Friday, March 26, 2021

DFA a1 problems during running - why?

 Low DFA a1 while running - a possible fix? 

I have previously noted DFA a1 decline associated with poor Movesense ECG waveform and have had multiple "complaints" of DFA a1 reading too low at easy intensity levels during certain forms of exercise in certain people.  The most common example being running. The scenario is usually as follows :

"My DFA a1 drops while running at trivial effort, way below what it should be and much different than with similar heart rates when biking."

My guess was that some type of core/trunk/diaphragm electro mechanical interaction was distorting the ECG waveform, making the peak of the R wave less able to be measured precisely. Since DFA a1 is an index of correlation properties and fractal patterns, a loss of true R peak timing will result in an artificial reduction in the complexity of these patterns.  Kind of like a dirty fingerprint on your camera lens - a loss of detail occurs and the picture is not going to be sharp. But perhaps the answer is from a different type of interference.

Why would running give us a distorted R peak?  Here is a clever and informative view of  chest wall motion during running:

As you can see there is quite a bit of motion and sensor bounce going on.  Whether this is part of the issue is unclear, but it does make one wonder.  In addition, the volume is rapidly changing, which could lead to variation in the ECG signal.


The question is what happens if we reduce the "pounding"?  Would the DFA a1 measurement normalize to where it should?

Thanks to a blog reader who kindly did some tests, we have some confirmation that this might be true.

He did a series of intensity ramps under three conditions, all using the same HRM (H10)

  • Bike
  • Stepper/elliptical
  • Running on a treadmill

What I have done is simply plotted the heart rate and DFA a1 over the course of the intensity ranges.  Normally, if your DFA a1 is in a .75 range cycling at a HR of 150 for example, we would expect similar behavior while running with the same HR.

  • The Run test in red clearly shows the early DFA a1 drop at low HR.
  • The green (stepper) and black (cycle) show similar and more realistic DFA a1 behavior.
  • The stepper/elliptical machine is using similar motor kinetics as the run, just without the impact.


His HRVT on the bike:

  •  The HRVT HR is about 146 bpm - a very reasonable figure with good R correlation


  • Early DFA a1 decline when running with a typical HRM chest belt occurs is a subset of individuals.  However, most do not have this issue. 
  • It is not an issue on the bike.
  • In this case, simply decreasing the ground force impact totally eliminated the effect.
  • It seems to be due to some sort of electromechanical interference related to impact.  Since there is no significant impact on the bike or elliptical stepper, the effect on DFA a1 does not occur.
  • Can this be compensated for?  I can't say, but the HRM designers and engineers may be able to adjust their algorithms to eliminate some of this.  The first step (no pun intended) is to confirm that the phenomena exists (done), why it happens (maybe) then hopefully find the solution.
  • In the meantime, trust the bike results but an alternate solution is using an elliptical trainer, stepper, staircase as workarounds.


  1. Bruce, do you think that a forehead sensor, like the Moov sweat, would suffer the same problem ?

    1. Unfortunately the optical HRMs just don't have the RR accuracy for a1 - but yes, that would be a good thing to have tried otherwise.

    2. Hello Bruce,
      I did everything I could, I took many test, varying the ramp, the power, the speed and the grade. In cycling I get consistent results, I can clearly see DFA declining and going up, with Kubios, and I know now where is my LT1.
      In running it is impossible for me: I also did a couple of brick workout, bike+run: LT1 at 132-135bpm, but when I go on the treadmill, at 121 my DFA is already at 0.60.
      You identified the issue, now: do you think that Exoskin should suffer the same problem when running ? I was thinking of buying one, if there are no other sensor that could help me.
      Or my I trust the bike LT1 and assume it is (more or less) the same for running ?

      thank you very much for all your work

    3. You certainly have done a good job with this. I would trust the bike data for running, it should not be much different. The hexoskin actually doesn't do the best at a1 either with it's lower sample rate. So just go with your bike threshold for running and future progress in your fitness.

