What mixture tuning is deciding
A spark-ignition engine burns an air-fuel mixture. The mass of air relative to the mass of fuel sets the torque the mixture can make, the fuel it uses, the exhaust temperature, and whether a three-way catalyst can convert carbon monoxide, hydrocarbons and oxides of nitrogen at the same time. Mixture tuning is the job of choosing that ratio for each operating point, then checking that the engine actually received it.
Bosch’s M-Motronic training text treats 14.7 kg of air to 1 kg of fuel as the stoichiometric mass ratio for the gasoline it discusses: the proportion at which the fuel and the oxygen can be fully consumed together. The same text says a three-way catalyst needs that stoichiometric mixture to work at high conversion, while maximum torque and lowest fuel consumption sit at different ratios. No single mixture is best at idle, on a light cruise and at wide-open throttle.
The page below is about reading and interpreting the mixture. Map names, axis units and which request wins are specific to the ECU. Worked ME7 request paths are on ME7 fueling and MAF calibration. A torque-path introduction that includes lambda is on MG1 torque, air, fuel and tuning basics.
AFR, lambda and equivalence ratio
Air-fuel ratio (AFR) is the mass of air divided by the mass of fuel. An AFR of 14.7:1 means 14.7 kg of air for each 1 kg of fuel. A lower AFR number is richer. A higher AFR number is leaner. The number is only meaningful once the fuel is named, because each fuel has its own stoichiometric AFR.
Lambda, written λ, cancels that fuel-specific scale. In the M-Motronic definition it is the inducted air mass divided by the air mass that stoichiometric combustion of the same fuel would require. MIT’s engine course defines the same relative air-fuel ratio, and defines the fuel-air equivalence ratio φ as its reciprocal:
λ = (actual AFR) / (stoichiometric AFR of the fuel in the tank)
AFR = λ × (stoichiometric AFR of that fuel)
φ = 1 / λ
Lambda 1 is stoichiometric. Lambda below 1 is rich: less air than complete combustion needs, which is the same thing as extra fuel. Lambda above 1 is lean: excess air. Equivalence ratio runs the other way, so φ = 1.18 is λ = 0.85. Loggers and gauges use both conventions. Confirm which one a channel uses before comparing two traces.
Lambda is the unit to compare across fuels and across ECUs. An AFR gauge that assumes gasoline at 14.7:1 will print a misleading AFR when the tank contains ethanol blend, even while the true lambda is correct. Multiply the measured lambda by the stoichiometric AFR of the fuel actually in use.
Stoichiometric ratio depends on the fuel
Bosch uses 14.7:1 for gasoline in the M-Motronic text, and gives the matching volume picture as roughly 9,500 litres of air for 1 litre of fuel. MIT’s combustion lecture puts a typical petroleum fuel near 14.6:1, with a range of about 14.2 to 15 as composition changes. A later engine study measured 14.42:1 on one gasoline sample. Treat 14.7 as the reference value in the Bosch text, and expect pump fuel to move inside that band.
Oxygen already bound in the fuel lowers the external air it needs. Oak Ridge’s comparison table gives anhydrous ethanol a stoichiometric AFR of 9.0 against 14.7 for gasoline. The US Department of Energy’s E85 handling handbook lists about 9 for ethanol, 14.7 for gasoline and about 10 for E85, and notes that the blend properties follow the hydrocarbon used to make it. Because retail E85 ethanol content varies, an AFR target copied from a gasoline table does not transfer. A lambda target does.
| Lambda | Mixture on the Bosch scale | AFR if stoich is 14.7 | AFR if stoich is 9.0 |
|---|---|---|---|
| 1.20 | Lean, at the upper end of the economy band Bosch gives as 1.1 to 1.2 | 17.6 | 10.8 |
| 1.10 | Lean, lower end of that economy band | 16.2 | 9.9 |
| 1.00 | Stoichiometric | 14.7 | 9.0 |
| 0.95 | Rich edge of the maximum-power band | 14.0 | 8.6 |
| 0.90 | Inside the maximum-power band | 13.2 | 8.1 |
| 0.85 | Rich edge of the maximum-power band | 12.5 | 7.7 |
| 0.80 | Richer than that power band | 11.8 | 7.2 |
| 0.75 | Richer still | 11.0 | 6.8 |
The AFR columns are λ multiplied by the cited stoichiometric ratio, rounded to one decimal. They describe that fuel only. The 0.80 and 0.75 rows are arithmetic, not a recommended target.
Power, economy and the catalyst want different mixtures
Bosch’s excess-air curves, for manifold-injection spark-ignition engines, place maximum power at lambda 0.85 to 0.95. The same passage places best economy with about 10 to 20 percent excess air, lambda 1.1 to 1.2. Past the lean limit for that engine the mixture stops igniting reliably and misfire begins. The limit is a property of the engine and its mixture preparation. Direct injection is called out separately: those engines can run stratified at excess-air factors extending to lambda 4, because the combustion mode is different from a homogeneous manifold-injected charge.
