DSPi Firmware

DSPi transforms a Raspberry Pi Pico or other RP2040-based board into a very competent and inexpensive little digital audio processor. It acts as a USB sound card with an onboard DSP engine, allowing you to make use of essential tools like room correction, active crossovers, parametric EQ, time alignment, loudness compensation, and headphone crossfeed.

It is my hope that the RP2040 and RP2350 will garner a reputation as the "swiss army knife of audio for less than a cup of coffee".

Feel free to join the official Discord server for development updates, discussion or to request assistance!

---

Table of Contents

• Key Capabilities
• Platform Support
• Audio Signal Chain
• Hardware Setup
• DSP Features
  • Matrix Mixer
  • Parametric Equalization
  • Crossover Filters
  • Loudness Compensation
  • Headphone Crossfeed
  • Volume Leveller
  • Per-Channel Preamp
  • Master Volume
  • I2S Output
  • Subwoofer PDM Output
• User Presets
• Developer Reference
  • System Architecture
  • Performance Tuning
  • USB Control Protocol
  • System Telemetry
  • Data Structures
  • USB Audio Loopback Capture (DSPI_LOOPBACK)
• Building from Source
• Detailed Specifications
• License

---

Key Capabilities

• USB Audio Interface: Plug-and-play under macOS, Windows, Linux, and iOS. Supports 16-bit and 24-bit PCM input at 44.1, 48, and 96 kHz.
• S/PDIF Input: 24-bit PCM stereo audio at 44.1 or 48kHz. Channel status bits and LG SoundSync are decoded. LG Soundsync volume and mute status is optionally used to control the DSPi user volume and output mute.
• 24-bit S/PDIF or I2S Outputs: Up to four independent stereo output slots (8 channels on RP2350, 4 channels on RP2040). Each slot can be switched at runtime between S/PDIF and I2S, enabling direct connection to any standard DAC. I2S slots share a common BCK/LRCLK and can optionally produce a 128×/256× master clock.
• Per-Channel Preamp: Independent gain control for each USB input channel (L/R), applied as PASS 1 of the DSP pipeline before any other processing.
• Matrix Mixer: Route either or both USB input channels to any output with independent gain and phase invert per crosspoint. 2x9 on RP2350, 2x5 on RP2040.
• Parametric Equalization: Up to 10 PEQ bands per channel with 11 filter types (peaking, low/high shelf, low/high pass, notch, all-pass, plus gentle first-order shelf and all-pass variants). 110 total filter bands on RP2350, 70 on RP2040. RP2350 uses a hybrid SVF/biquad architecture for superior low-frequency accuracy.
• Active Crossovers: Up to 4 dedicated crossover filters per output channel: Linkwitz-Riley, Butterworth, and Bessel low/high-pass up to 8th order (48 dB/oct), for multi-way speaker and subwoofer integration.
• Volume Leveller: RMS-based, stereo-linked, soft-knee upward compressor that lifts quieter content toward a target level without ever amplifying loud passages. Optional 10 ms lookahead, configurable speed and max-gain ceiling, with a -6 dBFS gain-reduction safety limiter.
• Loudness Compensation: Volume-dependent EQ based on the ISO 226:2003 equal-loudness contour standard. Automatically boosts bass and treble at low listening levels to maintain perceived tonal balance.
• Headphone Crossfeed: BS2B-based crossfeed with interaural time delay (ITD) reduces unnatural stereo separation for headphone listening. Three classic presets plus fully custom parameters.
• Master Volume: Device-side output ceiling (-128 to 0 dB, with a true-mute sentinel) applied at the very end of the signal chain, independent of USB host volume and DSP processing. Two persistence modes: stored independently of presets (default — survives reboots, unaffected by preset switching) or saved/restored as part of each preset.
• Per-Output Gain & Mute: Independent gain and mute controls for each output channel.
• Time Alignment: Per-output delay (up to 85ms) for speaker/subwoofer alignment with automatic latency compensation between S/PDIF/I2S and PDM output paths.
• Subwoofer Output: Dedicated mono PDM output channel with a high-performance 2nd-order delta-sigma modulator, enabling direct subwoofer output without the need for a second DAC.
• Dual-Core DSP: EQ processing is split across both cores on both platforms for maximum throughput when multiple outputs are active.
• Configurable Output Pins: All output GPIO pins (including I2S BCK/MCK) can be reassigned at runtime to suit custom PCB layouts, no reflashing required.
• 10-Slot Preset System: Save, load, and manage up to 10 complete DSP configurations with user-defined names. Includes per-channel naming, configurable startup slot, and bulk parameter transfer for fast state synchronization.
• Diagnostics: Per-channel peak/clip metering, USB PHY error counters (CRC, bit-stuff, timeout, overflow, sequence), buffer fill statistics, S/PDIF DMA starvation counters per output slot, and CPU load reporting per core.
• Firmware Update via USB: A vendor command reboots the device into the UF2 bootloader, allowing the host app to push new firmware without a physical BOOTSEL press.

