Determining CP + W' correctly for use in the W'bal model

[Nathan Townsend]

Recently I've made several posts outlining why W'bal goes negative if you use FTP instead of CP in the model.  This occurs mainly because FTP underestimates CP.

In most cyclists (excluding elite and pros), CP lies in the range 20-30MMP.  So you simply do a TT in this duration range, and I recommend this as a doublecheck of your CP, but it does not give you the value of W'.  For that you need to conduct short TTs also.

I do not recommend the 3min AOT test because during the last 30sec of this test you are still recruiting type IIx fibres.  These fibres are fatigued but they still produce some force. Therefore, the potential to overestimate CP exists (especially for well trained cyclists) because this essentially violates the concept of CP, which is that type IIx fibres are not being recruited.  Secondly, getting access to the lab equipment is difficult to perform this test.

The recommended "gold standard" to estimate CP is to conduct 3-5 time to exhaustion tests as constant power in the range 2-15min.  Again, the problem with this is that not everyone has access to a cycle ergometer that has a constant power mode.  Secondly, TTe tests are notoriously variable due to psychological factors.  At the point of exhaustion, if I were to offer you $10,000 to continue for another 30 s. I reckon you'd give it a red hot go.

A practical field method has been validated in a couple of studies (see links below).  These use standard self-paced TTs instead of TTe tests over the same duration. There are some important points to consider though when conducting self-paced TTs.  

1.  Most important is that the TT is done as evenly paced as possible. Going out too hard could cause premature fatigue, whereas going out too easily might lead to some W' being left in the tank at the end  (ie: you do not attain VO2max)

2. The key to estimating CP properly is that we want the shortest and longest duration which elicits VO2max.  2min might just not be long enough in all cases, so I would recommend 2.5 or 3min to be sure.  Anything over 15min (if evenly paced) begins to approach the maximum duration that humans can sustain which elicits VO2max ie: a TTe test lasting >20-25min will not induce VO2max.  Task failure occurs in combination with other reasons such as increased central fatigue, that are less present at shorter durations.  So to be safe, I believe it is wise to cap the long end at around 15min.  For this reason also it should be obvious that the 3/20min test has the potential to underestimate CP.

3. The durations should be evenly spaced.  Thus duration of 3,8,14min would be preferable to 2,5,15min for example.

Lastly, a small modification that I believe could be used to prevent an underestimation of W' by a conservative pacing error, is simply to commence your "end spurt" about 30sec earlier than you otherwise normally would.  If at the end of the TT, the power is still rising and especically if it is above the average power, this would indicate that you have W' left in the tank (thus inducing a small underestimation).  However if the power is decreasing at the end, and it is below average power, then it implies you've expended virtually all that you can.   If the pacing is done very evenly though, we do not expect a large end spurt to occur anyway. 

If you have 3-5 tests in the range 3-15min you can then model CP + W' using several methods including the nonlinear 2p model which does a hyperbolic curve fit, or the 2p linear models with fit a linear regression to either power vs 1/time (s), or work vs time (s).   You could do all three models and take the average. 

Another reason why the 3/20min test is potentially problematic is because you are restricted to the linear models only.  In this case the 3min test has a large effect on the value of CP.  This makes a little sense physiologically, but can be a misinterpretation of the data.  For example, lets say we have two cyclists: A: 3MMP = 450; 20MMP = 300, and B: 3MMP = 400; 20MMP = 300.  In this case, cyclist A will have a lower CP by about 10W despite the fact the highly aerobic 20MMP test result is the same in both.  Cyclist A will also have a much larger W' in order to compensate.   This is a valid explanation, however the exact same results could occur if cyclist A had a higher CP, and only a marginally higher W'.  This problem starts to get ironed out if you use the nonlinear hyperbolic model and especially if you use 4 or 5 evenly spaced TTs to fit the data.

So there you go.  In a single training session you could go find a decent bit of road where you can hammer out a 14-15min TT, then do all three tests with an approx 30min recovery gap in between.  You could then go an do a 25MMP to doublecheck the CP value or even include in the average of the other CP model estimates.  For standard interval sessions  conducted under similar environmental conditions as the test method of CP+W', you should find that exhaustion will occur when W'bal = 0 +/- 1.5 kJ.  It's not perfect though and Phil's paper using short intervals and recoveries (20s on / 10sec off) revealed a larger error in W'bal at fatigue. 