Emissions close the gap on a warmed-up road engine. Bosch states that a three-way catalyst depends on a stoichiometric mixture, and that the window in which CO, HC and NOx are converted together is narrow, about lambda 0.99 to 1. Once the engine is warm, the M-Motronic strategy holds idle and part-throttle at stoichiometry for that reason. Enrichment and enleanment remain available for cold running, wide-open throttle and, on later systems, component protection.
Untreated exhaust follows the mixture. Rich of stoich, carbon monoxide rises. Lean of stoich, oxides of nitrogen rise until combustion temperatures and misfire turn the trend. Those are untreated-gas trends in the Bosch curves. A lit three-way catalyst changes what leaves the tailpipe, provided the mixture stays inside its window.
Reading a narrowband sensor
A switching lambda sensor, the narrowband type, compares oxygen at its exhaust electrode with oxygen at a reference electrode. Bosch describes a special ceramic that becomes conductive when hot, and a voltage that steps when the mixture crosses lambda 1. In the M-Motronic account the lean side, lambda above 1, is about 100 mV, and the rich side, lambda below 1, is about 800 mV. The voltage jumps between those levels at lambda 1. The curve is a step. It does not encode how rich or how lean the mixture is once it has left the step.
The ceramic has to be hot before that step exists. Bosch gives about 350 °C exhaust gas for an unheated sensor and about 200 °C for a heated sensor. Heated sensors in that generation are described as ready in about 30 seconds. A cold sensor, an open circuit or a short can sit at a mid-scale voltage that the ECU rejects as implausible. A voltage frozen in the middle of the scale is not evidence of a perfect mixture. The useful narrowband observation is whether the signal is switching across the step, held on the rich side, or held on the lean side, after the sensor is hot.
Closed-loop control with this sensor is a two-sided correction. Bosch describes an initial jump followed by a ramp: the ECU lengthens injection when the sensor says lean and shortens it when the sensor says rich. Gas travel time from the valve to the sensor delays the report, so the correction is always late. Bosch describes the result as a continuous oscillation whose period follows that transport delay. Seeing the voltage swing between the lean and rich levels at warm idle is the loop working.
Use a narrowband trace to answer one question: is the warmed engine being held on the stoichiometric switch? Use a wideband, or the ECU’s own wideband channel, when the question is the value at wide-open throttle. Sensor generations, heaters and connectors are compared on Wideband oxygen sensor types.
Reading a wideband sensor
A wideband lambda sensor adds an oxygen pump cell to the Nernst cell. Bosch Mobility describes it as measuring oxygen across a wide range so that lean and rich mixtures can be controlled, including lean petrol concepts, diesel and gas engines. The Bosch Motorsport LSU 4.9 data sheet gives a monotonic measuring range from lambda 0.65 to free air, for gasoline, diesel and E85. The controller, or the ECU’s lambda interface, converts the pump current into lambda. Read that converted value. Raw pump current is only meaningful with the sensor’s characteristic and the controller’s scaling.
The same LSU 4.9 sheet limits normal operation to exhaust pressure at or below 2.5 bar, with decreasing accuracy above that, and to exhaust temperature below 930 °C in continuous operation, briefly below 1,030 °C. A sensor screwed into a pre-turbine boss on a boosted engine can see pressure and temperature outside that window. A reading taken there needs the controller’s pressure compensation, or it needs to be treated as degraded. A bung after the turbine sees lower pressure and is the usual place for an external meter, with the penalty of a longer gas delay.
Placement decides what the number means:
- A sensor in one runner reports that cylinder’s exhaust, which is the right place to find a single-cylinder mixture fault and the wrong place to judge the engine average.
- A sensor just upstream of the catalyst, in the collected stream, reports the mixture the engine delivered. That is the control sensor.
- A sensor downstream of a lit catalyst reports the gas after the catalyst has stored and released oxygen. It is the catalyst monitor. It is a poor meter of the combustion mixture.
- A single tailpipe probe after the collector averages the banks. One lean bank can hide behind a rich bank.
Follow the controller manufacturer's calibration procedure: free-air calibration is required by some devices, while others use a factory trim resistor. For example, the AEM X-Series inline manual specifies resistor calibration as its default and offers free-air calibration as an option. Set the display to lambda or understand which stoichiometric AFR it uses. Log it against rpm, load and throttle on the same time base as the ECU, then shift the traces mentally by the transport delay before calling a spike a mixture error.
Individual cylinders and exhaust temperature
A sensor in a shared collector samples the combined exhaust stream. It can conceal differences between cylinders, so an acceptable overall lambda is not proof that every cylinder has the same mixture. A sensor in an individual runner provides a more local reading, but exhaust leaks, misfire, sensor operating limits and gas transport delay still affect interpretation. Bosch specifies that the sensor location must give a representative sample and stay within its operating limits. Bosch Motorsport: LSU ADV installation.