---

Platform Support

Both platforms are fully tested and production-ready. The RP2040 reaches 307.2 MHz with a slight voltage bump; the RP2350 hits the same frequency at the same voltage. Clock is fixed (no rate-dependent switching), and PIO dividers are integer at every supported sample rate. The RP2350 offers significantly more processing headroom thanks to its hardware floating-point unit, enabling more output channels and a hybrid SVF/biquad filter architecture for improved low-frequency accuracy.

---

Audio Signal Chain

DSPi processes audio in a linear, low-latency pipeline:

RP2350 (11 channels, 9 outputs):

USB Input (16/24-bit PCM Stereo, 44.1 / 48 / 96 kHz)
or
S/PDIF Input (24 bit PCM Stereo, 44.1 / 48 kHz)
    |
PASS 1: Per-Channel Preamp (independent L/R gain) + USB Volume
    |
PASS 2: Master EQ (10 bands per channel, Left/Right)
    |
PASS 2.5: Volume Leveller (RMS upward compression, optional)
    |
PASS 3: Headphone Crossfeed (BS2B + ITD, optional) + Master Peak Metering
    |
        Loudness Compensation (volume-dependent EQ, optional)
    |
PASS 4: Matrix Mixer (2 inputs x 9 outputs, per-crosspoint gain & phase)
    |
PASS 5: Per-Output EQ -> Gain/Mute -> Delay -> Output Gain × Master Volume
    |
    +-- Out 1-2 --> S/PDIF or I2S slot 0 (data: GPIO 6 default)
    +-- Out 3-4 --> S/PDIF or I2S slot 1 (data: GPIO 7 default)
    +-- Out 5-6 --> S/PDIF or I2S slot 2 (data: GPIO 8 default)
    +-- Out 7-8 --> S/PDIF or I2S slot 3 (data: GPIO 9 default)
    +-- Out 9   --> PDM Sub               (data: GPIO 10 default)
                  (I2S BCK/LRCLK shared on GPIO 14/15 default; optional MCK on GPIO 13 default)

RP2040 (7 channels, 5 outputs):

USB Input (16/24-bit PCM Stereo, 44.1 / 48 / 96 kHz)
or
S/PDIF Input (24 bit PCM Stereo, 44.1 / 48 kHz)
    |
PASS 1: Per-Channel Preamp + USB Volume
    |
PASS 2: Master EQ (10 bands per channel, Left/Right)
    |
PASS 2.5: Volume Leveller (RMS upward compression, optional)
    |
PASS 3: Headphone Crossfeed (BS2B + ITD, optional) + Master Peak Metering
    |
        Loudness Compensation (volume-dependent EQ, optional)
    |
PASS 4: Matrix Mixer (2 inputs x 5 outputs, per-crosspoint gain & phase)
    |
PASS 5: Per-Output EQ -> Gain/Mute -> Delay -> Output Gain × Master Volume
    |
    +-- Out 1-2 --> S/PDIF or I2S slot 0 (data: GPIO 6 default)
    +-- Out 3-4 --> S/PDIF or I2S slot 1 (data: GPIO 7 default)
    +-- Out 5   --> PDM Sub               (data: GPIO 10 default)
                  (I2S BCK/LRCLK shared on GPIO 14/15 default; optional MCK on GPIO 13 default)

