Encoder: Incremental or Absolute — How to Choose

Incremental or absolute encoder: the short answer

If the axis has to know its position after power is removed, without homing again, take an absolute encoder. If you only need speed or a relative count, and the machine passes a reference point on every start anyway, incremental is enough. Everything else in encoder selection is secondary to that one question.

Then comes the duller part, and it is where orders go wrong: interface, resolution, shaft mechanics. The most common reason a unit comes back is not «too few pulses» but a receiver that does not understand the signal. So our order of work is: first check what your drive or PLC accepts, only then pick the speed sensor.

One boundary up front, so you do not search twice. This page is about which encoder to buy. Closing the feedback loop in a drive is covered separately in how variable frequency drives work with an encoder. Every figure below comes from the datasheet of a part listed in our encoder section, not from a catalogue blurb. One more boundary: an encoder counts shaft rotation. When the measurement is a linear distance to a surface and there is no shaft in the task, it is a distance sensor that solves it, not an encoder.

What really separates the two types

Incremental does not know where it is

An incremental encoder outputs pulses; the receiver counts position. Cut the power and the count resets, so the machine has to find zero again. Take Omron E6B2-CWZ6C at 500 ppr: the E6B2-C series datasheet gives phases A, B and Z, a phase difference between A and B of 90°±45° (1/4 T ± 1/8 T), an NPN open-collector output, and a supply from 5 V −5% to 24 V +15% with ripple up to 5%.

Phase Z is one pulse per revolution, your zero. The receiver works out direction from the order of A and B, and nothing beyond those three lines exists on the output. Hence the arithmetic: for an incremental sensor to be useful at all, the machine needs a way to reach its reference. No limit switch and no permission to make a trial move means incremental does not fit, whatever its pulse count.

Absolute reports a code at every instant

Sick AFM60A-S4NB018x12 (part no. 1059039) reports position the moment it powers up. Its datasheet gives 262,144 steps per revolution, that is 18 bit, plus 4,096 revolutions counted, that is 12 bit. Together, a multiturn device of 18 bit × 12 bit. Error limits 0.03°, repeatability standard deviation 0.002°.

Do the arithmetic with us, because this is where most people draw the wrong conclusion. 360° divided by 262,144 steps is 0.00137° per step, roughly 5 arcseconds. The error limit in the same datasheet is 0.03°, about 22 steps. Resolution and accuracy are different properties, and 18 bit does not mean 18 bit of accuracy. Buying bits for positioning accuracy makes little sense: mechanics, mounting and the stated error limit bite long before the resolution does.

The second useful number is 4,096 revolutions. On a leadscrew with a 5 mm pitch that is 20.48 m of absolute travel with no extra hardware. For most gantries and hoists that is plenty, which means no battery-backed turn counter in the design.

Side by side

The interface matters more than the resolution

You do not choose the interface; the receiver does. Read the datasheet of your drive or module: it states plainly which encoders it evaluates. If your input only understands HTL, an absolute device with EnDat will not run, and no amount of bits will change that.

Something needs saying bluntly here, because we field this question weekly. The Siemens 6SL3055-0AA00-* parts sitting in our Siemens section are not encoders. Per their datasheets they are Sensor Modules for SINAMICS: they evaluate an external sensor's signal and pass it to the drive over DRIVE-CLiQ. They have no shaft and cannot count revolutions on their own. Do not buy one instead of an encoder; buying one alongside an encoder is often exactly right. So you do not have to read the table blind, the short bridge from signal to part number: a resolver is the 5AA3, sin/cos 1 Vpp or EnDat is the 5BA3, TTL or HTL is the 5CA2.

Two practical readings of that table. First, HTL in its bipolar form runs 300 m against 100 m for TTL, so on a long cable route the signal type decides more than the brand does. Second, in the IP67 modules the line to the measuring system itself is capped at three metres, thirty at the absolute limit — the module goes next to the motor, not in the cabinet, and that belongs in the layout while the cabinet is still on paper.

A word on servo, separately. There the feedback arrives with the motor: a resolver or absolute device is already on the shaft, and what you buy is the amplifier and its interface, not the encoder. If that is your side of the problem, look at servo motors and servo drives — selection there starts from the motor-and-drive pair, and this page will not help you much.

Not sure which signal your input accepts? Send us the drive model and the part number of its expansion card and we will check the datasheets ourselves. Our details are on the contact page.

How many pulses per revolution you actually need

Resolution is limited by the maximum output frequency, not by preference. The E6B2-C datasheet states the formula outright: maximum electrical response speed in rpm equals the maximum response frequency divided by the resolution, times 60. Let us put real numbers in.

1. 500 ppr, NPN output (E6B2-CWZ6C). Maximum response frequency 100 kHz. 100,000 / 500 × 60 = 12,000 rpm. But the series mechanical limit is 6,000 rpm, so the bearing is the bottleneck here, not the electronics.
2. 1,024 ppr, NPN output. 100,000 / 1,024 × 60 ≈ 5,859 rpm. Already below the mechanical 6,000, so the electronics has become the limit.
3. 2,000 ppr, NPN output. 100,000 / 2,000 × 60 = 3,000 rpm. Half of the rated mechanical speed traded away for resolution.
4. 1,000 ppr, PNP output (the E6B2-CWZ5B version). For PNP the datasheet gives 50 kHz, not 100. 50,000 / 1,000 × 60 = 3,000 rpm. Same housing, same resolution, half the speed ceiling — purely because of the output type.

