Reconstructing a proprietary Real-Time Engine through non-invasive kernel instrumentation.
In the high-stakes world of automotive HIL simulation, the real-time engine is the gatekeeper. Our system relied on Red Hawk, a proprietary vendor engine. To gain true strategic independence, the mission was to replace this "black box" with a custom-engineered in-house solution.
I was strictly forbidden from viewing the vendor's source code. Logic had to be reconstructed for FPGA-driven hardware purely by observing data patterns at system boundaries.
To understand the "black box", I transformed the FPGA kernel driver into an instrumentation probe. By intercepting the IOCTL interface, I recorded how the engine translated abstract parameters into microsecond-level hardware commands.
Catastrophic execution time. 4x slower than vendor baseline. Unusable for real-time loops.
Full functional parity at 38% higher performance than the proprietary engine.
Evicted all runtime malloc/free calls for deterministic scheduling.
Replaced floating-point math with integer-based wheel patterns.
Optimized PCIe payload packing at the instruction level.
Compiler-orchestrated SIMD for tooth sequence generation.
Don't guess—measure. Kernel-level logs provided an unvarnished truth that documentation hides.
Functional correctness is the foundation. Performance is a second-pass architectural discipline.
Modeling a complex hardware controller requires understanding not just 'what' it does, but 'when' it does it.