KTH7812-X-N-QN16 advantages in collaborative robot joint modules

Why collaborative robot joints use magnetic encoders

Joint modules need angle feedback. An optical code disc will not fit into a joint only tens of millimeters across, and it suffers from dust and oil, so magnetic sensing is the mainstream route: a magnet fixed on the shaft, the chip on the PCB, reading across an air gap. The strengths of KTH7812-X-N-QN16 hit the pain points of a joint directly: 16-bit absolute angle means position is known at power-up with no homing; INL ±0.35° and noise 0.015°; 1μs data update is far more than enough for a 4kHz control loop; -40~125℃ covers the temperature rise inside a joint; and the QFN 3×3mm package fits inside the joint housing.

Magnet selection and temperature drift control

The magnet is the source. Pick it wrong and nothing downstream can be saved. The on-axis mounting condition in the datasheet is diameter 10~30mm, thickness 2.5~5mm, surface field 30~150mT. Two materials: NdFeB is cheap and strong but has a temperature coefficient of -0.120%/℃; samarium cobalt is expensive at -0.035%/℃, so its angle drift under the same temperature rise is one third of NdFeB. Joints running continuously commonly reach 70~80℃ at the housing. When the machine spec demands accuracy within ±0.5°, that difference eats a large part of the budget, so the material must be fixed at the selection stage.

Air gap and eccentricity tolerance margins

On-axis mounting allows an air gap of 1.0~5.0mm; the 1mm lower limit is tight, and beyond 5mm the signal-to-noise ratio drops. Where the structure is compact and the magnet cannot slip over the shaft end, use off-axis mounting: the chip can sit away from the shaft axis, but the field is no longer axisymmetric, so factory calibration is mandatory. Large hollow shafts use multipole magnetic rings, and the smaller the pole pitch, the more sensitive to eccentricity. Three mounting styles correspond to three structure types. Fix the structure type first, then talk accuracy figures — doing it in reverse order means tuning parameters for nothing.

KTH7812-X-N-QN16 key specifications

No homing, but calibration is still required

Being absolute, the chip knows its position at power-up, with no homing like an incremental encoder. But "no homing" does not mean "no calibration": magnet-to-chip relative position carries assembly error, and the zero offset plus the second-harmonic error from eccentricity must be measured out on the production line. The method is simple: rotate the joint a full turn on a fixture, sample the deviation between chip output and an external reference at equal divisions, fit the zero offset and first/second-order compensation coefficients, and write them into the drive. One station, two or three minutes — far cheaper than rework later. Bind the calibration data to the serial number: keep a readable copy of the zero offset, the compensation coefficients and the magnet batch recorded at assembly. Otherwise, swapping same-model joints two years later means recalibrating.

Field-strength alarms as a backstop for loose magnets

Magnets are fixed with adhesive or clips, and long-term vibration and temperature swings can shift or detach them. KTH7812-X-N-QN16 has built-in field-strength upper and lower limit alarms, three threshold levels each for mgh and mgl: for example, mgh=1 triggers above 34mT and clears when it falls back to 28mT; mgl=0 triggers below 18mT and clears when it rises back to 24mT. Wire the alarm bits into the drive fault word, and the system can shut down before angle distortion grows when a magnet ring comes loose — an order of magnitude cheaper than investigating after a customer reports "arm jitter". Choose thresholds by the actual air gap; for projects with a large air gap, move up one level.

ABZ and UVW interface configuration points

If the drive needs angle fusion or line-side calibration, read the 16-bit angle directly over the four-wire SPI, in parallel with ABZ/UVW — no extra parts, and the line-side calibration bench no longer needs an ABZ counter. ABZ line count is programmed via PPT(9:0), 1~1024 pulses/rev, with a maximum output frequency of 16MHz; ABZLIMIT can step it down to prevent edges from being missed at high speed with a large line count. UVW handles commutation, 1~8 pole pairs set to the motor's pole count — set it wrong and the motor jitters or even reverses. Write line count, pole pairs and direction into the line-side configuration table and do not let the field change them. Also be clear about single-turn versus multi-turn: KTH7812-X-N-QN16 outputs single-turn 16 bits; multi-turn position is accumulated in drive software and lost on power failure. Power-loss retention means either a battery-backed counter or homing at power-up via mechanical limit stops.

What if you use an optical encoder, resolver or Hall array instead

Optical encoders are bulky, fear dust and oil, and will not fit in a joint. Resolvers meet the accuracy but are expensive and complex to decode. Hall switch arrays lack resolution. The balance a magnetic encoder strikes between size, cost and accuracy is a major reason joint modules choose it — the price is that magnet, air gap and calibration must all be done properly together.

FAQ

Q: Is the accuracy still ±0.35° INL once installed in the machine?A: No. That is the device's own linearity error. After assembly there is still magnet placement error and eccentricity harmonics. Landing within ±0.5° at machine level counts as passing, and line calibration is what gets you there.

Q: NdFeB or samarium cobalt for the magnet?A: If accuracy must stay within ±0.5° and the joint reaches 70~80℃, choose samarium cobalt — its drift is one third of NdFeB. On a tight budget with looser accuracy, NdFeB works.

Q: How do you get multi-turn position from single-turn absolute angle?A: Accumulate turns in drive software, which loses count on power failure. Add a battery-backed counter, or home at every power-up using mechanical limit stops.

Q: What if ABZ lines share a duct with the motor's three-phase wiring?A: Differential routing, twisted pairs, ferrite rings, shield grounded at one end. Software filtering adds phase lag, so the control loop bandwidth has to come down accordingly.

Q: How to pick within the same series?A: KTH7813 has noise of 0.004° but a 16.3ms response — good for force control at low speed. KTH7815 responds in 0.03ms for high-speed spindles. KTH7816 is SSI+PWM. The order: fix the interface first, then bandwidth, then noise.

KTH7812-X-N-QN16 and related magnetic sensor parts ship from Shenzhen Manduic Technology Co., Ltd., an authorized distributor of CONNTEK — samples, volume supply, datasheets and selection support for angle sensor grades included.

For samples, product datasheets, detailed adaptation solutions or technical support, please contact us:

Tel: +86-150-1290-5940

Email: sales@manduic.com

Website: www.manduic.com