Signal Chain Details

1. Input (USB): 16-bit or 24-bit PCM stereo audio at 44.1, 48, or 96 kHz. Bit depth is selected via USB alt setting; sample rate via the USB Audio Class rate-set request.
2. Input (S/PDIF): 24-bit PCM stereo audio at 44.1 or 48kHz. Channel status bits and LG SoundSync are decoded. LG Soundsync volume and mute status is optionally used to control the DSPi user volume and output mute.
3. Per-Channel Preamp (PASS 1): Independent gain control for the USB Left and Right input channels in dB. Applied at the very start of the DSP chain so its setting affects all downstream processing.
4. Master EQ (PASS 2): Up to 10 bands of parametric EQ per channel (Left/Right). Supports all PEQ filter types (see Parametric Equalization).
5. Volume Leveller (PASS 2.5): Optional feedforward, stereo-linked, single-band RMS compressor with soft-knee upward compression — quieter content is boosted toward a target level while content above the threshold passes through untouched. Configurable speed, max-gain ceiling, and noise gate. Optional 10 ms lookahead. A -6 dBFS gain-reduction safety limiter prevents output overshoots.
6. Headphone Crossfeed (PASS 3): Optional BS2B crossfeed that mixes a filtered, delayed portion of each channel into the opposite channel. Uses a complementary filter design with interaural time delay (ITD) via an all-pass filter. Three presets (Default, Chu Moy, Jan Meier) plus custom frequency and feed level. ITD can be independently toggled. Master peak metering taps into this stage.
7. Loudness Compensation: Optional ISO 226:2003 equal-loudness EQ that adapts to the current volume level. At low volumes, bass and treble are boosted to compensate for the ear's reduced sensitivity. Configurable reference SPL and intensity. Driven by the USB host volume position so it remains correct regardless of master-volume attenuation downstream.
8. Matrix Mixer (PASS 4): Routes the two USB input channels (Left/Right) to all output channels. Each crosspoint has independent enable, gain (-inf to +12 dB), and phase invert. Outputs can be individually enabled/disabled to save CPU. RP2350 has a 2x9 matrix (9 outputs), RP2040 has a 2x5 matrix (5 outputs).
9. Output EQ (PASS 5): Independent 10-band EQ per output channel on both platforms. Ideal for crossover filters and per-driver correction. On RP2350, filters below Fs/7.5 use SVF topology for superior low-frequency accuracy; higher frequencies use traditional biquad.
10. Per-Output Gain & Mute: Independent gain (-inf to +12 dB) and mute for each output channel.
11. Time Alignment: Per-output delay for speaker alignment, up to 85 ms (4096 samples at 48 kHz). Automatic latency compensation between S/PDIF/I2S and PDM output paths.
12. Master Volume: Device-side output ceiling, -128 to 0 dB with a true-mute sentinel at -128. Folded into the per-output multiplier at PASS 5 so it's effectively free CPU-wise. Independent of the USB host volume — the two multiply together. Does not affect loudness-compensation behavior.
13. Outputs: Each numbered slot is configurable as either 24-bit S/PDIF or 24-bit I2S (left-justified, MSB-first). I2S slots share a common BCK/LRCLK clock pair (LRCLK is always BCK + 1 due to a PIO side-set constraint). An optional master clock (MCK) at 128× or 256× Fs can be routed to a separate GPIO. PDM subwoofer is always on its own dedicated output and pin.

---

Hardware Setup

Flashing the Firmware

1. Download the latest DSPi.uf2 release for your board.
2. Hold the BOOTSEL button on your Pico while plugging it into your computer.
3. A drive named RPI-RP2 will appear.
4. Drag and drop the .uf2 file onto this drive.
5. The Pico will reboot and appear as a "Weeb Labs DSPi" audio device.
6. Download and launch the DSPi Console application to control the DSPi.

Wiring Guide

RP2350 (Pico 2) — up to 8 output pins:

RP2040 (Pico) — up to 6 output pins:

Warning: RP2040 input pins are not 5v tolerant, use only 3.3v external devices such as optical receivers.

Notes: S/PDIF input requires either a Toshiba TORX141 optical receiver or a suitable receiver circuit. Possible receiver circuits are described by ST Microelectronics here. S/PDIF output requires either a Toshiba TX179 optical transmitter or a simple resistor divider. I2S output is a standard 24-bit-in-32-bit left-justified frame — wires straight into most I2S DACs. PDM output is a 1-bit logic signal that requires a resistor and capacitor to form a low-pass filter for conversion to analog audio.

PCM1808 ADC Module on Pico 2 (I2S Input)

Use the PCM1808 in slave, standard-I2S mode: MD1 = LOW, MD0 = LOW, and FMT = LOW. In DSPi Console, select I2S as the input source, set RX data to GPIO 4, keep BCK on GPIO 14 (GPIO 15 is LRCLK), enable MCK on GPIO 13, and set the MCK multiplier to 256x. That gives the PCM1808 SCKI/MCLK frequencies of 11.2896 MHz at 44.1 kHz, 12.288 MHz at 48 kHz, and 24.576 MHz at 96 kHz.

PCM1808 modules vary: some expose separate VDD/3V3 and VCC/5V pins, while others have a regulator and only expose VCC. Follow the module silkscreen for power, keep all digital I/O at 3.3 V logic, and always share ground with the Pico 2.

