Files
rspektrum/README.md
T
tyler 4c3c6a955a perf: unblock long-file loading; add collision detection and jump nav
Three things, all surfaced while working a multi-node protocol issue on a
5.7-hour capture.

**Loading was frame-paced, not compute-bound.** The STFT overview advanced
a fixed 200 segments per frame, so the frame limiter — not the FFT — set
the pace: 478k segments at ACTIVE_FPS meant over a minute spent waiting
between frames rather than computing. The tell was absurd: backgrounding
the window, which skips presenting entirely, loaded the same file in
seconds. The progress bar was slower precisely because you were watching
it. The overview is now computed in one blocking call after presenting the
loading panel, so focused load matches the unfocused speed. The panel says
the window will stop responding and drops the percentage, which could not
animate and would have read as a hang. ACTIVE_FPS 30 -> 60 while here;
idle still parks at ~0% CPU through the event-wait path.

Background work also no longer stops when the window loses focus. Pending
work now counts as "active" regardless of focus, so the loop doesn't block
in PollInputEvents waiting for input that isn't coming, and an unfocused
frame with work outstanding skips the draw pass entirely rather than
throttling the high-res fill to the refresh rate.

**Collision detection.** Annotations that overlap in both time and
frequency are flagged, merged into contiguous regions, and drawn as red
bands confined to the band the overlap occupies (padded, so a narrow
overlap stays findable) rather than spanning the full axis and hiding the
signal being pointed at. N / Shift+N and sidebar buttons jump between
regions, centring each without disturbing the current zoom.

Point markers are excluded: control events and assertions have no band and
zero duration, and treating a missing band as "whole spectrum" — which is
how they are *drawn* — made every marker collide with whatever it sat
inside. That was 38% of the reported collisions on a real capture. Only
things that actually occupy the air can interfere. Counts verified against
an independent reference implementation on two captures.

**Scope waveform via min/max summary.** The envelope rescanned every
visible sample every frame — ~245M reads, near 1 GB of memory traffic, on
a multi-hour file, measured at ~60 ms/frame for a few hundred pixel
columns. It now draws from 1024-sample buckets built once at load (60 ms,
1.9 MB), measured at ~0.07 ms/frame. Keeping both extremes per bucket
means single-sample transients still show at full zoom-out, which plain
decimation would drop; verified that no column ever understates a true
peak. Zoomed in past a bucket it falls back to raw samples, which is cheap
there by definition.

Also fixes ComputeCollisions never running for files opened through the
file browser, and moves the collision panel above the annotations dropdown
where it isn't pushed off the bottom of the sidebar.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V8ZWfr5XZyyDttvkhJUgHN
2026-08-12 13:58:58 -07:00

13 KiB
Raw Blame History

rspektrum

rspektrum is an interactive spectrogram viewer for inspecting radio captures and arbitrary audio. It loads a WAV file, computes a short-time Fourier transform (STFT), and draws the result as a zoomable, pannable timefrequency image. Its distinguishing feature is support for mLnL annotations — labelled regions (TX frames, assertion outcomes, impairment fires, …) carried inside the WAV file itself — which it overlays on the measured spectrogram so you can compare what a modem intended to transmit against what actually hit the air.

You can box a time/frequency region, hear it back through a bandpass filter, and export either the picture (PNG) or the isolated audio (WAV). rspektrum runs three ways: a native desktop app (C + raylib), a headless command-line renderer, and a WebAssembly build in the browser.

Click for Video Demo

rspektrum spectrogram view with mLnL annotation overlay


What it's for

The primary use case is reviewing captures from the mLink radio stack: a WAV recording of an over-the-air signal with an embedded mLnL chunk describing what the modem/daemon believed it was transmitting at each instant. rspektrum renders those annotations on top of the measured spectrogram, frame by frame, so intent and reality sit side by side.

It also works as a general-purpose spectrogram tool for plain WAVs with no annotations. See mlnl_chunk_spec.md for the annotation format.


Features

  • STFT spectrogram — selectable colormaps, adjustable dB floor / dynamic range, absolute (dBFS) or relative amplitude scaling.
  • mLnL annotation overlay — labelled boxes from the WAV's embedded annotation chunk; hover a box (or its region on the scope) for per-frame detail (sequence, channel, rate, scheduling offset…).
  • Zoom & pan the time/frequency view.
  • Region selection — box a time and frequency range with the mouse.
  • Filtered playback — play just the selected region, band-limited to the selected frequency box via an FFT bandpass. What you hear is what you'd export.
  • Waveform scope — toggleable time-domain view beneath the spectrum.
  • Marker / ruler and a spectrum slice (PSD) readout.
  • Export — save the view as a PNG, or the selected region as a WAV.
  • Headless render mode — produce an annotated PNG from the CLI with no window, no GL, and no X server. Pure CPU; runs in CI, containers, or over SSH.
  • Broad input — WAV directly (8/16-bit PCM, 32-bit float; stereo downmixed to mono); other formats transcoded via ffmpeg if it's on PATH. Drag-and-drop.
  • Cross-platform — Linux/desktop, Windows, and a WebAssembly build.

