Add SpO2 derivative metric

[nitin710]

Description

Add SpO2 metric utilizing MAXIM's SpO2 algorithm, but reduce it's resource(primarily memory) requirement.

Useful links

• https://www.analog.com/en/technical-articles/guidelines-for-spo2-measurement--maxim-integrated.html

Current SpO2 algorithm

• https://github.com/EmotiBit/EmotiBit_MAX30101/blob/master/src/spo2_algorithm.cpp

SpO2 Algo explained

• Find the valley (using existing peak detector)
• For each of consecutive valleys
  • Record the max value for each channel(IR and RED) (this is the point of peak reflection)
  • Record the index of the max https://github.com/EmotiBit/EmotiBit_MAX30101/blob/d92268f897fb5f4269a5a03fb26f3ffb27d14bd8/src/spo2_algorithm.cpp#L169-L172
• Find ac component for a pulse(both channels) https://github.com/EmotiBit/EmotiBit_MAX30101/blob/d92268f897fb5f4269a5a03fb26f3ffb27d14bd8/src/spo2_algorithm.cpp#L174-L179
• Use SpO2 formula https://github.com/EmotiBit/EmotiBit_MAX30101/blob/d92268f897fb5f4269a5a03fb26f3ffb27d14bd8/src/spo2_algorithm.cpp#L180-L185
• Find median of all SpO2 calculated in the 100 sample interval https://github.com/EmotiBit/EmotiBit_MAX30101/blob/d92268f897fb5f4269a5a03fb26f3ffb27d14bd8/src/spo2_algorithm.cpp#L190-L196
• Return value from the SpO2 LUT https://github.com/EmotiBit/EmotiBit_MAX30101/blob/d92268f897fb5f4269a5a03fb26f3ffb27d14bd8/src/spo2_algorithm.cpp#L198-L201

Suggested tweaks to the algorithm

Step	Current Approach	Purpose	Requirements	Notes on current approach	Suggested approach
Store 100 samples in a buffer	2 100 uint32 buffers are used	to store 4 secs worth of data(freq=25Hz)	2 x array[100]	These buffer sizes grow linearly with frequency
Calculate DC mean of 100 samples (only on 1 channel)	take avg of 100 points	Remove DC component	1 byte	The Assumption being that both channels of sensors are syunced in time
Subtract DC mean from actual data (only on 1 channel)	Use 1 new array of 100 data points to store DC removed signal	To process only AC component	1x array[100]
invert the signal	multiple by -1	To use existing peak detector(to find valleys. Peak = valleys in inverted signal)	NA
Smooth the signal	4 point moving avg	to prevent erroneous peak detection	NA
Calculate Threshold	Specify bounds for peak detection algo	?	1 byte
Detect valleys(using peak detector)	use defined peak detector algo	find valleys (valleys in the OG signal)	NA
calc HR using dist. b/w valleys	average all inter valley sapcing and use Fs to calc HR in bpm	get HR	Not m/m intensive	double check the algo
Perform bounds check on valley locs	simple array bounds check	All valleys must have a valid index, [0,BUFFER_LEN). If a valley lies, outside these bounds, there has been an error, and algo returns with invalid	NA
Perform AC/DC signal content analysis for SpO2	defined algo	Find N and D, when LUT_index = N/D, and LUT[i] = SpO2	2 adjacent valleys, and the peak in the middle of the valleys
Post-process SpO2	post processing	find median of SpO2 values
return result	-	Use the Look-up-table to get SpO2 value

[nitin710]

Requirements to port the algorithm

• 1 SpO2 measurement requires measurements of 2 valleys and a peak (between the valleys)
• Valley 1:
  • i1 : location of valley 1 (we can always consider this to be 0)
  • x[i1]: signal value at valley
• Valley 2:
  • i2 : location of consecutive valley wrt to i1
  • x[i2]: value at valley 2
• Peak between valleys
  • i_m: location of max value between valleys wrt i1 and i2
  • x[i_m]: value of the max value between peaks

synopsis

• we need 6 values:

  • 2x location and values of valleys
  • 1x location and value of peak between valleys

• Signal -> BPF -> wait for a valley -> wait for peak -> record peak value and relative location to first peak -> wait for valley -> record valley loc wrt valley 1 and value of signal at valley

• perform the calculation

Implementation

• 1x BPF
  • preconditioning the signal for valley/peak detection

[nitin710]

❓ Where would the SpO2 live in the oscilloscope?