Weight Measurement
Any bird weighing event is recorded and stored to memory as a new log entry. In addition, continuous offset measurements and calibration data are logged. The most important logs are also transmitted wirelessly via SIGFOX (if coverage is available).
Bird Weight
Anisca Bird continuously measures the weigth present on the wooden perch (“Detection”). It does so in a low-power fashion until the measured weight is above WAKE_UP_THRESHOLD. The device then attempts to read the RFID tag of a potential bird present. If successful, the weight is measured continuously (“Acquisition”) for as long as it stays above WAKE_UP_THRESHOLD.
Detection
During detection, the following hard-coded constants are used:
Low-power detection interval: 1 second
WAKE_UP_THRESHOLD: 50 grams 1
RFID read attempts: max. 10x in 3 second intervals. If unsuccessful after this, system goes back to sleep.
Acquisition
After a bird was detected and a valid RFID was read out, its weight is sampled continuously. Multiple samples are taken into account because the bird may be moving and hence an instantaneous measurement would not provide a reliable weight value.
The final raw weight value is obtained as a result of 3 acquisition steps:
Sub-samples: The load cell ADC is sampling the weight at 10 Hz.
Samples: Always 10 sub-samples are averaged to create one sample. A standard deviation on the 10 sub-samples is calculated.
If the sub-sample standard deviation for a given group of 10 exceeds MAX_SUB_STDEV, the sample gets discarded. This ensures only samples where the bird was sitting still will be considered.
If a sample is MAX_SAMPLE_DIFF larger than the previous sample, then the previous sample gets discarded. This ensures all samples to be on a relatifely flat line and removes outliers.
Median: A maximum of N_SAMPLES is collected as described in points 1 & 2. If the bird leaves before reaching N_SAMPLES, a smaller number of samples will be available.
The median value over all collected samples is calculated. This is the retained raw weight value.
The standard deviation over all collected samples is calculated. This is the retained raw stddev value.
Hard-coded parameters:
Processing
The raw weight and raw stddev values obtained above are unitless ADC values. They need to be converted to grams using calibration data in order to present meaningful weight data. This is done as follows:
The last recorded raw offset value is subtracted from the raw weight, in order to receive the net raw weight
net raw weight and raw stddev are divided by the most recently measured calibration slope to obtain weight and standard deviation of the sample in grams.
Logging
For more details on what data is logged or transmitted, see: Weight.
Offset measurement
The raw offset is measured in the exact same way as the bird weight, except that it happens in a periodic manner whenever the weight detection stays below the WAKE_UP_THRESHOLD.
The following parameters are hard-coded:
N_OFFSET_SAMPLES: 11
Offset acquisition & logging period: 3 minutes
Offset uplink period (transmitted via SIGFOX): 1-2 hours
Calibration
The calibration is a linear regression between raw ADC values and the weight placed on the perch in grams. We denote the following terms:
Slope (a): the multiplication factor between grams and raw ADC value
Intercept (b) the calculated raw ADC value when zero weight is placed on the perch. Theoretically, this is the offset value during calibration.
R^2 the R-square of the linear regression.
When performing a calibration routine, both a and b are calculated automatically by the device and stored for internal use. All subsequent weight measurements are processed using these most up-to-date calibraion values.
It is known that the calibration values can change over time as the physical state of the sytstem changes. Therefore, frequent re-calibrations of the system are advised.
Temperature Dependency
It is known that notably the calibration slope is temperature dependent. This dependency is not compensated for by the device. It is adviseed to perform a temperature compensation of the logged weight values once the relationship between temperature and slope has been established in a post-processing step.
Footnotes