Distance Detection

The goal of the distance detector is to produce highly accurate distance measurements while maintaining low power consumption by combining the features of the A121 sensor with powerful signal processing concepts, all wrapped with a simple to use API.

The full functionality can be explored in the Exploration Tool. Once the desired performance is achieved, the configuration can be carried over to the embedded version of the algorithm, available in the C-SDK.

Introduction

The purpose of the distance detector is to detect objects and estimate their distance from the sensor. The algorithm is built on top of the Sparse IQ service and has various configuration parameters available to tailor the detector to specific use cases. The detector utilizes the following key concepts:

1. Distance filtering: A matched filter is applied along the distance dimension to improve the signal quality and suppress noise.

2. Subsweeps: The measured range is split into multiple subsweeps, each configured to maintain SNR throughout the sweep while minimizing power consumption.

3. Comparing sweep to a threshold: Peaks in the filtered sweep are identified by comparison to one of three available threshold methods.

4. Estimate distance to object: Estimate the distance to the target by interpolation of the peak and neighboring amplitudes.

5. Sort found peaks: If multiple peaks are found in a sweep, three different sorting methods can be employed, each suitable for different use-cases.

Distance Filter

As the sensor produce coherent data, samples corresponding to the location of an object will have similar phase, while the phase of free-air measurements will be random. By applying a filter in the distance domain, the noise in the free-air regions will be suppressed, resulting in an improved SNR.

The filter is automatically configured based on the detector configuration as a second order Butterworth filter with a cutoff frequency corresponding to a matched filter.

Subsweeps

The measurement range is split up into multiple subsweeps to allow for optimization of power consumption and signal quality. The profile, HWAAS and step length are automatically assigned per subsweep, based on the detector config.

• A shorter profile is selected at the start of the measurement range to minimize the interference with direct leakage, followed by longer profiles to gain SNR. The longest profile used can be limited by setting the parameter max_profile. If no profile is specified, the subsweeps will be configured to transfer to the longest profile (without interference from direct leakage) as quickly as possible to maximize SNR. Longer profiles yield a higher SNR at a given power consumption level, while shorter profiles gives better depth resolution.

• The step length can also be limited by setting the parameter max_step_length. If no value is supplied, the step length is automatically configured to appropriate size, maintaining good depth resolution while minimizing power consumption. Note, the algorithm interpolates between the measured points to maintain good resolution, even with a more coarse step length.

• HWAAS is assigned to each subsweep in order to maintain SNR throughout the measured range as the signal strength decrease with the distance between the sensor and the measured target. The target SNR level is adjusted using the parameter signal_quality.

  Note, higher signal quality will increase power consumption and measurement time.

  The expected reflector shape is considered when assigning HWAAS to the subsweeps. For planar reflectors, such as fluid surfaces, select PLANAR. For all other reflectors, select GENERIC.

In the Exploration Tool GUI, the subsweeps can be seen as slightly overlapping lines. If the measured object is in the overlapping region, the result from the neighboring segments is averaged together.

Thresholds

To determine if any objects are present, the sweep is compared to a threshold. A peak is defined as a middle point that has greater amplitude than its two neighbouring points. For an object to be detected, it has to yield a peak where all three points are above the threshold. Three different thresholds can be employed, each suitable for different use-cases.

Fixed amplitude threshold

The simplest approach to setting the threshold is choosing a fixed threshold over the full range. The amplitude value is set through the parameter fixed_threshold_value. The fixed amplitude threshold does not have any temperature compensation built in.

Fixed strength threshold

This threshold takes a fixed strength value and converts to the corresponding amplitude value. The purpose is to produce a threshold that is able to detect an object of a with a specific reflectiveness, independent of the distance to the object. The strength value is set through the parameter fixed_strength_threshold_value. The fixed strength threshold does not have any temperature compensation built in.

Recorded threshold

In situations where stationary objects are present, the background signal is not flat. To isolate objects of interest, the threshold is based on measurements of the static environment. The first step is to collect multiple sweeps, from which the mean sweep and standard deviation is calculated. Secondly, the threshold is formed by adding a number of standard deviations (the number is determined by the parameter threshold_sensitivity) to the mean sweep. The recorded threshold has a built in temperature compensation, based on the internal temperature sensor.