Individual exhaust-gas-temperature (EGT) probes provide another view of cylinder behaviour, but temperature is not a direct measurement of lambda. Compare like-for-like probe positions and types under similar operating conditions, and investigate a difference alongside lambda, ignition, injector operation and engine condition; do not assume one temperature identifies the cause or proves a safe mixture. Haltech documents both individual-cylinder EGT monitoring and an example comparing mixture and ignition timing. Its configuration distinguishes an individual-cylinder sensor from one in a collector: EGT sensor setup.
Where the number in a log comes from
Several channels get called “the lambda reading”. They answer different questions.
| Source | What it reports | What it cannot show |
|---|---|---|
| ECU requested lambda | The target the calibration selected for this moment, after the active requests and limits | Whether the injectors delivered it |
| ECU measured lambda, from its own wideband | The oxygen the control sensor saw, scaled by that ECU | A true value while the sensor is cold, failed, or outside its pressure range |
| Narrowband voltage | Which side of lambda 1 the exhaust is on, once the sensor is hot | A numeric mixture at wide-open throttle |
| Short-term and long-term fuel trim | How far the ECU is currently correcting, and how far it has learned to correct, the pilot injection | The absolute mixture. A trim near zero can sit on top of a target of 1.00 or, on a wideband system, on top of a rich target |
| External wideband | An independent lambda at the bung you fitted | The ECU target, unless you log that as well. A post-catalyst bung does not read the engine mixture |
| Exhaust-gas analyser | Lambda calculated from CO, CO2, HC and O2, when the analyser is given the fuel’s hydrogen, carbon and oxygen fractions | A fast trace. It is a slow, diluted sample, and it follows the probe location with the same catalyst problem as any other sensor |
On a closed-loop petrol ECU the usual OBD presentation of fuel trim is a percentage correction on top of the pilot quantity. Positive trim means the controller is adding fuel because the measured mixture was lean of the target. Negative trim means it is removing fuel because the mixture was rich of the target. Some tuning suites invert the sign. Read the channel definition for the logger in use.
Bosch’s account of adaptation is the reason both trims exist. Closed-loop correction always arrives one gas-transport delay late, so a new operating point is wrong until the loop catches up. The learned pilot correction is there so the next visit to that region starts closer to the target. A large learned correction at idle therefore says the base metering is offset in the region where the engine spends its time at low airflow. It does not, by itself, say what the wide-open-throttle mixture will be.
Closed loop, open loop and the delay
Closed loop means the measured oxygen is being allowed to change the injected quantity so that measured lambda moves toward the target. On the narrowband systems Bosch describes, that target is the switch at lambda 1, and the authority to do it waits until the sensor is hot and the coolant has passed a temperature threshold. Cold engines are run richer than lambda 1 so that they will idle; the loop stays off until then.
Open loop means the injection is coming from the air measurement, the injector model and the requested lambda, without that feedback. Wide-open throttle on the older narrowband strategies is open loop, because the switching sensor cannot measure a rich target. A wideband control sensor can hold a rich target in closed loop. Seeing “closed loop” in a log at full load means feedback is active. It does not mean the target is 1.00. Read the requested-lambda channel.
The delay is part of every reading. Bosch ties the closed-loop oscillation period to the time the gas takes to travel from the chamber to the sensor. On a fast throttle opening the manifold wall film also moves: higher manifold pressure slows vaporisation, fuel condenses on the walls, and the mixture goes lean until that film is rebuilt. The inverse happens on a sudden closing. Judge a wide-open-throttle lambda after the throttle and the boost have been steady long enough for that transport and wall-film lag to settle. The lean spike at the start of the pull is the transient Bosch describes.
Overrun is a third state. Bosch interrupts fuel on trailing throttle, chiefly so a misfire does not dump unburned mixture into the catalyst and overheat it. Measured lambda then rises sharply because the exhaust is mostly air. Exclude that section when you average a pull or an idle log.
Cold start, warm idle and part load
Cold start needs extra fuel. Bosch’s reasons are physical: poor vaporisation, poor mixing, and fuel condensing on the inlet walls, all of which remove fuel from the charge that reaches the cylinder and leave the burning mixture lean unless more is injected. Post-start enrichment continues briefly. Warm-up enrichment then tapers as temperature rises. There is also a second warm-up pattern in the same text: a lean warm-up with retarded ignition, used to heat the exhaust, as against a rich warm-up with secondary air injected after the valves. A cold lambda trace can therefore be rich or deliberately lean according to the catalyst-heating strategy. Compare it with the requested lambda for that mode.
Once the engine is at normal temperature, Bosch’s standard concept for idle and part-throttle is stoichiometric. Idle here means the engine makes only enough torque to cover its own friction and the accessories. The three-way catalyst’s window, about 0.99 to 1, is the band that loop is trying to stay inside. On a narrowband system the voltage keeps switching and the short-term trim keeps moving. On a wideband system the measured lambda sits on the target, with a small ripple from the same transport delay. Both are healthy closed-loop behaviour.
