The benefits of full-spectrum data for analyzing bat echolocation calls. Full-spectrum (with SonoBat call trending) and zero-crossing interpretations of the same Myotis californicus call signal in the presence of noise.
The most commonly used methods for processing bat echolocation calls, full-spectrum analysis and zero-crossing, provide different interpretations of acoustic signal content. Understanding how these two approaches extract information content from acoustic data can help users interpret results and select how and when to use each system. Chris Corben developed the zero-crossing system that became known as Anabat to brilliantly enable the analysis of bat echolocation calls in the day when computers used floppy disks and simply could not handle the enormous data load of ultrasound. The digital points needed to fully represent sound increases with higher frequencies (more vibrations per time). A few seconds of bat ultrasound recording would fill a floppy disk. Chris adapted zero-crossing to rapidly extract and distill ultrasound data down to about a thousandth of the digital points otherwise needed for full resolution and enabled widespread investigation into the realm of bat echolocation.
Field investigations of bat echolocation supported by zero-crossing have tremendously expanded our understanding of bat ecology and activity, but have also revealed the flexibility and intraspecific variety of bat calls and the challenges to readily identify bat species by their calls. While zero-crossing extracts the basic time-frequency content of a signal, full-spectrum adds the dimension of amplitude changes within bat calls, and in contrast to zero-crossing that only retains the dominant frequency at any time, full-spectrum data retains simultaneous multiple frequency content of a signal at any time to interpret the full acoustic soundscape. Indeed, the time-amplitude information from full-spectrum data does enhance species discrimination. But perhaps the greater benefit of full- spectrum data comes from the higher quality and higher resolution time- frequency analysis it provides compared with that from zero-crossing and this further enhances confident species identification. In addition, the full soundscape information from full-spectrum provides metrics for quality control essential for automated analysis.
Bat echolocation call data begins with vibrations of the bat’s larynx imparting pressure waves on the air (i.e., sound) that propagate through the air at high velocity (i.e., the speed of sound). A sensor at a fixed point will oscillate sympathetically as these waves pass. Plot of pressure oscillation in Plot of voltage oscillation in air per time measured from a detector circuitry per time fixed point. acquired from microphone.
Sound waveform The strength, or amplitude, and the period of the oscillations of the acquired electrical signal corresponds to the amplitude and frequency of the original acoustic signal. amplitude time Both zero-crossing and full-spectrum analysis begin with this same electrical signal data.
Zero-crossing analysis Divide by 8– count every 8 period oscillations Measure the time to make every 8 oscillations; this corresponds to the average frequency over the interval of those 8 oscillations. ..... . frequency Then plot those avg. frequencies per time. time
Zero-crossing analysis Period counting is independent of signal strength. Strong or weak signals with ..... . the same frequency content frequency would plot equivalently. time
Multiple frequency content and zero-crossing analysis Multiple signal sources contribute to real world soundscapes, e.g., cascading water, vehicles on a roadway, wind blowing past vegetation or structures, insects, and perhaps bats. Pressure oscillations from multiple sources interact and combine to form a single signal. This example soundscape has two high frequency sources from bats (blue and green), and a stronger lower frequency source. In the combined signal received by the microphone, the stronger lower frequency signal overpowers the higher frequency signals and controls the zero axis crossing, rendering the bat signals undetectable by zero-crossing.
Multiple frequency content and zero-crossing analysis Zero-crossing analysis can only detect the dominant, i.e., strongest, frequency content of any signal. Any other signals in the soundscape remain invisible to zero-crossing analysis. Full-spectrum analysis can access multiple frequency content to reveal the entirety of bat calls even when the signal strength of all or part of the a call falls below other signals in the soundscape. Full-spectrum analysis would reveal the bats in this signal.
Full-spectrum analysis Full-spectrum analysis by computer requires digital full- spectrum data, i.e., a digitized representation of the complete acoustic waveform. amplitude time The highest frequency resolved by a digital representation of a waveform depends upon how many digital samples of the waveform were taken per time. Because a signal oscillates once up and once down in each period, you basically need to sample at a rate of twice that of the highest frequency desired (the Nyquist frequency). That means 300,000 samples per second to resolve bat calls up to 150 kHz; much more data than required by zero-crossing– thank you to Moore’s Law for making this practical.
Full-spectrum analysis Full-spectrum analysis extracts frequency and amplitude content by sampling overlapping snippets (windows) of the waveform. amplitude window sections time Full-spectrum analysis assembles these to construct a representation of the entire soundscape by first passing the signal through bandpass frequency filters and repeating this process for each frequency band.
Full-spectrum analysis An example of a full-spectrum amplitude rendered soundscape. frequency band time A sonogram displays a flat plot of this type of data with the amplitude mapped in color…
Full-spectrum analysis A sonogram generated from full-spectrum data displays a high resolution rendering of the time, frequency, amplitude, and multiple frequency content of a signal. kHz msec Sonogram of two overlapping Leptonycteris calls.
Full-spectrum & zero-crossing analysis compared Sonogram of two overlapping Leptonycteris calls displayed beside the same signal processed by zero-crossing (Z-C) divide by 8. Note how Z-C can only track one signal at a time, and jumps between signals to whichever has the most power in a time interval.
Full-spectrum & zero-crossing analysis compared The scattered points at the beginning and end result from Z-C interpretation of lower amplitude background noise, which Z-C cannot readily discard because it cannot interpret relative amplitude of the signals that generated these points.
Full-spectrum & zero-crossing analysis compared Same signal with scattered “noise” points removed. In practice, zero-crossing users can specify the removal of extraneous points to clean up the display. Software recognizes unwanted points by distance from previous and next points in sequence.
Full-spectrum & zero-crossing analysis compared Free-tailed bat, Tadarida brasiliensis Full-spectrum processing of a call with a SonoBat generated time- frequency trend line (yellow trace). The overlain magenta points display the zero-crossing processed interpretation of the same signal. The zero-crossing processed time-frequency trend jumps off the primary call signal as the call amplitude diminishes and becomes overpowered by the amplitude of the call’s echo.
Full-spectrum & zero-crossing analysis compared With strong signals and no confounding additional signals or noise, full-spectrum time-frequency trending and zero- crossing produce similar results.
Full-spectrum & zero-crossing analysis compared Although too numerous to see individually at this scale, the full- spectrum processing of this call provided 238 points for delineating the time-frequency trend line compared with 34 in the trend from zero-crossing the same signal at divide by 8. Full-spectrum processing provides higher resolution and higher quality faithful renderings of the time-frequency domain of the calls. Full-spectrum enables full-resolution time-frequency trends.
Full-spectrum & zero-crossing analysis compared Because lower frequencies have fewer signal oscillations per time, zero- crossing will generate fewer trend points per time for lower frequency signals. The strongest part of this Euderma maculatum call only generated 5 trend points along the call’s fundamental with divide by 8. Full-spectrum analysis will process all calls from the same detector and with the same data format with equivalent resolution. Although the reduced number of waves at lower frequencies also limits resolution, the over- lapping windows of full-spectrum processing still provides more trend pts.
Full-spectrum & zero-crossing analysis compared Reducing the division ratio improves zero-crossing call resolution of low frequency signals. Divide by 4 reveals 13 trend points for this Euderma maculatum call compared with 5 at divide by 8.
Recommend
More recommend
