Add DShot module #12854

bkueng · 2019-08-31T12:15:24Z

This brings #12634 from @Igor-Misic and adds a new output module that uses the low-level DShot interface. The PR includes the commits from #12790.

The dshot module is implemented as drop-in replacement for fmu, which means it handles the same modes so that other functionality is not lost - the duplication is however not nice.

I've flown this on the Omnibus BetaFPV tiny whoop with DShot600.

DShot is enabled on OmnibusF4SD and Pixracer.
Caution: Before flying this on a Pixracer, we need to double-check that there are no DMA resource conflicts.

Missing/next steps:

Support for F7 (including DMA-able memory and flushing caches). @Igor-Misic is that something you're still interested in/ already looking into?
Enable on all other boards.
More testing, also on boards with IO.
DShot Telemetry

Igor-Misic · 2019-08-31T14:38:14Z

@bkueng, first congrats on this amazing work you have done so far, and for the first DShot PX4 flight :)

I went through F7 DMA drivers (Nuttx DMA for F7) and I think this may work just as it is. F4 and F7 share the same registers for DMA. As you helped me on Slack to understand FMUv5 boards I added defines to skip that unsupported IO Processor (F1 MCUs) building Dshot for them. Firmware is building for F7 boards, and it should use their own drivers (nuttx/arch/arm/src/stm32f7/smt32_dma.h) for DShot drivers.
Unfortunately, I don't have any F7 board to run and check it.

dagar · 2019-08-31T14:44:52Z

The dshot module is implemented as drop-in replacement for fmu, which means it handles the same modes so that other functionality is not lost - the duplication is however not nice.

One idea I was going to explore was continue splitting all the px4fmu functionality into separate minimal modules and rely on the underlying channel allocations in io_timer to prevent conflicts. Pin ownership would then become first come, first served.

drivers/px4fmu split into:
      -> drivers/pwm_out
      -> drivers/input_capture
      -> i2c and peripheral reset -> drivers/boardctl (or almost anywhere else...)

Instead of modes the fmu module (now simply pwm) would try to take the required pins when loading a mixer. If there's a conflict we can tell the user exactly what's happening (trying to load a mixer with 6 outputs and pins 5 & 6 already taken for gpio, etc) and either completely abort, or only use a subset.

Pros

smaller easier to understand modules
new standalone input_capture module could be used for fmu-v5 CAP pins as well
pwm module could potentially then be consolidated with linux_pwm_out (each linux board implementing a tiny PWM hal)
get rid of awkward rcS pin configuration
https://github.com/PX4/Firmware/blob/f0ee0b5d495224cdc517c216bd304094df9f0817/ROMFS/px4fmu_common/init.d/rcS#L354-L369

Cons

likely harder to convey the overall pin configuration to a user for configuration in QGC than if we went for a single unified px4fmu
additional constraints on timer boundaries we might need to respect

hamishwillee · 2019-09-05T03:48:42Z

@bkueng Can you please ping me when Dshot is in a documentable form - I presume this is not going into PX4 v1.10?

In the docs we we explicitly state that we don't support it in the docs - so we would need to remove that. If there is any explicit config or different hardware set up we'd want to capture that too.

bkueng · 2019-09-05T14:21:47Z

@bkueng, first congrats on this amazing work you have done so far, and for the first DShot PX4 flight :)

Thanks! The larger portion of the credits goes to you, without the lower level parts it would not be possible.

I went through F7 DMA drivers (Nuttx DMA for F7) and I think this may work just as it is.

Indeed, it does! I expected more work would be needed.

@dagar yes that might work.

@hamishwillee sure. The short answer: as soon as this goes in it's usable (we need to enable it on all boards though).

Updates

added F7, enabled KakuteF7
added (indirect) telemetry via UART and publish esc_status
added CLI commands, for example:
- let the ESC beep: dshot beep1 -m 1
- permanently reverse motor direction (no more ESC resoldering :)
```
dshot reverse -m 1
dshot save -m 1
```
- retrieve ESC infos: dshot esc_info -m 2:

nsh> dshot esc_info -m 2
INFO  [dshot] ESC Type: #TEKKO32_4in1#
INFO  [dshot] MCU Serial Number: 480022-001751-564e34-323420
INFO  [dshot] Firmware version: 32.60
INFO  [dshot] Rotation Direction: normal
INFO  [dshot] 3D Mode: off
INFO  [dshot] Low voltage Limit: off
INFO  [dshot] Current Limit: off
INFO  [dshot] LED 0: unsupported
INFO  [dshot] LED 1: unsupported
INFO  [dshot] LED 2: unsupported
INFO  [dshot] LED 3: unsupported