Constant False Alarm Rate (CFAR) threshold (default)

A final method to construct a threshold for a certain distance is to use the signal from neighbouring distances of the same sweep. This requires that the object gives rise to a single strong peak, such as a fluid surface and not, for example, the level in a large waste container. The main advantage is that the memory consumption is minimal. The sensitivity of the threshold is controlled through threshold_sensitivity. As the CFAR threshold is formed based on each momentary sweep, any temperature effects on the signal are implicitly accounted for by the algorithm.

Reflector Shape

The expected reflector shape is considered when assigning HWAAS to the subsweeps and during peak sorting.

The reflector shape is set through the detector configuration parameter reflector_shape.

For a planar reflector, such as a fluid surface, select PLANAR. For all other reflectors, select GENERIC.

Reflector Strength

The reflector strength characterize the reflectiveness of the detected object. The detector reports a strength number for each estimated distance.

The strength is estimated using the RLG equation, peak amplitude, noise floor estimate and the sensor base RLG. More information on the RLG equation and base RLG can be found here.

The estimated strength is used by the detector when sorting the estimated distances according to their relative strengths. It can also be used by the application to infer information about a certain distance estimate. For example, a highly reflective object such as a metal surface will typically have a higher strength number than a less reflective surface such as a wooden structure.

Ideally, the strength estimate is agnostic to the distance of the object. However, due to close range effects, the strength tends to be under estimated at short distances (< 1m).

The strength is reported in dB.

Peak Sorting

Multiple objects in the scene will give rise to several peaks. Peak sorting allows selection of which peak is of highest importance.

The peak sorting strategy is set through PeakSortingMethod, which is part of the detector configuration.

The following peak sorting options are available.

Closest

This method sorts the peaks according to distance from the sensor.

Strongest (default)

This method sorts the peaks according to their relative strength.

Note, the reflector shape is considered when calculating each peak’s strength. The reflector shape is selected through detector configuration parameter reflector_shape.

Note, regardless of selected peak sorting strategy, all peaks and the corresponding strengths are returned by the distance detector.

Detector Calibration

For optimal performance, the detector performs a number of calibration steps. The following section outlines the purpose and process of each step. Note, which of the following calibration procedures to perform is determined by the user provided detector config. For instance, the close range measurement is only performed when measuring close to the sensor.

To trigger the calibration process in the Exploration Tool gui, simply press the button labeled “Calibrate detector”. If you are running the detector from a script, the calibration is performed by calling the method calibrate_detector.

Noise level estimation

The noise level is estimated by disabling of the transmitting antenna and just sample the background noise with the receiving antenna. The estimate is used by the algorithm for various purposes when forming thresholds and estimating strengths.

Offset compensation

The purpose of the offset compensation is to improve the distance trueness (average error) of the distance detector. The compensation utilize the loopback measurement, where the pulse is measured electronically on the chip, without transmitting it into the air. The location of the peak amplitude is correlated with the distance error and used to correct the distance raw estimate.

Close range measurement calibration

Measuring the distance to objects close to the sensor is challenging due to the presence of strong direct leakage. Direct leakage is the static component of the measured signal, visible for the first couple of centimeters, resulting from reflections from components close to the sensor such as lens and PCB, as well as the energy propagating directly from Tx to Rx. One way to get around this is to characterize the leakage component and then subtract it from each measurement to isolate the signal component. This is exactly what the close range calibration does. While performing the calibration, it is important that the sensor is installed in its intended geometry and that there is no object in front of the sensor as this would interfere with the direct leakage.

The calibration is only performed if the logic is enabled through the parameter close_range_leakage_cancellation and the start_m is set to a value lower that ~20 cm when using CFAR threshold and ~11 cm for the other thresholds. The reason for CFAR requiring a greater distance is to initialize the threshold with data, free from direct leakage.

The close range measurement calibration is only valid in the range of +-15 °C from where it was calibrated.

Recorded threshold

The recorded threshold is also recorded as a part of the detector calibration. Note, this calibration is only performed if the detector is configured to used recorded threshold or if close range measurement is active, where recorded threshold is used.

Detector Calibration Update

To maintain optimal performance, the sensor should be recalibrated if calibration_needed is set to True. A sensor calibration should be followed by a detector calibration update, performed by calling update_detector_calibration.

The detector calibration update carries out a subset of the calibration steps. All the calibration steps performed are agnostic to its surroundings and can be done at any time without considerations to the environment.

Temperature Compensation (Recorded Threshold)

The surrounding temperature impacts the amplitude of the measured signal and noise. To compensate for these effects, the recorded threshold has a built in compensation model, based on a temperature measurement, internal to the sensor. Note, the effectiveness of the compensation is limited when measuring in the close range region.

