Edited by: John Gregory Harris, University of Florida, United States
Reviewed by: Soumyajit Mandal, Case Western Reserve University, United States; Yi-Wen Liu, National Tsing Hua University, Taiwan
*Correspondence: Jacob Isbell
This article was submitted to Neuromorphic Engineering, a section of the journal Frontiers in Neuroscience
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
The use of echolocation for navigating in dense, cluttered environments is a challenge due to the need for rapid sampling of
Bat echolocation is the unusual ability by bats to emit an ultrasonic sound pulse and measure the time until echoes begin to arrive (for estimating range) combined with the more general ability of mammals to determine the direction of sound. The ultrasonic frequencies used by bats are difficult to detect by most animals and have short wavelengths (~ 3–17 mm) that produce detectable echoes from small insects. To localize the direction of echoes, bats (e.g., the big brown bat) have been shown to rely primarily on the use of interaural level differences produced by the head and pinnae, a common strategy for small mammals (Grothe et al.,
A typical operational assumption in echolocation is that all of the sounds following an emitted pulse are echoes from the
In close-quarters maneuvering, a high sampling rate is desirable when the angle to nearby objects is changing rapidly. Little is known about what bats do when a high pulse rate is needed to maneuver near objects in an environment that produces long-delay echoes, a situation that produces echo aliasing. Big brown bats have been shown to alternate between pulsing rapidly and pulsing slowly. Pulsing rapidly gives a clearer picture for close ranges while pulsing slowly gives a clearer picture for long ranges (Petrites et al.,
The sonar system used in the work presented here consists of two custom modified MaxBotix® sonar transducers (shown in Figure
The A two transducer sonar head mounted on a hobby servomechanism that was used for the experiments in this paper. The sonar modules are a custom 40 kHz system modified from a high-power Maxbotix sonar.
The sonar system executes four repeated steps: pulsing, sampling, processing, and communicating. As part of the cycle, there is an added delay interval that is used to reject aliased echoes (discussed in section Adaptive Delay). A few of these steps are shown in Figure
An oscilloscope readout of two pulse-echo cycles (without aliasing) showing the transducer envelope voltage (
At low pulse rates, the echoes are monitored for a period of time associated with the maximum range of the sonar and an extra delay would be added after transmitting the recorded data. In the case of fast pulsing where a target is being tracked, once the target echo is received, a short data burst is transmitted and the next cycle is initiated. After detecting the target echo, the tracking window (in time) is updated and the intensities of both transducers are compared to rotate the servo motor to center the target echo. At this time, temporal windows before and after the target echo are monitored to detect if other echoes are about to overlap with the target echo. This information is used to initiate the various reactive strategies to avoid interference with target tracking (described in sections Adaptive Delay, Movement, and Beam Shaping).
There are occasions when the echo from the target disappears completely due to interference or occlusion by an object in the foreground. The tracker continues to search for the target at the same range for up to three cycles after the object disappears. If the object does not reappear, it will begin looking for a new target at a pre-specified acquisition range.
For the purposes of this study, the tracker is programmed to initially find the target at a single, pre-specified range (about 33 cm) and then follow it in range and in the horizontal plane by turning the sensor head to center the object. Centering is accomplished by rotating the sensor head until the detected amplitudes of the target in the two transducers are approximately equal. Only horizontal angles are considered. Since the echo amplitude is logarithmically compressed, the difference between left and right outputs corresponds to a ratio of the two received amplitudes. This ratio (invariant to echo amplitude) can be mapped to a specific angle. This mapping is defined by the spatial sensitivity and placement angles of the receivers and is found empirically. The ratio is monotonic and allows for reasonable angle measurements over a range of ±30°. Outside of this region, only one transducer will produce a significant response, allowing only a coarse approximation of direction. The response of our system at various angles is discussed in the Beam Shaping section.