Known Issues

Problems I came across while testing:

Sometimes (quite rare) upon disarming the ESCs re-initialize, so there is something about the signal they don't like. I see this on 2 completely different quads (F7 vs F4 and different ESCs), so it's unlikely a HW-problem.
Telemetry and sending of commands is not 100% reliable. Sometimes a command is not executed, and telemetry requests regularly and randomly time out. A cross-test with betaflight sending commands on the same HW does not show that issue.

hamishwillee · 2019-09-05T22:50:34Z

@hamishwillee sure. The short answer: as soon as this goes in it's usable (we need to enable it on all boards though).

And the long answer? :-) Because I take that to mean that these instructions will work without any change http://docs.px4.io/master/en/peripherals/pwm_escs_and_servo.html

bkueng · 2019-09-17T18:30:39Z

This can be tested on Pixracer now, e.g. by setting DSHOT_CONFIG to 1200.

mhkabir · 2019-09-23T18:46:13Z

@bkueng How can we get this working on Pixhawk 4?

src/drivers/dshot/module.yaml

bkueng · 2019-09-24T18:02:36Z

@bkueng How can we get this working on Pixhawk 4?

I enabled it on v5, but only for the first 4 outputs (the next ones conflict with px4io
serial DMA (UART8_RX)). I had to disable RX DMA on the GPS port - can you test GPS as well?

mhkabir · 2019-09-24T18:13:55Z

OK, let me have a look , thanks :)

platforms/nuttx/src/px4/stm/stm32_common/dshot/dshot.c

bkueng · 2019-09-27T12:07:32Z

I've flown this quite a bit on Pixracer and KakuteF7 (Kopis 2 frame with telemetry), so that I'm confident it's working.

Flight Performance

On both vehicles I've flown (250 size), I noticed considerably smoother flight performance - both audibly and visibly. Prop-wash handling is noticeably improved.
Part of this is certainly due to the 1kHz rate loop, and I have not (yet) checked how much of it is to attribute to DShot.

Boards

DShot is enabled on these boards:

Pixhawk (v3): enabled on all outputs, disabled RX DMA on TEL4 and IO debug port
Pixhawk 4 (v5): enabled only on first 4 outputs, disabled RX DMA on GPS port
Pixracer (v4): enabled on all outputs, disabled RX DMA on FrSky telem port
OmnibusF4SD: enabled on all outputs, disabled RX DMA on USART1
KakuteF7: enabled on all outputs, disabled RX DMA on UART4 (GPS port)

Issues

Issues from above:

Sometimes (quite rare) upon disarming the ESCs re-initialize, so there is something about the signal they don't like. I see this on 2 completely different quads (F7 vs F4 and different ESCs), so it's unlikely a HW-problem.

The problem is the flash-based param saving: in some cases the entire CPU stalls for ~300ms, which causes DShot not to update anymore (on PWM and Oneshot, the timers continue to trigger). If in-air, the motors might stop, so the param auto-saving is now deferred until disarming for flash-based param storage.

Telemetry and sending of commands is not 100% reliable. Sometimes a command is not executed, and telemetry requests regularly and randomly time out. A cross-test with betaflight sending commands on the same HW does not show that issue.

Fixed by updating the timer config before triggering the DMA update.

Logs

With a Pixracer:

Other

CPU load is about 1% higher and ~1KB more RAM is used with DShot vs PWM (RAM is due to the DMA buffers)
If for some reason the DShot signal stops updating while motors are spinning (e.g. actuator topic is not updated), the motors slowly stop and disarm.
There is a pending refactoring from @Igor-Misic to use NuttX DMA methods, but it still has issues, so we can merge that separately.
Once we use ITCM RAM on F7, we'll need to use DMA memory allocators for the DShot buffers.

@PX4/testflights can you test this on Pixhawk (v3), Pixracer and Pixhawk 4 on vehicles with ESCs that support DShot? You can follow the instructions here: https://github.com/PX4/px4_user_guide/pull/578/files#diff-58ea316b3b439c16b17ab4f924adc37a. Be careful, the vehicle might need some re-tuning.

I'm not sure if we want to pull this into v1.10?

dagar · 2019-09-27T14:44:21Z

Once we use ITCM RAM on F7, we'll need to use DMA memory allocators for the DShot buffers.

