home / articles / speed density on a factory honda ecu
Build series ยท Fuelling
Speed density on an ECU that was built around an airflow sensor
Every decision an engine computer makes about fuel starts with one question: how much air is in the cylinder? Honda answered it for a naturally aspirated engine. Under boost, that answer is wrong in several ways at once.
Why the factory air model fails under boost
The Rebirth's ECU measures air with an airflow sensor in the intake. That works well for the engine Honda built. Add a turbo and four things break:
- A ceiling on air. The software caps the air per cylinder at the most a non-turbo engine could ever hold. With the stock settings that cap starts to bite just above atmospheric pressure. Past it, fuel stops rising while air keeps rising. This is the classic "goes lean above 100 kPa" behaviour.
- Load that stops at atmosphere. The ECU's idea of engine load is clamped at barometric pressure. Around two dozen fuel, ignition and enrichment tables are indexed by that load, so none of them ever see boost.
- A sensor out of range. The airflow sensor tops out, and readings near its limit are thrown away as faulty.
- Fuel pressure. The factory model assumes the pressure across the injectors never changes. On this fuel system, boost in the manifold reduces it, which cost about 9 % of fuel at 10 psi.
I saw the first one in my September logs: across 4,500 to 5,500 rpm the fuel delivered was a flat line while the table underneath it varied by 18 %.
What speed density means
Speed density does not measure airflow. It calculates it. The ECU takes manifold pressure, engine speed and intake temperature, and looks up how well the engine fills at that point, its volumetric efficiency. From those it works out the mass of air in each cylinder.
For a turbo engine it is the natural approach, because pressure is the thing that changes. It also depends completely on the pressure sensor being able to read the whole range, which is why that came first.
Getting there
The car arrived with a crude arrangement: airflow sensor below a fixed pressure, a table above it, and a hard switch between the two. For the first weeks I worked within that, refitting the boosted part of the table from logged drives.
Full speed density went on the car on 3 November 2026. The airflow sensor is still plugged in but the ECU no longer uses it. There are two efficiency tables, because this engine has two cam modes that fill very differently. The same calculation also feeds the automatic gearbox its estimate of engine torque, so the transmission now knows about boost as well.
Before the first flash, the new model was checked against 3,503 rows of logged driving. In 1,003 of 1,004 boosted rows it fuelled within 1 % of the proven file.
Five things the road taught me
1. The idle was rich
On the first drive the idle fuel correction was pinned at minus 18 %. The lowest corner of the table had never had real data. I had estimated it, and estimated badly.
2. The hiccup at the cam switch
At light load the car stumbled. The logs showed it happened on every one of eight switches into the economy cam: calculated air dropped by a third to a half, and the mixture swung lean. The table for the new cam was being selected instantly, while real manifold pressure took up to a quarter of a second to catch up. Carrying the airflow smoothly across the switch cut the dip to between 5 and 9 %.
3. Lean after pressure rises
Nine of sixteen events still ran lean for about half a second after manifold pressure rose. The model was reacting to pressure it had already missed. I added a lead that acts only on rising pressure.
4. Lean on tip-in
Of 84 part-throttle tip-ins in one drive, 68 were lean, and they stayed lean for more than a second after pressure had settled. That pattern is fuel wetting the intake walls instead of reaching the cylinder. I added a correction triggered by the throttle. It helps. I am not yet satisfied with it.
5. A new fuel pump changed everything
On 7 November I fitted a stronger fuel pump. Boosted fuelling went from 8 to 11 % lean to 10 to 15 % rich overnight, because the old pump had been sagging and my table had quietly been compensating for it. The whole boosted region had to be refitted.
The evening it went lean
On 11 November the worst pull of the night read 27 % leaner than commanded at the top of the boost range. I put the previous file back on the car that night.
It was not a fault in the code. I found three causes: a dip in the efficiency table in one band of pressure and engine speed, boost air being calculated at a fixed temperature instead of the real one, and a day-to-day shift of 4 to 8 % that I still cannot explain. I filled the dip, switched the temperature correction on, and made the boosted cells deliberately richer. That file went on the car on 12 November.
My rule from that night: better rich than lean.
What is not right yet
My own notes say it directly: the boost fuelling has not converged. It was lean, then rich after the pump, then close, and each refit rests on one or two drives. Part-boost fuelling has been within 3 % since early October, so the trouble is at the top.
The car's own mixture sensor is also a limit. It updates only about five times a second and cannot read richer than a certain point, so on very rich pulls the true number is unknown.
This is why the ECU now learns its own fuelling. A table fitted from a handful of drives will always lag behind an engine that changes with the weather.
By Haris, Hex and Boost.
Have a car the usual tools cannot touch?
Tell me the car, the engine, the gearbox and what you want it to do. I reply to every serious enquiry.