The range of objects is determined by the time when the echo is received (i.e., time-of-flight). In practice, this is a very stable measurement that is minimally affected by noise. The echo amplitude, however, is very sensitive to factors such as the shape and orientation of the target, interfering reflections and echoes, and positioning of the transducers. At high repetition rates, a reverberant room can become filled with sound, introducing significant background interference. To avoid wild oscillations in the servo motor pointing, the system is restricted to moving a maximum step of 5° between echoes.
Once an object is found at the pre-specified acquisition range, it is labeled as the target and tracked. In the next pulse cycle, the sonar will expect to receive an echo within 6.3 cm of the previous target range. By restricting the temporal size of the tracking box, all echoes other than the target are ignored allowing the system to track a single object in the midst of other objects. Analog-to-digital sampling is performed with a period of an eighth of a millisecond and thus the range resolution is 2.1 cm/sample.
In the rapid pulse mode, the maximum detection range for the sonar system is limited by the interpulse interval. If an object has an echo time that is greater than one pulse period, it is detected by the system in the next pulse cycle. It is then perceived as having an echo time that is one pulse period less than it actually is. Since this distortion is caused by sampling related to each pulse, we call it aliasing. This is demonstrated in Figures
Aliasing visualized. In this cartoon example, each timeline has pulses (represented by tall lines) and received echoes (represented by shorter lines). Each pulse and its echoes are given a unique color. From top to bottom, the interpulse interval decreases until a new pulse occurs before all echoes from the previous pulse are received, shown in the bottom timeline. The echo is misinterpreted as a closer object associated with the latest pulse. This is the aliased echo, and is labeled with an asterisk.
Transducer envelope of pulses and echoes at different repetition rates demonstrating aliasing in the bottom graph. The outgoing pulse peaks at 0.4 V, overlapping echoes from two closely-spaced PVC pipes are seen peaking at 0.2 V, and a single loud echo made by a square poster board is seen peaking at 0.25 V. The interpulse interval is decreased in each graph until a new pulse occurs before all echoes from the previous pulse are received, causing an aliasing condition where the poster board incorrectly appears at short range.
Two strategies specific to problem of aliased echoes overlapping the target echo are presented: First, by using an
These strategies may not always work, particularly if the aliased object is close in range to the target and the sonar beam is too wide for the movement strategy to avoid illuminating the aliasing object. In this case,
The range at which the aliased echo appears is dependent on the time between sending pulses. To control this, a variable delay period is inserted before sending the next pulse. Increasing this delay shortens the aliased echo time, making it appear to move closer to the sonar. Decreasing the delay increases the aliased echo time, making it appear to move away from the sonar (an example is shown in Figure
Manipulating the received time of an aliased echo. The tall line represents the pulse and the short lines represent the echoes. The echoes associated with a given pulse are the same color. The top timeline shows an alias (white) that is close to interfering with the first dark echo, the target. The introduced delay is increased (in the bottom timeline) to shift this aliased echo away from the target in time. Similarly, an alias on the other side of the target can also be shifted away by decreasing the delay (not shown).
The alias rejection system introduces a delay interval with a maximum of 3 ms into the timeline. The interval length is changed in eighth millisecond increments based on where the aliased echo appears relative to the tracked target. If an aliased echo is within 5 range samples, or 10.7 cm, of the target echo, the delay interval will be changed to repel the aliased echo. For an aliased echo that appears closer than the target echo (i.e., in between the target and the sonar system), we increase the delay to move the aliased echo away from the target echo; an aliased echo further away than the target echo decreases the delay. If the delay reaches its maximum amount or if it is decreased to zero, the delay value is reset to 1.5 ms (half of its maximum value). This will cause an aliased echo to jump to the other side of the target echo, being shifted by 12 range samples. If there is an aliased echo detected on both sides of the target, the delay is shifted by a large amount, equivalent to 11 range samples, in an attempt to clear both aliased echoes away from the target echo. This process is summarized below.
Large delay
While we have assumed a relatively isolated target object to track, a real second object in close proximity to the target cannot be “rejected.” In this case, the alias rejection system would continuously shift the delay, resulting in oscillations of the delay shifting and resetting when the delay interval reaches its limits. To prevent these oscillations, additional code is used to recognize authentic (i.e., non-aliased) echoes.
