Yet another mapped points question

[junits15]

Hey everyone,

I'm slowly working through the coyote cookbook and I have two questions that i still haven't been able to figure out so I figured I'd ask here and see what the general opinion is!

1. I understand the function of mapped points and blending on my mustang (2019 GT) I get that the first 14 points are for when the IMRC is open and the second 14 points are for when the IMRC are closed. I understand that weights are applied to each table to get a weighted sum of any given cell across all 28 tables which generates the requested timing/torque/cam angle. The thing I don't understand is how they are functioning, or more specifically where these weights come from? See the image below:
Screenshot 2023-03-29 182129.png

This is midway through a pull I recorded on the street. MP23 is at 61.54% weight, and the rest of the tables are at other smaller weights that total up to 100%. What logic is generating these weights? Why is the ECM putting over half the weight on table 23? Moreso, over half the weight is from 23, and the rest is all small percentages from the other 13 tables in the upper half of the mapped poibts. Most of them contributing under 5% to the final value (seems weird to even bother with these small values). What is commanding all these small weighs to be tacked on? I still have a lot of confusion regarding the HDFX system in general and I'm trying to put the puzzle together bit by bit, but this is something that's been bugging at me for a while.

Am I correct in understanding that the blending weights are based off the current cam angle?


2. This is less important but something I've been wondering, I've been told that the stock Gen3 coyote can be made to hit MBT on 93, I personally haven't seen it on a stock tune but I've seen it get close during some e30 testing I did a few weeks ago. If that is true, how is it possible that people are gaining an additional 40-50hp by switching to e85 ethanol? Wouldn't you just immediately hit MBT then be totally out of room to adjust timing for more power?

Thanks in advance, I know these are likely really basic questions but I appreciate everyone's patience and help immensely!

[engineermike]

You have to envision the mapped points as locations on an xy coordinate system. It measures the actual cam positions and plots the coordinates. Then it calculates the distance to each mapped point location and weights the closest ones high and the farthest ones low. It would take more lines of code to get it to weed out the low/far points. Actually, if they wanted it to use less points, they would just run it on a snap line in OP mode like gt500, Roush, and some ecoboost. I actually run mine like that.

I haven?t seen any gen3 able to run mbt at wot on 93. That said, e85 has some power advantage by itself even without more timing. I think there?s an ethanol efficiency table that shows that.

[junits15]

You have to envision the mapped points as locations on an xy coordinate system. It measures the actual cam positions and plots the coordinates. Then it calculates the distance to each mapped point location and weights the closest ones high and the farthest ones low. It would take more lines of code to get it to weed out the low/far points. Actually, if they wanted it to use less points, they would just run it on a snap line in OP mode like gt500, Roush, and some ecoboost. I actually run mine like that.

I haven?t seen any gen3 able to run mbt at wot on 93. That said, e85 has some power advantage by itself even without more timing. I think there?s an ethanol efficiency table that shows that.

So I was kind of envisioning it like a 4d object we have load x rpm x timing x mapped point to make some 4d...thing

But from what you're saying it seems like that is the wrong way to go about it, that there is a separate mapping that links cam positions to mapped points with a 2d function. The closer you are to a mapped point on this mapping the higher the weight applied to it in the calculation.

I've seen snap line mentioned, I assume that just locks to individual mapped points instead of blending them? Is that just for ease of tuning?


As far as the last point, I'm going to take your word for this, the place I heard that MBT could be achieved on 93 isn't exactly an ideal place to be getting tuning info lol. But thats good to hear, because it aligns with what I've been reading so far and what i expect to see. Would straight e85 be enough to reach MBT?

[engineermike]

The load and rpm go into all the calculations for timing, SD, and torque, but that's specific to each mapped point. The outcomes are then blended as a function of the weighting. For instance, if MBT at a given load and rpm for mapped point 0 is 20 deg and for MP 1 is 30 deg and the blending is 70% 0 and 30% 1, then the resulting MBT will be 27 deg at that measured cam timing. IMRC position adds a second layer and it can become 3D because the IMRC could be open or closed or somewhere in between.

Think of it this way: if you lock it into 0,0 cam timing with IMRC locked closed, then it will always run mapped point 0 and, as such, all of the SD, timing, and torque data will come from the MP0 tables and no other. Once you move cams from 0,0, it starts blending in the SD, timing, and torque data from other points.

