Shane Spahr // Site
ready
Themes
Site administration
path/portfolio/projects/rf-math-radar
surfaceRF Math & Radar
statusportfolio surface ready

RF Math & Radar

My RF work connects signal processing, mathematical models, and hardware preservation: extracting useful measurements from radio signals, understanding what those measurements can establish, and keeping serviceable equipment usable.

This collection brings together passive FM radar, an open controller concept for replacing some legacy ratheon radar systems, and ties into RF imaging with WiFi beamforming, channel measurements, and gausian splatting. The records constain research, design studies, and archived code sketches. Bench validation and measured performance remain to be documented here.

note I have been informed that I can't (and shouldn't) publish my instructions, equations, or code for active or passive radar imaging because it's export controlled technology.

Passive FM radar · Active radar · RF imaging · RF mathematics

Passive FM radar and SDR processing

The passive-radar investigation began with using FM broadcasts as an illumination source. The SDR reference receiver captures the direct broadcast while a second, coherent channel captures reflections. That project and its field results explore refactoring standard processing stages with reusable C++ components.

The design covers channel selection and rational resampling, adaptive clutter cancellation, cross-ambiguity processing, range–Doppler displays, and target tracking. The code includes adaptive-filter modules, configuration calculations, and dynamic source generation utilizing a library of filter classes and general algorithmic signal processing definitions.

The processing path I sought to recreate was:

  1. Capture aligned reference and observation channels with their sample-rate and tuning metadata.
  2. Select the broadcast channel and resample both streams consistently.
  3. Estimate and subtract direct-path leakage and stationary clutter.
  4. Compare the remaining signal with delayed and frequency-shifted reference copies to form a delay–Doppler map.
  5. Evaluate detections and tracks against known test signals before interpreting field recordings.

The related GNU Radio notes explore FIR/IIR filtering, decimation, matched filtering, and implementation boundaries between Python experiments and streaming blocks. VOLK supplies optimized SIMD math kernels; profiling those kernels is a potential integration step after a correct reference implementation exists. The archive does not establish a measured speedup or a completed GNU Radio module.

Active radar: preserving a legacy ratheon system

The ratheon project started with an analog radome and a gently used C120 multifunction display. My goal was an open replacement controller and acquisition path that lets the existing scanner feed a modern display or an open navigation application.

The architecture rrequired separatinh deterministic acquisition from host software: an FPGA or soid state timing engine (v2) associates sampled radar video with pulse timing and azimuth, then sends structured radar spokes to a renderer. A spoke carries an angle, range samples, a timestamp, and acquisition metadata. Control and status handling are documented alongside the data stream.

The first milestone was to characterize the actual scanner interface and replay a recorded sweep through a host-side renderer. That established the signal and timing model before it was necessary to design the final converter, FPGA, buffering, or transport. LightHouse integration was a red-herring compatibility investigation; an open display adapter was the better candidate endpoint. Neither interface compatibility nor a working replacement board is established by the archived discussion.

This project also motivated a broader exploration of PCB mapping, automated probing, optical inspection, and IC state-change sensing. Those foundations helped develop a longer-term (but still theoretical) instrumentation direction, with small, repeatable design, analysis, and measurement tasks (time permitting.)

RF imaging with WiFi and beamforming

Beamform imagaing, and the later RF imaging project outline an experiment in recovering narrower spatial information from RF measurements. The starting data was obtained by Fourier series expansion of doppler signals derrived from WiFi channel state information (CSI): complex channel estimate circuits which allow amplitude and phase to be examined by contrasting antenna pairs, frequency bins, and time.

The plan was to utilize RF captures paired with optical or depth observations, but that integration was never developed. However, preprocessing established repeatable amplitude and phase baseline measurements, and accounting for timing and phase offsets, it allowed for extracting features associated with motion and changing propagation paths.

My notes propose several approaches to the next steps:

  • Beamforming and direction estimation using a calibrated, tuned, and directional antenna arrangement.
  • AI/ML analysis for phase differences and Doppler features that could help paint changes over time.
  • Sparse inverse methods for Tomographic spatial reconstruction.
  • RF paired with image/depth data to support learned reconstruction and evaluations against simpler hardware only baselines.

The difficult part was obtaining trustworthy, aligned measurements with cheap retail equipment. Computer assisted synthetic scenes and ray tracing offered candidate ways to explore assumptions, but their relationship to real harware captures did not represent the reality of signal capture and analysis. Multi-antenna 3D reconstruction plan extensions seem out of reaach based on recent results.

Mathematics that connects the projects

I want the derivations and code to preserve the geometry, units, timing conventions, and assumptions behind each result.

Delay and range

For a colocated transmitter and receiver, an echo's round-trip delay gives R = c × Δt / 2. In passive bistatic processing, delay is measured relative to the direct broadcast: ρ = R_T + R_R − L = c × Δt, where R_T and R_R are the transmitter-to-target and target-to-receiver distances, and L is the transmitter-to-receiver baseline. Here c is propagation speed and ρ is excess path length. A passive delay bin therefore does not directly give distance from the receiver. These conventions follow the MathWorks radar geometry reference.

Sampling and coherent processing

For N samples at the actual processing rate f_s, the coherent processing interval is T = N / f_s. An unpadded Fourier grid over that interval has frequency spacing Δf = 1 / T. Increasing the observation time gives a finer Doppler grid, subject to the signal remaining coherent. Rational resampling changes the actual rate, so the configuration must carry the resulting duration and bin scales through the pipeline. Manning's range–Doppler derivation develops the corresponding cross-ambiguity calculation.

Sample spacing, plotted bin spacing, and the ability to distinguish two targets are separate quantities. A denser plot alone is not evidence of improved physical resolution.

Cancellation and inverse problems

In the passive-radar code study, clutter cancellation fits delayed reference-signal components to the observation channel and examines the residual. The engineering question is how filter length, regularization, and adaptation affect both interference rejection and target preservation.

In imaging, the corresponding question is which scene properties the measurements constrain. Calibration errors, multipath, limited bandwidth, and antenna geometry belong in the model. Reconstruction quality needs comparison with known scenes and held-out captures, alongside uncertainty and failure cases.

Source history

This overview draws on the archived blog site pages “FM Radar Project Details” and “GNU Radio's VOLK Optimization” (October 2023), “RF Imaging with WiFi” (February 2020), and “IC State Change Sensing” (February 2019), together with the curated RF-imaging engineering notes. The logs preserve design exploration and code drafts.

Browse all projects · Return to Portfolio