An engine management system for gasoline spark ignition engines.
The firmware is flexible to most engine configurations, and with 16 outputs it is suitable for up to an 8 cylinder engine with sequential fueling and ignition.
Features:
- 16 high precision outputs with 250 nS scheduling accuracy usable for fuel or ignition
- 24x24 floating point table lookups
- Speed-density fueling with VE and lambda lookups
- Supports 8 12-bit ADC inputs when using AD7888
- Supports 16 GPIOs
- Configurable acceleration enrichment
- Configurable warm-up enrichment
- Boost control by RPM
- Frequency sensor inputs
- PWM outputs
- Fuelpump safety cutoff
- Mostly runtime-configurable over USB interface
- Check Engine / Malfunction Indicator control
Currently the only two decoders styles implemented are even-tooth and missing-tooth wheels. Wheels of each type are supported on both cam and crankshaft, and both support an optional cam sync input. Each wheel configuration has a tooth count and an angle per tooth. For a cam wheel it is expected that this totals up to 720 degrees -- for example a 24 tooth wheel on the cam is 24 total teeth with 30 degrees per tooth. A crank wheel totals 360 degrees. Without a cam sync, the current wheel angle has a random phase at the granularity of the wheel.
This decoder is configured with a tooth count and angle per tooth. Two decoder-specific options:
rpm-window-size- RPM for table lookups is derived from an average this many teethmin-triggers-rpm- Number of teeth to have passed since sync-loss to re-establish an rpm, before (optionally) waiting for a cam sync pulse.
This wheel can be used without a cam sync, but will have random phase to the granularity of the wheel, and is only useful on low resolution wheels. For example, a Ford TFI distributor has a rising edge per cylinder: It would be configured with 8 teeth, 90 degrees per tooth. An ignition event could be configured for each cylinder, and the random angle is not an issue due to the use of a distributor.
An example of a even tooth wheel with a cam sync is the Toyota 24+1 cam angle sensor - 24 cam teeth, 30 degrees per tooth, and an additional 1-tooth wheel. The first tooth past the cam sync is at 0 degrees.
This decoder is configured with a tooth count and angle per tooth. The tooth count includes the missing tooth, so a 36-1 wheel is 36 teeth. If the wheel is crank-mounted without a cam sync, cam phase will be unknown and events should be symmetric around 360 degrees (batch injection and wasted spark). The first tooth after the gap is at 0 degrees.
See the runtime configuration section for details on the runtime control interface. Almost everything is configuable through that interface, but to use a different engine configuration (read: not my Supra), it is probably best to start with modifying the source code.
Static configuration is in config.c. The main configuration structure is
config, along with any custom table declarations, such as the provided example
timing_vs_rpm_and_map. Pin configurations are platform specific and are
documented in the platform source.
| Member | Meaning |
|---|---|
num_events |
Number of configured events |
events |
Array of configured event structures, see Event Configuration below |
freq_inputs |
Array of configured frequency inputs |
decoder.type |
Decoder type. Possible values in decoder.h |
decoder.offset |
Degrees between 'decoder' top-dead-center and 'crank' top-dead-center. TFI units by default emit the falling-edge trigger 45 degrees before they would trigger spark in failsafe mode |
decoder.trigger_max_rpm_change |
Percentage of rpm change between trigger events. 1.00 would mean the engine speed can double or halve between triggers without sync loss. |
decoder.trigger_min_rpm |
Minimum RPM for sync. Should be just below the slowest cranking speed. |
sensors |
Array of configured analog sensors. See Sensor Configuration below. |
timing |
Points to table to do MAP/RPM lookup on for timing advance. |
ve |
Points to table for volumetric efficiency lookups |
commanded_lambda |
Points to table containing target lambda |
injector_pw_compensation |
Points to table containing Voltage vs dead time |
engine_temp_enrich |
Points to table containing CLT/MAP vs enrichment percentage |
tipin_enrich_amount |
Points to table containing Tipin enrich quantities |
tipin_enrich_duration |
Points to table containing Tipin enrich durations |
rpm_stop |
Stop event scheduling above this RPM (rev limiter) |
rpm_start |
Resume event scheduling when speed falls to this RPM (rev limiter) |
fueling.injector_cc_per_minute |
Injector flow rate |
fueling.cylinder_cc |
Individual cylinder volume |
fueling.injections_per_cycle |
Number of times an injector is fired per cycle. 1 for sequential, 2 for batched pairs, etc |
fueling.fuel_pump_pin |
GPIO port number that controls the fuel pump |
ignition.dwell_us |
Fixed (when in fixed dwell mode) time in uS to dwell ignition |
Certain inputs are used as frequency inputs which may also act as the decoder wheel trigger inputs. Each one has a selectable edge and type.