Validity and reliability of critical power field testing.

http://www.ncbi.nlm.nih.gov/pubmed/25260244

High agreement between laboratory and field estimates of critical power in cycling.

http://www.ncbi.nlm.nih.gov/pubmed/24022574

[Mark Liversedge]

Hi Nathan,

Thanks for taking the time to post here, this is excellent stuff. For the 3.2 release we're updating the website to include some 'practical' advice to go alongside the science section. I will reuse this post to advise on the very best way to estimate CP and W'.

Mark

[Nathan Townsend]

On Wednesday, 8 April 2015 12:28:17 UTC+3, Mark Liversedge wrote:

Hi Nathan,

Thanks for taking the time to post here, this is excellent stuff. For the 3.2 release we're updating the website to include some 'practical' advice to go alongside the science section. I will reuse this post to advise on the very best way to estimate CP and W'.

Mark

 

My pleasure Mark.  Viva la open source software!

One last caveat, I mentioned in another thread that the veloclinic plot is an excellent method also to doublecheck the CP estimate.  So for example, out you go and do 3-5 TTs in the range 2.5 - 15min, and you add these efforts to your overall P(t) curve.  Then you tweak the value of CP as per Mike's recommendations (http://veloclinic.com/veloclinic-plot-w-cp-subtraction-plot/).  What I expect will occur with this plot is that if CP is tweaked correcly, then the little mini peaks that occur in the curve will correspond to the 3-5 full on TT max efforts you added to the overall P(t) curve when coming up with the original CP estimate.  However, if you have a lot of really maximal efforts within that range of 2.5 - 15min... and I mean REALLY maximal.... like falling off the bike and writhing on the ground in agony post effort, then what will happen is that the curve will appear flatter with less variation across the top (which thus corresponds to your true value of W').

I'm really very impressed by what Mike came up with here and I'm not sure if people really understand how significant this advancement is to the CP concept. 

So now we have 3 methods to check or tweak the CP value following an initial series of TTs:

1. Veloclinic plot "shape"

2. W'bal at exhaustion = 0 +/- 1.5 kJ

3. A TT in the range 20-30min.  

de Lucas et al (2013) reported a mean(SD) duration of 22.9 +/- 7.5 min during constant load cycling @ CP in trained cyclists (CP = 295 +/- 39W). 

Time to exhaustion at and above critical power in trained cyclists: The relationship between heavy and severe intensity domains

http://www.sciencedirect.com/science/article/pii/S0765159712000639

[Nathan Townsend]

http://veloclinic.com/veloclinic-plot-w-cp-subtraction-plot/

Sorry one last thing on this, look at the peak value of W' that occurs when CP is underestimated.  It is around 30 kJ.  Again, going back to the 3/20 test we can see that 30 kJ is a typical value when using this test, but it is way above typical values we see for the 2p nonlinear model which is around 10-20 kJ.  

It should be easy to understand why this occurs from a physiological perspective, if your CP is set too low, then if you cycle 5% above this value but say 1% below your real value, then according to the CP model, you will be expending W' (and thus increasing its value), but in reality you are still at a physiological steady state and therefore you are not rapidly developing peripheral muscle fatigue and not expending W'.  Then when it comes to the end spurt in this long interval, you raise the power above CP and truely expend all of your W', thus adding to the erroneous W'exp accumulated previously and massively inflating your W' estimate.

[Pete from AUS]

Is there still more work needed to improve the W'Bal model?

[mike veloclinic]

Nathan,

Thanks for laying out a standardized approach to determining W' and CP. Regarding the peaks on the Veloclinic plot yes you are absolutely right that each peak should correspond to a tested fresh fall off the bike effort. Attached is my n=1 personal validation study and you can clearly see the 4 peaks of 4 maximal efforts and a sprint effort. Each effort was done fresh and continued to the point of desperate flailing accompanied by embarrassing primordial throat sounds. I am planning on getting the plot introduced formally into the scientific literature once we've collected enough data to do so. 

I want to add a couple of observations on using the VC plot:

1. I view it as a diagnostic plot. The actual parameters should come from fitting the CP model to the test data directly and then using VC plot to visualize if there are problems with your fit as compared to your full PD curve.

