purer songs. The relative intensity of the first and second harmonics in C. saundersii is similar to that in the Gryllidae (Bennett-Clark, 1971 ; Counter, 1976).

Comparison within each individual shows that the LPA consistently over-estimates the peak frequency, relative to the Fourier analysis, by about 60 Hz. This is probably due to the smoothing effect which the LPA has on the data. With a sharply skewed distribution such as the power spectrum shows, smoothing would tend to shift the peak in the direction of normality, i.e. toward higher frequencies. The sharpness of tuning (Q) of the main peak, estimated from the log power spectrum, is about 15. The average value of Q from part C of each pulse in the LPA is consistently about half this value (Table 1). This difference also is probably due to the smoothing effect of the LPA, which generates wider bandwidths than the original data. The Q value of 15 for the main peak in C. saundersii is similar to that of calling songs in Gryllidae but somewhat lower than that observed in Gryllotalpidae and some Tettigoniidae (Bennet-Clark, 1971; Counter, 1976, 1977). Nocke (1971) gives higher values for Gryllidae but these are based upon resonances obtained experimentally in the soundradiating structures of the wing and not on frequency spectra of the calling songs.

No significant difference between free and tethered cicadas could be found in any of the parameters studied. Thus, from the Fourier analyses, the mean peak frequency and standard deviation for free and tethered cicadas are respectively 873 ± 18·56 Hz (n = 6) and 853 ± 27· 4 Hz (n = 4). Similarly, the first harmonic fell below the main peak by 17-8 dB (S.D. = 1·27) in free cicadas and by 17·05 (S.D = 1·6) in tethered cicadas.

In conclusion, the computer analysis shows that the calling song of C. saundersii approximates to a modulated pure tone, that the pattern of sound is entirely consistent with its being generated by the excitation and decay of a resonant system, and that tethering the cicada makes no significant difference to the physical characteristics of the song.