Add EDR Amplitude and EDR frequency to EmotiBit stock

[nitin710]

Description

• Add data type record EDR events
  • An Electrodermal response is a physiological activity, which measures and logs a significant change in EDA over a short period of time.
  • External stimulus can generate a physiological response, and this response can be logged using EDR

Lit Rev

• 

TypeTag considerations

• EDA can also be a Skin Conductance Response
• A signal not considered at this time is EDA recovery
• We are also not sending TimeOfOnset explicitely.
  • it is implicit on the Frequency
• Electrodermal response amplitude
  • EC - Electrodermal change?
  • SA - Skin conductance Amplitude
• Electrodermal response frequency
  • EF - Electrodermal Response Frequency
  • SF - Skin conductance Frequency
• Electrodermal Response Rise Time
  • ET - rise time?
  • SR - Skin conductance Rise Time
  • May want to steer clear of R in case we want to add Recovery

[nitin710]

Draft 1

Issues

• check the issue mentioned here
  • https://github.com/EmotiBit/EmotiBit_FeatherWing/issues/172#issuecomment-1028125966)

    Proposed changes

    
    in processElectrodermalResponse()
    {
    // static variables
    float* data;
    size_T dataLen;
    edrAmplitudeHighPassFilter // highPass digital Filter  // used to segment out the sharp changes
    edaAmplitudeLowPassFilter // low pass digital Filter  // used to provide a baseline that works as a reference for threshold
    edrFrequencyFilter // lowPass digital filter
    static float threshold; // maybe this should be a percentage of the current EDA value? ~ 0.1
    static uint16 interResposeSampleCount; // to count number of samples passed between EDR events
    float interResponseTime;

// Load latest EDA signal dataLen = dataDoubleBuffer[EDA]->getData();

// process EDA for(dataLen) { lowPassfilteredSignal = edaAmplitudeLowPassFilter (data[i]); highPassfilteredSignal = edrAmplitudeHighPassFilter(data[i]); if((highPassfilteredSignal) / lowPassfilteredSignal > threshold) { // add (highPassfilteredSignal - lowPassfilteredSignal ) to _outDataPacket

// calculate the time between response
interResponseTime = lowPassFilter( interResponseSampleCount/samplingFreq);

// Add (1/interResponseTime) to _outDataPackets

} else { interResponseSampleCount++; } }

## Questions
1. Should we consider a `+ve` and `-ve` swing in EDA to count as a EDR?
    - If yes, how do we represent that? 
      - Do we just send a `+ve` and `-ve` value?

[nitin710]

@produceconsumerobot what do you think of the draft? Still trying to think of a way to bypass typeTag

[nitin710]

Notes from Discussion

• Time of event occurrence is the Time calculated from the time recorded in dataDoubleBuffer
• Send rise time as well
• EDR Amplitude is the peak of the response
• EDR Frequency depends on the time-onset of a response
• On threshold crossing
  • we store the time.
  • flip a bool
  • update and send the freq
  • in the bool subState
  • we wait for a drop on the EDA
  • rise time = samplePeakTime - onsetTime
  • amplitude = peak
  • send both
  • flip the bool back
• we might want to band pass filter EDA to remove noise on EDR events
  • 0.5Hz to 1.5Hz
  • Freq. of the lowPass(reference to the threshold) = start with 0.5(can choose to make it lower)

[nitin710]

Draft 2

in processElectrodermalResponse()
{
float* bufferData;
uint32_t timestamp;
size_T bufferSize;
edrAmplitudeHighPassFilter // for bandpass @ 0.5Hz
edaAmplitudeLowPassFilter // for bandpass @1.5Hz
edaAmplitudeBaselineFilter // @0.1Hz  // used to provide a baseline that works as a reference for threshold
edrFrequencyFilter // lowPass digital filter @1Hz
static float threshold; // ~ 0.1 (10%)
static uint16 interResposeSampleCount; // to count number of samples passed between EDR events
float interResponseTime;
static const float timePeriod = 1/ samplingFreq;
static bool onsetDetected;
static float baseEdrAmplitude;
float riseTime;

// Load latest EDA signal
bufferSize = dataDoubleBuffers[EDA]->getData(&bufferData, &timestamp, false);

