Direct Digital Lambda (DDL)
Micro-Controlled Wide-Band Exhaust-Gas-Oxygen Sensor

Copyright © 2000-2002 Bernd Felsche , Perth, Western Australia.
This page discusses the use of a microcontroller (µC) close-coupled to a wide-band lambda probe (O2 sensor) to divest details of sensor operation from an engine management system which really only wants to get feedback about the air-fuel-ratio.

Digital techniques employed in the proposed system greatly simplify analogue circuitry otherwise required to drive the sensor and to provide more latitude for control, diagnostics, testing, calibration and tuning.

Cell temperatures are measured directly by alternating-current pulses. The previous approach was to use the heater resistance as an indicator of cell temperature. Direct measurement is preferable as cell temperature determines cell performance characteristics.

The proposed method arises primarily from DIY-WB discussions in the DIY-EFI mailing list (and off-list with other list members) and is a logical extension of prior art and therefore not patentable in a sane world.

A DIY-WB project exists that provides an interface to a NTK/Honda-type wide-band O2 sensor. Take a look at that if you're simply looking for a convenient, working, analogue interface.

The owner of the Copyright © hereby grants free use of the idea on the understanding that any development be contributed to DIY-EFI projects.

The author would like to thank the following DIY-EFI list members for taking the time to discuss issues and to provide pointers to further sources of information.

A separate DDL mailing list has been established.

Disclaimer

I am not an electronics engineer. I'm a mechanical engineer.
Any information on this page is used at your own risk.

DDL Advantages


A Wide-Band O2 Sensor

Very little hard, detailed information about how the sensors are supposed to be driven is readily available. A summary of relevant information is collected here. (Updated October 26th, 2001)

Circuit Details

A sample design is composed of an Atmel AT90S4433 in conjunction with a Bosch wide-band lambda sensor. The recently-announced Atmel ATMega8 has more on-chip features and higher clock rates, so provides a ready upgrade path if the chip AT90S4433 proves to be marginally insufficient.

DDL Circuit Schematic
Click image for larger view, or shift-click to download either the gzipped Postscript (56k) or the Adobe-PDF (112k) version.
A sample board layout illustrates the low component count and small size.
Gzipped Cadsoft Eagle schematic and board are available.

The circuit shown is for discussion only and is incomplete! There is no software yet to operate the sensor.

The µC is an 8-bit RISC-based CPU with 6-channel, 10-bit ADC, with a total of 20 I/O pins, UART, one 8-bit timer/counter, one 16-bit timer/counter, a watchdog timer, in-built PWM on the 16-bit counter, an analogue comparator and internal brown-out detection. There's 4K bytes of flash for program memory, 256 bytes of EEPROM and 128 bytes of SRAM. Flash and EEPROM are in-system-programmable via a serial peripheral interface (SPI). Various power-saving modes are supported; keeping SRAM alive with only a few micro-amps.

Pump Cell Drive

The pump cell is driven by a virtual H-bridge switched by pulse-width-modulation (PWM) from two pins (PD4 and PD5) of the µC. A member of the DIY-EFI suggested the drive method; employing a 74LS244-type bus driver chip to provide the necessary current sources and sinks as the µC cannot source sufficient current. Another contributor suggested pairing (otherwise unused) outputs of the driver for improved drive performance.

Later generations of the µC are capable of sourcing sufficient currents, but the voltage that that would present to the µC's input pins at certain stages would be in excess of the rated input.

A 74AS244 is shown, but just about any 74*244 equivalent that will operate from (approximately) 3V should work at the low speeds of the circuit. The low voltage is chosen so that the maximum voltage presented to the ADC inputs of the µC will be below the 5V rail.

An AC part is a reasonable option. The driver must be able to sink or source at least 10 mA. The VOH and VOL do vary between versions, but an acceptable component should be available, given that the NTK controller only outputs ±2V. As the µC is capable of taking relative measurements, the voltage losses through the driver can be compensated. Component choice is reduced to a question of supply voltage, reasonable availability and moderate efficiency.

Pump current is reversed by alternating drive pins on the input to the driver, and gating the latch to alternate the current flow direction. The latch is biased to 5V so that there is no drive current unless the µC is actually running. A RESET causes all I/O pins to tristate to high-impedance.

The driver is powered from a reduced-voltage rail. This allow the Nernst cell voltage, which floats above the pump cell's bias be measured almost all the time, even when the potential on the pump cell has been reversed. A maximum of (about) 3V is applied at the pump cell; the Nernst cell voltage may be 1.1V higher. The AVR's ADC and analogue comparator require that the voltage to be measured remains below AREF; its analogue reference voltage (and full-scale for the ADC) and below Vcc + 0.5V.

Average pump current (Ip) is determined by the nett pulse-width of current flow through the bridge.

Various strategies are immediately obvious and the advantages of pulse-width modulation (PWM) in conjunction with alternating current through the pump cell are discussed in the literature.