Custom Pin Assignments

All default pin assignments above work out of the box, but every output pin — including I2S BCK and MCK — can be reassigned at runtime through the DSPi Console application. No reflashing required. This is useful when designing custom PCBs or adapting to boards where the default GPIOs are inconvenient.

Pin assignments are saved to flash and restored automatically at boot. A few GPIOs are reserved and unavailable for output use: GPIO 12 (UART TX) and GPIOs 23-25 (power control and LED). LRCLK is always pinned to BCK + 1 due to a PIO side-set constraint.

---

DSP Features

Matrix Mixer

The matrix mixer routes the USB stereo input to all output channels. RP2350 has a 2x9 matrix (9 outputs), RP2040 has a 2x5 matrix (5 outputs). Each crosspoint (input/output pair) has:

• Enable/Disable: Route active or inactive.
• Gain: -inf to +12 dB per crosspoint.
• Phase Invert: Polarity flip for driver alignment.

Each output channel also has:

• Enable: Disabled outputs skip all processing (EQ, delay, conversion) to save CPU.
• Gain: Per-output gain (-inf to +12 dB).
• Mute: Soft mute per output.
• Delay: Per-output time alignment.

Output Availability: Core 1 is shared between the PDM subwoofer modulator and the EQ worker that processes higher-numbered S/PDIF outputs. PDM and EQ worker modes are mutually exclusive:

RP2350:

RP2040:

When the PDM subwoofer is active, Core 1 is fully dedicated to the delta-sigma modulator, so higher-numbered S/PDIF outputs are unavailable. When PDM is off, Core 1 runs as an EQ worker processing those outputs in parallel with Core 0.

Common Configurations (RP2350):

Common Configurations (RP2040):

Parametric Equalization

Each band can be set to any of these 14 types. Every type uses a corner/center frequency; Q and gain apply only where listed (other types ignore them).

Frequency ranges from 10 Hz to 0.45xFs; Q from 0.1 to 20. First-order types are single-pole (6 dB/oct) and have no resonance.

On RP2040, all filters use biquad IIR (Transposed Direct Form II) with Q28 fixed-point arithmetic. On RP2350, the firmware uses a hybrid SVF/biquad architecture: filters below Fs/7.5 (~6.4 kHz at 48 kHz) use the Cytomic SVF (linear trapezoid) topology for superior numerical accuracy at low frequencies, while higher frequencies use traditional TDF2 biquad. Flat bands (and zero-gain peaking/shelf bands) are automatically bypassed for zero CPU overhead.

Channel Layout:

RP2350 (11 channels):

RP2040 (7 channels):

Crossover Filters

In addition to the parametric bands above, every output channel has up to 4 dedicated crossover filter bands for building active multi-way speaker and subwoofer systems. Crossover filters are low-pass or high-pass only and are tuned by corner frequency alone (Q and gain are not used). Three families are available:

Each crossover filter is built from a cascade of up to four biquad sections. On RP2350 the same hybrid rule applies per section (sections below Fs/7.5 use the Cytomic SVF for low-frequency accuracy). Crossover bands exist on output channels only, separate from and in addition to that channel's 10 parametric bands. A typical 2-way active speaker drives one output through a high-pass crossover (tweeter/mid) and another through a low-pass (woofer); a subwoofer output usually takes a low-pass.

Loudness Compensation

Based on the ISO 226:2003 equal-loudness contour standard. At low listening volumes, the human ear is less sensitive to bass and treble frequencies. Loudness compensation applies a volume-dependent EQ curve to maintain perceived tonal balance across all listening levels.

• Reference SPL: Configurable (40-100 dB). Set this to the SPL where your system sounds tonally balanced at full volume.
• Intensity: Adjustable from 0-200% of the standard ISO curve.
• Implementation: Precomputed coefficient tables for all 91 volume steps, double-buffered for glitch-free updates.

Headphone Crossfeed

Implements Bauer Stereophonic-to-Binaural (BS2B) crossfeed with a complementary filter design that reduces unnatural stereo separation for headphone listening. Each channel receives a lowpass-filtered, time-delayed mix of the opposite channel, simulating the acoustic crossfeed that occurs with loudspeaker listening.

• Complementary Design: Direct path = input - lowpass(input). Guarantees mono signals pass through at unity gain with no coloration.
• Interaural Time Delay (ITD): First-order all-pass filter adds ~220us of delay to the crossfeed path, modeling sound traveling around the head for 60-degree stereo speaker placement. ITD can be independently enabled/disabled.
• Presets:

Volume Leveller

A feedforward, stereo-linked, single-band RMS dynamic range compressor that maintains consistent perceived volume across content with varying loudness.