// process EDA
for(bufferSize)
interResponseSampleCount++;
{
  lowPassfilteredSignal = edaAmplitudeLowPassFilter (data[i]);
  highPassfilteredSignal = edrAmplitudeHighPassFilter(lowPassfilteredSignal); // creates a bandpass by cascading low and high
  if(!onsetDetected)
  {
    if( highPassfilteredSignal / lowPassfilteredSignal > threshold)
    {
      // set onset detected.
      // record onset time(by back-calculating based on frequency)
      // set base edr amplitude
      baseEdrAmplitude = data[i]

      // calculate the time between response
      interResponseTime = lowPassFilter( interResponseSampleCount * timePeriod);
      // Add interResponseTime  _outDataPackets
      // interResponseSampleCount  = 0;
    }
  }
  else  // after onset has been detected
  {
    if(data[i] < data[i-1]) // peak detected
    {
         amplitude = data[i-1] - baseEdrAmplitude;
    }
    // calculate the rise time
   riseTime = interResponseSampleCount  * timePeriod;
   // send TimePeriod
   // onsetDetected = false; 
  }
}

• Add acquireData.EdrAmplitude in the code as well to be able to toggle it on and off

[produceconsumerobot]

• from Boucsein et al 2012

[produceconsumerobot]

SCR duration is also a somewhat commonly used term, e.g. from Handler et al. 2012 An EDA Primer for Polygraph Examiners

[produceconsumerobot]

from Setz et al 2010 Discriminating Stress From Cognitive Load Using a Wearable EDA Device

[produceconsumerobot]

From Boucsein's 2012 book

[nitin710]

A Guide for Analysing Electrodermal Activity (EDA) & Skin Conductance Responses (SCRs) for Psychological Experiments https://www.birmingham.ac.uk/Documents/college-les/psych/saal/guide-electrodermal-activity.pdf

[nitin710]

@produceconsumerobot as discussed, we may want to move the EDR:FREQ to a periodic channel. Let's pool that discussion into this issue.

Early draft:

[nitin710]

Moving processElectrodermalActivity() into EmotiBitEda

• sampling rate exists in the EmotiBitEda() class
• dataDoubleBuffers[(uint8_t)EmotiBit::DataType::EDA] is replaced with the class private member DoubleBufferFloat

Proposed solution for addPacket

• AddPacket() needs
  • timestamp
  • typetag
  • data
  • datalen
• We can move all the mathematics inside EmotiBitEda
• in EmotiBit.cpp
  • in processAperiodic()
  • onsetTime = emotiBitEda.getOnsetTime();
  • responseFreq = emotiBitEda.getResponseFreq();
  • dataLen = emotiBitEda.getDataLen();
  • addPacket(onsetTime, EmotiBitPacket::TypeTag::ELECTRODERMAL_RESPONSE_FREQ, &responseFreq, dataLen, 4); // 4 = precision

[nitin710]

Typetag Nomenclature

possible typetags between E series and S series	Signal name	Typetag(S series)	Typetag(E series)
EDA	EA	EA
EDL	EL	EL
EDR	ER	ER
Skin Conductance Response Amplitude	SA	EC(electrodermal change)
Skin Conductance Response Frequency	SF	EF
Skin Conductance Response Rise Time	SR	ET
Skin Conductance Response Recovery Time	ST	E?
Skin Conductance Response Duration	SD	ED
Skin Conductance Response Latency	SL	E?

S- series

Reasons to have S series

• The literature seems to annotate this physiological phenomenon as "Skin conductance response", so it feels like a fit
• We are also reporting every data in uS, which is conductance
• Electrodermal activity (EDA) is the umbrella term used for defining autonomic changes in the electrical properties of the skin. The most widely studied property is the skin conductance, link
• if people are interested in Resistance, they may use conductance and derive resistance in post-processing.

note: when we add Skin conductance duration, need to resolve conflict with type tag SD-Card percentage.

Reasons not to have S series

• Conflict with typetags in Factory test, but we can keep them isolated by enforcing serial(setup related) and wifi comms(data related) to be separate.

E-series

Reasons to Have E series

• Legacy
  • We have had EDA EDR and EDL before
• E for Electrodermal has an intuitive meaning

Reasons not to have it

• Outstanding namespace conflicts
  • Recovery
  • Latency
• Electrodermal Activity is an umbrella for various signal, and we are trying to represent, specifically, skin conductance phenomenons
  • SCRs are a subset of EDA representation
  • Using SCRs may make sense in trying to zone in the meaning of the signal

[nitin710]

Here U1 represents a bandpass filtered eda, which is being used to determine the SCR derivatives. bandpass [0.3-1] Hz