Intelligent Services
Rand Dannenberg, Ph.D. (Owner/CEO)
Materials Science & Engineering
Optics
Physics
Artificial Intelligence / Machine Learning
Rand Dannenberg, Ph.D. (Owner/CEO)
Materials Science & Engineering
Optics
Physics
Artificial Intelligence / Machine Learning
Consulting Philosophy
"To provide technical results to my clients in a timely manner, and help them get answers. To get at the truth for my clients, undistorted by belief or salesmanship, using the scientific method at its purest, and doing so in the most informed state possible, while seeking the scrutiny and criticism of the informed."
Overview
Optics for Space & Astronomy Applications, Virtual Reality / Augmented Reality, Stray Light Simulations, Sensor Performance Simulation, Autonomous Vehicle Sensors
RaNDTek LLC is a consulting company, located near Los Angeles.
An initial consultation is pro bono (free). The most meaningful things I have worked on began this way.
I support everything from fundamental physics, to manufacturing management, to business development.
I am a U.S. Citizen, and I employ other contractors and experts who are also, when needed.
Services & Qualifications
I help solve engineering and manufacturing problems.
I help with business development plans in the SoCal/Los Angeles area.
I offer consulting services in the areas of materials engineering, optics, and applied physics, using simulations, custom codes, optical design, data reduction, materials analysis, and literature summary.
The problems I have been given have ranged from the "completely standard", where someone skilled was just needed to do the work, to the "difficult and arcane", where no solution, explanation, or approach had existed, prior.
I help capture production processes by generating documentation within quality systems, then provide training to staff.
I assist attorneys and companies with patent applications, claims, and the many related specifics of the process.
I have assisted companies in product liability lawsuits as an independent analyst, finding the root cause.
Specific Skills & Background
Astronomy, Physics, Materials Science, Optics, Nanotechnology, Multi-Physics Simulation, Focal Plane Arrays, Rugates, Programming, Ferroelectric-Pyroelectric and Bolometric Sensors, Optical Coatings, Optical Materials, Optical Thin Films, Electro-Optics, Fiber Optics, R&D and Project Supervision, Marketing, Sales, Nanotechnology, Product Liability Advisement, Technical Literature Survey, Electrochemistry, Surface Modification.
I use a number of modeling and simulation codes.
Artificial Intelligence, Neural Networks, Deep-Learning, Computer Vision:
Keras
Tensorflow
Theanos
PyTorch
Molecular Modeling & Electronic Structure Calculation using Density Functional Theory:
CASTEP
Gaussian
dmol3
Computational Mechanics Modeling:
DEFORM
Simufact
Geomagic
iSight
Some Ansys
Data manipulation and specialized calculations:
MATLAB
Python
Visual Basic
C++
Mathematica
MathCad
Use of Large Language Models for the development of custom code and programs.
I have become an expert at the prompting of LLM's to produce small codes and functions for calculation, data analysis, plotting, and condensation.
Optical system design and analysis and Stray Light:
Zemax
FRED
Code V
LightTools
Coherent Beam Propagation:
RSoft
Lumerical FDTD
LightTrans
Comsol Multiphysics/RF
Thin film filters and coatings:
TF Calc
Essential Macloed
Filmstar
the ellipsometric code W-VASE
SBIR Proposals and Proposal Assistance
I have written, won, and program managed many DoD, NASA, and several DoE SBIR proposals.
I can offer a service to assist you with your proposals, with technical writing, literature, or discussions with TPOC's.
Fundamental Physics Publications
I enjoy thinking about and publishing results on problems relevant to the foundations of modern physics. Links to recent papers are shown below:
Software Development for Astronomy & Astrophysics
I developed custom software for analysis of absorption spectral data from the Hubble Space Telescope (HST) Space Telescope Imaging Spectrograph (STIS):
https://www.stsci.edu/hst/instrumentation/stis
This work is was done in conjunction with Prof. John Webb at the University of Cambridge in the UK, and the University of New South Wales in Sydney. Dr. Webb studies variations in the fine structure constant near white dwarf stars, and cosmological variations with data from quasars.
The software, being developed in Python, takes the raw spectrum (green), automatically detects the absorption lines (magenta), then determines the lower frequency continuum spectrum from the absorption of the photosphere of the star (red). This allows much more accurate determination of the wavelength and species of the absorption lines.
This process is automatic and takes less than one second to execute, much faster, and less subjective than the manual selection and fitting methods using the older, classic astronomy code, Iraf, which takes many days.
Software Development for Astronomy & Aerospace
There are many applications in astronomy, and astronomy-related observing, in which a sequence of images must be de-noised, either in real-time, or during analysis after collection. The ability to do this well, and extract better quantitative information, will have a meaningful impact on fundamental science, and defense related applications.
I developed RaNDTek LLC proprietary Machine Learning / AI-based algorithms and image-processing-data-reduction techniques for space applications, and earth-based astronomy. The marriage between the intended applications and AI has never been explored. The prototype enviroment is Python, with the Deep-Learning Libraries Theano, TensorFlow, Keras, and PyTorch. These algorithms "learn the noise", and can do so dynamically in operation for corrected signal or image extraction.
The strategy is to be backwards-compatible with all existing hardware, resulting in enhanced performance with no more than a software modification, yet agile enough to be able to apply the principles to new observing hardware with additional features.
The most basic description of AI, Machine Learning and Neural Networks, is that they are fitting, filtering, or interpolation functions to data, given some number of inputs.
The most basic question in getting a well-conditioned AI, is:
"What function, how many parameters, what training data, and how much, is best for the application?"
The guiding principle is:
"Find the inputs that have the highest probabilty of producing a unique output."
One must do this with the understanding that the probability of a unique output is never unity, and changes with the application.
