/EMG-Signal-Processing-Library

A real-time signal processing library for EMG sensors. View the README.md to see raw vs. processed signals!

Primary LanguageC

EMG Signal Processing Library (graphed signals below!)

  • This is a specialized real-time signal processing library for EMG signals
  • This library provides the tools to extract muscle effort information from EMG signals in real time
  • Most of the algorithms implemented run in constant time with respect to sampling rate
  • Currently supports the following languages: C, Python

alt text

A comparison of a raw EMG input signal vs. processed output signal (sample_frequency_ = 1000, range_ = 0.1, min_EMG_frequency_ = 25, max_EMG_frequency_ = 300, remove_low_frequency_ = HIGH_PASS_FILTER_ON, reference_available_ = REFERENCE_UNAVAILABLE)

alt text

Another comparison of a raw EMG input signal vs. processed output signal (sample_frequency_ = 1000, range_ = 0.1, min_EMG_frequency_ = 25, max_EMG_frequency_ = 300, remove_low_frequency_ = HIGH_PASS_FILTER_ON, reference_available_ = REFERENCE_UNAVAILABLE)

C DOCUMENTATION

1) Setting up

Copy the following .c and .h files into your working directory:

  • emg
  • ecg
  • peak_to_peak
  • moving_average
  • queue

This library was developed using the GCC compiler with the -std=c11 option. It is advisable to also use this option for compiling.

2) EMG PROCESSING

  • Use the EMG module to extract muscle effort information from an EMG signal
  • This is provides the basis for recognizing gestures through EMG signals. For example, the Myo armband recognizes hand gestures by determining how hard each muscle group in the forearm is flexing.
  • The module processes the EMG signal using the following steps:
    1. Filter high frequency noise from signal, and subtract a reference signal from the actual signal if one is provided
    2. Filter low frequency noise from signal and normalize signal (if HIGH_PASS_FILTER_ON is specified in the constructor)
    3. Frequencies that are low and high frequency noise are determined by the "min_EMG_frequency_" and "max_EMG_frequency_" constructor arguments
    4. Rectify the signal
    5. Take a moving average of the signal
  • The resulting filtered signal is indirectly proportional to the effort exerted by the muscle the sensor is attached to.
typedef enum {
	REFERENCE_AVAILABLE,
	REFERENCE_UNAVAILABLE,
	HIGH_PASS_FILTER_ON,
	HIGH_PASS_FILTER_OFF
} EMG_OPTIONS;

An enum of the options which must be specified to create a new EMG struct.

EMG *new_EMG(	uint16_t sample_frequency_, 
				float range_, 
				uint16_t min_EMG_frequency_, 
				uint16_t max_EMG_frequency_, 
				EMG_OPTIONS remove_low_frequency_, 
				EMG_OPTIONS reference_available_);

The constructor. Returns a pointer to the newly created EMG struct.

  • sample_frequency_ * The constant frequency of which data will be given to the EMG data structure (frequency in which filter_EMG will be called). * Specified in Hz

  • range_

    • The number of data points used to calculate moving average, specified in number of seconds (0.2-0.5 is a good value for this).
    • Data points used in moving average = range_ * sample_frequency_
    • Note that the greater the range_ value, thr greater the latency of the filter

  • min_EMG_frequency_

    • Components of signals (noise) below this frequency will be removed.
    • This argument is specified in Hz
    • A good value for this is 40

  • max_EMG_frequency_

    • Components of signals (noise) above this frequnecy will not be accounted for
    • This is also used in the peak to peak calculations which help remove low frequency noise.
    • This argument is specified in Hz
    • A good value for this is 150

  • remove_low_frequency_

    • This argument specifies whether or not to remove low frequency noise.
    • Can be either HIGH_PASS_FILTER_ON (remove low frequency noise) or HIGH_PASS_FILTER_OFF
    • If the EMG signal does not pass through an analog high pass filter, it is highly recommended to enable this feature
    • Disabling this feature decreases latency slightly

  • reference_available_

    • This specifies whether a reference signal will be provided (in which case filter_EMG_r will be called instead of filter_EMG)
    • Can be either REFERENCE_AVAILABLE or REFERENCE_UNAVAILABLE
void free_EMG(EMG *self);

The destructor. Takes an EMG pointer as argument.

int  filter_EMG(EMG *self, int data);

This function is used to input the most recent data point and returns the most recent filtered EMG value. High and low frequency noise is removed from the signal. The signal is then rectified with respect to it's centre. A moving average filter returns the completely filtered EMG signal. The returned signal is indirectly proportional to the effort exerted by the muscle. This function must be called at the frequency specified in the constructor.