• Upward compression: Boosts content below the threshold while leaving content above the threshold completely untouched. No makeup gain needed.
• RMS-based detection: Tracks root-mean-square envelope, which correlates with perceived loudness better than peak detection.
• Soft-knee: Gradual transition between full boost and unity gain for transparent, artifact-free behavior.
• Stereo-linked: The louder of the two channels determines gain for both, preserving the stereo image.
• Gain-reduction safety limiter: -6 dBFS ceiling enforced via gain reduction (instant attack, 100 ms release) rather than hard clipping. Rarely engages since loud content passes through at unity.
• Optional 10 ms lookahead for smoother transitions.
• Configurable: speed (attack/release), max-gain ceiling (cap on how much quiet content can be lifted), and gate threshold (below which the leveller stops boosting to avoid amplifying silence/noise).

The leveller sits at PASS 2.5 — after Master EQ, before crossfeed. Independent of Loudness Compensation; both can be enabled together without conflict.

Per-Channel Preamp

Each USB input channel (Left and Right) has an independent preamp gain in dB, applied at PASS 1 before any other processing. Useful for trimming channel imbalance, attenuating hot inputs ahead of EQ, or matching levels across sources. A legacy single-value command remains for backward compatibility (sets both channels to the same value).

Master Volume

A device-side output ceiling applied at the very end of the signal chain, independent of USB host volume.

• Range: -128 to 0 dB. -128 is a sentinel for true silence (mute).
• Independent of USB host volume: the two multiply together. The host slider operates within whatever range master volume permits.
• Independent of DSP processing: loudness compensation, EQ, leveller, and crossfeed are all driven by the raw USB volume, not the master volume — their behavior is unchanged regardless of the master setting.
• Two persistence modes (selectable at runtime, persists across reboots):
  • Mode 0 — Independent (default). Master volume is a stand-alone device setting. The app calls a save command to capture the current value into the directory; that value is applied at every subsequent boot. Preset save/load do not touch master volume — switching presets never moves the volume.
  • Mode 1 — With preset. Master volume is part of each preset. Saved with the preset, restored on preset load, like any other DSP parameter. Useful when different presets target speaker setups with different sensitivity / maximum-output requirements.
• Default at first boot: -20 dB (configurable via MASTER_VOL_DEFAULT_DB in firmware).

I2S Output

Each output slot can be switched at runtime between S/PDIF (default) and I2S, independently per slot. A single device can drive a mix — e.g., slot 0 as I2S into a DAC chip, slot 1 as S/PDIF over Toslink to an external receiver, all from the same audio pipeline.

• I2S format: 24-bit data, left-justified, MSB-first, 32-bit frames. Drop-in to most standard I2S DACs (PCM5102, ES9038Q2M, etc.).
• Shared clocks: All I2S slots share a single BCK/LRCLK pair. LRCLK is always BCK + 1 (PIO side-set hardware constraint).
• Optional MCK: When enabled, a 128× or 256× Fs master clock is generated on a configurable GPIO. Required by some DACs that don't have an internal PLL. At 96 kHz, only 128× is selectable due to PIO clock-divisor limits.
• Sample-aligned start: I2S slots can be brought up together so multiple DACs stay phase-locked.

The DSP pipeline is identical for both output types — only the final encoding differs (BMC/NRZI for S/PDIF vs. raw left-justified PCM for I2S).

Subwoofer PDM Output

The subwoofer output uses a high-performance software-defined delta-sigma modulator running on Core 1.

• Modulation: 2nd-Order Delta-Sigma
• Oversampling Ratio: 256x (12.288 MHz bit clock at 48 kHz)
• Dither: TPDF (Triangular Probability Density Function) with noise shaping
• DC Protection: Leaky integrator design preventing DC offset accumulation

The objective was to use as much of Core 1 as necessary to produce an output that could be used full-range while sounding perfectly fine, even if it will only be used to feed a subwoofer. This implementation is very stable and without pops, clicks or idle tones.

---

User Presets

DSPi includes a 10-slot preset system that stores complete DSP configurations in flash. A preset is always active — there is no "no preset" state.