For snap lines, you draw a line between 2 mapped point locations on your XY coordinate system. If the actual cam angles are near that line, then it uses data from only the 2 points on either end of the line, but still interpolates based on distance to the endpoint. So if you make a snap line between MP0 and MP1, then any time it's close to that line, it will only weight 0 and 1 and no other. I find this makes tuning much easier because now you only have to worry with 2 points at a time rather than 12 or 26. Really, the correct way to tune the engine is to lock onto a specific mapped point and fully calibrate everything, then move on to the next point. This, as you can imagine, is super time consuming and I bet Ford has an automated process for doing it.

Straight E85 is good enough to reach MBT at 20 psi boost, so yea I think the stock NA motor could get there on it.

[junits15]

The load and rpm go into all the calculations for timing, SD, and torque, but that's specific to each mapped point. The outcomes are then blended as a function of the weighting. For instance, if MBT at a given load and rpm for mapped point 0 is 20 deg and for MP 1 is 30 deg and the blending is 70% 0 and 30% 1, then the resulting MBT will be 27 deg at that measured cam timing. IMRC position adds a second layer and it can become 3D because the IMRC could be open or closed or somewhere in between.

Think of it this way: if you lock it into 0,0 cam timing with IMRC locked closed, then it will always run mapped point 0 and, as such, all of the SD, timing, and torque data will come from the MP0 tables and no other. Once you move cams from 0,0, it starts blending in the SD, timing, and torque data from other points.

For snap lines, you draw a line between 2 mapped point locations on your XY coordinate system. If the actual cam angles are near that line, then it uses data from only the 2 points on either end of the line, but still interpolates based on distance to the endpoint. So if you make a snap line between MP0 and MP1, then any time it's close to that line, it will only weight 0 and 1 and no other. I find this makes tuning much easier because now you only have to worry with 2 points at a time rather than 12 or 26. Really, the correct way to tune the engine is to lock onto a specific mapped point and fully calibrate everything, then move on to the next point. This, as you can imagine, is super time consuming and I bet Ford has an automated process for doing it.

Straight E85 is good enough to reach MBT at 20 psi boost, so yea I think the stock NA motor could get there on it.

So the closer a cam angle pair is to a specific mapped point the heaver that point is weighted? And a snap line still blends but blends between less points? Where are those relationships defined? I feel like if I could visually look at how the relationship between cam angle and MP's is defined in the calibration file this will start to make sense to me.

That does indeed sound super time consuming, I suppose a whole day on a dyno could do it but that feels overkill for just learning. The Coyote Cookbook explains a method where we select the OP table (I don't have an active OP table in my 2019 so I guess I'd select MP28 or something), and then calculate the difference between our reference table and every other table. So MP1 at some RPM and load is X% away from MP28 at the same cell. Then tune while locked to the reference table (MP28 or OP or whatever) and use the relationships we found earlier to populate all the other tables.

I'm still reading the book so this may just be a later chapter but I appreciate the help always!

[engineermike]

So the closer a cam angle pair is to a specific mapped point the heaver that point is weighted?

Yes, exactly.

And a snap line still blends but blends between less points?

A snap line forces it to blend only the 2 points at either end of the snap line, and no other points.

Where are those relationships defined?

Not sure what you mean by this. The relationships are defined by the "location" of the mapped points on an XY coordinate system, and the snap lines and points are defined in the tune. For the snap lines, they are defined simply by from point [left column] to point [right column].

I feel like if I could visually look at how the relationship between cam angle and MP's is defined in the calibration file this will start to make sense to me.

Same here, but there is no such visualization tool in HPT Editor. I have on many occasions used graph paper or excel to plot the different points and lines, then add in the commanded angles at different loads and speeds plus add OP cam timing since it doesn't command points but angles instead.

The Coyote Cookbook explains a method where we select the OP table (I don't have an active OP table in my 2019 so I guess I'd select MP28 or something), and then calculate the difference between our reference table and every other table. So MP1 at some RPM and load is X% away from MP28 at the same cell. Then tune while locked to the reference table (MP28 or OP or whatever) and use the relationships we found earlier to populate all the other tables.