A warm-idle log that supports that conclusion has the coolant at operating temperature, the sensor heater finished, closed-loop status set, measured lambda at the stoichiometric target, and the two banks close to each other where the engine has two control sensors. The learned trim can sit away from zero. That means the pilot quantity is offset and the loop is covering it. A trim that has run to the end of its authority means the loop can no longer hold the target, and the idle mixture will then drift off lambda 1.
Part-load on the same warmed catalyst strategy stays at lambda 1 through the light-throttle range. Some direct-injection calibrations leave that homogeneous stoich mode and run stratified and lean. Bosch allows excess-air factors out toward lambda 4 for direct injection into the chamber. A cruise lambda well above 1 can be the requested stratified mode. Confirm the requested lambda and the combustion mode before treating it as a fault. Homogeneous engines with a three-way catalyst stay near lambda 1 in this region because that is where the catalyst works.
Idle airflow also explains a pattern that looks like a mixture fault. Unmetered air, from a vacuum leak or crankcase ventilation that enters downstream of the air meter, is a large fraction of the small idle airflow, so measured lambda goes lean and the trim adds fuel. The same leak is a small fraction of wide-open-throttle airflow, so the wide-open reading can look acceptable while idle trim is large. The lean direction at idle is the clue.
Wide-open throttle
Wide-open throttle is the operating point at which the throttle no longer restricts the fresh charge and the engine is making the torque that charge can support. Bosch states that supplementary enrichment may be required there, and that the enrichment is what produces maximum torque and power. The excess-air curve puts that maximum, for manifold injection, at lambda 0.85 to 0.95. On the 14.7 gasoline scale that is about 12.5 to 14.0 AFR. On the ethanol 9.0 scale it is about 7.7 to 8.6 AFR. The lambda band is the one that transfers.
That band is a combustion result in the Bosch text, not a map value to paste into every ECU. The delivered mixture still has to match the request. A useful full-load record contains requested lambda, measured lambda, engine speed, a load or cylinder-charge channel, throttle, intake temperature, coolant temperature, and, where they exist, fuel pressure, injector duty and knock correction. On a boosted engine add boost and the exhaust or component temperature the ECU actually uses.
Read the settled part of the pull. The opening lean spike is the wall-film delay. A sample taken while the gearbox is shifting, or while the throttle is still moving, mixes two operating points. After the trace has settled, measured lambda should lie on the requested lambda for that speed and load. A request inside 0.85 to 0.95 is in the band Bosch associates with maximum torque for manifold injection. A request nearer 1.00 leaves that band. Later calibrations sometimes hold lambda 1 over more of the load range and enrich only at the highest loads, so compare the measurement with the active request rather than with the power band alone.
Richer than the power band costs torque and raises fuel use and carbon monoxide. Leaner than the request, at this load, is the direction that moves the engine away from the torque mixture and toward higher exhaust temperature. The temperature mechanism is the one the turbocharger section below takes from the ME7 function description.
Why a boosted engine usually runs richer at full load
Pressure charging raises the fresh-air density in the cylinder, and Bosch notes that maximum torque scales with that charge density. Lambda is a ratio, so the same lambda at a higher air mass is a larger fuel mass. The fuel pump, the rail pressure and the injector flow have to supply that extra mass. If they cannot, measured lambda climbs even though the requested lambda was left unchanged. That lean drift is a delivery limit. Injector duty that is already near the whole cycle, or fuel pressure that falls as airflow rises, is the supporting evidence. Fuel-system capacity is covered on Fuel system upgrades.
The target itself often becomes richer as well, for a reason that is separate from simply feeding the extra air. A translated Bosch ME7.1 function description for exhaust-temperature control (ATR) and lambda component protection (LAMBTS) states the mechanism directly: an excessively high exhaust temperature can be lowered by enriching, because more fuel enters the cylinder than stoichiometric combustion needs, the unused fuel vaporises on the cylinder walls, and the exhaust temperature falls. The parts named are the exhaust manifold, the turbocharger and similar components. Enrichment in that description is applied only while temperature is high. It is removed again when the temperature falls back through the threshold. It is a downward control of temperature.
In that ME7.1 structure the component-protection target comes from a map against engine speed and relative cylinder charge, and it is enabled only after modelled temperatures at the manifold, the catalyst or the cylinder head have crossed their thresholds and a delay has elapsed. Full-load enrichment can carry the base rich target. Component protection, and where fitted a measured exhaust-temperature control, can demand a still richer target when that base enrichment is not enough. The active request can therefore be richer than the full-load map alone. This behaviour is described for ME7.1. Other ECU families use their own maps and thresholds. The underlying temperature effect is the one that text gives.
