Understanding Noise Sources in 6-Axis Force-Torque Sensors

Why Noise Matters

In a 6-axis force-torque sensor, the noise floor determines the smallest force or torque change the system can detect. A sensor with a ±1,000 N range and 44 mN noise (RMS) at 100 Hz has a dynamic range of roughly 88 dB — meaning it can resolve signals four orders of magnitude smaller than its full scale. That resolution is only useful if you understand and manage the noise sources that compete with your signal.

Source 1: Thermal Noise in Strain Gauges

Every resistive element generates Johnson-Nyquist noise — random voltage fluctuations caused by the thermal agitation of charge carriers. In a Wheatstone bridge configuration with four 350 Ω strain gauges at room temperature, this baseline thermal noise is on the order of a few nanovolts per root-hertz. It sets the fundamental physics limit of the measurement.

In practice, thermal noise is rarely the dominant contributor in a well-designed sensor. But it becomes relevant in high-resolution, low-bandwidth applications — such as precision assembly or micro-force measurement — where other noise sources have been suppressed.

Source 2: Amplifier and ADC Noise

The analog front-end — instrumentation amplifiers and analog-to-digital converters — adds its own noise. Input-referred noise of the amplifier is typically specified in nV/√Hz and gets multiplied by the signal bandwidth. A 24-bit sigma-delta ADC operating at 500 Hz introduces quantization noise, though modern oversampling architectures push this well below the amplifier noise floor.

The key parameter here is the signal-to-noise ratio (SNR) of the entire analog chain. In AXIOM sensors, the electronics are integrated directly into the sensor body, which shortens the analog signal path and eliminates cable-induced noise pickup between the gauge and the digitizer — a common problem in sensors that rely on external electronics boxes.

Source 3: Electromagnetic Interference (EMI)

In industrial environments, electromagnetic interference from motor drives, power inverters, and switching power supplies couples into the sensor signal path. This appears as periodic noise at the switching frequency (typically 8–20 kHz) or as broadband noise from arc welding and plasma processes.

IP67 sealing and a metal sensor housing provide some shielding, but the most effective countermeasure is keeping the analog signal path short and fully differential. Single-cable integration — where the digital signal leaves the sensor body already converted — eliminates the most vulnerable segment of the signal chain entirely.

Source 4: Mechanical Crosstalk

In a 6-axis sensor, force applied along one axis ideally produces zero output on the other five. In reality, manufacturing tolerances, gauge placement accuracy, and elastic body geometry cause small amounts of crosstalk — typically 1–2% of full scale in well-designed sensors.

Crosstalk is not noise in the electronic sense, but it appears as an unwanted signal that varies with the loading condition. Factory calibration using a full 6×6 cross-coupling matrix compensates for this, but residual crosstalk after calibration contributes to the effective noise floor under multi-axis loading.

Source 5: Temperature Drift

Strain gauges have a temperature coefficient of resistance (TCR) and a temperature coefficient of gauge factor (TCGF). As the sensor body heats up — from ambient changes, motor heat conduction, or process heat — the bridge balance shifts. This manifests as a slow drift in the zero-load output.

Temperature compensation is handled at two levels: material selection (matched TCR gauges on the same substrate) and digital compensation in firmware. For applications with rapid thermal transients — such as sensors mounted near welding torches or autoclave environments — the compensation bandwidth matters as much as the compensation accuracy.

Practical Implications

When evaluating a force-torque sensor, noise is often reported as an RMS value at a specific output rate. A noise specification of 44 mN at 100 Hz means the standard deviation of the force reading under zero-load, static conditions, at that bandwidth. Higher output rates (250 Hz, 500 Hz) generally show higher noise because less averaging is applied per sample.

For real-world integration, consider:

• — Match the output rate to your control loop. Using 500 Hz when your controller runs at 100 Hz wastes bandwidth and increases noise without benefit.
• — Filter in software only after understanding the raw noise spectrum. Aggressive low-pass filtering hides noise but also hides real contact events.
• — Mount rigidly. Mechanical vibration transmitted through a compliant mount adds noise that no amount of digital filtering can remove cleanly.
• — Re-zero periodically in applications with thermal drift. A 30-second re-zero cycle eliminates accumulated offset without interrupting operation.