For example, if you are not visually convinced that there is a clear W' plateau (at least 3+ peaks of similar height) then the number or quality of the tests is probably inadequate to give you reliable parameters estimates.

2. With the VC plot take advantage of the ability to consider data well outside the recommend range or sub-maximal data. Since the VC plot is not doing a "fit" of the data it doesn't matter if you include data outside the CP model valid range or submaximal data. In fact, a good confirmation that you have a good fit is that sub-maximal data, or data from efforts shorter than 1-2 minutes or longer than 45+ minutes should fall clearly below the test data peaks in the 3-15 min ish test range. One of the shortcomings of the Critical Power is the narrow range of validity. For a typical W' of about 20,000 J the time range only translates into a power spread of about 90 watts or about 35% change for a typical CP. That's not a whole lot considering the noise in the system. If you can use the plot to stretch the range considered out to 1-60 minutes the power spread bumps up to something more like 250-300+ watts reducing the effect of the noise in the system. 

A minor testing tip to add is that for the shorter intervals try flying starts with a lead in just below your expected CP. With the lead in you will get the VO2 kinetics up without depleting significant W' decreasing the Wilkie effect caused by VO2 lag. From the attached jpeg the peak around 600 watts was an effort just shy of 1 minute done with the flying start. 

A last tip that is a general one, remember that models are dumb. They don't know if the phenomena found in your data is due to the physiologic mechanism that you are trying to interrogate or from a combination of other unrelated factors. For example as Nathan has previously pointed out, TT efforts longer than 15-20 minutes tend to show a mixture of central fatigue and incomplete peripheral fatigue. However, that doesn't mean you won't find a hyperbolic shape out -past the valid range of the Critical Power model. So always take a minute to step back and consider whether the results make a damn bit of sense. 

One caveat to consider as well,  be skeptical if you can stay on the CP model curve past 30-40 minutes don't necessarily dismiss your CP estimate just because you can hold it past 30-40 minutes. As Nathan has previously discussed, TT's longer than 15-20 minutes result in a shift from peripheral fatigue (and development in VO2max) to less peripheral fatigue and the development of central fatigue. If you are developing central fatigue and can not deplete W' (ie achieve full peripheral fatigue) there can not be enough power to keep you up on the Critical Power model curve. However there will be a transition zone were you will stay above CP though below the CP model curve because you can no longer fully deplete W'. At some point though, either because there is too much central fatigue to recruit enough motor units to achieve CP and/or because the fatigue resistant motor units have also fatigued you ultimately fall below Critical Power as well. I think it would be an interesting research question to start to understand how the relationship between central fatigue and percent of W' can be depleted as well as how much central fatigue can be tolerated before power falls below CP and when that happens for most trained individuals. 

mike 

[Jens]

Mark, if CP is higher than FTP we should be allowed to set each value seperately. CP to get the best out of the W'bal model and FTP to get the TSS right.

Jens

[Nathan Townsend]

On Friday, 10 April 2015 02:54:00 UTC+3, Pete from AUS wrote:

Is there still more work needed to improve the W'Bal model?

Do humans know everything there is to know about the universe?

Answer this question and understand why I asked it, and you have the answer to yours  ;-) 

In particular, we have no mathematical model which specifically attempts to describe the effect of central fatigue, fatigue induced locomoter muscle inefficiency and substrate depletion on the decline in both CP and W' that inevitably occurs over time.  These could occur at different rates during intermittent exercise vs constant load.  We also need to better describe what happens at altitude.  I'm presenting a paper on this at ECSS this year and conducting a follow up study starting data collection in a couple of weeks. We also need to understand what happens during exercise in the heat.  There is no doubt that CP decreases and probably W' at some stage also, but unlike altitude (in which CP decreases immediately, whereas W' stays about the same), in the heat it is likely that it will take time before core temp rises and sustainable power takes a dive.  Thus a time dependent effect occurs in the heat. 

[Nathan Townsend]

On Friday, 10 April 2015 10:22:15 UTC+3, mike veloclinic wrote:

Nathan,

 

A minor testing tip to add is that for the shorter intervals try flying starts with a lead in just below your expected CP. With the lead in you will get the VO2 kinetics up without depleting significant W' decreasing the Wilkie effect caused by VO2 lag. From the attached jpeg the peak around 600 watts was an effort just shy of 1 minute done with the flying start. 