I don't follow the explanation, but you can't populate other tables using one table. The only way to make it accurate is to lock to one table and calibrate it, then lock to the next, and the next. If you're using more than one table at a time and you need to add or subtract timing, then it's a guessing game as to which to add or subtract from. Many tuners, even big ones, just make all the spark tables all the same if they don't have OEM data to work off of, such as the case of supercharged coyotes at loads >1.1.

[junits15]

Yes, exactly.
Not sure what you mean by this. The relationships are defined by the "location" of the mapped points on an XY coordinate system, and the snap lines and points are defined in the tune. For the snap lines, they are defined simply by from point [left column] to point [right column].
.

Sorry for the confusion, what I had meant by that question was where in the calibration file do we define what cam angle pairs go to which mapped points (approximately)? So if there is an XY coordinate plane and (if I'm understanding right) on one axis is say exhaust cam angle, then the other axis would be intake cam angle and the plane would then be populated with mapped points. Is that right?

Or is it hard coded? Like we cannot adjust this relationship so there's no table for it.

Yes, exactly.
I don't follow the explanation, but you can't populate other tables using one table. The only way to make it accurate is to lock to one table and calibrate it, then lock to the next, and the next. If you're using more than one table at a time and you need to add or subtract timing, then it's a guessing game as to which to add or subtract from. Many tuners, even big ones, just make all the spark tables all the same if they don't have OEM data to work off of, such as the case of supercharged coyotes at loads >1.1.

Yeah My fault I wasn't super clear on it, basically from what I understood (and bear in mind I may be wrong) they used one table as a "reference point". So on a Gen2 they used the OP table, then he loaded all of the tables into a spreadsheet and used it to calculate, cell by cell, the percent difference every other table had from the reference table.

So for example at load 1, 6k RPM, he would calculate the percentage difference every other mapped point (at load 1 6k RPM) was from the reference table. So like MP20 would be x% away at that particular load and RPM, and MP18 would be y% away at the same load and RPM, and so on and so forth. Then he would repeat this for every single load and RPM combination across all of the tables. Then from my understanding we would lock to the reference table and tune the reference table like normal. Once the reference table was tuned we would take the results of the tuning and use our spreadsheet and previously defined relationships to then populate the other tables.

Now that I think about it, it sounds like alot of headache that likely wouldn't work, it would probably be easier to just disable most of the MPs and run like that


EDIT: He did that for the TORQUE tables and not for the timing tables, I guess it wont translate

[junits15]

mapped points config.png

Is this where we link mapped points to cam angles?

[engineermike]

Sorry for the confusion, what I had meant by that question was where in the calibration file do we define what cam angle pairs go to which mapped points (approximately)? So if there is an XY coordinate plane and (if I'm understanding right) on one axis is say exhaust cam angle, then the other axis would be intake cam angle and the plane would then be populated with mapped points. Is that right?

Or is it hard coded? Like we cannot adjust this relationship so there's no table for it.

The coordinate is the IVO and EVC for that mapped point, as defined in tables 38150 and 38151. So if Mapped Point 25 the IVC is -20 and EVO is 20, then the X-Y coordinate is -20, 20.

Yeah My fault I wasn't super clear on it, basically from what I understood (and bear in mind I may be wrong) they used one table as a "reference point". So on a Gen2 they used the OP table, then he loaded all of the tables into a spreadsheet and used it to calculate, cell by cell, the percent difference every other table had from the reference table.

So for example at load 1, 6k RPM, he would calculate the percentage difference every other mapped point (at load 1 6k RPM) was from the reference table. So like MP20 would be x% away at that particular load and RPM, and MP18 would be y% away at the same load and RPM, and so on and so forth. Then he would repeat this for every single load and RPM combination across all of the tables. Then from my understanding we would lock to the reference table and tune the reference table like normal. Once the reference table was tuned we would take the results of the tuning and use our spreadsheet and previously defined relationships to then populate the other tables.