More boost, a hotter intake charge or a long time at high load pushes those modelled temperatures up. The same ME7.1 sheet states that a loss of ignition-angle efficiency raises exhaust temperature, and that mixture enrichment can be used against that rise. As the temperatures cross the protection threshold, the richer protection target becomes the one the engine is asked to follow. A car calibrated at a lower boost can therefore need a richer wide-open target once boost, intake temperature or sustained load is raised. How much richer is what that engine’s temperature and knock behaviour decide. The function sheet does not publish a universal offset per unit of boost.
Knock is a further, separate path. Bosch defines knock as end-gas igniting before the flame reaches it, when the spark is too far advanced, with a sharp rise in chamber temperature and pressure. Some calibrations add mixture enrichment as part of the knock response. On ME7 that path is distinct from component protection; see the request table on the ME7 fuelling page. Enrichment is not a substitute for a plausible ignition map. Timing and knock are covered on Ignition tuning. Charge temperature, which sets how soon the protection path and the knock path both become active, is covered on Charge cooling, intercoolers and water injection.
Ethanol changes the arithmetic and the cooling. Oak Ridge lists ethanol’s latent heat at 923 kJ/kg against roughly 350 to 500 kJ/kg for gasoline, and notes that ethanol exhaust runs cooler than gasoline exhaust in the same engine conditions. Those properties are a reason a high-ethanol blend can hold similar hardware temperatures at a lambda closer to the torque band than gasoline needs. Confirm that on the temperature log for the engine in question. The lambda is still the value to log. The gasoline AFR number is not.
Richer than the request is a different finding
A richer target is a calibration choice: the ECU asked for it, for torque or for temperature. A measured lambda below that target means the engine received more fuel, or less air, than the model used when it calculated the injection. Enriching the map further would move the request away from the engine and would not repair the metering error.
| What the log shows | Reading |
|---|---|
| Measured lambda matches the active request after the transient | Delivery and the model agree at this operating point |
| Measured lambda below the request | Richer than asked. Extra fuel, or less air than the air meter reported |
| Measured lambda above the request | Leaner than asked. Missing fuel, or extra air |
| Idle trim adding fuel, wide-open lambda on target | The offset is in the low-airflow metering. A vacuum leak fits this shape |
| Idle near lambda 1 with modest trim, wide-open lambda drifting up as boost rises | Fuel delivery is running out as air mass rises |
| Requested lambda steps richer part way through a long pull, and measured lambda follows | A protection or temperature path has become the active request. The base full-load map is no longer the target |
| One bank richer than the other at the same moment | That bank’s fuel, air or exhaust sample differs. A tailpipe average will conceal it |
Typical reasons for a mixture richer than the request are a fuel pressure higher than the model, injectors that flow more than the constant the ECU is using, an air meter that under-reads, or an injector that is not closing. Typical reasons for a mixture leaner than the request are the opposite air-meter error, low fuel pressure, an injector at full duty, or air entering the exhaust ahead of the sensor. An exhaust leak upstream of the sensor adds oxygen to the sample, so the indicated lambda rises. The engine can be richer than the gauge says.
Change one of those causes at a time and repeat the same settled pull. A new requested-lambda number, written into a map before the measurement matches the old request, mixes a metering fault with a target change and the next log cannot separate them.
Diesel lambda stays lean of 1
The numbers above are spark-ignition numbers. Bosch’s diesel injection text draws the contrast on purpose: a diesel has to run with overall excess air, because locally rich regions make soot. For the turbocharged diesels in that text, full-load lambda lies between about 1.15 and 2.0, and idle or no-load lambda rises above 10. Those figures are the overall cylinder masses. Local lambda inside the burning spray is what sets ignition and soot, and it is richer than the average.
Later passenger-car diesels with exhaust-gas recirculation sit at lower idle lambda than that older no-load figure, because recirculated gas displaces air. One published road-test example, with the recirculation valve operating, showed idle lambda around 2.9 to 3, against about 6.5 to 7 on the same engine with recirculation out, and about 1.33 at maximum load, where the smoke limit was described as holding lambda at or above about 1.3. That is one vehicle. It agrees with the Bosch statement that a turbocharged diesel at full load is still lean of 1, and it shows that a modern idle number can be far below the older “above 10”.
Bosch’s electronic diesel control description uses a wideband sensor for this directly. A smoke-limit map, with air mass and engine speed, produces a desired lambda. The maximum fuel quantity is calculated from that desired lambda and the measured air, and a lambda controller trims the quantity so measured lambda meets the desire. Richer on a diesel means a lower lambda, more smoke, more particulate-filter loading and higher exhaust temperature. It is still normally above 1. A petrol wide-open target of 0.85 applied to a diesel is a different combustion regime from the one the smoke limiter is built to hold.
Readings that lie
The sensor reports oxygen where it is fitted, after it is hot, inside its pressure range. Several ordinary conditions produce a confident number that is not the combustion lambda.