| Member | Meaning |
|---|---|
type |
Can either be TRIGGER or FREQ. Some platforms can only use certain pins as trigger inputs |
edge |
Determines which edge is counted. RISING_EDGE, FALLING_EDGE, BOTH_EDGES |
Event configuration is done with an array of schedulable events. This entire
array is rescheduled at each trigger decode, allowing any angles (0-719) to be
used for events, as long as a trigger event occurs between any two fires of a
given event (e.g. for FORD_TFI, for 8 events, as long as the decoder fires
twice in 720 degrees, which of course it does). Currently there are two types
of event:
| Event | Meaning |
|---|---|
IGNITION_EVENT |
Represents spark ignition event. Will start earlier than specified angle because of dwell, will stop at the angle specified minus any advance. |
FUEL_EVENT |
Represents fuel injection event. These events will attempt to end fueling at the specified angle, but rescheduling will preserve the event duration rather than stop angle |
The event structure has these fields:
| Member | Meaning |
|---|---|
type |
Event type from above |
angle |
Base angle for an event |
output_id |
OUT pin to use for this output |
inverted |
Set to one if active-low |
Sensor inputs are controlled by the sensors config structure member, and is an
array of configured inputs. The indexes into the array for various sensors are
hardcoded, use the provided enums to reference the sensors. Each sensor entry
is a structure describing what pin to use, and how to translate the raw sensor
value into a usable number:
| Member | Meaning |
|---|---|
pin |
For ADC input, specifies which ADC pin. For FREQ inputs, specifies which frequenty input to use (hardware specific) |
method |
Processing method, currently supported are linear interpolation and thermistors |
source |
Type of sensor, currently supported are analog and frequency based. |
params |
Union used to configure a sensor. Contains range used for calculated sensors, and table for table lookup sensors. |
params.range.min |
For processing, value that lowest raw sensor value reflects |
params.range.max |
For processing, value that highest raw sensor value reflects |
params.therm |
For processing, Bias resistor and A, B and C parameters of the Steinart-Hart equation |
fault_config.min |
Raw sensor value, below this indicates sensor fault |
fault_config.max |
Raw sensor value, above this indicates sensor fault |
fault_config.fault_value |
During sensor fault, use this fallback value |
lag |
Lag filtering value. 0 means no filtering, 100 will effectively never change. |
window.total_width |
When method is windowed, total stride of degrees for a single averaging |
window.capture_width |
When method is windowed, window inside of total stride to average samples for |
window.offset |
When method is windowed, offset of capture window inside total window |
For method SENSOR_LINEAR, the processed value is linear interpolated based on
the raw value between min and max (with the raw value being 0 - 4095 for 12-bit
ADCs).
The onboard ADC is not used. Instead an external ADC is connected to SPI2 (PB12-PB15). Currently a TLC2543 or AD7888 ADC is supported.
Tables can be up to 24x24 float values that are bilinearly interpolated. Use
struct table to define a table, with the following relevent fields:
| Member | Meaning |
|---|---|
title |
Table title |
num_axis |
1 or 2, for 1d vs 2d lookup |
axis[0] |
Only axis if 1d table, horizontal axis if 2d. |
axis[1] |
Vertical axis for 2d table. |
axis[?].name |
Axis Title |
axis[?].num |
Number of columns/rows in the axis |
axis[?].values |
Value for each column in the axis |
data.one |
Array of data points for 1d table |
data.two |
2d array for 2d table. First index represents vertical axis. |
For a new table to be configurable over the console, it must be declared
externally such that it can be directly referenced in console.c.
The EMS is uses a cbor-based (https://cbor.io/) binary RPC protocol for its console. For hosted mode, standard out and in are used to exchange messages, and on the stm32f4 target, the console is exposed via a USB virtual console. Details of the binary protocol are described in INTERFACE.md.
Requires:
- Complete arm toolchain with hardware fpu support.
- GNU make
- check (for unit tests)
Before trying to compile, make sure to bring in the libopencm3 submodule:
git submodule update --init
To run the unit tests:
make PLATFORM=test check
To build an ELF binary for the stm32f4:
make
obj/stm32f4/viaems is the resultant executable that can be loaded.
You can use gdb to load, especially for development, but dfu is supported. Connect the stm32f4 via
USB and set it to dfu mode: For a factory chip, the factory bootloader can be
brought up by holding BOOT1 high, or for any already-programmed ViaEMS chip, the
get bootloader command will reboot it into DFU mode. Either way, there is a
make target:
make program
that will load the binary.
The platform interface is also implemented for a Linux host machine.
make PLATFORM=hosted run
This will build viaems as a Linux executable that will use stdin/stdout as
the console, and event/trigger updates will be sent to stderr . The test
trigger that is enabled by default will provide enough inputs to verify some
basic functionality. Full integration testing using this simulation mode is
planned.
The hosted implementation uses threads to simulate interrupt handling and the timers/DMA. It attempts to schedule these two threads with realtime priority. If you experience difficulty keeping sync due to expired triggers (due to scheduling latency), make sure the threads have permissions to be realtime, or otherwise constrain it to a single cpu.
The current primary hardware platform is an ST Micro STM32F407VGT microcontroller. There are a few non-production-ready hardware designs available under https://github.com/via/tfi-board, though I would recommend anyone attempting to use those designs contact me. The STM32F4-DISCOVERY board can be used with this firmware, with appropriate extra hardware for vehicle interfacing.
Note: These are subject to change with the next hardware design, which will be moving to better support a 64 pin chip and more flexible PWM outputs.
| System | Pins |
|---|---|
| Trigger 0 | PA0 |
| Trigger 1 | PA1 |
| Frequency | PA0, PA1, PA2, PA3 |
| Test Trigger | PB10, PB11 |
CANL |
PB5 |
CANH |
PB6 |
| USB | PA9, PA11, and PA12 |
| ADC | PB12, PB13, PB14, and PB15 |
| OUT | PD0-PD15 |
| GPIO | PE0-PE15 |
| PWM | PC6, PC7, PC8, and PC9 (currently all fixed at same frequenty) |