Calibration Performed by Jesse Barber, STELLA Calibration and Validation Lead

### Acquisition Process and Best Practices

The laboratory calibration set-up for the STELLA instrument uses the integrating sphere of the NASA 618 Field optics Cal Lab and a Spectralon panel with reflectance of 99% in the spectral range we are interested in. The light from the sphere was directed on to the spectralon panel, placed 18.5 inches from the source, and reflected onto the STELLA detectors approximately 12 inches away at an angle of approximately 25 degrees, see photos for more information.

The STELLA was placed at this location to minimize the angle above, but not shadow the spectralon with the body of the STELLA instrument.

The angle of the spectralon panel was eyeballed to split the difference between the source and the detector. This was done to account for the large field of view (40 degrees) of the STELLA instrument, but also minimize the lighting gradient from one side of the panel to the other. If setting this up yourself, or with a different detector, etc. It is far more important to ensure that the lighting gradient on your reflective surface is negligible; in other words, prioritize lighting the panel evenly by keeping it directly facing your source. When putting together a set-up such as this and placing your detector somewhere between your source and reflector, it is very important that your detector not be collecting from a shadowed part of the reflector.

The sphere was left on for a minimum of 30 min to ensure the lamps had warmed up sufficiently for each data acquisition.

The vast majority of data was acquired in the averaging mode, where the unit saves a number of spectra and then averages them together. In this calibration process we used an average of 20 spectra during each data collect session for each unit and each reflector. All this data is saved and transferred to the sub sheets in the excel document.

When convenient, I performed the same measurements with my laboratory working standard detector, an ASD Fieldspec4. The data for which has not yet been matched to that of the STELLAs.

### Data Processing

The goal of this calibration is currently to verify the repeatability and consistency of a suite of STELLAs against a stable source over a few months timeframe. Future data analysis will aim to determine how accurate the irradiance calibration is by cross comparison to an ASD Fieldspec4.

Table A is a collection of all of the average data from each day for each STELLA.

Table A-1 takes each STELLA and averages them over all the data collected for each, as well as a total average across all data for all STELLAs.

Table B is the table of standard deviation. The standard deviation of the bulk of Table B is calculated by using the 20 datapoints from each data collect. Giving a standard deviation for each data collect.

Table B-1 is the standard deviation calculated by taking into account the complete set of data points taken by all STELLAs.

Table B-2 uses the complete set of data taken by each STELLA to calculate a total Standard Deviation for each STELLA.

Beyond Table B we have a series of tables to explore the various standard deviations divided by various averages. This gives us a coefficient of variance for each wavelength for various cases.

% Average Standard Deviation Per STELLA /Total Average for each STELLA is a measure of the variability across a single STELLA in relation to only itself.

% Average Standard Deviation Per STELLA / Total Ensemble Average is a measure of the variability across a single STELLA in relation to the entire ensemble. If my understanding of the STELLA ensemble is correct, this is the least useful measure.

Finally, % Ensemble Standard Deviation / Total Ensemble Average is a measure of the variance across the entire ensemble of STELLA.

There are two graphs in the document, depicting the average irradiance response read by each STELLA, along with the average of the whole ensemble and the standard deviation of that whole ensemble. These two graphs are depicting the visible bands and the infrared bands respectively.

The X axis is described as “Band”, with a band legend below the graphs. Because there are some idiosyncrasies with Excel, and overlap between the graphs, I decided to go with this “band” way of depiction and split visible and infrared into two graphs.

Visible Band |
Visible Peak Wavelength |

1 | 450 nm |

2 | 500 nm |

3 | 550 nm |

4 | 570 nm |

5 | 600 nm |

6 | 650 nm |

Infrared Band |
Infrared Peak Wavelength |

1 | 610 nm |

2 | 680 nm |

3 | 730 nm |

4 | 760 nm |

5 | 810 nm |

6 | 860 nm |

### Concerns, Problems, Caveats, and Future Work

There are a few concerns, struggles, mistakes, and points that I would like to bring up here for anyone looking at this data in the future.

-The first and most obvious thing is that this data was collected using the first generation of STELLA with AS7262 and AS7263 sensors. We have since moved on to STELLA 1.2 and 2.0 with plans moving fast toward a STELLA 3.0 with a totally new sensor. Because of this, this data collected may have a limited life in the STELLA project, but should be applicable to any other devices that use this sensor.

The other problems that arose from using the first batch of STELLAs are that some of the original number succumbed to corrosion due to poor quality solder and were unable to continue data collection. So there are some UIDs in the early data that are not continued in the later data. We have since identified and dealt with this problem. The other problem, which is more of an annoyance for the data scientist, is that the data output structure has changed over the months that the calibration took place, due primarily to input that I and another teammate had. The data structure currently has no plans to be changed in the future, but if one wants to go back to look at the old data, they must keep in mind that the structure is subtly different.

-While data was collected on May 3rd, due to an oversight and over reliance on muscle memory, instead of the full 4 lamp brightness that is used in nearly all data collects, I only used 2 lamps. This doesn’t make the data bad per se, but it precludes it from joining the rest of the data analysis. I have still included it for completeness, but it the source is approximately half the brightness that it should be.

-In the current software/firmware package in averaging mode, the STELLA calculates both the average and standard deviation of the set that was just taken. However, because these are original 1.0 STELLAs, only one has the cache space to calculate the standard deviation. I do not know why this one has just that little bit more memory to be able to calculate it while the others can’t. The new STELLAs do not have this problem as they have more memory to work with.

-There is other data from our Lab’s ASD Fieldspec4 that has taken hyperspectral radiance data from the same viewpoint as the STELLAs have. This data needs to be convoluted down to match the bands of the STELLA. At the moment, I do not have much time to do this, and the attempt that I made does not match what was expected. I think this is because I neglected to use a discretized version of the integral. We will see if this is the reason when I make sure the calculation process is correct. If not, we will be releasing calibration coefficients of the ensemble average for the STELLA to bring the data to truth.