Ha, I think I may have to give you some credit in a future paper (or indeed we just collaborate and you're a co-author!).  This is a great idea, which I will look at implementing, but it also serves another purpose which is that standard VO2 kinetics could measured at the onset of the test prior to the TT phase.  However, it still leaves Wilkie's correction absent from a W'bal model.

In case people don't quite understand what we're talking about here see the article below.  Essentially, one of the issues with the CP model is that it assumes instant VO2 kinetics, ie: when you step up to a higher workload from beneath to above CP, the model assume you are instantly producing ATP aerobically commensurate with CP.  In reality this doesn't happen, there is a lag, and during that lag you're using W'bal to make up the shortfall in energy.  So typically W' is underestimated compared to MAOD (see below).  If you conduct the TT though starting from a higher workload, this effect is reduced because O2 kinetics are already up to speed.

Validity of the two-parameter model in estimating the anaerobic work capacity

http://www.ncbi.nlm.nih.gov/pubmed/16261386

[Nathan Townsend]

To recap and summarise additional info from above discussion:

1.  Do 3-5 evenly spaced time trials in the range 3 - 15min.  I would not do more than 3 TTs in one session, and always allow 30min recovery between efforts (I'm open to suggestion about the order but I'm tempted to suggest the 7-8min test first followed by the 12-15min test, and finish with the shortest.  The reasoning here is that if there is any priming effect of the 7-8min effort, then your 12-15min should be solid.  Then mentally a 3min effort at the end should be easier to deal with than a long TT). 

2. Use a false flat if available, otherwise I would recommend a long course outdoor velodrome (which usually does not have sharp bends).  It is a nice controlled and safe environment. 

3. Add a 4-5min period of steady state cycling immediately prior to the commencement of the effort.  I would be cautious of getting too close to CP for this effort, but also you don't want it to be too far below CP.  Therefore, I think around 80-85% CP should be about right.  

4.  Even pace as much as possible.

5. Ideally, power should be declining as you finish the TT, such that it is below the average power you were sitting on for the bulk of the TT. (eg: if you find yourself holding 430W for most of the 3min effort, but in the last 30sec you're giving it absolutely everything you've got and the power is slipping downwards through about 350-400w, then you're not at risk of underestimating W' due to too conservative pacing).  

If all is well then you should see 3-5 fairly even peaks in the veloclinic plot where x = average power of each individual TT effort.  Adjust CP as necessary and conduct additional testing if required.  A standard practice is to check the SEE of the 2p nonlinear hyperbolic curve fit.  If the SEE for CP is >5W and/or W' is > 1.5 kJ then either repeat the outlier test, or do another test at a different duration. 

[Nathan Townsend]

On Sunday, 12 April 2015 13:52:07 UTC+3, Nathan Townsend wrote:

To recap and summarise additional info from above discussion:

4.  Even pace as much as possible.

Actually I think a positive split is preferable here (ie: power is gradually declining throughout the test).  Thus a fast start, but not "all-out".

see below.....

Influence of all-out and fast start on 5-min cycling time trial performance.

http://www.ncbi.nlm.nih.gov/pubmed/19727014 

Effects of starting strategy on 5-min cycling time-trial performance.

http://www.ncbi.nlm.nih.gov/pubmed/19724963

[Rob Manning]

So reading through all these posts (and Nathan and Doc, thank you so much for taking the time to post these.  It takes my old ex phys knowledge and sort of turns it upside down a little bit, forcing me to revisit everything.) it becomes clear that we're looking for 3 to 5 self paced TTs, evenly spaced, between 2 and 15 minutes.  So for example, 2, 5, 8, 12, 15 minutes, for example?  Extra data points would remove the heavy reliance on any single point.

With GC, the modeling only asks for 2 points.  

Is there any way (mark, more question for you) to create a model that will take into account all these points, fit the curve to these points and extrapolate out?

Further, it seems fairly simple in terms of training setup when you consider this approach:

CP is 20-30MMP (either tested or extrapolated in the above conditions?)

Ignore FTP for <60 minute efforts (does it even have value at that point?)