  2. I've been puzzling over this exact issue for over 2 weeks. At first, I thought it was due to running injury, but I could never explain why my cycling VT1 HR was so much higher than my running VT1 HR using DFA. This makes so much more sense. Thank you!

  3. I think I might be observing this but while cycling. Sudden drop of a1 even at very low intensity (50% FTP). It happens 80-90% of the time (hence I know it actually might work as expected with my setup). I use Polar H10 over Bluetooth, FatMaxxer, almost no artifacts reported at auto. After reading here it might be some kind of interference I think I'll try experimenting with different HR strap positions.

    1. You might try rotating the H10 a couple of inches to the left. That gets a stronger signal and in some people, a more symmetric waveform. If you have Fatmaxxer, you can take a look at the signal voltage depending on belt position (developer mode and preferred device both needed to get an ECG snippet). Let me know if this helps.

    2. I did experimented a bit, observing ECG waveform. Moving sensor slightly to the left in fact gives symetric waveform (contrary to central position). Now I need to see if it makes any difference with a1. I'll report my findings.

      I won't lie, it makes a little bit worried that the problem might be not equipment but actually my heart/ANS. Although it is puzzling that recent ramp test gave me results that made perfect sense (HRVT1 at ~75% of FTP), and since then all I got was quite nonsensical data (continuous 0.5-0.6 at 50-60% FTP). Especially that my training volume and intensity weren't that high lately, so excessive fatigue should not be the case.

    3. Your experience is the same as mine with sensor placement - I now keep it a little to the L.
      I would be happy to take a look at your ramp tests - send me a link to the original one that made sense and a repeat test(using the same protocol so we can compare). You can post it here or send it to me via twitter.

    4. Report 1: I did an indoor ride with sensor rotated 15mm to the left (position I figured out by looking at EKG waveform). And It was the first ride in a long time from which I got data that makes sense. Before a1 usually was dropping below 0.75 at 150W (50% FTP), and at 180W (60%) a1 stayed around 0.5. Today I did two sets of 15 minutes long steady state intervals. The result: 165W - average 0.86 and 1.03, 180W - 0.82 and 0.97, 195W - 0.78 and 0.84. All of those were significantly below my assumed VT 1, RPE easy (1-2 out of 10).

      Now I need to do apples to apples comparison. Change H10 position every 15 minutes or so and see if it affects a1 while holding the same prescribed wattage.

    5. Here are editable spreadsheet graphs on Google Drive. I included only indoor trainer sessions done in air conditioned room for better comparability. I observed similar thing (a1 seemed much too low) on every outdoor ride lately though.

    6. I had a look at everything and will report out your data over the next day or two. Except for the double ramp you did, all looks good. Since I don't have the "raw" RR, we don't know what the true artifact situation looks like, which could have been an issue with the double ramp. If you are riding outdoors in high heat/humidity, a1 can be driven lower.

    7. Here you are - your data is on the bottom of the page. Nice looking ramps and histograms.

    8. Thank you, I'll take a look.

      I added another graph for a ride I did today (switching sensor position during a ride) with raw data from Fatmaxxer included. I also included the raw data for mnetioned double ramp test (10-08, I only have features and rr saved from that day). I observed the difference between two positions varying between 9% and 40% (comparing consecutive 10 min. steady state intervals).

      As of artifacts I rarely get any at all. I remember 2 or 3 premature beats which I actually felt and saw them being filtered out by Fatmaxxer. Similarly with artifacts from moving a strap today while recording.

    9. You're surely right about my outdoor rides. I assumed that heat played big part in much lower a1 I observed, hence I didn't used them in my comparisons. And the weather was rather brutal during my outdoor rides lately (either 32C av. temp. or very high humidity). -> "The results from this study show that exposure to high ambient temperature produces a significant increase in HR through an increase in sympathetic and decrease in parasympathetic drive."

  4. Thanks for the reference. I did do a look at skin/core temps in a previous post - interesting subject. I'll look at your new data shortly.