- Sensor still cold. Bosch’s switching voltage is not valid below the temperatures given above. Discard the start of the log.
- Exhaust leak before the sensor. Added air raises indicated lambda.
- Probe after the catalyst. The catalyst consumes oxygen and oxidises residual fuel. The downstream gas is the catalyst monitor.
- Pressure the sensor was not compensated for. The LSU 4.9 sheet treats accuracy as falling once exhaust pressure exceeds 2.5 bar.
- Fuel the gauge was not told about. An AFR display locked to 14.7 mis-states E10, E85 and ethanol. The lambda channel does not have this error.
- Averaging across overrun or a gearchange. Fuel cut drives lambda high. A tip-in drives it lean for the wall-film delay. Both distort a mean taken over the whole trace.
- One cylinder, read as the engine. A probe in a single runner, or one bank of a V engine, is a local sample.
- Controller and ECU on different time bases. A wideband logged on its own cannot be compared with a requested-lambda channel until the traces are aligned.
What to record before changing a mixture map
- Record the engine, the fuel in the tank, whether it is turbocharged, the ECU hardware and software, and a recoverable original calibration. Mixture targets from another engine are a hypothesis.
- Confirm the sensor type and its position. A narrowband voltage answers only the stoichiometric question. A wideband after the catalyst answers a catalyst question.
- Log requested lambda, measured lambda, short-term trim, learned trim, coolant temperature, intake temperature, engine speed, load or cylinder charge, and throttle. Add boost, fuel pressure, injector duty and the protection-temperature channel where they exist. Log both banks where both exist.
- At warm idle, expect a stoichiometric target, closed loop, measured lambda on that target, and trims that still have authority left. Investigate a lean idle with a large positive trim before touching a full-load map.
- On a settled wide-open pull, compare measured lambda with the active request after the tip-in spike. For a homogeneous petrol engine, Bosch’s maximum-torque region is lambda 0.85 to 0.95. A boosted engine can be commanded richer than that when component temperature requires it.
- If measured lambda and the request disagree, repair the air measurement or the fuel delivery before editing the target. If they agree and exhaust temperature or knock correction is still unacceptable, the target and the ignition map are the calibration question, decided on this engine.
- Repeat the same idle and the same pull after one change. Keep the log with the calibration that produced it.
Carburettor engines face the same ratio, set with jets and needles instead of a lambda target. That hardware is covered on Carburettor tuning. How a remap fits into the wider air, fuel and protection calibration is covered on ECU remapping.
A narrowband can only hold lambda 1
A switching sensor has one calibrated point. In the M-Motronic account the voltage sits near 100 mV when lambda is above 1, jumps at lambda 1, and sits near 800 mV when lambda is below 1. Closed loop uses that jump as the setpoint. It adds fuel on the lean side and removes fuel on the rich side, and the exhaust delay makes the mixture walk back and forth across the step. The average of a healthy oscillation is lambda 1, inside the catalyst window of about 0.99 to 1.
That is the whole measurement. A voltage held near 800 mV means richer than lambda 1. It does not say whether the mixture is 0.98 or 0.80. A voltage held near 100 mV means leaner than lambda 1, and not whether it is 1.05 or 1.40. Wide-open-throttle enrichment, a lean target, and a diesel smoke-limit lambda all sit off the step. A narrowband cannot check them. The number comes from a wideband, or from a gas analyser that calculates lambda from the exhaust gases.
Using a wideband oxygen sensor
A wideband keeps measuring once the mixture leaves lambda 1, because the controller reports pump current rather than a switch. On the LSU 4.2 characteristic that current is about zero at lambda 1, negative when rich and positive when lean. That is what makes a rich wide-open target, or a lean one, a number instead of a pegged voltage. The sensor generations, heaters and connectors are on Wideband oxygen sensor types. The practical rules below are the ones those sheets actually impose.
- Match the controller to the sensor. The LSU 4.2 and LSU 4.9 technical informations name the AWS box, the LA4 and the CJ125 as examples, and they give those two units different voltage conversions for the same pump current. LSU ADV has no trim resistor and is specified with the LT4 ADV evaluation unit. An NTK element is a different sensor. A menu entry for the wrong generation produces a confident, wrong lambda.
- Let the controller run the heater. Both LSU sheets forbid a direct 24 V supply unless a converter is fitted. While condensation is possible the 4.2 sheet limits effective heater voltage to 3 V and the 4.9 sheet to 2 V. Water on the hot ceramic cracks it.
- Put the sensor where the gas is representative and still the engine’s mixture. Ahead of the catalyst, in the collected stream, answers the fuelling question. Downstream of a lit catalyst answers a catalyst question. One runner answers one cylinder.
- Follow the installation limits for that sensor. The LSU 4.2 sheet requires the connector end upward, at least 10° off horizontal, so liquid does not pool on the element. It also requires the reference-air path, including the connector’s compensation hole, to stay clear of oil and wax. A blocked reference shifts the lambda.