  • self
    • A A pointer to the EMG struct
  • data
    • The most recent raw data point
int  filter_EMG_r(EMG *self, int data, int reference_data);

This is the same as filter_EMG, except a reference data point must also be specified. This can only be called if REFERENCE_AVAILABLE was specified in the constructor.

  • reference_data
    • The most recent reference data point

3) ECG PROCESSING (requires further development)

  • This library provides a tool to derive the BPM indicated by an ECG signal
  • The module uses the following steps to achieve this:
    • Initialization:
      1. Find the average peak to peak values (the difference between the maximum and minimum values over a small peroid of a signal) of the signal over the first 3 seconds of operation
      2. If the autodetect threshold feature is enabled, also find the average of the greatest peak to peak values over this time and calculate the threshold by dividing this value by the average value and multiplying by a hard-coded fraction (feel free to change this to tune the autodetect feature)
    • Normal operation:
      1. When the latest peak to peak value of the signal surpasses the average found during initialization multiplied by the threshold, that instant is registered as a heart beat.
      2. Heart now cannot be detected for 1/2 times the interval between the last two beats detected
      3. Calculate the number of data points since the last heartbeat was detected and store the number
      4. Take the average of the number of samples between the last 3 heartbeats. Use this value to calculate beats per minute (BPM) with the formula 60*sample_frequency/average_number_of_samples
typedef enum {
	REFERENCE_AVAILABLE,
	REFERENCE_UNAVAILABLE,
	AUTODETECT_THRESHOLD_ON,
	AUTODETECT_THRESHOLD_OFF
} ECG_OPTIONS;

An enum for the options that must be specified when creating a new ECG struct.

ECG *new_ECG(uint16_t sample_frequency_, 
	 		 float pkpk_threshold_ratio_, 
			 ECG_OPTIONS reference_availability_, 
			 ECG_OPTIONS autodetect_);

This is the constructor for the ECG struct. It returns a pointer to the newly created struct.

  • sample_frequency_
    • This is the constant frequency in which data will be given (the frequency which get_BPM or get_BPM_r will be called)
  • pkpk_threshold_ratio_
    • This is the ratio of peak to peak voltage during ventricular depolarization and the PR interval.
    • This value differs slightly between each ECG setup and must be determined by experimentation.
    • Alternatively, you can use the autodetect_ feature, which allows the software to calculate this ratio itself but does not guarantee perfect consistency
    • 2.0 is a good value for this
  • reference_availability_
    • Specifies whether the filter will be used with a reference signal (which means get_BPM_r will be called instead of get_BPM).
    • The value of this argument can either be REFERENCE_AVAILABLE or REFERENCE_UNAVAILABLE
  • autodetect_
    • Specifies whether pkpk_threshold_ratio_ will be specified by the user, or calculated automatically.
    • This argument can either be AUTODETECT_THRESHOLD_ON or AUTODETECT_THRESHOLD_OFF.
    • Note that autodetect is not perfect, so it's typically more accurate to manually set an experimentally obtained pkpk_threshold_ratio_.
void free_ECG(ECG *self);

The destructor for the ECG struct. Takes an ECG struct as its argument.

void initialize_ECG(ECG *self, 
					int data);

This function is automatically called every time get_BPM or get_BPM_r is called, for the first 3 seconds of operation. This is necessary to initialize the BPM detection feature. It is not necessary for the user to call this function.

int get_BPM(ECG *self, 
			int data);

This function returns the BPM indicated by the ECG signal supplied to it.

  • self
    • A pointer to the ECG struct
  • data
    • The most recent data point of the signal.

Note that it takes 5-6 seconds of operation to achieve full accuracy. For full accuracy, this function must be called at the frequency specified when the ECG struct was created

int get_BPM_r(ECG *self, 
		      int data, 
			  int reference);

This function is the alternative to get_BPM, when a reference was specified as available.

  • reference
    • The most recent reference data point of the signal

Python DOCUMENTATION

  • The Python implementation of this library provides the same features as the C implementation
  • The Python implementation uses true objects with member functions, as opposed to structs
  • Most functions and variables have the same name as the C implementation
  • Constant time is not guaranteed in the Python implementation, since some native Python functions are used