9/23/2016

Garmin Vivosmart HR (7) ~ Quantify the offset Part 2

Data on September 20, 2016

An example with a similar trend

   In the previous post, I showed a data with big offsets during some period.  This post shows a good data which has smaller offset.  The plot above shows the trend is almost the same between the chest band based heart rate sensor and the embedded wrist based heart rate sensor.  I applied the same method to calculate the offset and plot.  The following is the plot for the offset distribution.



 The offset distribution
   This plot shows majority of samples are in a group that the offset is within 2.   It is hard to tell if there is any difference on the first plat.   However, there is a portion the offset is bigger in the beginning where the heart rate value has a big transition.  The reason is not very clear with the data I have.  But it may be the following reasons:
  • In many cases, the reading in the beginning of the training is not very accurate
  • It is probably the offset is bigger when it has a big transition
   We might need more data to tell the clear reason.  I plan to correct more data and analyze them.
   It is clear that even if the heart rate trend looks a similar, but there are some big offsets.   It is hard to tell where the offset is coming from.   I guess it is worse to analyze the data more.

Actual amount of offset?

   Now, to check how much offset we need to consider to use data, here is another plot that indicates the amount of offset and total number of samples for each offset group.

 Distribution of samples for each offset amount

   The data shows more than 92% samples are less than 2 offset.  The other observation is that a section which has a big offset is only beginning, the rest of parts are really stable and matched.  Based on those indications, it is worth to evaluate to analyze the offsets to quantify the offset.
    Based on the data I posted last week, it seems that data for outdoor running has some bigger offset than the indoor training in a gym, such as elliptical, stair master and so on.

Other finding

   When I analyzed the data, I found out the sampling is not synchronized between the two devices.  Especially the data sampling rate between the two devices is not the same.   Even the same device in the different training log has some variation.  The overall sampling rate is something around 3 ~ 7 seconds.  But I have no idea how a device defines the rate.
   As an example, Vivosmart HR data on September 21 has every 4.1 seconds as an average, the minimum is 1 second and the maximum is  25 seconds.  On the other hand, Forerunner 910XT has every evey 4.2 seconds as an average, the minimum is 1 second, and the maximum is 10 seconds.  The entire training data is for 1 hour.  The total number of sampls for Vivosmart HR is 910 and 898 for Forerunner 910XT.

  In addition, the heart rate data for Vivosmart HR is 143 [bpm]  per minutes as an average, 76 [bpm] for minimum and 169 [bpm] for the maximum.  Forerunner 910XT, it shows the average heart rate is 146 [bpm] , the minimum is 80 [bpm] and the maximum is 168 [bpm].  I think they are close enough

No comments:

Post a Comment