What is gyro bias stability?

The 30-second answer

A gyroscope measures rotation rate. When it is perfectly still it should read zero, and it does not. It reads some small non-zero value, and that value slowly wanders while the unit runs.

Gyro bias stability is how much it wanders, quoted in degrees per hour.

It matters because that error is integrated. A gyro reading a steady 1 °/hr when it is actually still will convince your system it has turned one degree after an hour. Heading error becomes position error in proportion to distance travelled, so bias stability decides how long the whole system stays useful.

Bias, and the three different things people call it

These get used interchangeably and mean different things. It is worth being precise when you ask.

Bias offset. The raw zero error, before any calibration. Largely removed at the factory, so rarely the interesting number.

In-run bias stability. How much the bias wanders during a single run, once it is warm. This is what “bias stability” normally means and it is the headline spec.

Turn-on to turn-on repeatability. How close the bias lands each time you power up. Different from in-run stability, and often worse. Poor repeatability means a long alignment routine before every mission, an operational cost that rarely appears on a datasheet.

How it is measured, and why that matters

The standard method is the Allan deviation. You log a long, still recording, then compute how much the averaged output varies at increasing averaging times, and plot it.

The curve falls at short averaging times, because averaging beats down noise. Then it flattens, and then it rises again as slow drift takes over. The bottom of that curve is the quoted bias stability.

Two things follow from this, and both are worth knowing:

It is a best case by construction. The minimum of a curve, measured on a bench, at constant temperature, on a unit that is not moving. Your vehicle is none of those things.

The averaging time at the minimum matters. If the minimum sits at 300 seconds, the figure describes behaviour over five-minute windows. If your outage is thirty seconds, a different part of the curve governs your result, and that part is usually dominated by noise rather than bias. See angular random walk below.

Bias stability is not the only number

Angular random walk, in degrees per root hour, is the gyro’s noise, and it sets short-term heading accuracy. Two units with identical bias stability and different ARW behave very differently in the first minutes of an outage, which is often exactly the window that matters.

As a rule of thumb: ARW dominates the first seconds and minutes, bias stability dominates from there on. Ask for both.

The grades

Three ways the number gets flattered

The Min column quoted as the spec. Many datasheets print Min, Typical and Max. Marketing quotes the Min. Ask which column the number came from. Our own manuals print 0.5 °/hr as the minimum and under 1.0 °/hr as typical at 25 °C, and both belong in the conversation.

Bench conditions left in a footnote. “Allan minimum, 25 °C, after a 30 minute warm-up” is not the same number your vehicle sees.

Silence about temperature. Bias moves with temperature. A unit quoted only at 25 °C is not telling you how it behaves across the range you will actually field it in. Ask for over-temperature data.

What OSCP publishes

Across the MK2 family the gyros are specified at 0.5 °/hr minimum in-run bias stability, typically under 1.0 °/hr at 25 °C, with angular random walk of 0.06 °/√hr on the MK2M2 and 0.03 °/√hr on the MK2E2 optical axis. Accelerometer bias stability is <15 µg across the family.

The MK2Z prototype reaches 0.005 °/hr on its optical Z axis. It is prototype hardware, quoted for evaluation rather than production.

Every unit ships with its own outgoing static test report, so you can see what your specific unit did rather than what the family does.

Frequently asked questions

Is lower always better?

For drift, yes. But a lower bias figure on a unit that is g-sensitive can still perform worse on a vibrating platform than a slightly higher figure on a unit with no moving parts in the sensing path. Bias stability is measured on a bench and your vehicle is not a bench.

What does bias stability actually cost me in metres?

Through heading, roughly: one degree of heading error puts you about 17 metres off track per kilometre travelled. Work out how long your gyro takes to accumulate a degree at its bias figure, and compare that with your outage.

Why do two IMUs with the same bias stability perform differently?

Usually angular random walk, g-sensitivity, or over-temperature behaviour. On our own road test, two units both publishing 0.5 °/hr minimum finished 5.7 m and 20.5 m from truth.