The CFAR threshold exhibits an indirect temperature compensation as the threshold is formed based on the sweep itself. As the sweep changes with temperature, so does the threshold accordingly.

The fixed thresholds (amplitude and strength) does not have any temperature compensation.

Result

The result returned by the distance detector is contained in the class DetectorResult.

The two main components of the distance detector result are the estimated distances and their corresponding estimated reflective strengths. The distances and the corresponding strengths are sorted according to the selected peak sorting strategy.

In addition to the distances and strengths, the result also contains the boolean near_edge_status. It indicates if an object is located close to start of the measurement range, but not resulting in a clear peak, but rather the tail of an envelope. The purpose of the boolean is to provide information in the case when an object is present, just outside of the measurement range. One example of when this becomes useful is the Tank reference application, which is built on top of the distance detector. If the tank is overflowing, the peak might end up just outside of the measured interval, but the tail end of the envelope would still be observable.

The result also contains the boolean calibration_needed. If True, the procedure, described in the section Detector Calibration, needs to be performed to maintain optimal performance.

Note, the sweep and threshold, presented in the distance detector GUI are not returned by the distance detector. These entities are processed and evaluated internally to the algorithm. The purpose of visualizing them in the GUI is to guide in the process of determining the detector configuration, such as selection of threshold strategy and sensitivity.

Hints and Recommendations

The purpose of this section is to provide information on how to configure the distance detector, as well as some practical aspects of the algorithm and overall application.

Configuration Hints

The following section contains hints and recommendations on how to configure the distance detector.

Several of the described parameters affect the sensor configuration and memory utilization. For a quantitative estimate on these numbers, please consult the Resource calculator, available in the Exploration Tool.

The distance detector has two predefined configurations, available in the application as presets, with the following design philosophies:

Balanced

A trade-off between SNR, radial resolution and power consumption. Here, a larger step length is used, reducing the number of measured data points. Also, the signal quality is set to a more moderate value, resulting in a lower HWAAS. Both aspects yield shorter measurement and lower power consumption. Lastly, a higher max profile is used, providing higher SNR per measured per measurement instance.

High accuracy

Optimized for better radial resolution and SNR, with a penalty on power consumption. Here, a lower step length is used, providing more data points to be processed for the distance filter, increasing the SNR through processing. Also, the signal quality is increased, resulting in more HWAAS. Lastly, a shorter max profile is used, providing better radial resolution.

These presets should be viewed as a starting point, from where a more tailored configuration can be developed.

The following points provide insight into the configuration process.

• Set start_m and end_m to the desired measurement interval.

• Measuring close to the sensor (sub ~6cm) requires close_range_leakage_cancellation to be enabled. This will trigger the close range calibration method. The calibration procedure requires a known environment and is valid in a temperature range of +-15 °C from the temperature where it was executed. For more details, see the section Close range measurement calibration under Detector Calibration.

  Due to these restrictions, it is advised to only use this mode when the use case allows for calibration in a known environment, and the possibility to redo the calibration when the temperature has changed more than 15 °C, indicated by the variable calibration_needed.

  If close_range_leakage_cancellation is disabled, the application will not perform the close range leakage cancellation. Measuring close to the sensor can result in artifacts from the direct leakage being visible as peaks in the sweep.

• The step length and profile are both automatically selected to yield a good trade-off between SNR and power consumption. The SNR can be improved by reducing step length through the parameter max_step_length, with a penalty on power consumption. The radial resolution can be increased by limiting the max profile used through the parameter max_profile, with a penalty on SNR.

• The reflector_shape should be set to PLANAR when measuring a planar surface. In all other cases, it should be set to GENERIC.

• Peak sorting determines the sort order of the detected objects. Whether to use CLOSEST or STRONGEST depends on the use case.

  Note, regardless of the selected peak sorting method, all detected distances are returned by the application.

• There are four threshold methods available. Which one to use is use case dependent. More information can be found under the section Thresholds.

  • CFAR - Suitable when the use case involve clear peaks such as a level measurement application. The method is robust over temperatures and does not required any consideration to the surroundings when calibrating.

  • FIXED_STRENGTH - Applies a threshold to the estimated strengths. This threshold is suitable when estimating the distance to a strong reflector in a cluttered environment.

  • FIXED - Applies a threshold to the sweep amplitude. This threshold detects objects based on their measured amplitudes.

    Note, for a given object, the amplitude reduce with distance as less energy is reflected back to the sensor, resulting in missed detections.