• 10 Preset Slots: Each slot stores the full DSP state: per-channel preamp, EQ bands, delays, loudness, leveller, crossfeed, matrix mixer, output gains/mutes, output type (S/PDIF or I2S), I2S clock configuration, pin assignments, master volume (used in Mode 1), and per-channel names.
• Per-Channel Names: Each channel can be given a user-defined name (up to 31 characters) that is stored with the preset.
• Startup Configuration: Choose which preset loads on boot — either a specific default slot or whichever slot was last active.
• Pin Config Inclusion: Optionally include or exclude GPIO pin assignments when saving/loading presets (default: include — pin layout travels with the preset).
• Master Volume Mode: Selects whether master volume is part of each preset (Mode 1) or stored independently in the preset directory (Mode 0, default). See Master Volume.
• Preset-Switch Mute: Audio output is briefly muted (~10 ms) during preset transitions to prevent audible glitches.
• Legacy Commands: The original save/load/reset commands (0x51-0x53) redirect through the preset system, operating on the currently active slot.
• Bulk Parameter Transfer: The complete DSP state can be read or written in a single USB control transfer (~2.9 KB) for fast synchronization with host applications.
• Auto-Migration: Older preset directories are upgraded transparently on first boot of new firmware — slot names, startup config, and other persisted state are preserved.

---

Developer Reference

System Architecture

• Core 0: USB communication, audio streaming, DSP processing (master EQ, crossfeed, loudness, matrix mixing, output EQ for S/PDIF pair 1), and control logic.
• Core 1 (three modes):
  • PDM Mode: Delta-sigma modulator for subwoofer output (when the PDM output is enabled).
  • EQ Worker Mode: Processes output EQ, delay, and S/PDIF conversion for higher-numbered outputs in parallel with Core 0. On RP2350: outputs 3-8. On RP2040: outputs 3-4. Activated when any of those outputs are enabled and PDM is disabled.
  • Idle Mode: When no outputs requiring Core 1 are enabled.
• PIO & DMA: Hardware offloading for S/PDIF encoding (PIO0) and PDM bitstream generation (PIO1) ensures zero CPU overhead for I/O.
• Math Engine:
  • RP2040: 32-bit fixed-point (Q28) processing with hand-optimized ARM assembly for the inner DSP loop.
  • RP2350: Single-precision float pipeline with hardware FPU. Hybrid SVF/biquad EQ — Cytomic SVF for low frequencies (below Fs/7.5), TDF2 biquad above. SVF provides superior numerical accuracy for low-frequency filters where biquad coefficient quantization becomes problematic.

Note: PDM mode and EQ Worker mode are mutually exclusive on Core 1. When the PDM output is enabled, Core 0 handles all S/PDIF output EQ processing. When PDM is disabled and higher-numbered outputs are active, Core 1 runs as an EQ worker for those outputs.

Performance Tuning

Both platforms run at a fixed 307.2 MHz system clock (VCO 1536 MHz / 5 / 1) so PIO dividers stay integer at every supported sample rate, eliminating sample-rate-dependent clock switching glitches.

The RP2040 reaches 307.2 MHz with a slight voltage bump above the 1.10 V nominal; the RP2350 is comfortable at the same voltage at this clock. The voltage step is applied before the frequency change. Sample rate changes do not retune the system clock, only the PIO dividers, so transitions between 44.1 / 48 / 96 kHz are seamless.

Flash access is also tuned: PICO_FLASH_SPI_CLKDIV is set to 6 to keep XIP and erase/program operations safely below the W25Q080's 104–133 MHz spec at this clock. On the RP2350, runtime QMI register management is handled by firmware/DSPi/flash_clkdiv.c since the bootrom does not honor the boot2 setting on that platform.

USB Control Protocol

Configuration is performed via Interface 2 (Vendor Interface) using Control Transfers under Windows and via Interface 0 under macOS. The device supports WinUSB/WCID for automatic driverless installation on Windows.

Request Table

REQ_GET_STATUS (0x50) - System Telemetry

The REQ_GET_STATUS request returns data based on the wValue field:

A starvation event means the S/PDIF DMA needed a buffer but the consumer pool was empty, so the firmware substituted a silence buffer for that transfer. This is a more direct output-side dropout signal than the older spdif_underruns USB-packet-gap heuristic.

Data Structures

Filter Packet (16 bytes):

struct __attribute__((packed)) {
    uint8_t channel;  // RP2350: 0-10, RP2040: 0-6
    uint8_t band;     // 0-9
    uint8_t type;     // 0=Flat, 1=Peak, 2=LS, 3=HS, 4=LP, 5=HP
    uint8_t reserved;
    float freq;       // Hz
    float Q;
    float gain_db;
}

Matrix Route Packet (8 bytes):

struct __attribute__((packed)) {
    uint8_t input;          // 0-1 (USB L/R)
    uint8_t output;         // RP2350: 0-8, RP2040: 0-4
    uint8_t enabled;        // 0 or 1
    uint8_t phase_invert;   // 0 or 1
    float gain_db;          // -inf to +12dB
}

Runtime Pin Configuration

Output GPIO pins can be reassigned at runtime without reflashing. This is useful for custom PCB layouts or when the default pin assignments conflict with other hardware.