Use CP or xMMP as the standard for which your training and interval work is to be done?

So for example, if my training calls for 4 minute efforts, I should be using my 4 minute MMP as the "benchmark" for codifying the intensity of those intervals, yes?  NOT my FTP or 20-30MMP/CP?

Isn't this just a hugely simpler way to train without using all the percentages and such?

-Rob

[Ruud Goorden]

Where have we seen this one before?

http://www.fredericgrappe.com/wp-content/uploads/2011/07/The%20record%20power.pdf

[Mark Liversedge]

On Friday, 17 April 2015 16:34:50 UTC+1, Rob Manning wrote:

So reading through all these posts (and Nathan and Doc, thank you so much for taking the time to post these.  It takes my old ex phys knowledge and sort of turns it upside down a little bit, forcing me to revisit everything.) it becomes clear that we're looking for 3 to 5 self paced TTs, evenly spaced, between 2 and 15 minutes.  So for example, 2, 5, 8, 12, 15 minutes, for example?  Extra data points would remove the heavy reliance on any single point.

With GC, the modeling only asks for 2 points.  

Is there any way (mark, more question for you) to create a model that will take into account all these points, fit the curve to these points and extrapolate out?

When you look at the CP chart it applies the selected model to the data available.

So (a) choose the model in chart settings (b) use a meaningfull date range (last 1 or 2 months) and the TTE values should be pretty obvious. See this tutorial: https://vimeo.com/100244204

If you want to go one step further you can update the metadata to add a checkbox for tests and set it for the test rides and then apply a filter in the chart to only plot for tests. See this tutorial: https://vimeo.com/107577461

We could also update the dialog tool to be more sophisticated for when there is no ride data and allow multiple points to be entered. Can look at that for v4.

Cheers

Mark 

[Mark Liversedge]

On Friday, 17 April 2015 21:33:58 UTC+1, Ruud Goorden wrote:

Where have we seen this one before?

http://www.fredericgrappe.com/wp-content/uploads/2011/07/The%20record%20power.pdf

Shhhh. Don't go trip-trapping across that bridge. Its already been done to death.

Mark 

[Nathan Townsend]

On Friday, 17 April 2015 23:44:31 UTC+3, Mark Liversedge wrote:

We could also update the dialog tool to be more sophisticated for when there is no ride data and allow multiple points to be entered. Can look at that for v4.

Cheers

Mark 

Hi Mark, I'm not 100% sure how the auto CP + W' calculator in GC works.  As you know, the problem with using field derived powermeter data recorded during training is that not ALL durations represent true maximal efforts because nobody does a true maximal over every duration in 1sec increments. So we get those little steps occurring whenever there is a maximal effort.  My understanding is that you created an algorithm in GC which detects the edge of those little steps and then uses those values to fit a hyperbola.  Is this correct?

In terms of using non-dedicated performance testing over specific durations, this method should be virtually bulletproof as long as there are 3-5 true max efforts in the range 3-15min and the algorithm does not look outside of that window. However, if you cannot be certain that 3-5 full maximal efforts exist within the 3-15min range, then you will always obtain an error in either CP or W', or both.  

I use a scientific graphing program called Graphpad PRISM to fit the hyperbola (and say for example, modeling VO2 kinetics), but for a coding guru such as yourself, then surely it should be possible to write a small, independent module that runs a least squares hyperbolic curve fit on manually entered values?

 

[Randolph Baral]

I would've thought that 'non-dedicated truly maximal efforts' are most likely to be achieved on hills (for most) which would then overestimate CP (for non-hill efforts).....please correct me if I'm wrong.

Other than that, the points on the steps most likely indicate maximal efforts and might be good to use for selection of, for example, a 3:43 or a 16:37 effort achieved in races that might be greater than dedicated testing could produce.

I have previously thought that the ability to select intervals based on gradient would be a useful feature (eg best 5min power with <2% gradient...or whatever time period or gradient)......maybe in version 5! :-)

I really appreciate the discussion.