We're currently using non-DMA capable memory in the heap on F7.

https://github.com/PX4/Firmware/blob/4f71c4e123a627ce0724b8065089c8e86038aebc/boards/px4/fmu-v5/nuttx-config/scripts/script.ld#L76

On both vehicles I've flown (250 size), I noticed considerably smoother flight performance - both audibly and visibly. Prop-wash handling is noticeably improved.
Part of this is certainly due to the 1kHz rate loop, and I have not (yet) checked how much of it is to attribute to DShot.

It would be interesting to compare a few flights at different loop rates. You can do this now with the IMU_GYRO_RATEMAX parameter. With a bit more work (drivers cleanup + drdy, standalone rate controller, optimization in a few key places) we should be able to get to 2 kHz "soon".
Related - betaflight/betaflight#7327 (comment)

And handle failures of up_dshot_init(). On Omnibus: reduces memory usage if dshot is enabled by ~1.0KB. The buffer is roughly 1KB in size.

No other functional change, just restructuring.

This reloads the timer configuration before triggering DMA. Without that, in rare cases, there were 17 bits sent instead of 16. The 1. bit (1. pulse) was always wrong (too much), the rest of the bits were the correct DShot packet that was meant to be sent.

But only on the first 4 FMU outputs, as the next ones conflict with px4io serial dma (UART8_RX) RX DMA is disabled on the GPS port as well (conflicts with TIM1).

This is especially important for DShot, that does not update when the CPU stalls.

Disables RX DMA on TEL4 and IO debug serial port

MaEtUgR · 2019-10-11T13:45:25Z

This is so awesome!!! 🚀

I come from flying the Holybro Kopis 2 with the new DSHOT support and it flies that good now! It's really confidence-inspiring.
Next time I'll bring a GoPro but here's the eye cancer stream such that you get an impression:
https://youtu.be/XDT9YaBVlEs
Here are the two flight logs:
https://logs.px4.io/plot_app?log=786298ac-1b19-4019-be56-130c64a1baf0
https://logs.px4.io/plot_app?log=4c7c0414-e84e-47d6-9116-6fda9c0cb67e

Thanks a lot to @bkueng @dagar @Igor-Misic and everyone else who helped for doing all these low-level improvements that will drastically improve any low latency actuated vehicle's flight performance!

thomasderflieger · 2019-10-15T06:23:29Z

Since being spoiled from flying racing-quadrocopters with DSHOT and telemetry for quite some time i am very happy to see those features being implemented into PX4. Thank you!

I'm currently building a VTOL with the CUAV V5 Nano Flight Controller. It has 3 Motors and 5 Servos, wich is also exactly the number of PWM outputs of the FC. It is similar to the Convergence-airframe, but i can also tilt the rear Motor.

I enabled it on v5, but only for the first 4 outputs (the next ones conflict with px4io
serial DMA (UART8_RX)). I had to disable RX DMA on the GPS port - can you test GPS as well?

Doas that mean if i enable DSHOT the first 4 outputs are going to use DSHOT or doas it depend on the motor-count. If it is enabled on the first 4 outputs, i could not use nr. 4, but then im missing an PWM output for one of my Servos.

The second question is regarding logging. Is it possible to log the ESC telemetry data and add it to the logfiles? That could help with analysis a lot.

hamishwillee · 2019-10-15T22:49:06Z

@thomasderflieger Docs are here: http://docs.px4.io/master/en/peripherals/dshot.html. You wire according to your mixer file/airframe reference: http://docs.px4.io/master/en/airframes/airframe_reference.html. Ie it is enabled for whatever motors you have connected, but not for servos.

@bkueng Would need to comment on logging.

bkueng · 2019-10-16T05:48:02Z

Doas that mean if i enable DSHOT the first 4 outputs are going to use DSHOT or doas it depend on the motor-count. If it is enabled on the first 4 outputs, i could not use nr. 4, but then im missing an PWM output for one of my Servos.

It's indepenent on the motor count and the current implementation does not allow for mixing DShot with PWM outputs on the FMU pins (on a board with an IO you could use PWM on the IO pins and DShot on the FMU pins).

The second question is regarding logging. Is it possible to log the ESC telemetry data and add it to the logfiles? That could help with analysis a lot.

It already is. It is the esc_status message.

thomasderflieger · 2019-10-16T07:30:49Z

It's indepenent on the motor count and the current implementation does not allow for mixing DShot with PWM outputs on the FMU pins (on a board with an IO you could use PWM on the IO pins and DShot on the FMU pins).

