Home of Brian M Anderson*
Leopard, QTAM, Incompressible Fluid Simulation, to Networks and Large Data Applications
This site is to showcase for some of my work and projects in both my undergraduate work in Computer Science at the University of Illinois, and my graduate studies for Electrical and Computer Engineering at the University of Arizona all accomplished while proudly serving an active duty role in the United States Air Force.
In my undergraduate work I focused on developing advanced numerical solvers for models such as atmospheric scattering and smoothed particle hydrodynamics which enforce null divergence incompressibility conditions. I have developed a inference based corrective solver which utilizes GNN’s to match the incompressibility correction scheme to the particle neighborhood configuration which speeds up the PCISPH algorithm, already a realtime, algorithm by a factor of 1000x making large scale dense fluid simulation for incompressible fluids possible.
In addition, I developed a compiled static language, Leopard, with a LALR(1) grammar as a superset of GoLang and mapped this grammar to LLVM IR for deployment across multiple and expanding ISA architectures including Apple Silicon and embedded devices. This work is not insignificant as it also provides a foundation for adapting a very versatile language with diverse targets including embedded and GPU architectures. This language was also purpose built to be the most flexible language ever built by enabling a standard operating mode with GC as well as a pointer acquisition handle mode for Non-GC routines with specified non-interrupt architectures while handling concurrency and generic modern language features.
While these applications had always been important to me, my continuous service in the United States Air Force introduced me to the RF communications and RF detection problem sets. This lead me to apply and be accepted into the Graduate ECE program at the University of Arizona where I am projected to graduate in 2026.
In my short time at the University I have already been responsible for ground-breaking work in RF Engineering, where I developed Quadrature Tensor Amplitude Modulation (QTAM 64x32) & parallel transmission protocol ‘tensor’.
All of this is to say that I’m a highly flexible and dedicated engineer capable across a diverse portfolio of problem sets from high-level software and architectural problem sets to low-level RF transmission and hardware engineering issues.
While I have proudly served 13 years as an active duty enlisted I am currently preparing to transition into a civilian and engineering role.
GNN Newtonian Incomprehensibility with Smoothed Particle Hydrodynamic Systems.
I utilize datasets simulated through PCISPH algorithms to train a large parameter GNN which is able to resolve particle error corrections for arbitrary particle neighborhood particle configurations. The GNN outputs enforce constant density fluid/solid conditions. Work is applicable to both fluid simulation as well as atom particle neighborhood configurations during state transitions which can model solid fracture and stress.
GNN Newtonian SPH Correction
24 GhZ Shortrange FMCW Detection Radar
I utilized an Analog Devices MMIC at 24 GHZ and a PLL with ramp synthesizer to engineer a demonstration FMCW detection radar. Utilized tools like Anysys HFSS to design and manufacture a 2 x RX / 2 x TX Patch Antenna Receiver.
This project was able to detect range with a 3 cm resolution and 0.5 degree angular separation with FMCW radar approach utilizing FMCW generated radar return frames with 128 sample frames and a 256pt FFT.
Radar processing made extensive use of ARM Cortex R4 - DSP functionality to store and process raw radar data blocks.
This was one of my very first projects exploring the development and deployment of RF radar solutions, further development would have involved the use of more powerful Low-Noise Signal Amplifiers for the Antennas as well as beamforming patterns. This project gave me alot of ideas about the detection space. And I walked away with a few goals for a next iteration of this problem set.
Utilizing a flexible Antenna with Hexagonal Antenna Structure and surface mounting onto a small parabolic surface.
Using beamforming techniques for calibration of the radar beam and combining with the parabolic form factor to emit more parallel beam energy.
Utilize PHEMT based amplifiers for low-noise amplification and separate the modules with RF Coax connectors for diversity in connections
{Leopard} - Programming Language 2024
While my work on Leopard is not completely done at the moment I do have the belief that this language will eventually be the most productive, useful, and significant programming language built since C. Giving the programmer the option to work between memory handling contexts, with the ability to address hardware in an un-managed context provides the ability to develop drivers in Leopard while providing C ABI access in the meantime.
Much of the work after completing this language will involve in providing native hardware support for peripheral devices and components such as interfacing with PCIe, USB, Displays, and Kernel level system software.
This language has been a very ambitous project overall and I do continue to work on it. For more details on the development you can find my work here:
https://github.com/andewx/leopard
{leopard} - was hand crafted by myself in 2023 as what was meant to be a mirror of golang with some added features that felt essential while keeping the best aspects of GoLang that is:
First Class Methods - Methods are more tightly integrated into the language where structs can expect to have their own method APIs.
Safe Pointer Managed Environment - We give the developer the control to compartmentalize their program between a safe GC memory arena and a managed pointer arena. Managed pointers define explicit control requirements on object handles where ownership is passed between contexts. Without explicit ownership and verification an object can not be used, similar to borrowers in other languages.
LLVM IR + Optional Embedded - Leopard language maps to the LLVM IR language like many modern languages and can therefore support many modern compute architectures
Method & Operator Overloading - We limit overloading to a finite set of common operators and allow parameter namespace method and function overloading. I felt this allows for a more expressive programming paradigm where certain libraries can be more grounded and expressive with their mathematical foundations. I also support Unicode operators so an expression such as ∂f(t)/∂t can be implemented by libraries.
Vector Expressions - Vector & Matrix operations are very common and neccessary programming constructs therefore we provided a means to declaring packed data types like float4, float3, mat4, matNN as core components of the language. Notably these datatypes can be implemented with packed bit representations and operations with these data types and other vectors automatically are matched with SIMD type parallel instruction architectures.
Hardware Mode Addressing - Hardware mode instructions circumvent the need to rely on C libraries for modern compute tasks and working with drivers
C ABI Calls: We can load C ABI symbol tables and call C functions while interfacing with C structures.
Most other GoLang features - While Leopard is not a strict superset of Golang it is a 95% match and draws inspiration from all of its best features like, channels, part of the generics approach, and it’s approach to modules.
100 Tb/s transfer data rates over the Air
Developed Tensor QAM representation where each constellation point holds an effective Z-component effectively scaling the QAM M signal representation by a factor of N.
We introduce a QPSK modulation for the IQ Z Component and use blind source separation along with a ANN detector to retrive the M symbol which now holds N^3 positions.
Typically we can assume that the QPSK modulation is limited to 8 levels however machine learning has promise to increase the number of detectable levels.
Utilizes a MISO Parallel Scaler Protocol ontop of the QTAM expansion which scales datathroughput by another variable factor K, improving data throughput over modern SOA solutions by overall factor of N^2 with N^4 total bitrate throughput.
Quadrature Tensor Amplitude Modulation
Effortlessly scale out QAM signal represntations from 64 QAM to 2048 QAM and higher for example theoretically 520K QAM Signal Density