The BMW WilliamsF1 team lifts the hood on its fifth-generation vehicle control system.

A modern Formula 1 (F1) car communicates an enormous amount of data to the engineering teams that support it. How the teams collect, analyze, and react to that data is vital to competitiveness. The collect-analyze-and-respond process is also subject to change, as the sport’s rule-makers aim to reduce onboard electronics to control costs and achieve closer racing competition.

Throughout the last decade, the electronics department at BMW WilliamsF1, led by Steve Wise along with engineers Dave Walker and Ian Cartwright, can claim to have been in the driver’s seat as onboard electronics have spread to every extremity of a Formula 1 chassis.

The team is now running its fifth-gener-ation VCM (vehicle control and monitor-ing) unit; the current Stage 5 VCM uses a Xilinx Virtex™-E XCV600-E FPGA along-side a Texas Instruments™ DSP. Shown in Figure 1, the VCM controls all aspects of the 2003 FW25’s chassis (other than the engine). And the team is continuing to extend and refine the unit’s capabilities.

Adapting to Rule Changes
Rules change quickly in F1 racing. Teams must be able to respond with a competitive solution that meets every new detail. The regulatory landscape has been especially volatile over the last decade. Driver aids such as active suspension, anti-lock braking, traction control, and launch control all have been banned at one time. Some driver-assist features have been granted a reprieve for this year – only to be completely outlawed for the 2004 season.

The ability to quickly add and remove functional blocks is therefore critical to each team’s ability to compete and to demonstrate compliance.

Successive generations of the BMW WilliamsF1 VCM have made great use of FPGAs to achieve rapid compliance. This flexibility has allowed BMW WilliamsF1 to migrate more and more sophisticated functions into hardware as new generations of FPGA silicon offer extra capacity and complexity.

Controlling an F1 Car
The VCM is a data logger and processor on a grand scale. Its vehicle control functions include overseeing the hydraulically actuated gear change, which calls for the VCM to initiate a gear shift either in response to a driver request or automatically.

Shifting Gears
The gear change sequence in a F1 gearbox requires precise control over the positions of the gearbox actuators, as well as controlling the clutch and coordinating engine revolutions per second with the BMW engine controller. A typical gear change takes less than 50 ms. The system also prevents the driver from damaging or over-revving the engine.

For the 2003 racing season, the VCM is allowed to initiate all gear changes with no input from the driver. But automatic gear changes will be outlawed for 2004, requiring the driver to initiate each gear change manually. Once the driver has initiated the change, the VCM will be allowed to handle the rest of the gear change sequence.

Traction
The VCM also handles traction control by performing calculations based on complex tire models to predict the amount of wheel slip required to achieve maximum traction and minimum tire wear. The VCM calculates control targets and sends this data to the engine controller.

Generating control signals for the hydraulically actuated differential also comes under the VCM’s domain – with the aim of providing maximum traction from each of the rear wheels to optimize stability of the car when cornering.

Launch Control
The use of “launch control” – which optimizes the car’s standing start as it leaves the grid – has received much press coverage in recent seasons. Incoming rules will outlaw this feature, but for the 2003 season, the launch start controller is a functional block within the VCM. Clutch and engine targets communicated by the VCM help the car to accelerate from the grid to 100 m/hr in less than three seconds.

Real-Time Feedback
The VCM displays driver information signals and warning indicators on the steering wheel. Even though a modern F1 steering wheel can cost upwards of $33,000, there is relatively little intelligence onboard – that is, until next year’s BMW WilliamsF1 team adds a Xilinx FPGA to perform some of the instrumentation processing functions locally.

Logging and Telemetry
The VCM logs around 220 channels of data at rates as high as 1 KHz. As many as 90 of these channels are signals from sensors used to analyze the car’s performance over an entire race or test distance. Some channels monitor the actions of the control software, while others monitor the driver’s inputs. An example of the logged data is shown in Figure 2.

Each of the Stage 5 VCMs fitted to the two FW25 cars of drivers Juan Montoya and Ralf Schumacher can log 256 MB of data generated by the cars during a race. This data is stored on a standard, commercial Compact-Flash™ (CF) card mounted permanently inside the VCM, which provides convenient and cost-effective non-volatile storage. The CF card contents are downloaded by wire link after the race for analysis.

A subsection of the logged data is also transmitted via serial link to a telemetry transmitter on the car. This transmits the data in real time back to the garage where it is displayed on a PC, enabling technicians to monitor a car’s performance from the pit.

Failure Analysis and Reporting
Embedded in the VCM are functions to detect failures of sensors on board the car. The output of these routines prevents failed sensors from being used by the control algorithms, and allows problems to be identified quickly.

The failure of a small sensor cannot be allowed to ruin a driver’s race. To prevent this, the VCM monitors inputs from multiple sensors and acts on the majority decision in each case.

The level of redundancy is determined according to the importance of the function, the vulnerability of the sensor, and the potential weight penalty. The VCM is capable of monitoring as many as 100 input sensors, while controlling numerous hydraulic systems.

This failure detection capability is augmented by pit-to-car telemetry. The race engineer can arbitrate against a sensor or set of sensors believed to be providing inaccurate signals.