The most notable difference between an authentic echo and an aliased echo is their reaction to a large shift in the interpulse interval, a delay jump. An alias will be moved a significant amount, while an authentic echo will not be moved at all. Although a real object can still move noticeably, at low speeds (<3 m/s) it will not jump more the one range sample at a time.
The alias rejection system makes large delay shifts in three different scenarios: when the delay interval reaches its maximum, its minimum, and when two aliases sandwich the target (one on either side). The system uses these events as triggers to look for an authentic echo that remains in the same location. This is especially appropriate since an authentic echo triggers an oscillation that causes the delay to jump when the interval reaches a maximum or minimum. If an object doesn't move after a delay jump, it is recognized as an authentic echo and will not activate the alias rejection system. This is similar to a technique used in radar where a map of stationary clutter is memorized and removed (Skolnik,
An alternative method to avoid sonar aliasing is to reposition the sonar beam such that objects in the background do not generate echoes. The effectiveness of this technique will depend on using a relatively narrow transmission beam. Depending on the species of bat, transmission beam widths can range from 22 to 90° (Jakobsen and Surlykke,
When the sonar moves around, different sides of objects are exposed to the sonar. In general, this will complicate a decision to change the sensing angle, since the acoustic properties of an object can change greatly from different perspectives. To demonstrate this, two different objects were used as the aliasing object in two different trials: a large 46 cm (1.5 ft) diameter cardboard tube and a 30 cm wide, open cardboard box. The sonar was moved around a target object to continue aiming the beam at the target at the same range, but resulting in different backgrounds (Figure
A third strategy for reducing the effect of aliasing and clutter objects is to shape the acoustic beam so that only the target object is ensonified. With the two-transducer system used in the study, this is performed by transmitting with both transducers to create an interference pattern that has peaks and nulls that can be used to reduce interference. Plots of the beam shape are shown for a single transducer, the two transducers firing synchronously, and the two transducers firing out-of-phase (Figure
Figure
The experimental configuration used is shown in Figure
Figures
Oscilloscope showing transducer voltage, delay, and tracking for an
Continuation of Figure
A target being sandwiched by aliased echoes.
Alias rejection via movement. The sonar system (speaker) is kept a constant distance from the target. The alias is located at a distance
Figure
Traces showing the echo response of the target and the alias at different angles. The target trace (blue) gives a baseline for comparison. The “Alias” traces reduce in amplitude as the angle increases. For larger distances
Polar plots of the different firing patterns. Top shows single transducer pulses from the left and right transducers. Middle shows the synchronous in-phase firing pattern. Bottom shows the synchronous out-of-phase firing pattern.
A best-case example of clutter reduction using beam shaping. Shown are two echo traces from the same scene with different beam shapes. Two objects are present, the first echo is the target object (~3.2 ms); the second echo is from the clutter object (~3.5 ms) which is circled. When in-phase firing is used, the clutter echo is greatly reduced in amplitude.
Clutter rejection using beam shaping. The sonar system faces a target that is 4 ft away. The clutter object is also 4 ft from the sonar but is rotated around the sonar system, changing its angle in the view of the sonar.
For the beam shaping study, the results are shown in Figure
These graphs show how a clutter object appears at different angles. The target and clutter objects are at the same range. Only the angle to the clutter object is changed. The top graph shows the ratio of the target and clutter amplitude. The bottom shows simulated data, where only one object was scanned across all of the angles. The ratio was computed using the echo at angle 0 (i.e., the target) and the other angles (i.e., the clutter object). The circled area shows that for angles less than 18° the synchronous firing has better clutter rejection.
The synchronous firing pattern has a higher target to clutter ratio than the left or right transducers alone. This only occurs for angles less than 18°. This is due to the side lobes of the interference pattern; once the clutter starts to enter these lobes it is no longer sufficiently rejected and a single transducer will yield a better target to clutter ratio. In between 6 and 18°, where the most benefit is seen, there is a 3.39 dB average increase in the signal to clutter ratio with the synchronous firing pattern compared to the next best single transducer.