The power of the Deep-Learning libraries is the pre-existing computational infrastructure for finding solutions, and there are multiple options.
Software Development for Star Tracking, Satellite Pointing, Navigation, and Geolocation
I combined the Artificial Intelligence approach discussed above, with novel centroiding schemes.
This is software that computes what the starfield should look like given a hardware set of camera-FPA and lens, and simulates the noise that the system must contend with. The time, sensor latitude/longitude on the earth, and sensor body attitude (altitude/azimuth/clocking angle) on earth, or (pitch/yaw/roll) quaternions in orbit or flight, can be computed from the noisy starfield, with some parameters known, and others inverted from it. The inversion problem is known as "The Plate Problem".
These codes are for centroiding star trackers using novel, more accurate RaNDTek LLC proprietary methods of centroiding that are far more robust than traditional methods in the face of noise.
The most accurate star trackers available are very large instruments, claiming ~ 0.2 microradians (1 standard deviation) of net pointing knowledge accuracy (Attitude Error Standard Deviation or AESD). What if that same performance could be approached with a substantial SWaP (Size Weight and Performance) gain? See Figure 3.
The algorithms will be investigated for GPS-Denied Navigation and Geolocation for ships, land vehicles, and with modifications for flight, aircraft navigation. Spacecraft/satellite pointing is the traditional application, for greater range radio communication, and free-space laser communication.
The first objective of the development is to make any centroiding-based star tracker more accurate, with no hardware changes: Fully backward-compatible with all existing hardware via a software product offering, made to all manufacturers of star trackers.
The second objective is: A tracker hardware product offering optimized to take the greatest advantage of the AI and specialized centroiding.
The RaNDTek LLC techniques provide an improvement in three ways. First, they centroid more accurately sub-pixel in the presence of noise. Second, the noise is reduced. Third, one can extract data from lower magnitude stars - therefore there are more stars per frame aiding the statistics. All result in higher accuracy net pointing knowledge.
The same schemes also can improve the performance of Wavefront Sensors in Adaptive Optics Systems, with analogous backwards compatible software, and optimized hardware offerings.
Figure 1. Simulation of a starfield frame with noise.
Figure 2. Simulation results for attitude 1 fps.
Figure 3. Tracker Performances
(BELOW)
AESD: Attitude Error Standard Deviation
CESD: Centroiding Error Standard Deviation
(UPPER RIGHT)
Minimum Expected performance improvements with RaNDTek LLC AI for 31 Star Tracker products.
(LOWER RIGHT)
RaNDTek's COTS Trackers used for algorithm testing
Figure 4.
(LEFT) Noise on a pixellated image signal in 1-D for 3 true locations.
(RIGHT) Two centroiding techniques - Center of Mass & Proprietary Method. (pixel center, half to boundary, boundary)
Figure 5.
(LEFT) A noisy and pixellated star image in 2-D.
(RIGHT) Two Centroiding Techniques - Center of Mass (CoM) & Proprietary Method. (pixel center, half to corner, corner).
(BOTTOM) Centroiding technique errors from truth in pixels: Center of Mass (CoM) & Proprietary Method with a factor of 5 improvement in CESD.
(BELOW) Histrograms of Six methods of advanced centroiding under development at RaNDTek LLC, compared to standard CoM and Gaussian Weigthed CoM.
Video 1. This video puts the de-noising technique into perspective. On the left is a noisy image of a star with an SNR ~1, and to the right is the extracted star image. Quantitative information on star postion is extracted from chaotic data. Since most centroiding trackers limit the magnitude of stars that can be used to those whose signals are 3-9 times the noise, operation at SNR=1 allows stars of 1-3 higher magnitude to be used.
Figures 6,7,8.
(COUNTER CLOCKWISE)
When Geolocating/Navigating from the starfield, one must solve simultaneously for the sensor attitude (az/alt/clocking) and the latitude/longitude on the earth, given a known time. This increases the AESD in sensor altitude and azimuth, but allows for more accurate geolocation. Below are geolocation results for the CESD of 0.111 pixels for the 60 Mpixel camera in Figure 3 at 1 fps, producing a Geolocation Error Standard Deviation GESD ~ 7 meters at a fixed true lat/long and constant clocking angle (not solved for), about as good as your smartphone GPS.
GESD, like AESD, is also proportional to the CESD .
AI / Machine Learning Based Large-Parameter Free-Form Optics
Free-Form Optics is essentially designing with surfaces that are parameterized with functions with a very large number of adjustable parameters. The large number of parameters makes it very difficult for a human designer to conceptually link trends in them to optical behavior, or targets. Further, optimization of a large parameter space can be slow.
However, AI's are very good at taking vast numbers of interacting and human-incomprehensible inputs, and at least internally, making sense out of them.
RaNDTek LLC is presently investigting the use of Convolutional Neural Networks, Multilayer Perceptrons, and other architectures applied to Free-Form. They are trained with variations of parameterized models. The training data may be the parameters themselves and the coherent fields (or any other performance metric), or they may be 2D arrays of surface sag and the performance metric. The latter approach frees the designer from the constraints of the initial parameterization in the final stage, and/or allows iterative reparameterization.
Figure 9. (BELOW) Free-Form Optical non-diffractive beam shaper for single-mode optical fiber onto high-speed photodiodes. The microlens sag is shown on the right, while the left image shows how the optic losslessly transforms the beam profile from a Gaussian (blue) to a Top-Hat (orange), and the target profile is shown as (green). The last step in the design was handled by a trained AI.
Stray Light Analysis for NASA-NRL PUNCH Mission Coronagraph
We will be doing the stray light analysis for this mission:
https://directory.eoportal.org/web/eoportal/satellite-missions/p/punch