VEMS Home Products Documentation Wiki Webstore
WikiHome ProjectPage DocsPage UnderDevelopment MembersPage SiteIndex RecentChanges Preferences
GenBoard/UnderDevelopment/FirmWare/MiscOutputs (2008-03-10 12:10:50)

Misc output related subpage of GenBoard/UnderDevelopment/FirmWare

Hysteresis for misc output would be nice. Either +-1 threshold or prevent output change for 100 msec after an output change (to prevent prelling)

Also some general method (see interpretere below) to define functions for misc output actuation

Interpreter

There are more and more stuff that needs flexible configuration, without firmware recompile, eg. misc outputs. Currently condition is wired for most common function, RPM,MAP,TPS ranges

I think we cannot avoid a simple language/interpreter. Maybe a postscipt-like (but binary) code: words of 16 bits that are

---

Michael's Ideas...

There are a few parts to this challenge. First of all it's necessary to cater to a potentially long list of functions. Adding canbus enabled switches allows for an impressive range of features. VSS encoders, clutch switches, ALS enable switches, boost knobs are all potential inputs. On the output side one can have as many outputs as there are spare channels. Potentially extra drivers could be available via CANbus.

First the outputs...

I assume at this point we're dealing with a finite number - limited by the number of spares you have. This could still number 8 or more.

Outputs may potentially be switched based on a number of parameters. Here are a few features I thought up:

fan control

Vtec control

fuel pump (maybe pre-defined)

Tach signal (maybe pre-defined)

PWM IAC

intercooler fan and/or intercooler water spray

Nitrous enable

shift light

misc output control

Table based PWM. - Jörgen

This is a function that allow more flexible Boost and launch control functionallity in addition to being useful for fuelpump and waterinjection control

Another sort of output may need to be driven from a table. TGV/TVIS/AVCS/IAB control, electronic throttle and others come to mind.

For TGV/TVIS/AVCS/IAB and possibly iVTEC the computer may control offsetting one or more cams and/or adding to the lifts at different RPM and load combinations. Electronic throttle control will be the way of the future as more and more new cars are drive by wire. I suspect this will require a completely dedicated implementation. For the above the outputs are most likely to be PWM based. I can confirm that AVCS (on Subarus) is PWM controlled. I believe the same with the new i-VTEC found on hondas. Certain RPMs and MAP values require certain PWM outputs in order to aid in spooling the turbo or creating lower RPM torque.

Storage

The code for doing these checks is fairly trivial. Efficient storage of the parameters is somewhat less trivial.

A dynamic size for these functions would require a lot of work. First you would need a means of specifying the size of each record and secondly you would need a means of properly shuffling things around should changes be made. Anyone remember Norton Defrag?

The current implementation is rather short sighted and wasteful. You pay the overhead even if you don't use a feature. Unfortunately this implementation is more difficult because of the dynamic memory space. Using static sizing is as wasteful or worse than the existing solution.

There is lots of space in eeprom to store these but eeprom is a bad solution because it lacks any quick access method. As a result we are forced to read the values into RAM. Removing these from the 'C' table would free up about 30b of memory if the table size were reduced accordingly.

Personally I would prefer to have them stuck in flash. This would provide quick reads where the parameters would not necessarily need to be stored in RAM. Unfortunately this brings about a challenge for updates. While technically possible each parameter would be limited to about 10,000 changes. We're probably okay with this. The other challenge is halting the processor long enough to perform an update without missing an ignition event or some other time sensitive event. While tuning I'm not sure if you really want to write your flash page 100 times as you adjust values.

All in all I suspect we would need about 20*8 bytes of memory, be it flash or SRAM.

---

If a block contains 8 instructions (16 bytes), that can be do more than the current misc1 misc2 implementation, basically enough for any function that comes to mind. The fixed size block makes the defragmentation issue go away (the same way they allocate fixed size pages in OS-es).

Besides range checks and alike, there will be some complex functions as well.

We can add a

Shiftlight

We need to change shiftlight code a bit so the configured RPM treshold is compared againds RPM + dRPM/dt*0.5sec (not just RPM). This means the light (beeper, whatever connected) is activated somewhat earlier in gear1 and gear2 (but gearbox switches not needed).

Michael's Proposal

This will be merged to GenBoard/UnderDevelopment/FirmWare/DynamicConfigRelated, which is about saving space and versatility:

Michael suggest ripping following out of the config table and shortening that table.

structure for the outputs

offset output type slotid
IGN0 COIL 255
IGN1 COIL 255
IGN2 COOLINGFAN  
IGN3 COOLINGFAN  
IGN4 UNUSED 255
IGN5 UNUSED 255
IGN6 UNUSED 255
IGN7 UNUSED 255
FET0 FUELPUMP  
FET1 TACH  
FET2 WATERSPRAY  
FET3 VTEC  
FET4 INJECTOR 255
FET5 INJECTOR 255
FET6 INJECTOR 255
FET7 INJECTOR 255
FET8 WBO2  
FET9 UNUSED 255
FETA UNUSED 255

Assumng no pre-defined outputs like WBO2 heater, fuel pump, frees 40 bytes SRAM. Clearly does not worth to start with thist, ripping tables from SRAM is simpler and saves 1000 bytes.