Building

You need only make and a C compiler (gcc or clang) plus the X11/OpenGL development headers (see below). raylib is vendored in this repo and compiled from source — there is no separate raylib install step, no premake, no network access required. A plain clone builds:

make                # release -> bin/Release/rspektrum   (-O3 -ffast-math, AVX2/FMA)
make DEBUG=1        # debug   -> bin/Debug/rspektrum     (-g, no optimization)
make run            # build + launch
make test           # build + run the DSP correctness tests
make bench          # FFT benchmark over mlnl_samples.wav
make clean

Useful overrides: make CC=clang, or make ARCH=-march=native to tune for your own CPU (the default -march=x86-64-v3 targets any ~2013+ x86-64 chip; drop it with make ARCH= for an older CPU).

System dependencies

The compiler needs the X11 and OpenGL dev headers (the runtime libs are already present on any desktop; only the -dev/-devel packages are usually missing). The X11 extension libraries (Xrandr, Xinerama, Xcursor, Xi) are opened at runtime via dlopen, but their headers are still required to compile.

If make stops with an error like fatal error: X11/Xlib.h: No such file or GL/gl.h: No such file, install the dev packages for your distribution:

Distro Command
Debian / Ubuntu / Mint sudo apt install build-essential libx11-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev libgl1-mesa-dev
Fedora / RHEL / Rocky sudo dnf install gcc make libX11-devel libXrandr-devel libXinerama-devel libXcursor-devel libXi-devel mesa-libGL-devel
Arch / Manjaro sudo pacman -S base-devel libx11 libxrandr libxinerama libxcursor libxi mesa
openSUSE sudo zypper install gcc make libX11-devel libXrandr-devel libXinerama-devel libXcursor-devel libXi-devel Mesa-libGL-devel
Alpine sudo apk add build-base libx11-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev mesa-dev

On Debian/Ubuntu the single metapackage xorg-dev pulls in all of the X11 -dev packages above, if you'd rather not list them.

Run make check-deps to probe for the required headers before building — it prints the install hint for your platform if anything is missing.

Web (WebAssembly) build

./build_web.sh      # emscripten; emits the WebAssembly bundle to bin/web/

Usage (desktop GUI)

./bin/Release/rspektrum [input.wav]

Load a file by passing it on the command line, dragging a .wav onto the window, or pressing O for the file browser. Try the bundled sample:

./bin/Release/rspektrum mlnl_samples.wav    # in-repo WAV with an embedded mLnL chunk

Controls

Input Action
O Open file browser
Mouse wheel Zoom both axes (preserves aspect ratio)
Shift+wheel Zoom the time axis only
Ctrl+wheel Zoom the frequency axis only
Alt+drag / middle-drag Pan the view
LMB drag Select a time + frequency region
Space Play / stop the selected region
Hover an annotation Tooltip with that frame's mLnL detail; lists every overlapping frame under the cursor
N / Shift+N Jump to the next / previous collision
P Show / hide the waveform scope
M Marker / ruler tool
S Spectrum slice (PSD)
E Export PNG
W Export selection as WAV
Home Reset view (fit all)
End Zoom to start
F11 Toggle fullscreen
F1 About / help
Esc Clear selection / close dialog

Most controls are also available as buttons in the left sidebar (colormap, floor, dynamic range, annotation opacity, grid, …).

Inspecting overlapping transmissions

When several stations are on the air at once their annotation boxes stack, and the one drawn last hides the rest. Two features address that:

  • Hover any pile-up and the tooltip lists every frame under the cursor — one row per frame with its own colour swatch, led by the fields that actually tell them apart (node, frame name, position in the PTT, channel). Deep piles are capped with a +N more count.
  • Collisions (sidebar toggle) highlights where transmissions genuinely overlap in both time and frequency. N / Shift+N, or the sidebar < prev / next > buttons, jump between them; each jump centres the region, keeps the current zoom unless the region needs more room, and reports its position (Collision 7/54 — 3 frames at 1284.95s).

A collision requires a real overlap in time and band, so two frames in different channels at the same instant are not flagged, and neither are zero-duration point markers (control, assertions), which annotate the run rather than occupy the air. Markers are drawn only across the band the overlap occupies, not the full frequency axis. Adjacent collisions merge into one region, so a busy stretch reads as a single span rather than dozens of bars.


Usage (headless render)

--render writes the spectrogram straight to a PNG with no window, no GL context, and no X server. It computes the STFT, colorizes the bitmap, bakes the annotation overlay onto it, and exports — all on the CPU — so it runs anywhere (CI, a bare SSH session, a container with no display):

./bin/Debug/rspektrum --render OUT.png INPUT.wav [options]

The output is the real spectrogram bitmap at native STFT resolution (not a screenshot of the UI), so it carries no sidebar/scope chrome — just the timefrequency image with the annotation overlay.

