More channels. Wider Doppler. Faster radar.

Install VectorWarp 0.1.1

Fedora 44

Add the signed repository once, then install with DNF:

If curl or GnuPG is missing, install curl and gnupg2 from Fedora first.

curl --fail --location --proto '=https' --tlsv1.2 \
  https://mickeyslaven.github.io/blah2-VectorWarp/install.sh --output vectorwarp-install.sh
less vectorwarp-install.sh
sudo bash vectorwarp-install.sh --repo-only
sudo dnf install vectorwarp
sudo systemctl enable --now vectorwarp-api.service

Open http://localhost:3000, configure your receiver, and choose Save & Restart. Update later with sudo dnf upgrade vectorwarp.

Ubuntu, Debian and other supported systems

Add the signed APT repository once, then install with APT:

If curl or GnuPG is missing, install curl and gnupg from your distribution first.

curl --fail --location --proto '=https' --tlsv1.2 \
  https://mickeyslaven.github.io/blah2-VectorWarp/install.sh --output vectorwarp-install.sh
less vectorwarp-install.sh
sudo bash vectorwarp-install.sh --repo-only
sudo apt update
sudo apt install vectorwarp
sudo systemctl enable --now vectorwarp-api.service

The explicit systemctl command enables the browser interface at boot and starts only that service. Open http://localhost:3000 on the installed machine, or http://<server-IP>:3000 from another device on your trusted network. On a fresh install, radar processing stays stopped until you configure it and choose Save & Restart. Update later with sudo apt update && sudo apt install --only-upgrade vectorwarp.

Each package includes Kraken, USRP and dual HackRF adapters, plus the source kit to build RSPduo support from Settings after installing SDRplay's API. Receiver hardware and external software are separate; RSPduo needs the locally installed SDRplay API. Follow receiver setup after installing.

Direct downloads

Prefer the installer above for automatic updates. For a manual installation, choose the package matching your OS version and architecture.

x86-64 covers Intel and AMD PCs. DragonOS uses its Ubuntu base; check /etc/os-release. Raspberry Pi OS Trixie uses Debian 13 ARM64. No 32-bit package is provided. For other systems, use the source installation guide.

gpg --show-keys --with-fingerprint vectorwarp-archive-key.asc
gpg --dearmor --output vectorwarp-release-keyring.gpg vectorwarp-archive-key.asc && \
  gpgv --keyring ./vectorwarp-release-keyring.gpg SHA256SUMS.asc SHA256SUMS && \
  sha256sum --check --strict --ignore-missing SHA256SUMS

Release notes and all assets · Package manifest

Faster than regular blah2

Same recorded IQ at its original rate. Same CPU budget on each host.

On the RTX 4050 workload, regular blah2 missed 22 of 24 measured intervals; VectorWarp GPU missed none. Accuracy qualification runs at startup; accepted steady GPU frames do not repeat CPU clutter or complex-map accuracy calculations.

At 200 ms CPI and ±2400 Hz, these hosts used about 25% less processing time in CPU mode and 64–71% less in GPU mode than original blah2.

GPU acceleration covers clutter FFT/filtering and delay–Doppler work; the small FP64 coefficient solve and other radar stages remain on CPU.

Earlier live array proof: at 527 MHz and 2.4 MS/s, a five-channel array GPU run at ±800 Hz and 200 ms CPI averaged 95.5 ms. This is live capacity evidence from the prior version with recurring CPU checks, not a new live run or a matched upstream ratio.

More radar per frame

VectorWarp also completes wider and five-channel configurations where regular blah2 has no equivalent mode or cannot safely represent the requested geometry.

Physical NVIDIA, AMD and Intel GPU checks compare complex maps to a CPU reference. Results are tolerance-validated, not bit-exact or a guarantee for every host.

Equal-range Doppler tests

All comparisons retain the standard 256-bin, 30.604 km excess-path window at 527 MHz and 2.4 MS/s. At this full range and ±4800 Hz, Strix GPU processing averaged 67.0 ms per 200 ms CPI and 325.5 ms per one-second CPI. VectorWarp CPU took 173.8 ms and 886.0 ms respectively; both modes met all 24 steady deadlines in each case. These are capacity results where upstream's Doppler buffer cannot safely represent the configuration.

The combined efficiency changes also reduced the Strix ±2400 Hz GPU workload from 69.7 to 44.7 ms versus the preceding VectorWarp version in this campaign. Heavier workloads still miss some deadlines; the full report includes all configurations, a separate before/after comparison and remaining processing costs.

On Raspberry Pi 4, the CPU workload at 200 ms CPI and ±800 Hz took 798.2 ms in VectorWarp, versus 905.1 ms in original blah2 and 905.9 ms in Off World Labs' ARM fork, with NEON FFTW enabled for all three. Pi comparison and workload limits →

A separate Pi 4 replay loaded a newer Mesa driver for the test and reduced processing from 797.2 ms CPU-only to 578.1 ms with GPU Automatic (27.5% less time). It still missed the 200 ms deadline; the driver was not installed system-wide. Pi GPU results and supported driver updates →