Now for the slots... The features I could dream up with required anywhere from 1 to 8 bytes of space. In order not to be too wasteful I suggest definining different size slots and reserving a finite number of each.

Most people will want about 6 channels for injectors and coils. That leaves about 14 potential outputs.

number reserved slot size slotids
6 4 0..5
2 8 6..7

The space here was arbitrarily chosen. It could be expanded as more memory is found through making code more efficient. These slots are expected to be permanently stored in EEPROM much the same as the config table is stored.

Finally a table would be created in flash with the following stucture.

slotsize function pointer
4 &COOLINGFAN(output channel,slotid)
8 &VTEC(output channel,slotid)
etc etc

Attaching a feature to an output channel would consist of looking at the slot size required then traversing the output list to find a slot of that size that is available. It would then be added to the list and the type assigned. The tuning interface or software would need to know about every different output type's settings and storage format so that data could be entered in the slut.

Then at given intervals the array would be traversed and each pointed to function called. The function itself would then be responsible for looking at the stored parameters in its slot and making a determination of what it should do.

Detaching a feature would be as simple as removing its entry from that output element in the outputs table.

For sanity the firmware should prevent assignment of outputs in the H table to slots that are not already reserved as type COIL or INJECTOR. Conversely when these values are changed from COIL or INJECTOR they will need to be removed from the H table.

Results

These limits could be adjusted with ease subject to memory availibility. The flexibility does not end here. Again, subject to memory availibility larger slots could be defined for items such as boost control or AVCS controllers/active diff lockup controllers.

The following config variables could then be removed.

uint8_t fan_temp;

uint8_t fan_hyst;

uint8_t fan_channel;

uint8_t engine_off_delay;

uint8_t pump_on_mintime;

uint8_t fuelpump_channel;

uint8_t misc1out_minrpm;

uint8_t misc1out_maxrpm;

uint8_t misc1out_mintps;

uint8_t misc1out_maxtps;

uint8_t misc1out_minmap;

uint8_t misc1out_maxmap;

uint8_t misc1out_channel;

uint8_t misc2out_minrpm;

uint8_t misc2out_maxrpm;

uint8_t misc2out_mintps;

uint8_t misc2out_maxtps;

uint8_t misc2out_minmap;

uint8_t misc2out_maxmap;

uint8_t misc2out_channel;

uint8_t act_wot_rpm;

uint8_t act_wot_channel;

uint8_t act_rpm_rpm;

uint8_t act_rpm_channel;

uint8_t boost_conf;

uint8_t boost_targetoffs;

uint8_t boost_minpressure;

uint8_t boost_pid_kp;

uint8_t boost_pid_ki;

uint8_t boost_pid_kd;

uint8_t boost_pid_ilimit;

uint8_t boost_channel;

uint8_t water_pump_temp;

uint8_t water_pump_hyst;

uint8_t water_pump_channel;

uint8_t tach_channel;

uint8_t tach_divider;

uint8_t shiftcut_conf;

uint8_t shiftcut_channel;

uint8_t shiftcut_time;

This would free 40 bytes of RAM (same amount of space reserved in EEPROM, but we have much space there). So this implementation could be completed without requiring any additional SRAM.

Storing in FLASH - handling "dirty" data and writing while engine is off

would save a lot of SRAM, but caching "dirty" config/tables and storing in flash has it's own problems (that we want to solve, within 3 months, because of ARM, but there are better things to do now to help the case).

Finally, it is technically possible to store all of the slots in flash. I don't like this solution because of the complexities of live updating the flash while the engine is running. This would suck when you're making rapid changes for tuning. Yes, changed slots could be buffered in RAM to an extent but then the system becomes very complex.

We surely need dynamic structures (similar to your proposal) in the future. However, MegaTune is not yet prepared for it. So for now, we must save SRAM with other ways. Notably, using tables directly from EEPROM, and scheduling EEPROM write to be directly after user-space calculations (and using a lockbit or examining storage.c state) to avoid interference.

EEPROM

I've just re-read the docs on reading/writing eeprom. It seems that

are the typical times to read and write the entire 4k of eeprom space. As a result it should be possible to implement the slots system directly in EEPROM. This solves all the memory problems, actually freeing some memory for other uses. The only side effect is that a changed value must be changed and saved in the same step.

Firmware and MegaTune interface can be written for anything. In short, changes in stable will not be tolerated unless they have MegaTune support, or the MegaTune vemsv3.ini change is trivial.

A separate branch is OK though.