This application uses on-board ADC to convert analog input from an EMG sensor and performs digital signal processing filtering to eliminate frequencies out of the (30hz, 300hz) interval and 50hz disturbance. Processed samples are then sent via serial communication to the PC in order to be plotted. The application has been developed on top of the Miosix embedded operative system which provided real-time operating system (RTOS) features.
- STM32F4x nucleo board (project has been developed using STM32F401re nucleo board)
- EMG sensor
- For embedded software compilation:
- Miosix toolchain (for cross compilation)
- GNU ARM embedded toolchain (linker script)
- Eigen C++ template library (linear algebra operations, already embedded in the project)
[ 1 ] Connect microcontroller to the PC with the USB cable and open serial communication to read incoming values.
[ 2 ] Connect sensor to analog pin PA0 and ground to GND.
The ADC has been configured in interrupt mode and executes conversions at a 1000hz rate (1000 samples each second). Conversions are triggered by an internal timer (TIM3) that outputs an event each 1ms to start the conversion. TIM3 has a 1000hz frequency thanks to pre-scaling internal clock and setting the auto-reload register accordingly.
USART serial interface is used to communicate processed values to the computer. In order to mitigate the bottleneck problem on serial communication, boudrate was set to 230400.
Digital signal is lively processed in order to remove AC noise, remove DC offset, remove 50hz disturbance and mitigate 1/f noise.
The applied DSP makes use of two IIR filters (bandpass and stopband) and Zero-phase filtering tecnhinque in order to avoid filtering delay. Zero-phase filtering is a non-causal procedure, so it cannot be done in real time. That is why blocks of 300 samples are converted each time making a piece-wise filtering on the whole incoming signal. Because of that, filter's initial conditions was necessary in order to avoid DC noise on interval borders.
Please note: both subplots have different ylim. The file containing this simulation can be found in the misc/DSPemgsignal.m file.
y[t] = b[0] x[t] + b[1] x[t-1] + ... - a[1] y[t-1] - a[2] y[t-2] - ...
Filtering is done twice for each filtering, one back and one forth inverting the signal on second pass. In the end, N samples are filtered 4 times. Twice for the two filters (bandpass and stopband).
The general form of a discrete filter is
where N is the order of the filter, a_i are denumerators coefficient and b_i are denominator coefficients.
7th order IIR filter (30hz, 300hz)
a_coeff: 1,-2.13183455555828,1.47978011393210,-0.679740843101842,0.584825906895303,-0.218461835750097,-0.0211926261278646
b_coeff 0.199880906801133,0,-0.599642720403399,0,0.599642720403399,0,-0.199880906801133
5th order IIR filter (50hz)
a_coeff 1,-3.78739953308251,5.56839789935512,-3.75389400776205,0.982385450614124
b_coeff: 0.991153595101663,-3.77064677042227,5.56847615976590,-3.77064677042227,0.991153595101663
TODO
[ - ] Possibilities for serial reading:
PuTTY(Windows, free)Serial(Mac, $39.99)ls -l /dev/ttyXwhere X is the tty relative to serial port
[ - ] Use command make all in the miosix-kernel folder for embedded software compilation.
[ - ] Use st-flash write main.bin 0x08000000 command to flash generated bin executable into the board.
[ - ] data.txt contains raw sampling values from the sensor that can be used to simulate incoming raw values form the microcontroller.
[ - ] DSPemgsignal.m Matlab file inside the misc/ folder can be used to plot either live receiving data from serial port using or simulated values.
[ - ] xxx.ipynb Python notebook file inside the misc/ folder can be used to plot either live receiving data from serial port using or simulated values.
[ + ] Feature extraction on filtered signal (AFB, DAMV, DASDV, DLD, DTM...) [ + ] Movement prediciton using machine learning (KNN, SVM) or neural netowkrs [ + ] Hardware abstraction construct