Combined

Hyperactive//2026 Radio

Experimental real-time, long-range, high-quality, wireless digital audio for Hyperactive 2026.

Audio quality

64 kbps stereo Opus at 48 kHz should be transparent for most program material — comparable to a good 128 kbps MP3, but with lower latency and better behaviour at the edges of the bitrate. The fixed-point build removes float entirely; no performance penalty on Cortex-M33 which has a hardware FPU but where the float Opus path was benchmarked to be slower than fixed-point.

End-to-end latency is approximately one frame (20 ms) of algorithmic delay plus radio airtime (~1.6 ms at 100 kbps for a 160-byte packet) plus any playout buffer on the receiver.

Radio

Regulatory (FCC Part 15.247)

No frequency hopping required. Part 15.247 has two compliance paths — FHSS is one, but digital modulation is the other: fixed frequency, up to +30 dBm EIRP, with two constraints:

• 6 dB bandwidth ≥ 500 kHz — requires high frequency deviation; at 100 kbps, ±200 kHz deviation gets you there (BW_6dB ≈ 2 × f_dev + bit_rate). Lower deviation (e.g. ±50 kHz) only yields ~200 kHz and does not qualify.
• PSD ≤ 8 dBm in any 3 kHz band — GFSK spectral peaks at ±f_dev must stay under this limit at the chosen TX power.

The exact deviation and power settings need to be verified against both constraints once the RF configuration is finalised. At +20 dBm with sufficiently high deviation the path is viable, but the numbers need to be checked.

Packet structure

Each 20 ms Opus frame encodes to roughly 160 bytes at 64 kbps. With ~10 bytes of framing (sequence number, network ID, CRC) the total packet is ~170 bytes, for an effective air rate of ~68 kbps — well within the 100 kbps channel.

Performance

MCU runs at 270 MHz (overclocked from the 150 MHz default; stable at stock 1.1 V).

Benchmarked on hardware against broadband LCG noise (worst case for Opus — no clear pitch, flat spectrum):

Frame budget = 20 ms = 5,400,000 cycles at 270 MHz.

The encoder leaves 11.5 ms of headroom per frame for I2S DMA and SPI radio management. The decoder uses only 16.6% of the frame, leaving ample margin for playout buffering and output DMA.

Firmware architecture

Written in Rust (no_std, Embassy async runtime).

ha_radio/          # firmware binary (bench + main)
embedded-opus/     # safe Rust wrapper around libopus
opus-sys/          # raw FFI bindings; builds libopus via cc crate

embedded-opus/build.rs compiles a host-side C probe at build time to determine the exact encoder and decoder state sizes, so buffers are sized precisely without a runtime heap.

Planned: dual-core pipeline

The encode headroom is comfortable at 270 MHz, but the dual-core split is still planned for clean separation of concerns once the ADC arrives:

• Core 0 — I2S DMA capture, SX1276 SPI, radio management
• Core 1 — Opus encode only

A pair of Embassy Channel queues (depth 2) connect the cores: PCM frames inbound, encoded packets outbound.

Build

cargo build --release --bin ha_radio
cargo build --release --bin bench

Requires a Rust nightly toolchain and the thumbv8m.main-none-eabihf target (see rust-toolchain.toml).