Now that I think about it, it sounds like alot of headache that likely wouldn't work, it would probably be easier to just disable most of the MPs and run like that


EDIT: He did that for the TORQUE tables and not for the timing tables, I guess it wont translate

If I'm understanding this, he's basically saying if you have a good stock tune, then you'd lock to one, perhaps central, mapped point. Then keep track of all your changes. If you add to this table at a particular load and speed, then you add to all the tables at the same amount (or a fraction of it) to all the other tables. I can see how that might speed up the process a little, assuming your changes are universal. What I've found is that things like SD and torque can be robbed from similar quality tunes (Roush or OEM). MBT as well. That only leaves Borderline needing to be locked to each mapped point and tuned. I've found it useful to use the LSPI Max Load limiter set to a row value to tune that row. You're still talking about dozens of "pulls" just to tune Borderline. This is why I cut down to 4 or 5 mapped points. I'd rather have 4 or 5 mapped points calibrated accurately than 25 with crap data.

I got the impression some time back that Mercedes basically did the same thing but they just did a "4-corners" method. You calibrate the engine at the 4 combinations of min and max IVC and EVO, then make those 4 mapped points the only ones enabled. Then command the cam angles you want and it will always interpolate between those 4. It's a pretty novel idea but never really wanted to take the time to do it since mine works pretty good as it is.

[junits15]

Here we go

This makes it alot more clear (IMO) lol

mapped points vs cam anglespng.png

Intake cam angle vs exhaust cam angle with data labels indicating what mapped point they go to

This allows easier visualization of the "space" we are operating in.

[engineermike]

You’re gettin it now, but swap the axes. Not that it really matters but most people put ivc on the x axis. You can also add in the snap lines and the op angles.

[junits15]

I think we have the same understanding, its an interesting approach and it probably works "ok" but I bet it still wouldn't be as good as hand tuning all of the points one by one. Essentially what he's banking on is that we can just shift the entire "contour" up or down by an equal amount. I might give it a try.

It sounds like what you did is very similar the Mercedes "4 corners" method. Tune a few points very well and let the tune interpolate the rest.

[junits15]

Holy shit it all makes sense, i get the snap lines now too.

There's a ton going on here lmao

[engineermike]

The amount of processing these pcm’s do the entire time you’re driving is truly mind-blowing.

[engineermike]

It sounds like what you did is very similar the Mercedes "4 corners" method. Tune a few points very well and let the tune interpolate the rest.

The difference is that I tune 4 or 5 points and always keep it on those points. The other method tunes 4 points and never keeps it on those points. The advantage is that you can run it at any cam angle at any time and never have to move or calibrate another map. If I decide to change my op evc then I have to move the entire mapped point.

[murfie]

The blending weights are based off the current cam angles physical position. If the percentages are taken literally, this may not represent the spot in, and weight of, the tables being used to derive the various commanded values.
You have figured out the x,y coordinate system and its all starting to make sense, great.
The snap to points and lines are there to simplify and not have to use the exact reported %weight of each mapped point, which may spread out a lot. This would make computations behind determining various commanded values intense at times. Instead the points and lines allows less points to influence any given condition and should simplify the calibration to just calibrating a set of points or lines (interpolation between two points) for the indicated VCT source. In 15+ mustangs OP mode switched from a single OP mapped point to using the reported % weight of each mapped point, making WOT not as easy to calibrate, but quick transitioning in and out of OP more tolerant and robust. So you don't have to rely on transient table calibration as much.
I think a lot of people take the % and try to spot calibrate with it,which does not work out well. You have to calibrate around the snap to points and lines assigned to the different VCT sources via the arrays.
Wether the weights are determined by an inverse distance, Gaussian, Epanechnikow, or some other kernel, I don't know. I don't think it matters.

The physical cam control is based on a phaser duty cycle feedforward tables and PID position error feedback loop control.
The array tables define which mapped point cam angles belong to each VCT source.
The distance tables define how far along that array it needs to be per load and RPM. This determines the target cam angles, but actual cam position comes down to position error into the PID, phase rate, oil temperature and phaser duty cycle.
The stock phaser control is good, but not enough to hold to perfect points or lines, nor move positions through the arrays as fast as RPM and load change distance and VCT source.
Phasers depend on good oil condition and do wear, get contaminated, age, lose performance and fail over time. So keep an eye on their controlled position error. They may not be even close to where the calibration wants them to be.