The adaptive delay system for alias rejection tackles a problem that most engineered sonar systems avoid at the cost of a lower sampling frequency. When overlapping echolocation cycles are unavoidable, some form of pulse labeling is most commonly used (Uppala and Sahr,
The biggest limitation of the adaptive delay system is that it can only deal with a small number of aliased echoes. The case when two aliases sandwich the target is dealt with, but if three or more aliases occur in the right spots, there may be no delay time that prevents the target from being obscured.
The movement strategy is much different from the other strategies since it cannot be done on a pulse to pulse basis. Moving the sonar to improve sensing also impacts the decisions of navigation that the sensing is intended to facilitate. These results provide more information to consider by the navigation system that must balance sensing and overall task goals. The basic geometry and the angular response of the sonar system suggest that lateral movement with respect to orientation of the sonar is most effective. Another consideration is that any change in sensing angle may, in fact, generate new aliasing problems as it turns to include new background objects. Note that this approach (like the pulse timing method presented in section Adaptive Delay) will have little to no effect for clutter objects that appear at the same range as the target.
This technique is a useful way to reduce the effect of aliasing and is the only strategy presented here that is also potentially effective for objects at ranges similar to the target. It is most effective for small angles off-center. Synchronous firing creates a loud central lobe down the central axis of the sonar head. This allows for objects at longer ranges to be detected. This study did not utilize the out of phase firing primarily because the target is assumed to be held in the center of view. The out-of-phase transmission pattern has its minimum in the center of view. If a different tracking algorithm was used that kept the offending clutter in the center, this firing pattern could also be useful in rejecting clutter.
This kind of interference pattern has also been observed to be used by certain bats (Hartley and Suthers,
While these three strategies have been presented and considered separately, they can be combined into an integrated approach. Adaptive delay and beam shaping can be used simultaneously; the delay can be changed independently of the beam shape. Movements to specifically reduce aliasing can also be made, although other factors will likely affect what actions are taken.
If an alias is detected, the adaptive delay approach can be used to prevent the target from being obscured. At the same time, a movement direction can be suggested based on the apparent angle of the alias. If the obstructing echo is determined to be a real object and not an alias (part of the adaptive delay code), then different beam shapes can be used depending on the apparent angle of the obstructing echo. If the angle is less than 18°, synchronous firing will be used. If the angle is greater than 18°, only one transducer will be used. This approach is summarized in Figure
Flowchart for integrating the three strategies. Once clutter is detected, beam shaping, and the adaptive delay can be used simultaneously. If the adaptive delay determines that the object is an alias, a movement direction will be suggested.
These strategies complement each other well. Together, they present a multi-pronged approach for dealing with the interference produced while using high pulse repetition rates. Each strategy is suited to a different situation and need not be used simultaneously.
Three different active strategies for dealing with echo aliasing are described that can allow the use of sonar at high sampling rates in cluttered environments. Although a time-domain attentional system is assumed to be able to focus on a specific range to track objects, echoes from clutter objects can overlap in time, obscuring or confusing such an attentional system. At very short interpulse intervals, echoes from the background arriving after the next pulse appear to be at a shorter range then they actually are. These “aliases” can overlap the target and interfere, causing a failure of the tracking system. A dynamic pulse-timing strategy is proposed that can effectively “push” or “pull” the aliased echoes away from the tracked target echo by decreasing or increasing the interpulse interval. This prevents aliases from interfering with tracking. We have also presented a method of avoiding or reducing aliases based on positioning, as well as a method of shaping the echolocation beam to reduce the effect of aliasing or clutter.
Bats have been shown to use several different strategies when encountering cluttered situations that require fast sampling. They have been observed to change the frequency content of consecutive pulses (Hiryu et al.,
TH: created the hardware and software template for the sonar system; JI: created the sonar mount and modified the software for this project; JI: performed the data collection and analysis with advice and guidance from TH. All authors discussed the results and contributed to the final manuscript.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
1PIC18F2620. (n.d.). Retrieved March 21, 2016, Available online