Testing with ESP32

• 25Hz left and 100Hz right (you can notice the empirical frequency estimate in the screenshot)

[{"info":{
"source_id":"EmotiBit FeatherWing",
"hardware_version":"V05c",
"sku":"MD",
"device_id":"MD-V5-0000006",
"feather_version":"Adafruit Feather HUZZAH32",
"feather_wifi_mac_addr":"60:0c:6b:7e:b9:94",
"firmware_version":"1.5.4.feat-100HzPPG.10-TC",
"firmware_variant":"EmotiBit_stock_firmware_PPG_100Hz",
}}]

[{"info":{
"source_id":"EmotiBit FeatherWing",
"hardware_version":"V05c",
"sku":"MD",
"device_id":"MD-V5-0000002",
"feather_version":"Adafruit Feather HUZZAH32",
"feather_wifi_mac_addr":"10:65:90:be:5b:c4",
"firmware_version":"1.5.4.feat-100HzPPG.10-TC",
"firmware_variant":"EmotiBit_stock_firmware",
}}]

Testing with M0

• 25Hz left and 100Hz right (you can notice the empirical frequency estimate in the screenshot)

[{"info":{
"source_id":"EmotiBit FeatherWing",
"hardware_version":"V05c",
"sku":"MD",
"device_id":"MD-V5-0000010",
"feather_version":"Adafruit Feather M0 WiFi",
"feather_wifi_mac_addr":"f8:f0:05:90:08:59",
"firmware_version":"1.5.4.feat-100HzPPG.10-TC",
"firmware_variant":"EmotiBit_stock_firmware_PPG_100Hz",
}}]

[{"info":{
"source_id":"EmotiBit FeatherWing",
"hardware_version":"V04a",
"sku":"MD",
"device_id":"MD-V4-0000648",
"feather_version":"Adafruit Feather M0 WiFi",
"feather_wifi_mac_addr":"f8:f0:05:90:10:10",
"firmware_version":"1.5.4.feat-100HzPPG.10-TC",
"firmware_variant":"EmotiBit_stock_firmware",
}}]

25Hz vs 100Hz HR plot

• Experiment setup: 3-mins sitting -> 3-mins standing -> 3 mins sitting
  

(minor ticks on X-Axis every 60 secs)

Conclusion

• Looks like the HR detected at 25Hz and 100Hz is similar in both "sitting" segments
• In the standing segments, the PPG sensor seems to be miss-detecting beats. As the observer of raw data stream, the ppg data looked very noisy.
  • Can it be because of the arm orientation being vertical, instead of flat (while sitting) ?

Raw data

• upload_data.zip

@produceconsumerobot Can you test this new FW that prints the AC_signal value on the serial monitor?
1.5.4.feat-100HzPPG.12.zip

AC Testing

Conclusions

• It seems like setting hard thresholds is very fraught because the infiltration of dicrotic notches in some cases (loose) can exceed the size of a full beat in other cases (super tight)
• Perhaps a better option to manage this would be to create a heartbeat reliability score (0-100) returned from checkForBeat(). This could be calculated using the following formulation:

acLpFreq = 0.1; // This sets how quickly acAmp renormalizes to new levels
filtAcAmp = lpFilter(acAmp, acLpFreq);
acAmpFactor = 4; // This sets the AC amplitude deviation that results in zero weight
hbReliabilityScore = constrain(1- abs(acAmp - filtAcAmp) / filtAcAmp ) / acAmpFactor , 0, 1);

Side note: someday could be useful to have a non-linear, e.g. a sigmoid, transformation on hbReliabilityScore

hbReliabilityScore could then be added to addPacket(beatTime, EmotiBitPacket::TypeTag::INTER_BEAT_INTERVAL, ..., (int8_t) (hbReliabilityScore * 100))
and used to create a HR weighted by hbReliabilityScore as follows:

ibiWeightFactor = 0.1; // This sets how quickly HR can change
ibiWeight = constrain(hbReliabilityScore * prevHbReliabilityScore * ibiWeightFactor, 0, 1);
ibiAvg = ibiAvg* (1 - ibiWeight ) + interBeatInterval * (ibiWeight );
HR = 1 / ibiAvg;

It could also be useful to calculate hrReliabilityScore = hrReliabilityScore * (1 - ibiWeight ) + hbReliabilityScore * (ibiWeight );
and include this with addPacket(beatTime, EmotiBitPacket::TypeTag::HEART_RATE, ..., (int8_t) (hrReliabilityScore * 100)) ; to codify the HR reliability as poor when the hbReliabilityScore is persistently bad and likely affecting HR estimation.

Super Tight

Summary

• As the signal gets weaker from over-tightening, noise can infiltrate and get detected as short heartbeats

Data

Super Loose No Motion

Summary

• Taking a deep breath commonly causes a dicrotic notch to get detected as a heartbeat and double the HR

Data

Steps to merge PR

• Make a release for the https://github.com/EmotiBit/EmotiBit_ArduinoFilters library
• Make a release for the https://github.com/EmotiBit/EmotiBit_EmojiLib library
• Merge and release XPlat utils PR: platformio compiling successfully with these changes EmotiBit_XPlat_Utils#23
• Merge Feat 100 hz ppg EmotiBit_MAX30101#8 to master
  • We need to release it after the main FW PR is released. This is because, the MAX has a breaking change and requires the latest FW to work.
• updated documentation Adds documentation on using PPG 100Hz EmotiBit_Docs#105

Moved to future PR

• Improve HR algorithm by adding dynamic bounds on beat detection as suggested in this comment.