Skip to content

RØDECaster Duo audio routing (1.7.6 and 1.7.7) ​

This note explains how the experimental Duo route layout, duo-23-v1, was derived from official firmware, and why an earlier draft had fewer paths.

All of it comes from static analysis and ARM64 emulation of official RØDE files. No Duo was contacted, flashed or used for audio. The note does not show that a real Duo produces sound on these ports in every USB mode or after every reconnect.

Sources ​

ItemValue
1.7.6 archivehttps://update.rode.com/rcp2-duo/1.7.6/os.tar.gz, SHA-256 47d2ccb0d613b223ca6d5ed57c62ae6ee32ba71b662be5fbfb4ad3f44c757daa, MD5 50fb0c434ea5d4b643f7ff5a38253fe5
1.7.7 archivehttps://update.rode.com/rcp2-duo/1.7.7/os.tar.gz, SHA-256 4f3c7a022f07fd5b05955846e6a9052ad6ae214e12c04cda2d4749d775a304ed, MD5 09393fef8a9f3b627a39801148d85d1d
1.7.6 rc_audio_mixer897b9e25919d826ade563cad89a5acd140541b274b74524f655273488089f48c (same as Pro II 1.7.6)
1.7.7 rc_audio_mixer5151d0eb843dab46f441da4c2cca66f1636bfe522b2302345c17f28ecbffd0dd (same as Pro II 1.7.7)
audio-start.sh, bothf27368e6d68296ebb9c6f22401f2fe00317643bee23c08b86d44d183408fa263

In both releases the mixer program is byte-identical to the Pro II's. The board files (hostname, device tree, kernel, modules and controller firmware) are the Duo's, so neither archive can be used on the other model.

The analysis used the 1.7.7 beta archive. The stable 1.7.7 archive, checked again on 2026-11-03, is byte-identical to it.

The evidence files are byte-for-byte copies from .deps/duo-investigation/:

  • docs/firmware-audits/rodecaster-duo-1.7.7.json is the hook audit run with --baseline-id rodecaster-duo-1.7.6 --candidate-id rodecaster-duo-1.7.7. Nine whole functions are relocation-equivalent across the releases: the callback, enqueue, manager, invoker, job body, Mixer::Process, the function copy constructor, JACK registration teardown and active registration. Both ARM64 job-lifetime fixture sets pass. The hook offsets therefore match the corresponding Pro II releases. Only the route layout and source enum differ.
  • docs/firmware-audits/rodecaster-duo-routes.json is the route audit described below, for both images, including the source-enum table.

Paths in duo-23-v1 ​

JACK port numbers start at 1 (JUCEJack:in_3). The mixer's internal packed buffer arrays start at 0, so packed index 4 is a different buffer from out_4. See why outputs were held back.

RC2FX pathSaved IDMixer ports
Mic 1, Mic 2 (combo inputs)duo-analog-1, duo-analog-2in_1, in_2
Headset micduo-headset-micin_3
Bluetoothduo-bluetoothin_5/in_6
USB 1duo-usb-mainin_7/in_8
USB 2duo-usb-secondaryin_9/in_10
Chatduo-usb-chatin_11/in_12
Gameduo-usb-gamein_13/in_14
Musicduo-usb-musicin_15/in_16
Virtual Aduo-usb-ain_17/in_18
Virtual Bduo-usb-bin_19/in_20
CallMe 1-3duo-callme-1..3in_29..in_34
Wireless 1, 2duo-wireless-1, duo-wireless-2in_35, in_36
Headphones 1duo-headphones-1out_3/out_4
Headphones 2duo-headphones-2out_1/out_2
Headsetduo-headsetout_8 (left), out_7 (right)
Monitorsduo-monitorsout_5/out_6
USB 1 sendduo-send-mainout_15/out_16 to usb1out:playback_1/2
Chat sendduo-send-chatout_17/out_18 to usb1-chat-out:playback_1/2
USB 2 sendduo-send-secondaryout_19/out_20 to usb2out:playback_1/2

The table is DUO_PATHS in crates/rc2fx-device/src/catalog.rs. Every saved ID starts with duo-, so a Pro II preset cannot bind a chain to a Duo socket that is wired differently. The four virtual USB inputs were added to the original 19 paths, as described in USB path qualification. SMART Pads and the extra per-channel multitrack USB outputs are not in the layout.

The saved-show source enum is duo-sources-v1 (named in crates/rc2fx-firmware/src/model.rs and projected in crates/rc2fx-device/src/stock_inputs.rs). It has seven faders, and source 2 is labelled "Headset mic". The vendor code also gives source ID 2 its own digital-gain and phantom bypass (1.7.6 0x40a27c, 0x3a20d8 and nearby).

Source-enum table ​

The model-aware source-change lookup (1.7.6 0x40a360, 1.7.7 0x40cfd0) is relocation-equivalent as a whole function across the two releases. The route audit runs the table initializer that the lookup reads (1.7.6 0x135730, 1.7.7 0x1342a8) and checks all 31 entries. Both releases build the same 31 triples of a source ID and two internal slots, ending with a -1 sentinel. The report has the full table and its addresses.

The internal slots are the mixer's own source indices. They are unrelated to JACK port numbers, and not every enum value is an RC2FX processing path.

How each group was found ​

Addresses are 1.7.6 ELF virtual addresses unless marked. The route audit lists the 1.7.7 equivalents.

