FFT example

[microbit-carlos]

Reviving the branch originally developed by Raphael and updated by Joshua.

This is not really mergeable on this state as the CMSIS libs are added as a submodule to this repo and it changes the float-abi compiler flag to hard (we need to check if we can compile the CMSIS DSP code with float-abit=softfp).

Maybe it's worth considering moving this as a "CODAL library" that can also be used as a MakeCode extension, as the MicroBitAudioProcessor class is a fairly self-contained DataSink.

I'm not sure if I've wired the right DataSource to MicroBitAudioProcessor as a I get a lot high frequency values in a relatively quiet room (do we have a DataSource that has filtered out noise?) and the calculated frequencies are twice as expected.

codal-microbit-v2/samples/fft_example.cpp

Lines 18 to 37 in e3319f4

	/*
	if (mic == NULL){
	mic = uBit.adc.getChannel(uBit.io.microphone);
	mic->setGain(7,0);
	}
	if (processor == NULL)
	processor = new StreamNormalizer(mic->output, 1.0f, true, DATASTREAM_FORMAT_8BIT_SIGNED, 10);
	if (fft == NULL)
	fft = new MicroBitAudioProcessor(processor->output);
	uBit.io.runmic.setDigitalValue(1);
	uBit.io.runmic.setHighDrive(true);
	*/
	// Code above commented out was from the original example, which can
	// probably be replaced with this single line, but we need to double check
	// the configuration because the fft calculation results are twice the frequency
	fft = new MicroBitAudioProcessor(*uBit.audio.splitter->createChannel());

Also, it looks like CMSIS needs CMake 3.13 min, and that's a newer version than the Yotta docker image, so the MakeCode builds will fail.

[github-actions]

Build diff

Base commit: be35cc1d13697a783e08658a94e3af2b6c1afc4e
Action run: https://github.com/lancaster-university/codal-microbit-v2/actions/runs/4630881119

     VM SIZE    
 -------------- 
  [NEW]  +116Ki    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/CommonTables/arm_common_tables.c
  [NEW] +1.83Ki    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/TransformFunctions/arm_cfft_f32.c
  [NEW] +1.33Ki    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/TransformFunctions/arm_cfft_radix8_f32.c
  +0.5%    +608    [section .heap]
  +8.5%    +456    /home/runner/work/codal-microbit-v2/codal-microbit-v2/source/main.cpp
  [NEW]    +454    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/ComplexMathFunctions/arm_cmplx_mag_f32.c
  [NEW]    +432    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/source/MicroBitAudioProcessor.cpp
  [NEW]    +420    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/TransformFunctions/arm_rfft_fast_init_f32.c
  [NEW]    +406    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/TransformFunctions/arm_rfft_fast_f32.c
  [NEW]    +190    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/StatisticsFunctions/arm_max_f32.c
  [NEW]    +156    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/TransformFunctions/arm_cfft_init_f32.c
  [NEW]    +144    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/CommonTables/arm_const_structs.c
  [NEW]     +74    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-microbit-v2/CMSIS_5/CMSIS/DSP/Source/TransformFunctions/arm_bitreversal2.c
  -2.6%     -40    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-core/source/types/Image.cpp
  -1.5%     -52    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-core/source/core/CodalFiber.cpp
  -0.1%     -66    [97 Others]
 -39.7%     -92    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-core/source/core/CodalCompat.cpp
  -7.5%    -156    /home/runner/work/codal-microbit-v2/codal-microbit-v2/libraries/codal-core/source/drivers/AnimatedDisplay.cpp
 -98.4%    -492    /home/runner/work/codal-microbit-v2/codal-microbit-v2/source/samples/MicrophoneTest.cpp
 -26.6% -5.69Ki    [section .text]
 -97.8% -6.12Ki    /home/runner/work/codal-microbit-v2/codal-microbit-v2/source/samples/OOB.cpp
   +29%  +110Ki    TOTAL

[microbit-carlos]

Next steps for this work:

• It's possible I haven't wired things correctly as the calculated frequencies are twice the expected value, so we need to figure out the issue there
• The additional resources used by this feature are too large, 100KBs is a huge chunk of flash, we need to look into how to reduce this
• I believe the FFT calculation time was measure by Joshua at around 10-11 ms, which is not bad, but if we want this to be constantly running in the background we might need it to yield a few times to ensure other things have a chance to run
  • Talk: https://www.youtube.com/watch?v=LzkN4j2DB_w
  • One of the reasons this might be constantly running in the background is the possibility to create an "event block" for MakeCode, so that things can happen when a specific frequency is detected

[microbit-carlos]

When using this branch for the first time, the CMSIS_5 git clone might sometimes produce the following error:

Downloading CMSIS/Utilities/Win32/PackChk.exe (2.9 MB)
Error downloading object: CMSIS/Utilities/Win32/PackChk.exe (7934974): Smudge error: Error downloading CMSIS/Utilities/Win32/PackChk.exe (7934974e4e929021a7413a609183ce61142c38d9c87e40925a0feec1bc958892): batch response: This repository exceeded its LFS budget. The account responsible for the budget should increase it to restore access.

Errors logged to '/Users/carlos/workspace/tmp/CMSIS_5/.git/lfs/logs/20251218T101326.591298.log'.
Use `git lfs logs last` to view the log.
error: external filter 'git-lfs filter-process' failed
fatal: CMSIS/Utilities/Win32/PackChk.exe: smudge filter lfs failed

This can be fixed by using the GIT_LFS_SKIP_SMUDGE flag when building the project:

# First delete the libraries folder so build.py clones all repos from scratch
rm -rf build/
# This flag skips trying to download any of the git LFS files, which we don't need for the FFT
GIT_LFS_SKIP_SMUDGE=1 python build.py