A notable benefit is the expected improvement in cell response around stoichiometric AFR where the average pump cell current is zero, but equal amounts of gas are pumped in and out. By using PWM techniques and alternating cycles of equal charge, the measurement cell is continuously pumping ions; in contrast with a constant-current drive which would rely on random diffusion to sense any variation in gas composition.

Reference Cell

The µC balances the voltage at the reference cell against the required value of 0.45V. The cell's voltage will switch just like a conventional lambda probe as the composition of the gas (partial pressure) at the Nernst cell changes in response to the action of the pump cell.

Measurement of the cell voltage is by subtracting two voltage measurements at pins PC2 and PC3 as the cell floats to provide for periodic current reversals for cell temperature sensing.

One of the previous limits in DDL was the ADC delay for determining the Nernst cell voltage. It came to mind during that discussion with list members that the AVR has a perfectly-adequate analogue comparator on-chip and that this could be used to detect a swing from stoichiometric (0.45V) almost instantaneously. The deviation from the nominal voltage, and its direction could be used to determine the pump current direction required to re-balance the Nernst Cell.

The schematic shows a 0.45V constant-voltage device "riding" on the virtual ground with its output going to one input of the comparator, and the Nernst signal going to the other comparator input. An AD580 was chosen as a voltage reference due to ready availability; the supply voltage from the regulated 5V rail would be marginal due to the necessary voltage drop in the device to maintain the nominal output voltage above the floating ground.

Sensing bandwidth increases correspondingly. Response to a voltage change at the Nernst cell is as little as 4 µC clock cycles when receiving an interrupt from the comparator, compared to several hundred cycles for ADC conversion and comparison. The comparator's effective hysteresis is less than 10 mV.

Heater Drive

A power-FET is used to PWM the current through the integrated heater of the sensor. The pulse-width is determined in software.

The approximate current through the heater is measured by the voltage drop across sense resistor R1. R1 is chosen so that a cold sensor in series with the sense resistor, will present a voltage below the clamp voltage at the µC's ADC input when the FET is turned on. R1 must also not be so low that it will cause excessive current to be drawn through shorted heater terminals.

µC heater current measurement can be used to continuously verify that the heater is still intact.

High-side heater drive is preferable but would be more expensive due to the charge pump required to turn the FET on hard from logic level. Suitable chips to drive low-cost, efficient FETs from logic level are either rare or expensive; or both. A relay or circuit breaker could also be employed to protect the circuit but an in-line, 3 Amp fuse on the battery lead feeding the controller is much simpler, though limiting the maximum heater power.

Current is approximate as there are insufficient ADC ports on the µC to measure the actual battery voltage. Approximate measurement is sufficient to get the heater into the nominal operating temperature range, within which closed-loop temperature control is expected to be by measurement of cell temperature.

Sensor temperature change is limited to 50 K/second by software. The heater power can be logically "capped" to never exceed that rate, even if the sensor were mounted in an insulated chamber.

A resistor is used to bleed off any charge at FET gate which might otherwise cause it to become wedged on, especially when the µC's output is tri-stated due to a reset.

The high-current diode protects the FET against voltage reversals. This may be superfluous for the resistive load of the heater.

A Zener diode and series resistor protect and clamp the analogue input to the µC in case the heater terminals are shorted. A small amount of current (approximately 50mA at full battery voltage) will flow through the heater with the FET turned off.

Analogue to Digital Conversion

A precision voltage reference could be used as a reference to the µC's internal analogue to digital conversion (ADC) circuitry. The voltage rails will then not have to be regulated as accurately, which is difficult anyway due to the pulse driving of the implementation.

The main current consumers are however isolated. The pump drive IC operates off its own regulated voltage rail and the heater is operated directly from the battery.

Interfacing

The engine's ECU interface can be by single-wire, but the ECU connector shown has provisions for power and ground as well. This connection is convenient as it's no more complex than a 3-wire setup for switching lambda sensors with heater. It's entirely possible that the existing wiring harness could be employed.

As the µC has a UART, the addition of a level-driver chip (e.g. MAX232) would allow the µC to communicate directly with a portable, external computer.

A serial-programming interface (SPI) is also brought out to a connector on the sample schematic. An external programmer could then update the firmware in the µC while it's in-circuit.

A header block is shown on the schematic to allow an in-system programmer to be connected without fear of damaging the sensor if connected. A CMOS switch could be used instead as described in Atmel documentation because the lines concerned are used only to source or to sink currents when measuring the temperature of the Nernst cell.


Mode of Operation

Lambda Sensing

By sensing the voltage of the reference cell, the µC decides on the width of current pulses to push and pull through the pump cell. Using the 16-bit timer-counter of the µC provides up to 10-bit PWM resolution in hardware, more than enough for application. The schematic shows a short from the OC1 pin to INT1 so that the µC can put PWM outputs on the correct transistor pair without additional external components. As well as a physical limitation of pump current, the maximum average pump current can be limited according to stored parameters within the µC's EEPROM and/or flash memory.