- Follow the controller’s own calibration procedure: some require free-air calibration, while others use a sensor trim resistor or specify another method. The AEM X-Series inline manual, for example, defaults to resistor calibration and offers free-air calibration as an option. Display lambda, or understand which fuel-specific stoichiometric ratio the controller uses for its AFR display.
- Account for pressure. The LSU 4.9 sheet allows 2.5 bar absolute continuously and 4 bar for a limited life, and a change in pressure shifts the signal on the 4.2 sheet as well. A pre-turbine bung on a boosted engine needs the controller’s pressure compensation, or the reading is degraded.
- Log the controller’s lambda against the ECU’s requested lambda, rpm, load and throttle on one time base. Judge the settled part of a pull, after the wall-film spike and clear of any fuel cut.
Exhaust-gas analysers
A gas analyser measures the chemicals in a sample of exhaust, then, on a four-gas or five-gas instrument, calculates lambda from them. It is slower than a wideband and it sees things a single pump current cannot separate.
The Kane Automotive manual divides the instruments by how many gases they measure. Every model in that range measures carbon monoxide and unburned hydrocarbons. A four-gas model adds oxygen and carbon dioxide. A five-gas model adds nitric oxide. Lambda, air-fuel ratio and a corrected CO figure are calculated from CO, HC, O2 and CO2. A two-gas instrument stops at CO and HC, so it can show a rich or a misfiring idle and it cannot calculate lambda. A three-gas workshop set is commonly CO, CO2 and HC. That combination is not one of the Kane models cited here. Without oxygen it can still show the CO and CO2 pattern below, and it cannot finish the oxygen balance.
The calculation used for that lambda is the Brettschneider equation, from Johannes Brettschneider at Bosch, published in Bosch technische Berichte, volume 6 (1979), number 4, pages 177–186. Bridge Analysers describes it as the balance of oxygen against the carbon and hydrogen in the fuel, using measured CO, CO2, HC and O2. At lambda 1 the balance is exact. Bridge’s worked examples treat 1.05 as 5 percent lean and 0.95 as 5 percent rich, then multiply by the stoichiometric ratio of the selected fuel. Their note on nitric oxide is that 1,000 ppm of NO is only about 0.05 percent oxygen in the balance, so a four-gas analyser is enough for lambda. The fifth gas barely moves the number. It is there for NOx, not for a finer mixture reading.
Bridge also makes a point that changes how the two instruments are compared. The calculated lambda is an oxygen-to-fuel balance, and they state that it does not depend on combustion efficiency. A misfire does not, by itself, move Brettschneider lambda. The hydrocarbons do move, which is how the analyser shows the misfire while the lambda still reports the mixture that was delivered. A wideband returns one oxygen-derived number and does not report HC, CO or NOx separately. An air leak is different: Bridge says extra air dilutes the sample and shifts lambda by about the same fraction. Five percent extra air makes a true lambda of 1.000 read about 1.050.
The sample has to be taken where the question is. Pre-catalyst gas is the engine. Tailpipe gas after a lit catalyst is the catalyst’s output. The sample hose, the water trap and the calibration gas are part of the measurement. A leak in the hose is the same error as a leak in the exhaust.
Tuning with CO and CO2
Before wideband meters were common, mixture was adjusted from the exhaust chemistry, usually at idle. The aim people still use is the highest CO2 the engine will make at that idle, with CO down at the low percentage the manufacturer specified for the test. This page does not give a universal CO percentage. The specification belongs to the engine and the test.
The chemistry is why both gases are required. Rich of stoichiometric, part of the carbon leaves as carbon monoxide instead of carbon dioxide, so CO rises and CO2 falls. Lean of stoichiometric, spare oxygen dilutes the sample, CO2 falls, and free O2 rises, while CO is already near the floor. CO has stopped saying how lean the mixture is. CO2 also falls when combustion is incomplete and hydrocarbons are high. A low CO2 on its own is therefore three different faults. CO says “rich” only on the rich side of the peak. O2 says “lean or an air leak” once CO has already gone quiet. Hydrocarbons say “unburned fuel”, which is the misfire Bridge separates from the lambda balance.
| What the gases do | Reading |
|---|---|
| CO2 at its peak, CO low, O2 low, HC low | Complete combustion near lambda 1. This is the CO/CO2 idle target. |
| CO high, CO2 down | Rich of that peak. CO is useful here. |
| CO already low, O2 high, CO2 down | Lean of the peak, or extra air. CO cannot see this side. |
| HC high, CO2 poor, Brettschneider lambda still near the delivered mixture | Incomplete combustion. Bridge’s point: lambda did not move just because the oxygen was not used. |
| Lambda from the four gases shifted lean by about the leak fraction | Air entering the sample. Bridge’s example is 5 percent air moving 1.000 to about 1.050. |
The benefit is that the gases separate those cases, and they do it with a tailpipe probe on an engine that has no wideband, including a carburettor. A four-gas lambda is also an independent check of a wideband, because it is calculated from the chemistry rather than from that sensor’s pump current. Carburettor hardware for the same adjustment is on Carburettor tuning.