Kernalfunction.jpg

For me, above 3500RPM at WOT, running regular spark followed the calibrated borderline values and knock advance wasn't succesfully adding. Running premium knock advance added and it pretty much followed the MBT calibrated values. With in +/-1* on both fuels. Its all below 3500 and high loads that octane made a big difference in where the spark ended up.
With ethanol percentages increasing the combustion reaction kinetics change. This in turn increases torque, and when you increase torque at high RPMs such as 7250, HP goes up by ~1.4x.

[junits15]

The difference is that I tune 4 or 5 points and always keep it on those points. The other method tunes 4 points and never keeps it on those points. The advantage is that you can run it at any cam angle at any time and never have to move or calibrate another map. If I decide to change my op evc then I have to move the entire mapped point.

Oh I see, you've simplified the blending, did you put all of the points on snap lines? Or is this what 38152 "Snap to Point" is for? I like this idea for simplicity.

The blending weights are based off the current cam angles physical position. If the percentages are taken literally, this may not represent the spot in, and weight of, the tables being used to derive the various commanded values.
You have figured out the x,y coordinate system and its all starting to make sense, great.
The snap to points and lines are there to simplify and not have to use the exact reported %weight of each mapped point, which may spread out a lot. This would make computations behind determining various commanded values intense at times. Instead the points and lines allows less points to influence any given condition and should simplify the calibration to just calibrating a set of points or lines (interpolation between two points) for the indicated VCT source. In 15+ mustangs OP mode switched from a single OP mapped point to using the reported % weight of each mapped point, making WOT not as easy to calibrate, but quick transitioning in and out of OP more tolerant and robust. So you don't have to rely on transient table calibration as much.
I think a lot of people take the % and try to spot calibrate with it,which does not work out well. You have to calibrate around the snap to points and lines assigned to the different VCT sources via the arrays.
Wether the weights are determined by an inverse distance, Gaussian, Epanechnikow, or some other kernel, I don't know. I don't think it matters.

The physical cam control is based on a phaser duty cycle feedforward tables and PID position error feedback loop control.
The array tables define which mapped point cam angles belong to each VCT source.
The distance tables define how far along that array it needs to be per load and RPM. This determines the target cam angles, but actual cam position comes down to position error into the PID, phase rate, oil temperature and phaser duty cycle.
The stock phaser control is good, but not enough to hold to perfect points or lines, nor move positions through the arrays as fast as RPM and load change distance and VCT source.
Phasers depend on good oil condition and do wear, get contaminated, age, lose performance and fail over time. So keep an eye on their controlled position error. They may not be even close to where the calibration wants them to be.

Kernalfunction.jpg

For me, above 3500RPM at WOT, running regular spark followed the calibrated borderline values and knock advance wasn't succesfully adding. Running premium knock advance added and it pretty much followed the MBT calibrated values. With in +/-1* on both fuels. Its all below 3500 and high loads that octane made a big difference in where the spark ended up.
With ethanol percentages increasing the combustion reaction kinetics change. This in turn increases torque, and when you increase torque at high RPMs such as 7250, HP goes up by ~1.4x.

Can I ask what "premium knock advance" means? Sorry a lot of this is new to me, I kind of picked an extremely complex car to start learning how to tune on lmao.

As far as what they're using to estimate points, I'm with you, there's no way to know for sure and its probably not relevent. Really I think what we can assume is that whatever interpolation algorithm ford is using requires at least 14 points to recreate the contour accurately enough for what they want the factory calibration to be able to do. Its a question of if that algorithm is accurate enough to reproduce MP12 when only given MP2 and MP6. I think the answer is that it isn't good enough or else ford wouldn't have done it.

Have you found that the error in commanded cam angle vs. actual cam angle is enough to cause degradation in performance?

I'm finding myself wishing that we could understand the relationship between the tables, it feels wrong that every single table really needs individual calibration.


Thanks for the responses I really appreciate it a lot! I'm enjoying learning this

[junits15]

AH!

this is sampling theory. If we assume a fixed load and RPM and we allow the variables controlling the cam angles to change between their extremes we have a contour. (theoretically) In order to recreate that contour there is a minimum number of samples we need given some linear time invariant system and the maximum rate of change of the contour. That number is likely 14.

Then you just do this for every load/rpm combination for every single desired quantity that you want to vary w.r.t cam angle. This doesn't help tune it but it helps me understand it lol.


