Offset distribution
Quantify the difference
First of all, this review is not verify the accuracy of the
sensor reading. The primary purpose is
to find out how much difference between the two sensors, especially if we can
replace the chest based sensor to the wrist based sensor. Last week, I showed the sensor reading plots
on the same scale. Overall, it seems to
be close enough in most of cases for running / hiking / cardio in a gym. However it is not enough to show how much
closer. Therefore, I have been working
on quantify the difference.
A method to quantify
One of ideas to show the difference is to check the offset
distribution. The distribution could
show what the offset looks like. The
problem to get the distribution is the both devices are not synchronized for
the sampling. So that I wrote a program
to fill a value if data for one of device is missed. The program find out the last valid sample
data and the next sample data. The
assuming value between the two samples are linear. Then, calculate the offset between the two
devices and find out the distribution based on the offset amount.
As an example, I used a training data for today. I did a “threshold run”.
- 10 minutes warming up
- Run in Heart Rate zone 4 for 4 minutes, run in Heart Rate zone 2 in 20 seconds – Repeat 3 times
- Cooling down 10 minutes
This is a part of Garmin full marathon training plan. This menu is “Week3/Day4”. This is a sort of an interval training.
Here is the offset distribution:
Offset
|
Number of samples
|
[%]
|
0 ~ 2
|
668
|
58.3
|
2 ~ 4
|
100
|
8.7
|
4 ~ 6
|
35
|
3.1
|
6 ~ 8
|
25
|
2.2
|
8 ~10
|
24
|
2.1
|
10 ~ 12
|
23
|
2.0
|
12 ~ 14
|
24
|
2.1
|
14 ~ 16
|
24
|
2.1
|
16 ~ 18
|
14
|
1.2
|
18 ~ 20
|
21
|
1.8
|
22 ~
|
212
|
18.5
|
This example shows the reading does not match well. Here is the plot.
Heart Rate plat for Threshold Run on September 21, 2016
Also, this is the offset looks like.
The offset distribution with Time
Based on the plot, it seems that Forerunner 910XT reading
which uses the chest based heart rate sensor does not get correct data in the beginning. After that section, the both reading is getting closer. However, there are a few spikes on reading by the Vivosmart HR. They are probably not correct readings. As long as people are aware of this behavior, they are not a big deal in my opinion. I also did a similar training last week and the data showed a similar result.
I will apply this scheme to other data to analyze more detail.
(To be continue)
No comments:
Post a Comment