Audio Output

The path

[CODE] Open firmwares in the Felucca family output audio through the chip’s ALNK0 block (I2S-style audio link, registers at 0x12E00), fed by double-buffered DMA:

  • The firmware owns one buffer split into two halves. Each half holds N stereo frames as int32_t pairs (L, R), 24-bit, left-justified.
  • When the DMA finishes one half, ALNK0 raises its interrupt (FM1_IRQ_ALNK0). The interrupt handler fills the half the DMA just left, while the DMA plays the other one.
  • Felucca uses 256-frame halves at 44.1 kHz: 5.8 ms per half. That is your hard deadline per block.
/* the shape of the ALNK0 handler in Felucca-family firmware (GPL-3.0-only) */
void fm1_alnk0_irq(void)
{
    uint8_t p = fm1_audio_pending();
    fm1_audio_ack_aux(p);
    if (p & FM1_AUDIO_HALF) {
        uint32_t half = fm1_audio_free_half();
        render_block(&abuf[half * HALF_WORDS]);    /* must finish inside one half: 5.8 ms */
        fm1_audio_ack_half();
        if (fm1_audio_free_half() != half)
            xruns++;                               /* the DMA moved on while we rendered: a dropout */
    }
}

The last check is a cheap, reliable dropout counter: if the free half has changed by the time you finish, you were late. Show it on screen during development (Debugging).

Bring-up

Felucca’s fm1_audio_init() drives four GPIOs on port C (PC0, PC1, PC2, PC6) around the ALNK0 set-up, with 5 ms waits: these switch the output stage on (PC6 low first, then high once the link runs). The exact role of each line is not written down anywhere we know of. [CODE] [?]

Stop the DMA (fm1_audio_stop()) before a crash screen or before entering the ROM’s download mode; otherwise the last buffer loops as a buzz. Zero the buffer before flash erases, which run with interrupts off. [CODE]

What is on the other end: disputed

Source Says Tag
Felucca’s hal/fm1_audio.h “ALNK0 I2S → external codec” [CODE]
FM-1-RE “internal audio DAC”, the stock firmware running its audio server at 44118 Hz with 64-sample blocks [RE]
FM-1 resource hub §1.2 an external codec on I2S: PC0 MCLK, PC1 BCLK, PC2 LRCK, PC6 the codec’s power/reset; the chip’s own DAC (DACL pin) as an auxiliary line. From board photos and JieLi’s reference schematics; no codec part number given. [?]

Both can be partly true (an I2S link to an on-chip or off-chip converter). The part number, whether there is a separate codec, and the true sample rate under each clock setting are in Open Questions. If you know, please add a photo or a measurement.

Sample rate

Felucca-family firmware assumes 44,100 Hz. The stock firmware’s 44,118 Hz suggests the rate is derived from a clock that doesn’t divide exactly. A measured rate (e.g. record a long sine and compare with the computer’s clock) would settle it. [?]

USB audio

[CODE] Felucca 1.0 and fm1-base can expose the master output as a USB Audio Class 1 input (16-bit stereo, 44.1 kHz, isochronous asynchronous IN), so a computer can record the FM-1 over the cable. fm1-base also supports the other direction (the computer plays through the FM-1). The USB packet service must run every 1 ms frame, so in Felucca-family firmware it runs nested inside the audio render from the 10 kHz timer interrupt.

kmplngj’s Felucca fork offers the USB audio input at 44.1 or 48 kHz (FELUCCA_UAC_48K, on by default; 44.1 kHz only up to 1.0.5.2), per its BUILDING.md. [CODE]

Choosing a block size

Shorter halves give lower latency and less time per block. Longer halves give more slack against an occasional heavy block (a note-on, a parameter change), at the cost of latency. Felucca’s 256 frames (5.8 ms) is the common choice. Whatever you pick, measure the worst block, not the average: see Performance.

Sources


Unofficial community wiki. Not affiliated with M-VAVE or JieLi. Text under CC BY-SA 4.0; code excerpts keep their own licences (License). Facts are tagged [HW] [SDK] [RE] [CODE] [?]: see Home. Found an error? See Contributing. Home