Randolph

[Mark Liversedge]

On Sunday, 19 April 2015 11:39:52 UTC+1, Nathan Townsend wrote:

Hi Mark, I'm not 100% sure how the auto CP + W' calculator in GC works.  As you know, the problem with using field derived powermeter data recorded during training is that not ALL durations represent true maximal efforts because nobody does a true maximal over every duration in 1sec increments. So we get those little steps occurring whenever there is a maximal effort.  My understanding is that you created an algorithm in GC which detects the edge of those little steps and then uses those values to fit a hyperbola.  Is this correct?

Yes.

To derive CP and W' parameters we perform a brute force fit against the points on the bests curve.

To limit the number of points we use to do this (otherwise it will take ages) we only look at specific durations. These durations are defined by the user in the CP chart settings. 

The brute force fit basically looks for minimal values of W'/CP as the duration gets longer.

I use a scientific graphing program called Graphpad PRISM to fit the hyperbola (and say for example, modeling VO2 kinetics), but for a coding guru such as yourself, then surely it should be possible to write a small, independent module that runs a least squares hyperbolic curve fit on manually entered values?

We do exactly that for the trend curve on long term metric charts (quadratic trend), its already available for us (LTMTrend2.h for any devs that want to play). It would be trivial to apply apply a brute force to a fitted curve in each duration instead of the actual data points. 

But I need a good reason to do that !

SO ...

Mike P, aka veloclinic, has shared some thoughts offline about how we account for PM accuracy. 

He highlighted the fact that we know powermeters are typically accurate within 2%. And since we always use the peak values in GC then our bests and estimates are likely to be skewed 2% higher than reality.

But I argue, that this skew is also present in the way we track performance. i.e. we look for the peak powers over time and these will also be skewed to the upper values. So effectively, when we consider peaks we will always be skewed by the relative accuracy of the power meter.

Alternatively whatever 'statistical' algorithm we apply to the PD data to establish our model inputs would also need to be applied when we extracted and presented peak powers. Since they will also be skewed to the power meter accuracy.

So, I figure, we're using a model and peak powers that are both skewed, which is far better than applying different approaches; i.e. comparing an estimate that attempts to take into account PM accuracy with a peak power that does not (assume this is how WKO4 will work).

WHICH LEADS ME TO ...

Why are you fitting a hyperbola to the TTE efforts and not taking the best one ?

Cheers

Mark

[Mark Liversedge]

On Monday, 20 April 2015 08:51:04 UTC+1, Mark Liversedge wrote:

The brute force fit basically looks for minimal values of W'/CP as the duration gets longer.

Of course I mean't maximal ! 

[Nathan Townsend]

 All excellent points Randolph.  Let me address them individually....

On Monday, 20 April 2015 09:57:42 UTC+3, Randolph Baral wrote:

I would've thought that 'non-dedicated truly maximal efforts' are most likely to be achieved on hills (for most) which would then overestimate CP (for non-hill efforts).....please correct me if I'm wrong.

I tested myself on hills and on the flat and there is no question this effect is real. There is no study in the literature which systematically addresses this point (although it is a damn good idea for a project).  At present the only advice I can offer is what I mentioned above.  Do the test on a false flat if possible.  This way it isn't too far away from flat terrain and it shoul dnot underestimate hill climbing ability too much. 

 

Other than that, the points on the steps most likely indicate maximal efforts and might be good to use for selection of, for example, a 3:43 or a 16:37 effort achieved in races that might be greater than dedicated testing could produce.

Yes, if you have an effort which is your local hotly contested strava segment and your PB is 3:43. Then it makes perfect sense to include this duration in a hyperbolic curve fit. It also makes sense to use your best efforts for modelling purposes rather than dedicated test day, because you might slightly underperform on test day, however as stated, you then risk overestimating performance capability on flat terrain if all your PBs are on hills.  I don't think there is any clear answer to these issues because well, the physical nature of reality is pretty damn complicated in general and we just don't have sufficient understanding or computing power to account for everything.  I also don't think it is necessary to overcomplicate things and so if you stick to a few basic guidelines then your estimates will be close and then you can tweak them slightly depending on the terrain.

 

I have previously thought that the ability to select intervals based on gradient would be a useful feature (eg best 5min power with <2% gradient...or whatever time period or gradient)......maybe in version 5! :-)

For sure, a bandpass gradient filter would be cool.  Mark, there is an idea for you.  It could be known as the "Baral gradient filter"

[Mark Liversedge]

:)

In v4 we are going to extract intervals from all rides to enable charting and analysis like this. I think It would be interesting to start another thread on the criteria we want to use. Some of the ideas include starting W'bal and starting altitude as well as W' consumed / meters climbed.