Flag Effect
-r, --render OUT.png Render to OUT.png and exit (no window/GL/X)
-a, --annotations Force the annotation overlay on
--no-annotations Force the overlay off
--annotation-opacity=V Overlay strength 0..1 (default 0.5)
--annotation-kinds=LIST Comma-separated kinds to draw (default: all)
--width N Resize output to N px wide (default: native STFT size)
-h, --help Usage

Annotation boxes are drawn outline + label only (no translucent fill): mLnL captures contain many overlapping full-band boxes whose fills would alpha-stack to opaque and bury the signal, so the outline marks each region while the spectrogram reads through.

# everything, brighter overlay
./bin/Debug/rspektrum --render /tmp/all.png mlnl_samples.wav --annotation-opacity=0.7

# only on-air frames and failed assertions
./bin/Debug/rspektrum --render /tmp/tx.png mlnl_samples.wav \
  --annotation-kinds=tx_frame,assertion_failed

Annotation kinds: tx_frame, tx_burst, control, channel_up, channel_down, assertion_passed, assertion_failed, impairment_fire, gain_change, unknown.

The hover tooltip only appears with a live mouse over a box, so it cannot show up in a static --render. To verify tooltip behaviour you need a real (or virtual) display driving the GUI — see below.


Driving the GUI headlessly (agents / CI)

The app can be run, screenshotted, and clicked on a virtual X display with no monitor or GPU (Mesa software GL under Xvfb). The full playbook lives in AGENTS.md; the working reference implementation is shot_input.sh.

The loop in one breath:

Xvfb :99 -screen 0 1280x800x24 >/tmp/xvfb.log 2>&1 &     # 1. fake screen
DISPLAY=:99 ./bin/Debug/rspektrum mlnl_samples.wav \
        >/tmp/app.log 2>&1 &                              # 2. run on it
sleep 2                                                    # 3. reach a steady frame
DISPLAY=:99 import -window root /tmp/shot.png             # 4. grab the frame

Prerequisites (Debian/Ubuntu): sudo apt-get install xvfb imagemagick xdotool (plus libgl1-mesa-dri and LIBGL_ALWAYS_SOFTWARE=1 if GL fails / frames are black). Synthesize input with xdotool against DISPLAY=:99 to exercise UI paths.


Technical notes

  • STFT — Hann-windowed, 2048-point FFT with 50% overlap by default; frequency resolution sampleRate / fftSize Hz per bin. Amplitude in dB.
  • Axes — X = time (s), Y = frequency (Hz, scaled to the file's Nyquist), colour = amplitude.
  • Long files — two things keep cost tied to what's on screen rather than to total duration. The spectrogram image is built for the visible segment range (capped at 8192 px wide), so a multi-hour capture renders at all — an unbounded full-file image exceeds the GPU texture limit and silently draws nothing — and zooming in genuinely re-renders at higher resolution instead of magnifying pixels. The scope draws from a precomputed min/max summary (1024-sample buckets) rather than rescanning every visible sample each frame, which on a 5.7-hour file is the difference between ~60 ms and ~0.07 ms per frame. Keeping both extremes per bucket means a single-sample transient still shows up when fully zoomed out.
  • Time zoom limit — the tightest visible window is derived from the STFT hop (fftSize / HOP_RATIO samples), not from a fixed fraction of the file, so time resolution does not degrade as files get longer: a 30-minute recording zooms in just as far as a 30-second one. At 48 kHz / 2048-point FFT the floor is ~85 ms across the viewport; a smaller FFT zooms correspondingly tighter. Past that point there are no further STFT segments to show, so the view would only interpolate.
  • Playback / WAV export share one processing path: the selected time span, FFT-bandpassed to the selected frequency box, peak-normalised.
  • mLnL parsing — walks the WAV's RIFF chunks for the four-CC mLnL chunk (UTF-8 JSON Lines); unknown chunks are skipped, so annotated files stay standards-compliant audio everywhere else.

Source layout

src/
  spectrogram.c        # entry point, main loop, CLI args, headless render
  stft.c / fft.c       # STFT + FFT
  render.c             # spectrogram, annotations, tooltips, scope
  ui.c                 # sidebar, file browser, buttons
  audio.c              # WAV load (ffmpeg fallback), bandpass, playback, WAV export
  mlnl.c / mlnl.h      # mLnL annotation chunk parser
  platform_*.c         # per-OS shims (linux / win32 / web)

See raylib_for_desktop_applications.md for the performance / idle-CPU lessons behind the desktop build, and AGENTS.md for the headless-testing playbook.

Known rough edges — behaviour that is unspecified or awkward rather than simply broken — are tracked in known_bugs.md.