REQ_SET_OUTPUT_PIN (0x7C) — IN transfer, returns 1-byte status:

• wValue = (new_pin << 8) | output_index
• RP2350: output_index 0-3 for S/PDIF outputs 1-4, 4 for PDM subwoofer
• RP2040: output_index 0-1 for S/PDIF outputs 1-2, 2 for PDM subwoofer
• S/PDIF outputs are automatically disabled and re-enabled during the pin change (~1ms audio dropout on that output only)
• PDM output must be disabled first (disable via REQ_SET_OUTPUT_ENABLE), otherwise returns PIN_CONFIG_OUTPUT_ACTIVE

REQ_GET_OUTPUT_PIN (0x7D) — IN transfer, returns 1 byte:

• wValue = output_index
• Returns the current GPIO pin number for that output

Pin assignments are stored in each preset and can optionally be included during preset save/load (controlled via REQ_PRESET_SET_INCLUDE_PINS).

USB Audio Loopback Capture (DSPI_LOOPBACK)

A build-flag-gated capture path for bench measurement and automated verification. When the firmware is compiled with DSPI_LOOPBACK, the device exposes a second USB Audio Class 1 function that streams output slot 0 back to the host as a recording input, so a host tool can play a signal into DSPi and record exactly what slot 0 produced. This folds the capability of the standalone Weeb Labs USBrx (an external S/PDIF-to-USB recorder) into DSPi itself, removing the second board and the S/PDIF jumper from the loopback rig. The feature is excluded from release builds; a production device never exposes the capture endpoint.

Building

The loopback build lives in dedicated build directories so release artifacts are unaffected:

# Configure once (from the repo root):
cmake -S firmware -B build-rp2040-loopback -DPICO_PLATFORM=rp2040       -DPICO_BOARD=pico  -DDSPI_LOOPBACK=ON
cmake -S firmware -B build-rp2350-loopback -DPICO_PLATFORM=rp2350-arm-s -DPICO_BOARD=pico2 -DDSPI_LOOPBACK=ON

# Build, then flash build-<plat>-loopback/DSPi/DSPi.uf2:
cmake --build build-rp2040-loopback -j
cmake --build build-rp2350-loopback -j

DSPI_LOOPBACK is a CMake option (default OFF); enabling it adds loopback.c to the target and defines the DSPI_LOOPBACK macro that gates every firmware change. With the flag off, the build is byte-for-byte identical to the normal firmware.

USB topology

The capture is a self-contained UAC1 audio function appended after the vendor interface, so the existing interfaces (0 Audio Control, 1 Audio Streaming, 2 Vendor) keep their numbers. The configuration descriptor grows from 207 to 309 bytes.

The host sees one USB device with both a playback output and a 2-channel capture input. On macOS these appear as two Core Audio entries that share the name "Weeb Labs DSPi" (one with output channels, one with input channels). The Microsoft OS 2.0 / WinUSB descriptors are unchanged: their Function Subset still scopes WinUSB to the vendor interface only, so adding the capture function does not disturb driverless vendor-interface binding on Windows.

Capture data path

output slot 0 (audio_pipeline.c, post-DSP, pre-encode)
  -> loopback_push_slot0() -> SPSC ring (1024 frames, ~8 KB, .bss)
      -> rate-matching servo (fill_audio_packet, per USB frame)
          -> iso async IN endpoint 0x81 -> USB host

• Tap. A single call in process_input_block() reads slot 0's finished, interleaved 24-bit samples from audio_buf[0] immediately before the buffer is handed to the output DMA. It runs after all DSP (EQ, crossover, matrix, gain, delay, mute) and before the output encoder, so it captures exactly what slot 0 sends out whether slot 0 is configured as S/PDIF or I2S. The tap is read-only: it copies samples out and never writes back, so it cannot perturb the firmware's inter-slot output alignment.
• Ring. A single-producer / single-consumer ring of 1024 interleaved stereo frames (two int32 per frame, ~8 KB in .bss). The producer is the audio pipeline; the consumer is the USB transfer-complete callback. Both run in core-0 thread context, so plain volatile head/tail indices are sufficient. On overflow (host not draining) the producer drops frames and the servo re-primes.

