Architecture

The desktop app and headless binary share the Rust server and receiver engine. Both serve the same React interface.

React client ↔ REST / WebSocket / MCP ↔ Server control plane
                                              ↓ commands
Radio / network / recording → DSP engine → audio, events, spectrum, IQ

Crate boundaries

apps/sdrmm owns CLI configuration and process lifetime. apps/desktop starts the server on an ephemeral loopback port and opens a Tauri WebView. Both isolate SoapySDR discovery in a short-lived child process.

One source of truth for wire types

Define REST bodies, WebSocket messages, settings, and patch types in crates/wire. OpenAPI schemas derive from those types; cargo xtask codegen generates TypeScript declarations.

The client reads device capabilities, channel descriptors, and the node palette from the server, keeping controls aligned with the running build.

Data plane and control plane

The DSP path uses command queues for settings and bounded snapshots or buffers for output. It performs no I/O, locking, allocation, or async work in hot processing.

The control plane owns HTTP handlers, SQLite, workspace reconciliation, subscriptions, recording indexes, and serialization. It may allocate or block as needed.

Spectrum, audio, and video use binary WebSocket frames; browser audio is Opus-compressed. Decoder events use typed JSON. Durable state is fetched through REST after WebSocket invalidations.

Coherent processing

Each capture block carries its first sample index, including gaps from reported hardware loss. Coherent processing taps each lane into a ring and selects the sample range common to all lanes. After a gap, it advances to the next shared index before applying calibrated delays and weights.

Beamforming sums weighted lanes into a normal capture ring. Channels, recorders, and scopes consume that beam through the ordinary single-lane path.

An Array node combines streams already owned by Device nodes. device-array provides logical ingress lanes; the engine forwards corrected IQ, coordinates tuning, and handles member recovery. The array adapter never opens hardware.

Media and recording outputs cross preallocated single-producer/single-consumer buffer pools. Workers allocate transport payloads and publish them. Full queues never block DSP: media loss is reported and recordings fail explicitly. Shutdown drains pending buffers. Some decoder algorithms still allocate variable-sized results.

channels depends on dsp, modem, and wire. Shared modem algorithms belong in modem. Allocation, throughput, and modem measurement tooling belongs in test-support crates outside the application dependency graph. cargo xtask check enforces boundaries; cargo xtask perf checks DSP throughput, allocation, decoder searches, and engine publication.

Workspaces and live engine state

The workspace graph describes desired state. Applying it binds saved Device references to discovered radios, restores settings, and reconciles channels and engine objects.

Saved references use backend, serial, key, and variant identity. Engine IDs are temporary and never stored in the graph. Disconnected radios retain their nodes and settings until reconnection.

Failure and backpressure

Queues are bounded. Overruns, dropped frames, recording faults, truncated exports, WebSocket lag, and reconnection state surface to clients. Slow consumers cannot block capture or grow memory without a limit.

Testing layers

CI builds release configurations without enumerating host radios. Test at the narrowest layer that proves the behaviour, adding end-to-end coverage for cross-layer workflows.

Standard tables and their provenance

Some decoder constants come directly from specifications:

Sources are ETSI EN 300 401 (DAB), TS 102 563 (DAB+), EN 300 421 (DVB-S), EN 302 307-1 and -2 (DVB-S2/S2X), TS 102 606 (GSE), and ES 201 980 (DRM).

Table values were cross-checked against welle.io (GPL-2.0-or-later) and GNU Radio's gr-dtv (GPL-3.0-or-later). This attribution concerns table verification, not copied decoder code. The 7,378 DVB-S2 accumulator addresses were transformed mechanically. The VL-SNR 896-bit seed and Walsh–Hadamard rows were transcribed from the standard; their sixteen generated patterns match gr-dtv's tables.

Tests check independent properties such as puncturing density, polynomial roots, published CRC values, and parity checks on encoded words. These checks help detect transcription errors.