What Your Data Logger Is Actually Measuring

A speed trace looks like a fact. It is a line on a screen, it has a value at every point, and it appears to describe precisely what the car did. In reality it is a series of separate measurements taken by a sensor with its own limits, recorded at a chosen rate, often smoothed, and then joined together by the software. None of that makes it unreliable. But the conclusions you can draw depend on understanding what happened to the signal before it reached you.

A trace is a set of samples

A logger does not record continuously. It takes a reading, waits, and takes another. The number of readings per second is the sampling rate, and it sets the finest detail the data can contain.

The arithmetic is worth doing once. At 250 km/h a car covers about 69 metres every second. Logged at 100 readings per second, that is one sample roughly every 0.7 metres. Logged at 10 per second, it is one sample every seven metres. A brake point read from a ten-per-second channel cannot be known to better than several metres, however precisely the cursor appears to sit on it.

Match the rate to the signal

Different channels change at different speeds, and each needs a rate to suit. A useful rule is that a signal must be sampled at least twice as fast as the quickest change you want to see, and in practice a good deal faster.

• Temperatures change slowly — a few readings per second is ample
• Driver inputs — steering, throttle, brake pressure — need tens of readings per second or more to show their shape
• Suspension movement happens very quickly and needs hundreds per second; sampled too slowly it is not just coarse but misleading

Logging everything at the highest rate is not the answer either: it fills memory, slows the analysis and adds noise to channels that did not need it. The right rate is the one that fits the question.

Two kinds of speed

Speed is usually available from two sources, and they disagree in informative ways. Wheel speed is measured at the wheel, responds instantly and is recorded at a high rate — but it reads low when a wheel locks under braking, high when it spins under power, and shifts slightly as the tyre's rolling radius changes with wear, pressure and speed.

GPS speed measures the car's movement over the ground, so it is unaffected by locking or wheelspin. It is typically updated less often, and can degrade under bridges, beside tall structures or in tree cover. Neither is simply correct. The difference between them is itself a useful channel: it shows wheel slip.

Distance, and why the delta depends on it

Comparing two laps means lining them up by distance, and distance is not measured directly — it is calculated by adding up speed over time. Any error in speed accumulates. Two laps of the same circuit will come out at slightly different lengths, partly because the driver took a different line and partly because of that accumulated error.

This matters because the time-delta between laps is computed from the alignment. If two laps drift apart by a few metres, the delta shows time appearing and disappearing where nothing happened. A sudden step in a delta trace is as likely to be an alignment problem as an event on track, and is worth checking before it is explained.

Before you interpret a trace, ask how it was measured.

Filtering has a cost

Raw sensor signals are noisy, and most are smoothed before display. Smoothing makes a trace easier to read, and it also changes it: peaks are lowered, sharp edges are rounded, and depending on the method the whole signal can be shifted slightly later.

A heavily filtered brake pressure channel will show a lower peak and a gentler release than the driver actually produced. When two channels are filtered differently, events that happened together can appear to happen in sequence. Where timing or peak values matter, look at the least filtered version available.

Zero and calibration

A sensor reports a voltage; the logger turns it into a number using a calibration. If that calibration is out, every value is out, consistently and invisibly.

• Steering should read zero with the wheels straight — an offset makes left and right corners look different
• Brake and other pressures should read zero with nothing applied
• Suspension position needs a known reference, set with the car in a defined condition
• Any sensor that has been replaced or disturbed needs checking again

A few minutes spent confirming zeros before the first run protects every conclusion drawn afterwards.

Calculated channels inherit everything

Many of the most useful channels are not measured at all but calculated from others: balance indicators, slip, brake bias, the delta itself. Each inherits the sampling, filtering and calibration of everything it is built from. A calculated channel built on a miscalibrated input looks exactly as convincing as one built on a good one.

A short checklist

• Know the sampling rate of each channel you are relying on
• Know which channels are filtered, and how heavily
• Check zeros before the first run of the day
• Check that lap distances agree before trusting a delta
• When a trace shows something surprising, rule out the measurement before explaining the car

None of this is a reason to trust data less. It is the reason data can be trusted at all: a measurement whose limits are understood is worth far more than one that is simply believed.