Calibration constants within the µC map the pulse width to a lambda value. This can then be transmitted to the engine management system as a bit-stream or fed into a serial DAC (not shown) for a conventional analogue interface to an engine management system.

Lambda values from 0.7 to 4.0 are usually considered interesting. These could be mapped linearly to an 8-bit serial data stream ranging from 1 to 254 (extremes used to indicate special conditions to the ECU) with each increment representing a change in lambda of 0.013. A digital output of 24 would then be equivalent to lambda=1.0 A change of 1% is detectable and hence it's possible to operate within a narrow window to assist catalytic conversion.

The nett pulse width for the pump cell is determined by a software providing for rapid response to perturbations and reducing long-term drift. Input to the control loop is the Nernst cell voltage which remains nominally-fixed at 0.45V.

A fake lambda value can be provided to external circuitry until the sensor reaches nominal operating temperature, or in the event of a sensor fault.

Cell Temperatures

An accurate indication of temperature is provided by the cells themselves. Cell resistance is calculated by passing a nominally-constant test current through the cell to measure its resistance.

US Patent 6,120,677 describes using the pump and reference cells' resistance to determine temperature. The pump cell drive described above is already capable of allowing the necessary measurements to take place. Pins PC3, PC4 and PC5 measure the voltage across the cell as well as the drop across the current-sense resistor.

The provision of three analogue inputs per sensor provides precision information about the current and voltage levels for determining resistance. Note that two cells share a common pin on the sensor's connector. As the circuit switches are driven to saturation during pulse-width modulation, the voltage levels at the source and sink are defined by those components, so they are known to some precision without any measurement having to take place.

Cell resistance can be determined by one analogue measurement at the junction of cell and sense resistor; which has the reduced likelihood of aliasing (although that is in any case minimal as the characteristics are unlikely to change appreciably in the 140 µs or so it takes to sample 3 values).

Manufacturing tolerances can be compensated for by a procedure after assembly. Setting jumpers on board can allow the µC to measure and store correction factors for resistances. An example of this is described below in the section on Calibration.

Component drift is another issue. It is probably a good idea to calibrate at nominal operating temperature.

The preferred PWM method (similar to that in Patent 5,312,538) can be over-loaded to perform resistance measurement without disrupting normal pump-cell operation. Nernst cell voltage sampling on the other hand, will have to be suspended for about 1.5 milliseconds during temperature measurement.

The µC's push-pull drive for the Nernst Cell shares one pin with the SPI which is no real compromise because the latter is used only for programming the device when it's off-line. A current reversal of equal charge is necessary on the Nernst cell as it will otherwise remain biased in one direction for a significant time. The reversal reduces the recovery time from around 16 milliseconds to 0.5 milliseconds.

As the pump cell is to be continuously driven by suitable pulses, the biasing problem is eliminated. Pump drive must be coordinated as the pump and Nernst cells share a single connector pin. It appears likely that the timing of resistance measurement of the Nernst Cell can be based on cycling of the pump cell, allowing the pump to continue operating.

Literature indicates that the cell temperature varies relatively slowly; of the order of 3°C per second, so such measurements aren't necessarily frequent or disruptive if the control logic takes the time constants into account.

Heater

When the system is started, the pulse-width can be ramped-up according to the cell temperatures as evident from their resistance. The AVR's 8-bit timer-counter is used to synthesise PWM as this particular µC supports only one PWM in hardware,

The pulse width for the heater is also to be determined by a software algorithm with knowledge of cell temperatures determining a gradient to define the heater output required.

Some literature indicates that the sensor should operate at a temperature above that of the exhaust gas for maximum precision. If the pump cell (exposed to exhaust gases) is at a greater temperature than the Nernst, then heater output has to increase because the exhaust gas is hotter; up to the limit of operating temperature.

If, on the other hand, the Nernst cell is slightly hotter than the pump cell, then there may be sufficient heating. However, a colder pump cell would indicate that the exhaust gas is cooling the pump cell (no surprises here) and that the heater output can be lowered as long as both cells are above the minimum nominal operating temperature.

The cell temperatures and heater drive can be combined to provide a metric of exhaust gas temperature. Although the pump cell is nominally at exhaust gas temperature as a first-approximation, it may be slightly cooler or hotter than the surrounding gas stream. And indication of which and its magnitude is determined by the temperature gradient, the heater power and thermal calibration constants.


Calibration

Calibration precedures can be triggered by external events; the DATAOUT line on the ECU port could be pulled low or high to cause the µC to start listening for commands on the UART (serial) interface.


Last updated July 15th, 2002
Copyright © 2000-2002 Bernd Felsche , Perth, Western Australia.