The drawback is what this method cannot see. CO alone is blind lean of stoichiometric. CO2 alone cannot tell rich from lean from a misfire. The CO2 peak is a coarse finder of lambda 1, not a reading of a wide-open target in the Bosch maximum-torque band of 0.85 to 0.95, and not a reading of a richer boost-protection target. The sample line is too slow for the tip-in spike and the settled part of a pull. Those points want the wideband trace.
The calculated lambda is only as good as the fuel constants. The Kane manual’s displayed lambda uses a nominal hydrogen-to-carbon ratio of 1.7261 and an oxygen-to-carbon ratio of 0.0176. Those are that instrument’s petrol assumptions. Ethanol already contains oxygen, so the constants have to change with the fuel or the lambda moves while the engine does not. The analyser also has to be calibrated, and the sample has to come from ahead of the catalyst when the question is the mixture the engine burned. Nitric oxide on a five-gas instrument is the emissions and temperature clue. Bridge’s figure says it hardly changes lambda.
Sources
- Bosch, M-Motronic Engine Management, technical instruction. Stoichiometric 14.7:1; lambda definition; maximum power at lambda 0.85 to 0.95; economy at lambda 1.1 to 1.2; direct-injection excess air out to lambda 4; three-way catalyst window about 0.99 to 1; warm idle and part-load at stoichiometry; wide-open-throttle enrichment for torque; cold enrichment; wall-film transient; overrun fuel cut; narrowband about 100 mV lean and 800 mV rich; sensor temperatures; closed-loop oscillation and adaptation. Accessed 2026-11-11.
- MIT OpenCourseWare 2.61, lecture 4, Combustion and thermochemistry (2017). Petroleum-fuel stoichiometric AFR about 14.6, range about 14.2 to 15; lambda as relative air-fuel ratio and equivalence ratio as its reciprocal.
- Oak Ridge National Laboratory, Technical Issues Associated with Intermediate Ethanol Blends, ORNL/TM-2007/37, Table 4.1. Ethanol stoichiometric AFR 9.0, gasoline 14.7; latent heat 923 kJ/kg for ethanol against 349 and 380–500 kJ/kg for gasoline; ethanol exhaust cooler than gasoline under the same conditions.
- US Department of Energy, Handbook for Handling, Storing, and Dispensing E85, Table 2. Air-fuel ratio by weight about 9 for ethanol, 14.7 for gasoline and 10 for E85, with blend properties depending on the hydrocarbon component.
- Szybist, DOE DEER 2012 presentation. One measured gasoline stoichiometric AFR of 14.42 and one ethanol value of 9.45, illustrating composition scatter around the reference figures.
- Bosch Mobility, switching-type lambda sensor. Step response at the stoichiometric ratio, stated as 14.7 parts air to 1 part fuel.
- Bosch Mobility, wideband lambda sensor. Wide-range oxygen measurement for lean and rich control on petrol, diesel and gas engines.
- Bosch Motorsport, Lambda Sensor LSU 4.9 data sheet. Measuring range lambda 0.65 to free air; gasoline, diesel and E85; exhaust pressure at or below 2.5 bar with accuracy falling above that; continuous exhaust temperature below 930 °C.
- Translated Bosch ME7.1 function modules ATR 1.60 and LAMBTS 2.120. Enrichment lowers exhaust temperature by evaporating excess fuel on the cylinder walls; component protection of the manifold and turbocharger; the richer target is enabled from modelled temperature thresholds. This is an ME7.1 function description, not a universal map.
- Bosch, Diesel Injection Principles. Turbocharged diesel full-load lambda about 1.15 to 2.0, idle and no-load above 10, because the engine must run with overall excess air.
- Bosch, Electronic Diesel Control. Wideband lambda used for diesel closed-loop control and for full-load smoke limitation from a desired lambda.
- Castrol Fastscan, EGR diagnostics. One road-test example: idle lambda about 2.9 to 3 with EGR and about 6.5 to 7 without, and about 1.33 at maximum load. A single vehicle, not a class value.
- Bridge Analysers, Lambda calculation, the Brettschneider equation. Johannes Brettschneider, Bosch technische Berichte, vol. 6 (1979) no. 4, pp. 177-186. Four gases required for lambda; 1,000 ppm NO is about 0.05 percent oxygen in the balance.
- Bridge Analysers, diagnosing fuel control. Brettschneider lambda does not move with combustion efficiency alone; a 5 percent air leak moves lambda 1.000 to about 1.050.
- Kane Automotive gas analyser manual. Two-gas measures CO and HC; four-gas adds O2 and CO2; five-gas adds NO. Displayed lambda uses nominal H/C 1.7261 and O/C 0.0176.