  • RECORDED - This threshold records the background clutter and is thereafter applied to the sweep as a threshold. The threshold is suitable when the environment consists of a several reflecting objects that should not be detected (clutter).

    The threshold has a built in temperature compensation, based internal temperature sensor, adjusting the threshold to keep a constant false positive rate.

    Note, the threshold is only valid as long as the background is static. A change in the clutter can result in undesired objects being detected.

• Generally, the CFAR or RECORDED are preferred when the ambient temperature is expected to change. The FIXED_STRENGTH and FIXED are fixed and has no temperature compensations built in.

• threshold_sensitivity controls the false positive rate for the CFAR and recorded threshold. The parameters should be tuned for each use case to achieve the desirable performance.

• signal_quality should be set so that desirable detection rate is achieved. A higher value corresponds to higher HWAAS and SNR, but also higher power consumption.

Use Case Scenarios

This section outlines how to use the distance detector in common scenarios.

Measuring close to the sensor

The energy propagating directly from the transmitting to the receiving antenna is referred to as the direct leakage. The direct leakage component is typically stronger than the component reflected of the object of interest, resulting in no clear peak from the object being visible in the sweep. This becomes an issue when measuring closer than ~6 cm from the sensor.

One way of alleviating this issue is to use Close range measurement calibration, described under Detector Calibration.

As stated, this mode comes with some limitations on temperature range and requirements on calibration environment. Both these aspects needs to be considered before the logic is enabled.

Measuring far from the sensor

As the distance between the sensor and a given object increase, the amount of energy reflected back to the sensor decrease. This makes it harder to detect objects at greater distances.

To maximize detection rate, the max_step_length can be reduced, signal_quality increased and max_profile set to the highest profile (profile 5). The first two parameters will affect the power consumption.

Multiple objects in the scene

Since the A121 sensor is a single channel sensor (one Tx and one Rx), multiple objects at the same radial distance will be reported as a single object.

If multiple objects are present at different radial distances from the sensor, with a reasonable separation, they will be reported individually.

The distance detector returns all the detected distances and their corresponding strengths. The result can thereafter be post-processed by the application, for instance based on the distances or strengths, to produce the desired result.

One example where multiple objects will be detected is when measuring the distance to an oil surface in a metal tank. As oil is typically somewhat transparent to 60GHz, some energy will be reflected of the surface, while some energy will travel through the oil and be reflected of the bottom of the tank, resulting in two peaks at two different distances. The strength of the oil peak can be significantly lower than the strength of metal bottom peak. The estimated strengths can therefore be used to determine which estimated distance corresponds to the oil surface.

Practical Considerations

The following section highlights aspects outside of the distance detector, contributing to the overall performance.

Using a lens

A plastic lens can be used to shape the radiation pattern, focusing the emitted power in the desired direction and reduce side lobes.

Focusing the energy in the desired direction will increase the SNR and improve the detection at greater distances.

The lens is typically made out of plastic and can in many cases be incorporated in the cover of the plastic casing.

Post-processing

The distance detector returns all detected distances and their corresponding strengths. Post-processing the result in the application can help determining the relevant distance.

Below are a few possible post-processing concepts outlined.

Strength

The strength of the measured target can be characterized as a part of the development process and then used to filter out the relevant object.

For instance, the strength value of a water surface can be characterized and then used to identify it in an environment with other reflectors such as a concrete structure in sewer level application.

Distance

The distances can be processed to identify the relevant distance.

For instance, in a water tank level measurement application, the greatest distance can be selected as the water level since it is known that the water is not transparent to 60GHz and therefor no objects will be detected below the water surface.

Distance variation

Looking at the variation over several distance measurements can help identifying the distance to a dynamic target, such as a stream of water.

For instance, in a sewer application, where the sensor is mounted at the top of the manhole, looking down towards the water. In such an environment, it is common to have several reflectors, cluttering the scene. By looking at the variation of estimated distances, it is possible to determine which distance corresponds to a stream of moving water.

Configuration Parameters

class acconeer.exptool.a121.algo.distance.DetectorConfig(*, start_m: float = 0.25, end_m: float = 3.0, max_step_length: int | None = None, max_profile=Profile.PROFILE_5, close_range_leakage_cancellation: bool = False, signal_quality: float = 15.0, threshold_method=ThresholdMethod.CFAR, peaksorting_method=PeakSortingMethod.STRONGEST, reflector_shape=ReflectorShape.GENERIC, num_frames_in_recorded_threshold: int = 100, fixed_threshold_value: float = 100.0, fixed_strength_threshold_value: float = 0.0, threshold_sensitivity: float = 0.5, update_rate: float | None = 50.0)

start_m: float

Start of measurement range in meters.

end_m: float

End of measurement range in meters.

max_step_length: int | None

If set, limits the step length.