Cheers

Mark

[Randolph Baral]

"Baral gradient filter"

Ha ha!  My ego doesn't need stroking that much! :-)

(something like 95% of eponyms are by males....women are less testoesterone fueled so as not to name things after themselves).

I track my top 10x 20min power and 5min power for each season....and make a mental note if I cracked the top 10 when on the flat....which I last weekend! (not quite OCD enough to actually note the terrain of the top 10)

damn good idea for a project

My initial thought was that I would be happy to provide data and I'm sure others here would be too....I wonder how much calibration/zero offset of power meters matters since the study would be looking for variance between hills and flat (as opposed to variance between the power meters).  Even a pilot study of 10 athletes with 10 data points (all terrains then sort out by gradient) at 2 different durations gives 200 data points which would have decent power.

Randolph

[Mark Liversedge]

On Tuesday, 21 April 2015 00:13:08 UTC+1, Randolph Baral wrote:

damn good idea for a project

My initial thought was that I would be happy to provide data and I'm sure others here would be too....I wonder how much calibration/zero offset of power meters matters since the study would be looking for variance between hills and flat (as opposed to variance between the power meters).  Even a pilot study of 10 athletes with 10 data points (all terrains then sort out by gradient) at 2 different durations gives 200 data points which would have decent power.

I'm thinking a scatter plot for ride and interval metrics.

In this case x shows avg gradient and y shows AP.

With the trend line to show any correlation.

We can then all share results quite easily ?

.. and I was looking for an excuse to add a metric scatter plot anyway :)

Mark 

[Pete from AUS]

For a whole ride?

[Mark Liversedge]

Yes at this point (v3.2) I think a LTM Scatter that plots metrics on an x/y is appropriate.

BUT THINKING ABOUT V4 ...

We need to think very carefully about the UI for 'Interval analysis' 

At this point we are planning to have a separate view for Intervals (Trends, Diary, Intervals, Activity, Train) and allow you to select intervals in pretty much the same way we select rides via a list.

Now I'm much more interested in working with 'sets' of intervals rather than individual ones and haven't got a clue how we are going to manage that in the 'intervals view sidebar'. I think we may need to do it with a 'query' type interface rather than a list; we will (hopefully) extract many thousands of intervals from a season's worth of rides. It will not be sensible to expect the user to select from a list.

But open to any ideas or references to other tools that do this well * !

Mark

* and yeah, we might apply the same approach to the Trends and Interval view....

[Armando Mastracci]

Another approach for shorter durations tests, since there will likely still be available W' at the end of the effort, continue at, say, 25W, above your expected CP for as long as you can and add that to your W'.

So for example, for a 3min effort at 400W, at the end of your effort you drop it down to 275W and you are able to hold that for 40s more, this would be one test in support of a CP of 250W and W' of 23.5kJ not 22.5kJ.

You'll find this approach helps address poor-pacing or underestimation you may have done.  Note that in general, you will find that the shorter the durations, the longer you can sustain efforts just above your CP afterwards.  

[Daniele Morbidelli]

I am also experiencing a negative values of W'bal during some high intensity workouts and I also believe it is probably due to an understimation of the CP value.

From what I understood it should lies in the range of MMP20/MMP30 and I read that you suggest some short TTs to determine it.

Does it mean the value calculated automatically by GC is a "different CP"?

In my case for instance, the values calculated by GC are:  CP=283W, W'=31,7KJ while best 30min is 300W.

[Brad]

What's the point of completing 3-5 time to exhaustion tests if the 2 parameter CP model is set up in GC to search an interval in the Anaerobic range and another interval in the Aerobic range? Say you completed 3 tests of 3, 8 & 14min would you set up the model to search for intervals between 120-180sec for the anaerobic range to include the 3min test and 780-900sec range to include the 14min test into the CP calculation and not include the 8min test? So how do you include the extra 1-3 intervals into the CP calculation?

On Wednesday, 8 April 2015 18:28:50 UTC+9:30, Nathan Townsend wrote:

Recently I've made several posts outlining why W'bal goes negative if you use FTP instead of CP in the model.  This occurs mainly because FTP underestimates CP.