Rate-matching servo

The capture IN endpoint is asynchronous to the host's USB frame (SOF) clock in every input mode: with USB input, DSPi is the feedback master on its own crystal; with S/PDIF or I2S input, the output runs on the external source clock. A rate-matching servo therefore varies the packet size each USB frame (implicit feedback), with no resampling:

frames_this_frame = nominal + correction
  nominal    = audio_state.freq / 1000          (samples per 1 ms USB frame)
  correction = clamp(KP * (ring_fill - target), +/- MAX_CORR)

A fractional-sample accumulator carries the remainder across frames so non-integer rates (44.1 kHz) stay exact. The stream primes to the target fill level before sending audio and emits silence on underrun.

Driver registration

The capture function uses a dedicated custom UAC1 class driver (loopback.c), registered alongside the playback driver. With DSPI_LOOPBACK defined, usbd_app_driver_get_cb() returns both drivers. Because both audio-control interfaces match the same class / subclass / alt-setting, each driver's open() is scoped by interface number: the playback driver claims interface 0 only and the loopback driver claims interface 3 only, so neither hijacks the other's function. All capture control requests (SET / GET_INTERFACE, the endpoint sampling-frequency control) and the isochronous transfers route to the loopback driver.

Resource cost

The capture ring and buffers add roughly 8.7 KB of BSS on both platforms (RP2040 129436 to 138120 bytes; RP2350 237404 to 246092 bytes). Because the firmware is built copy_to_ram, this BSS competes with the program image in RAM; the RP2040 loopback build has about 6 KB of headroom, which is acceptable for a debug build and is part of why the feature is excluded from release. Capture rates are capped at DSPi's operating range (44.1 / 48 kHz), which also keeps the extra isochronous endpoint within the RP2040 USB controller's DPRAM budget.

Host usage

Any class-compliant recording tool can open the DSPi capture input. The repository's verification harness drives it directly: python3 -m tools.dspi_test.run --group audio plays signals into DSPi, records slot 0 from the capture, and checks the measured response against the firmware's filter math. See tools/dspi_test/test_harness.md for the full rig guide and Documentation/current_architecture.md ("USB Audio Loopback Capture") for additional firmware detail.

Files: firmware/DSPi/loopback.c / loopback.h (ring, driver, servo), with DSPI_LOOPBACK-gated additions in usb_descriptors.c / .h (descriptor block), usb_audio.c (driver registration and interface scoping), audio_pipeline.c (the slot-0 tap), and CMakeLists.txt (the build option).

---

Building from Source

To build the firmware yourself, you'll need a standard Raspberry Pi Pico C/C++ development environment.

1. Install Prerequisites

Ensure you have the following tools installed:

• CMake (3.13 or newer)
• Arm GNU Toolchain (arm-none-eabi-gcc, etc.)
• Python 3 (for Pico SDK scripts)
• Git

2. Clone the Repository

Clone the project recursively to include the Pico SDK and other submodules:

git clone --recursive https://github.com/WeebLabs/DSPi.git
cd DSPi

If you already cloned without --recursive, run:

git submodule update --init --recursive

3. Build the Firmware

You can build for either the standard RP2040 (Raspberry Pi Pico) or the newer RP2350 (Raspberry Pi Pico 2). The build system uses separate directories to avoid conflicts.

Option A: Build for RP2040 (Standard Pico)

mkdir build-rp2040
cd build-rp2040
cmake -DPICO_BOARD=pico -DPICO_EXTRAS_PATH=../firmware/pico-extras ../firmware
make

Output: DSPi/DSPi.uf2

Option B: Build for RP2350 (Pico 2)

mkdir build-rp2350
cd build-rp2350
cmake -DPICO_BOARD=pico2 -DPICO_EXTRAS_PATH=../firmware/pico-extras ../firmware
make

Output: DSPi/DSPi.uf2

4. Flash the Device

1. Hold the BOOTSEL button on your board while plugging it in.
2. Drag and drop the generated .uf2 file onto the RPI-RP2 (or RP2350) drive.

Alternatively, an already-running DSPi can be put into bootloader mode without a button press by sending REQ_ENTER_BOOTLOADER (0xF0). The DSPi Console application uses this for one-click firmware updates. See Documentation/Features/firmware_update.md for the protocol details.

---

Detailed Specifications

In-depth specs for each subsystem are kept under Documentation/Features/. These are the authoritative source for protocol formats, wire layouts, edge cases, and host-app integration patterns.

---

License

This project is licensed under the GNU General Public License v3.0. See LICENSE for details.