If no argument is provided, the step length is automatically calculated based on the profile.

Reducing the step length increases SNR through more efficient distance filtering, while increasing the measurement time and the processing load.

max_profile: Profile

Specifies the longest allowed profile.

If no argument is provided, the highest possible profile without interference of direct leakage is used to maximize SNR.

A lower profile improves the radial resolution.

close_range_leakage_cancellation: bool

Enable close range leakage cancellation logic.

Close range leakage cancellation refers to the process of measuring close to the sensor(<100mm) by first characterizing the direct leakage, and then subtracting it from the measured sweep in order to isolate the signal component of interest.

The close range leakage cancellation process requires the sensor to be installed in its intended geometry with free space in front of the sensor during detector calibration.

signal_quality: float

Signal quality (dB).

High quality equals higher HWAAS and better SNR but increases power consumption.

threshold_method: ThresholdMethod

Threshold method

peaksorting_method: PeakSortingMethod

Sorting method of estimated distances.

The distance estimates are sorted according to the selected strategy, before being return by th application.

reflector_shape: ReflectorShape

Reflector shape.

num_frames_in_recorded_threshold: int

Number of frames used when calibrating threshold.

A lower number reduce calibration time and a higher number results in a more statistically significant threshold.

fixed_threshold_value: float

Value of fixed amplitude threshold.

fixed_strength_threshold_value: float

Value of fixed strength threshold.

threshold_sensitivity: float

Sensitivity of threshold.

High sensitivity equals low detection threshold, low sensitivity equals high detection threshold.

update_rate: float | None

Sets the detector update rate.

class acconeer.exptool.a121.algo.distance.ThresholdMethod(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Threshold methods. CFAR Constant False Alarm Rate. FIXED_AMPLITUDE Fixed amplitude threshold. FIXED_STRENGTH Fixed strength threshold. RECORDED Recorded threshold.

CFAR = 1

FIXED = 2

FIXED_STRENGTH = 3

RECORDED = 4

class acconeer.exptool.a121.algo.distance.PeakSortingMethod(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Peak sorting methods. CLOSEST sort according to distance. STRONGEST sort according to strongest reflector.

CLOSEST = 1

STRONGEST = 2

class acconeer.exptool.a121.algo.distance.ReflectorShape(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Reflector shape.

GENERIC Reflectors of any shape. PLANAR Planar shaped reflectors facing the radar, for example water surfaces.

GENERIC = 4

PLANAR = 2

property exponent: float

Detector Calibration

class acconeer.exptool.a121.algo.distance._detector.Detector.calibrate_detector(self)

Run the required detector calibration routines, based on the detector config.

class acconeer.exptool.a121.algo.distance._detector.Detector.update_detector_calibration(self)

Do a detector calibration update by running a subset of the calibration routines.

Once the detector is calibrated, by calling calibrate_detector(), a sensor calibration should be followed by a detector calibration update.

Detector Result

class acconeer.exptool.a121.algo.distance._detector.DetectorResult(*, distances: ndarray[Any, dtype[float64]] | None = None, strengths: ndarray[Any, dtype[float64]] | None = None, near_edge_status: bool | None = None, calibration_needed: bool | None = None, temperature: int | None = None, processor_results: List[ProcessorResult], service_extended_result: List[Dict[int, Result]])

distances: ndarray[Any, dtype[float64]] | None

Estimated distances (m), sorted according to the selected peak sorting strategy.

strengths: ndarray[Any, dtype[float64]] | None

Estimated reflector strengths (dB) corresponding to the peak amplitude of the estimated distances.

near_edge_status: bool | None

Boolean indicating an object close to the start edge, located outside of the measurement range.

calibration_needed: bool | None

Indication of calibration needed. The sensor calibration needs to be redone if this indication is set.

A sensor calibration should be followed by a detector calibration update, by calling update_detector_calibration().

temperature: int | None

Temperature in sensor during measurement (in degree Celsius). Notice that this has poor absolute accuracy.

processor_results: List[ProcessorResult]

Processing result. Used for visualization in Exploration Tool.

service_extended_result: List[Dict[int, Result]]

Service extended result. Used for visualization in Exploration Tool.