Thats a pitty, was looking forward to using dshot for my VTOL, my current flight controller just has main-ports though. I assume that the Telemetry wont work without using dshot either, since the ESC's need a "telemetry-request message"?

It already is. It is the esc_status message.

Great, thanks!

bkueng · 2019-10-16T08:02:03Z

Thats a pitty, was looking forward to using dshot for my VTOL, my current flight controller just has main-ports though. I assume that the Telemetry wont work without using dshot either, since the ESC's need a "telemetry-request message"?

Exactly. Yeah, that's a drawback, but with all configuration options and timer restrictions, it would become really complex.

ndepal · 2019-10-16T16:40:09Z

It's indepenent on the motor count and the current implementation does not allow for mixing DShot with PWM outputs on the FMU pins (on a board with an IO you could use PWM on the IO pins and DShot on the FMU pins).

Just do be sure, does that mean that it's not possible to use camera triggering (GPIO) on a PixRacer together with DShot?

bkueng · 2019-10-17T07:19:26Z

Just do be sure, does that mean that it's not possible to use camera triggering (GPIO) on a PixRacer together with DShot?

That is possible. Do you have a setup where you can test it?

ndepal · 2019-10-17T07:45:07Z

yes we could probably test that pretty easily

ndepal · 2019-10-17T07:56:18Z

src/drivers/dshot/dshot.cpp

+	}
+
+	// publish when motor index wraps (which is robust against motor timeouts)
+	if (motor_index < _telemetry->last_motor_index) {


When we were testing dshot with just one ESC connected we could not get telemetry info on esc_status because of this line. I don't know if 1 ESC is an actual use case (planes?), but if it is, would the following be better?

if (motor_index <= _telemetry->last_motor_index) {

Good point. I don't see any issues with that. Can you do a PR?

ndepal · 2019-10-17T08:49:32Z

That is possible. Do you have a setup where you can test it?

A quick bench test with camera triggering configured on pins 56 on a PixRacer (GPIO, active low) appears to work just fine.

bkueng · 2019-10-17T09:13:33Z

Thanks for confirming

dagar self-requested a review August 31, 2019 14:44

dagar added the Admin: Enhancement (improvement) 💡 label Aug 31, 2019

dagar mentioned this pull request Aug 31, 2019

[WIP] Dev dshot #12634

Closed

weekly-digest bot mentioned this pull request Sep 1, 2019

Weekly Digest (25 August, 2019 - 1 September, 2019) #12858

Closed

bkueng force-pushed the fmu_refactoring_output_module__dshot branch from ce17e99 to 23003e9 Compare September 5, 2019 14:01

weekly-digest bot mentioned this pull request Sep 15, 2019

Weekly Digest (8 September, 2019 - 15 September, 2019) #12959

Closed

bkueng force-pushed the fmu_refactoring_output_module__dshot branch from ca321e4 to 51763a8 Compare September 17, 2019 18:28

bkueng force-pushed the fmu_refactoring_output_module__dshot branch from 51763a8 to ea90eae Compare September 17, 2019 18:59

weekly-digest bot mentioned this pull request Sep 22, 2019

Weekly Digest (15 September, 2019 - 22 September, 2019) #13002

Closed

bkueng mentioned this pull request Sep 23, 2019

add a documentation page for DShot PX4/PX4-user_guide#578

Merged

hamishwillee reviewed Sep 24, 2019

View reviewed changes

src/drivers/dshot/module.yaml Outdated Show resolved Hide resolved

hamishwillee reviewed Sep 24, 2019

View reviewed changes

src/drivers/dshot/module.yaml Outdated Show resolved Hide resolved

bkueng force-pushed the fmu_refactoring_output_module__dshot branch from ea90eae to 4ffbe0c Compare September 24, 2019 17:58

Igor-Misic reviewed Sep 24, 2019

View reviewed changes

platforms/nuttx/src/px4/stm/stm32_common/dshot/dshot.c Outdated Show resolved Hide resolved

bkueng force-pushed the fmu_refactoring_output_module__dshot branch 5 times, most recently from 049b662 to b8710af Compare September 27, 2019 06:02

bkueng added 10 commits October 10, 2019 11:56

dshot: reduce static buffer size

ac6de52

And handle failures of up_dshot_init(). On Omnibus: reduces memory usage if dshot is enabled by ~1.0KB. The buffer is roughly 1KB in size.