Board check ​

0x2572f0 (1.7.7 0x259640) reads /boot/extlinux/extlinux.conf and returns 1 for the Duo's "mini" device tree. The branches below depend on that result.

Analog inputs and headset mic ​

In Mixer::Process (0x657250, 1.7.7 0x657470), the Duo branch at 0x657698 jumps to 0x659008, which keeps the third hardware input and silences the fourth. The Duo's combo inputs are the first two capture ports, and the headset mic is the third.

Headphones and headset ​

The headphone branch 0x658830 jumps to 0x659014. Headphone 1 (mixer struct +0xc2040) writes packed outputs 2 and 3 (out_3/out_4). Headphone 2 (+0xc2048) writes packed 0 and 1 (out_1/out_2). The code at 0x659154..0x65918c copies headphone 1's left channel to packed 7 and its right channel to packed 6. The headset therefore carries the headphone 1 mix with the buffers swapped: out_8 is left and out_7 is right.

Digital inputs and USB sends ​

The JACK port-event handler (0x3716e0, 1.7.7 0x374410) connects ports by literal name. The route audit runs its real instructions in Unicorn, stubbing JUCE strings, allocations and the JACK connect call, and records the arguments passed to connect. Both releases produce exactly these connections:

Mixer portUSB endpoint
JUCEJack:out_15, out_16usb1out:playback_1, _2
JUCEJack:out_17, out_18usb1-chat-out:playback_1, _2
JUCEJack:out_19, out_20usb2out:playback_1, _2

The same handler connects the inputs (usb1in, usb1-chat-in, usb2in, btphone, RodeRfSPI), listed in the Duo reference. CallMe uses the same input ports as on the Pro II.

Monitors ​

The mixer constructor stores three monitor routing fields. The Pro II default store (0x66b494, 1.7.7 0x67488c) sets 10, 4 and 7. The Duo override (0x66f77c, 1.7.7 0x678ad4) sets 8, 4 and 5. The route audit finds each store by its instruction pattern and runs it for both board results.

The job body at 0x41bcb0 (1.7.7 0x41ecf0) then copies the monitor left and right samples into packed outputs 4 and 5. Run in Unicorn against eight hardware output buffers, it writes the expected samples into packed 4 and 5 (out_5/out_6) and leaves the headphone and headset buffers alone, in both releases. The 1.7.7 constructor differs at the instruction level from 1.7.6's, so the field stores were located one by one.

Why outputs were held back ​

An earlier Duo layout had 15 paths and left out Monitors and the three USB sends, so RC2FX offered no processing on those outputs. They still played the stock mix, because RC2FX leaves ports outside its layout alone.

The outputs were left out because of a numbering question. The monitor evidence above is in packed buffer indices, and which out_N a packed index reaches depends on which hardware buffers the JACK active mask (0x61cc90) enables. The mapping from mask to port was unknown, so the outputs were excluded until it could be established.

The USB sends do not depend on that mapping. The port-event handler connects named mixer ports directly to named USB endpoints, so the emulated connect calls answer the question. For Monitors, emulating the constructor stores and the job body fixes the producer side at packed 4 and 5, which is out_5/out_6 when the Duo's eight hardware outputs are active.

The native pipeline adds a runtime guard. It binds every path by port name, without falling back to packed indexes, and processes a stereo output only when both of its named buffers appear in the vendor callback. That prevents processing the next packed buffer or only one channel. It does not check which physical socket a port is wired to, and both names being present does not prove the stereo mix is correct. As with every output, the Duo's hardware outputs and USB sends are processed in the audio callback.

Open questions ​

  • Monitors in every mode. The monitor fixture assumes eight active hardware buffers. The shared startup callback also contains the Pro II's balanced-monitor rewiring, and nobody has observed a real Duo's JACK graph at startup or after a reconnect. The out_5/out_6 mapping is the firmware's intended producer mapping, backed by the runtime guard. If the guard does not see both buffers, Monitors stays unprocessed.
  • Live behaviour. The fixtures stub JUCE strings, allocations, events and JACK connections, and skip PAC/BTI hint instructions.
  • Hardware. No flashing, recovery, audio loopback, timing, CPU load or USB mode test has been done on a Duo.

Reproducing ​

Both commands read the official archives and change nothing. Set up a virtualenv with tools/firmware_audit/requirements.txt first.

sh
python -m tools.firmware_audit \
  --baseline PATH/duo-1.7.6-os.tar.gz --candidate PATH/duo-1.7.7-os.tar.gz \
  --baseline-id rodecaster-duo-1.7.6 --candidate-id rodecaster-duo-1.7.7 \
  --output duo-runtime-audit.json

python -m tools.firmware_audit.duo_routes \
  --baseline PATH/duo-1.7.6-os.tar.gz --candidate PATH/duo-1.7.7-os.tar.gz \
  --output duo-routing-audit.json

Both stop on any hash, instruction or fixture mismatch. Each report has a limitations field describing what it does not cover.

Verification receipt ​

These offline checks were run when the layout was added, with no device:

  • The firmware, device and installer unit tests pass.
  • The native library tests pass. They include a stereo output with only one channel observed, where both channels are left untouched, and a missing out_6, which is not replaced by out_7.
  • cargo xtask mixer-build passes its AArch64, glibc 2.35 and GCC 11 checks.
  • A temporary image built from the official Duo 1.7.7 archive keeps the boot and controller content and headers, the exact mixer and launcher hashes and the Duo device tree. It carries the runtime opt-in marker and fits the 275 MiB size limit.