Extra Processing Horsepower
BMW’s engine designers would say that a racing car can never have too much horsepower. BMW WilliamsF1’s electronic designers have a similar opinion about processing capability: More MIPS (million instructions per second) allow them to consolidate functions within the VCM, control more aspects of the car, and use more sophisticated models to generate accurate control signals.

Despite the rule-makers’ attempts to simplify and reduce the cost of F1 technology, the processing demands on the VCM continue on a steeply upward trend.

Paradoxically, rule changes for 2004 prohibiting pit-to-car telemetry, which are intended to reduce the complexity of electronics, will drive the processing load higher in future years. The decision-making performed by human operators in the garage must now migrate to the car itself.

Dave Walker and his team are faced with an ever-increasing demand for processing power. But they do not enjoy a corresponding increase in thermal budget or physical size limits. The growing list of duties for the VCM, originally used only as a pure data logger, has increased processing requirements.

Hardware Solution
With 220 channels to monitor, as well as numerous control algorithms to execute, Dave Walker says processor cycles are far too precious to waste on basic input functions, such as filtering or controlling A-to-D conversions.

Among the potential solutions to reducing processor load, the BMW WilliamsF1 team did not favor a microcontroller because of the need to write software and maintain the code base. In addition, the software inspection demands imposed by the FIA make the use of a microcontroller even less desirable.

A pure hardware solution is more optimal, according to Walker. Before the advent of modern multi-MFLOP (mega floating-point operations per second) DSPs, it was only possible to meet the timing demands for data logging and controlling high-speed hydraulic actuators by using hardware to accelerate certain functions.

The team has used field programmable gate arrays from a variety of vendors since the first-generation design, but BMW WilliamsF1 has now been working exclusively with Xilinx for a number of years. The current Stage 5 VCM uses the Xilinx Virtex™-E XCV600-E FPGA alongside a Texas Instruments™ DSP.

Although processing capabilities of DSPs have advanced dramatically since the first versions of VCM entered service, the team finds the hardware processing capabilities of the companion FPGA a valuable resource, which allows it to focus on using the DSP for pure number crunching.

There will never be a surfeit of processing power, as this can always be consumed by the increasingly sophisticated models provided to the VCM. These models demand more accurate control outputs for gearbox, differential management, and communications with the engine. Today’s XCV600-E is around 85% utilized. The team plans to migrate to a more complex, next-generation Virtex-II Pro™ FPGA in the VCM incarnation for 2004.

The team also plans to use Xilinx Titanium and Platinum technical support packages, which will include on-site design and training by expert Xilinx engineers.

Accordingly, the sixth-generation BMW WilliamsF1 VCM will perform even more processing, with a more powerful DSP chip alongside a more complex Xilinx FPGA. The Virtex-II Pro™ platform FPGA will also be able to offload additional DSP functions by virtue of its onboard multiply-and-accumulate (MAC) block to deliver a greater boost in effective system MIPS.

For 2004, the Virtex-II Pro device will host new programmable filtering and communication protocol control functions. The platform FPGAs will also manage UARTs and PWMs (pulse-width modulators) that were formerly implemented in separate components on the printed circuit board.

This ability to progressively soak extra discrete logic into the FPGA – at the same time as assimilating software routines – has enabled the VCM’s footprint to shrink significantly over its lifetime.

Conclusion
For those determined to win, Formula 1 has always demanded the utmost speed and nimbleness – not only on the racecourse, but also in terms of technical innovation. Rule changes introduced at the start of 2003 have taxed all teams’ abilities to respond quickly and effectively.

At the vehicle control level, engineers must to be able to add and remove functional blocks from the VCM within a short time-frame to meet FIA regulations. The teams must also be able to incorporate performance modifications as the current season progresses.

Fast feature swapping – combined with the ability to consolidate more of the low-level hardware and discrete logic into an FPGA to save weight, space, and power dissipation – is crucial to BMW WilliamsF1’s ability to demonstrate compliance with FIA rules. Moreover, Xilinx FPGA reprogramability enables the team to innovate quickly to maintain a competitive edge over rival teams.

To learn more about the Xilinx partnership with the BMW WilliamsF1 team, go to www.bmw.williamsf1.com and click on “Xilinx Joins the Team” under Partner News. To get the latest Formula 1 standings, go to www.formula1.com.

As of this writing (July 8), the BMW WilliamsF1 team had back-to-back wins, taking first and second place at both the European Grand Prix at Nurburgring and the French Grand Prix at Magny-Cours. Earlier this year, the team posted its first win of the season in the Monte-Carlo Grand Prix at Monaco.

According to a July 7 Formula 1 press release: “Williams has suddenly vaulted past both McLaren and Ferrari to establish itself as the team most in form ... If proof were needed of the growing Williams menace, in the last five races McLaren, initially the 2003 series leader, has scored 34 points; Ferrari has scored 55; and Williams has scored 70.”

In overall standings, BMW WilliamsF1 was trailing the series leader Ferrari by just three points.

Printable PDF version of this article with graphics.  (08/21/03) 315 KB