dshot: check if DMA still in progress

343b5c7

No other functional change, just restructuring.

kakutef7, kopis airframe: enable dshot & telemetry

479cb6d

mixer_module: extend printf status output

3aa3943

parameters: update perf counter for flash-based param_save_default()

24bd023

dshot: allow boards to enable dshot only on a subset of the timers

940a224

fmu-v5: add dshot support

e9cd447

But only on the first 4 FMU outputs, as the next ones conflict with px4io serial dma (UART8_RX) RX DMA is disabled on the GPS port as well (conflicts with TIM1).

parameters: defer auto-saving to after disarming for flash-based params

6b191aa

This is especially important for DShot, that does not update when the CPU stalls.

fmu-v3: enable DShot

69a5561

Disables RX DMA on TEL4 and IO debug serial port

bkueng force-pushed the fmu_refactoring_output_module__dshot branch from 0a96148 to 69a5561 Compare October 10, 2019 09:57

julianoes approved these changes Oct 10, 2019

View reviewed changes

bkueng merged commit 803a719 into master Oct 11, 2019

bkueng deleted the fmu_refactoring_output_module__dshot branch October 11, 2019 06:14

weekly-digest bot mentioned this pull request Oct 13, 2019

Weekly Digest (6 October, 2019 - 13 October, 2019) #13171

Closed

ndepal reviewed Oct 17, 2019

View reviewed changes

ndepal mentioned this pull request Oct 17, 2019

dshot telemetry: enable telemetry publishing for setups with only one ESC #13217

Merged

mrpollo mentioned this pull request Nov 4, 2019

v1.10.0 Release Notes + Changelog #13366

Closed

bys1123 mentioned this pull request Nov 26, 2019

D-Shot ESCs and Motor Docs PX4/PX4-user_guide#560

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add DShot module #12854

Add DShot module #12854

bkueng commented Aug 31, 2019

Igor-Misic commented Aug 31, 2019

dagar commented Aug 31, 2019 •

edited

hamishwillee commented Sep 5, 2019

bkueng commented Sep 5, 2019

hamishwillee commented Sep 5, 2019

bkueng commented Sep 17, 2019

mhkabir commented Sep 23, 2019

bkueng commented Sep 24, 2019

mhkabir commented Sep 24, 2019

bkueng commented Sep 27, 2019

dagar commented Sep 27, 2019

MaEtUgR commented Oct 11, 2019

thomasderflieger commented Oct 15, 2019

hamishwillee commented Oct 15, 2019 •

edited

bkueng commented Oct 16, 2019

thomasderflieger commented Oct 16, 2019

bkueng commented Oct 16, 2019

ndepal commented Oct 16, 2019

bkueng commented Oct 17, 2019

ndepal commented Oct 17, 2019

ndepal Oct 17, 2019 •

edited

bkueng Oct 17, 2019

ndepal commented Oct 17, 2019

bkueng commented Oct 17, 2019

Add DShot module #12854

Add DShot module #12854

Conversation

bkueng commented Aug 31, 2019

Igor-Misic commented Aug 31, 2019

dagar commented Aug 31, 2019 • edited

Pros

Cons

hamishwillee commented Sep 5, 2019

bkueng commented Sep 5, 2019

Updates

Known Issues

hamishwillee commented Sep 5, 2019

bkueng commented Sep 17, 2019

mhkabir commented Sep 23, 2019

bkueng commented Sep 24, 2019

mhkabir commented Sep 24, 2019

bkueng commented Sep 27, 2019

Flight Performance

Boards

Issues

Logs

Other

dagar commented Sep 27, 2019

MaEtUgR commented Oct 11, 2019

thomasderflieger commented Oct 15, 2019

hamishwillee commented Oct 15, 2019 • edited

bkueng commented Oct 16, 2019

thomasderflieger commented Oct 16, 2019

bkueng commented Oct 16, 2019

ndepal commented Oct 16, 2019

bkueng commented Oct 17, 2019

ndepal commented Oct 17, 2019

ndepal Oct 17, 2019 • edited

Choose a reason for hiding this comment

bkueng Oct 17, 2019

Choose a reason for hiding this comment

ndepal commented Oct 17, 2019

bkueng commented Oct 17, 2019

dagar commented Aug 31, 2019 •

edited

hamishwillee commented Oct 15, 2019 •

edited

ndepal Oct 17, 2019 •

edited