This is very cool stuff

[engineermike]

Oh I see, you've simplified the blending, did you put all of the points on snap lines? Or is this what 38152 "Snap to Point" is for? I like this idea for simplicity.

You can't NOT put all the mapped points on snap lines, since the snap lines are defined by the points on each end of them. The snap line is from MP(X,Y) to MP(X,Y), so by definition, both points are on the line. I personally do command OP cam timing only on snap lines, just like Roush, most Ecoboost, and Predator do.

I'm finding myself wishing that we could understand the relationship between the tables, it feels wrong that every single table really needs individual calibration.

Consider this...the Mapped Point located at IVO 30, EVC 0 is typically used for optimum stability at idle, due to the minimal overlap and smooth idle characteristics. But if you're targeting a higher torque, since it's a torque-based system, then the load needed to achieve that torque is higher, the manifold pressure is higher, so the throttle blade will need to be more open. If you compare this to IVO -20, EVC 20, then the load, manifold pressure, and MBT are different, which results in a different resulting throttle angle. Also different at 30, 50, and -20, 0, and so on. Each mapped point is a full calibration for an engine running that cam timing. You can draw some correlations around torque vs IVC and borderline knock vs IVC, but some of it isn't that intuitive. If you're trying to create tables using existing known good tables, then don't even bother. I've looked for correlations across platforms and there is little real consistency. For instance, if you compare the change in borderline timing vs rpm in a Predator vs a Roush vs a stock Gen3, you see they don't always go the same directions or follow the same trends. One thing I have notices is that for OEM high production engines, the cals seem to be far more developed with more mapped points and more significant figures than they are for specialty vehicles, which makes sense. More accurate data makes for higher fuel economy and less quirks, but also takes much more development money to create.

[murfie]

Can I ask what "premium knock advance" means? Sorry a lot of this is new to me, I kind of picked an extremely complex car to start learning how to tune on lmao.

As far as what they're using to estimate points, I'm with you, there's no way to know for sure and its probably not relevent. Really I think what we can assume is that whatever interpolation algorithm ford is using requires at least 14 points to recreate the contour accurately enough for what they want the factory calibration to be able to do. Its a question of if that algorithm is accurate enough to reproduce MP12 when only given MP2 and MP6. I think the answer is that it isn't good enough or else ford wouldn't have done it.

Have you found that the error in commanded cam angle vs. actual cam angle is enough to cause degradation in performance?

I'm finding myself wishing that we could understand the relationship between the tables, it feels wrong that every single table really needs individual calibration.


Thanks for the responses I really appreciate it a lot! I'm enjoying learning this

I don't like saying the octane rating numbers as they often vary per method of determining them, and are meaning less for what the stations are actually serving. I instead used what the station calls them regular, mid-grade, and premium. (87, 89, and 92 in my area).
I have not seen the normal variability in cam phaser control amount to anything meaningful in performance. When they fail or are no where near what is commanded that changes things.
I don't know what goes into the decision of how many points to use for an OEM calibrated vehicle. I know their patents imply their could be any number of mapped points only limited by computer power. It might have something to do with the VCT modes being used to simplify goals.

You can understand the relationship between the tables by what mapped point they belong to, the respective mode array that mapped point is in. The modes are controlled by load. Load comes directly from the MAF sensor and RPM sensor on your 19 mustang.

Stability limited (any mode enabled by a "max load" table):
Stability, fuel economy
This is not just idle, but also cruising targeting the angles that will achieve the best fuel economy right up until the stability limit of combustion and the torque request enabling decel fuel shut off.

optimal power (any mode enabled by a "min load" table):
Best drivability, optimal power
This is heavy throttle to full throttle acceleration, targeting the angles that will achieve highest power for highway passing, hills, and sport.

Going from stability limited operations to optimal power or vise versa, theres transient performance.

Then theres emissions reduction which is mainly for cold start to get the catalytic converter to temperature. Manual launch mode...

Its the IMRC open/ closed/ transition calibration in the 15+ mustangs that makes this all look more complicated than it truly is. They improve the combustion stability limit for fuel economy. Drivabiliy mode enables at a load where IMRC is going to be open, unless you are at high elevation. At high elevation the cams target angles for power much sooner because it willbe needed. The vacuum used to close the IMRC won't be as great for a given load and they will open sooner at elevation.