SigTekX
>_ GPU-Accelerated Signal Processing Engine v0.9.5
System Purpose
SigTekX is a high-performance research platform architected for real-time ionospheric scintillation analysis. It bridges the gap between low-level CUDA optimization and high-level Python research workflows, providing sub-millisecond processing latency for ULF/VLF radio signal monitoring.
Core Performance
- Sub-millisecond end-to-end latency
- Zero-copy lock-free streaming architecture
- CUDA/cuFFT accelerated spectral analysis
- 100% real-time compliance at 100 kHz
Real-Time Streaming Executor
The Streaming Executor is the core of the engine's real-time capabilities. It employs a lock-free Single-Producer Single-Consumer (SPSC) ring buffer to decouple data acquisition from GPU processing.
Zero-Copy Memory Architecture
Direct DMA Access
The streaming critical path utilizes a peek_frame() API for direct DMA from Pinned Host memory, reducing the H2D latency floor.
Lock-Free SPSC Design
Data ingestion and processing are synchronized via atomic head/tail pointers to prevent thread contention.
Benchmarked Performance
| METRIC | VALUE | SPECIFICATION |
|---|---|---|
| Mean Latency | 160.1 μs | RTX 3090 Ti | NFFT=4096 |
| Throughput | 4,549 FPS | ~37.3M samples/sec |
| Real-time Compliance | 100% | Zero deadline misses at 100 kHz |
| Accuracy (SNR) | >123 dB | IEEE 754-2019 FMA Compliant |
Methods Paper Roadmap
Dual-Plane Architecture
Decoupling the Data Plane (C++/CUDA) from the Control Plane (Python).
Custom Stage Ecosystem
Integration of Numba CUDA kernels and PyTorch models directly into the data plane.
Performance Targets
Validation of a Real-Time Factor (RTF) ≤ 0.33 for ionospheric workloads.
Embedded Deployment
Transitioning workstation GPUs to power-constrained embedded systems (NVIDIA Jetson).