In most cyclists (excluding elite and pros), CP lies in the range 20-30MMP.  So you simply do a TT in this duration range, and I recommend this as a doublecheck of your CP, but it does not give you the value of W'.  For that you need to conduct short TTs also.

I do not recommend the 3min AOT test because during the last 30sec of this test you are still recruiting type IIx fibres.  These fibres are fatigued but they still produce some force. Therefore, the potential to overestimate CP exists (especially for well trained cyclists) because this essentially violates the concept of CP, which is that type IIx fibres are not being recruited.  Secondly, getting access to the lab equipment is difficult to perform this test.

The recommended "gold standard" to estimate CP is to conduct 3-5 time to exhaustion tests as constant power in the range 2-15min.  Again, the problem with this is that not everyone has access to a cycle ergometer that has a constant power mode.  Secondly, TTe tests are notoriously variable due to psychological factors.  At the point of exhaustion, if I were to offer you $10,000 to continue for another 30 s. I reckon you'd give it a red hot go.

A practical field method has been validated in a couple of studies (see links below).  These use standard self-paced TTs instead of TTe tests over the same duration. There are some important points to consider though when conducting self-paced TTs.  

1.  Most important is that the TT is done as evenly paced as possible. Going out too hard could cause premature fatigue, whereas going out too easily might lead to some W' being left in the tank at the end  (ie: you do not attain VO2max)

2. The key to estimating CP properly is that we want the shortest and longest duration which elicits VO2max.  2min might just not be long enough in all cases, so I would recommend 2.5 or 3min to be sure.  Anything over 15min (if evenly paced) begins to approach the maximum duration that humans can sustain which elicits VO2max ie: a TTe test lasting >20-25min will not induce VO2max.  Task failure occurs in combination with other reasons such as increased central fatigue, that are less present at shorter durations.  So to be safe, I believe it is wise to cap the long end at around 15min.  For this reason also it should be obvious that the 3/20min test has the potential to underestimate CP.

3. The durations should be evenly spaced.  Thus duration of 3,8,14min would be preferable to 2,5,15min for example.

Lastly, a small modification that I believe could be used to prevent an underestimation of W' by a conservative pacing error, is simply to commence your "end spurt" about 30sec earlier than you otherwise normally would.  If at the end of the TT, the power is still rising and especically if it is above the average power, this would indicate that you have W' left in the tank (thus inducing a small underestimation).  However if the power is decreasing at the end, and it is below average power, then it implies you've expended virtually all that you can.   If the pacing is done very evenly though, we do not expect a large end spurt to occur anyway. 

[Mark Liversedge]

On Wednesday, 12 August 2015 13:15:35 UTC+1, Brad wrote:

What's the point of completing 3-5 time to exhaustion tests if the 2 parameter CP model is set up in GC to search an interval in the Anaerobic range and another interval in the Aerobic range? Say you completed 3 tests of 3, 8 & 14min would you set up the model to search for intervals between 120-180sec for the anaerobic range to include the 3min test and 780-900sec range to include the 14min test into the CP calculation and not include the 8min test? So how do you include the extra 1-3 intervals into the CP calculation?

Spot on Brad. Nathan, Damien and I have been discussing the needed improvements to the way GC works to match this.

We are looking at this now with a view to a number of related improvements in v4.0

Mark 

[Brad]

There's going to be alot to look forward to in v4.0 by the sounds of it.

So I guess for now it's only worth completing a 2.5-3min and a 13-15min MMP test to calculate your CP

[Mark Liversedge]

On Wednesday, 12 August 2015 14:08:19 UTC+1, Brad wrote:

There's going to be alot to look forward to in v4.0 by the sounds of it.

So I guess for now it's only worth completing a 2.5-3min and a 13-15min MMP test to calculate your CP

There's always more to come :) 

[Brad]

In CP chart settings under the 2 parameter model you have search interval Anaerobic and Aerobic ranges.  Could someone please explain what interval it would choose if you have the aerobic range set to 900-1200sec and have a 15mmp and a 20mmp, which one will it select for the data point that it uses in it's calculation for the CP? Is that even how it works by choosing one MMP duration within the selected interval range?