That fourth item is the trap behind most «wrong unit» orders. Choosing between NPN and PNP looks like a wiring question, yet here it also halves your speed. We covered the input logic itself in switching NPN and PNP on a variable frequency drive, and the controller wiring in connecting sensors to a PLC.

The rule of thumb we use ourselves: for pump or fan speed control, 100–500 ppr covers the job with room to spare. For gantry positioning, go absolute and read the error limit rather than the bit count. For winding and cut-to-length, count pulses per millimetre of travel instead of per revolution — that number follows from the roller diameter, and it almost never matches the catalogue figure.

A worked example, figures illustrative. A roller of 100 mm diameter gives a circumference of 3.1416 × 100 = 314.16 mm per revolution. A 1,000 ppr encoder on that roller yields 1,000 / 314.16 ≈ 3.18 pulses per millimetre, so a count step of 0.31 mm. If you need a 0.1 mm step, that calls for roughly 3,142 ppr or a smaller roller. Put your own diameter in, then check the speed ceiling with the same datasheet formula: at 3,142 ppr and 100 kHz you get about 1,910 rpm, which may already fall short of your line speed.

Encoders and variable frequency drives: where the boundary sits

A drive does not always accept an encoder, and that is the first thing to check. Most closed-loop vector drives work with incremental TTL or HTL through an add-on card, and that incremental TTL/HTL card input does not evaluate an EnDat or SSI signal: an absolute device on either interface needs a separate evaluation module. In SINAMICS systems the modules from the table above fill that gap: 6SL3055-0AA00-5CA2 evaluates TTL/HTL and SSI, 6SL3055-0AA00-5BA3 handles sin/cos 1 Vpp and EnDat 2.1.

What closed loop buys you, and when a drive needs it at all, is written up separately in how variable frequency drives work with an encoder. If your drive is a Veichi AC300 or AC310, that input is added by the AC300-PG1 encoder card: an incremental ABZ input at 5 V for closed-loop vector control, with a 12 V version in the same series. For another drive series, send us the model and we will tell you whether such a card exists for it.

Seven selection mistakes we see most often

• Incremental where homing is not allowed. A lift, a rotary table with limited travel, an axis under load. After a power loss the machine has nowhere to move to find its reference. This calls for absolute, and it is not a budget question.
• Paying for bits. 18 bit against an error limit of 0.03° is about 22 steps of error. Resolution beyond accuracy helps control smoothness, not positional accuracy.
• NPN and PNP mixed up. The E6B2-CWZ6C output is NPN open-collector: sink current up to 35 mA at an applied voltage up to 30 V DC, residual voltage up to 0.4 V at that same 35 mA. The CWZ5B version is the reverse — PNP, source current up to 35 mA. A PLC input wired for the opposite logic will not count at all.
• Mounting straight onto the shaft, no coupling. The E6B2-C datasheet allows 30 N radial and 20 N thrust loading, with a starting torque up to 0.98 mN·m. The AFM60A has more margin — 80 N radial, 40 N axial, bearing lifetime 3 × 10⁹ revolutions — and still will not tolerate misalignment. The E69-C06B coupling shipped with the E6B2 exists for exactly this.
• Enclosure rating wrong for the environment. The E6B2-C is IP50 per IEC 60529. That is neither washdown nor outdoors. The AFM60A gives IP65 on the shaft side and IP67 on the housing side, and its datasheet adds the condition explicitly: with the mating connector fitted. Without it, there is no rating.
• Inrush current forgotten. On power-up the E6B2-C draws an inrush of roughly 9 A for about 0.3 ms. On a shared 24 V rail with several sensors that is a real cause of dips and false triggers on neighbouring channels. We worked through the same logic on inductive sensors in testing a proximity sensor with a multimeter.
• The low end of the temperature range. The E6B2-C starts at −10 °C, the AFM60A at −40 °C. One more line from the AFM60A datasheet belongs here: its 9,000 rpm operating speed comes with an instruction to allow 3.3 K of self-heating per 1,000 rpm when you work out the operating temperature range. For an unheated store or an outdoor installation that, not the resolution, is the parameter that decides.

What we have in encoders

Nine parts in the section: incremental Omron units for general work, an absolute Sick device with PROFINET for positioning, and Siemens evaluation modules for SINAMICS.

Two further parts sit in the section and we deliberately quote no figures for them, because we did not have the official datasheet for those exact part numbers in hand while preparing this page: Omron E6C2-CWZ5B and Sick DFS60B-S4AA05000, which Sick itself classifies under incremental encoders. We confirm their parameters before an order rather than lifting them from a catalogue description. Write to us: we will send the datasheet and confirm the resolution before payment.

Related reading, if the job turns out not to be about revolutions at all: inductive sensors for presence and position without rotation, and the general sensor catalogue. Brand views: Sick and Omron.

Frequently Asked Questions

Can an absolute encoder be replaced with an incremental one?

Only if the machine can safely reach a reference point after every power-up. An incremental device stores nothing: it counts pulses from the moment power is applied. Where homing is impossible or unsafe, the substitution does not work, and extra pulses change nothing about it.

How many pulses per revolution does a pump or fan need?

For speed control, 100–500 ppr is usually enough. Check the speed ceiling with the datasheet formula: maximum response frequency divided by resolution, times 60. For the E6B2-CWZ6C at 500 ppr that gives 12,000 rpm electrically, so the mechanical limit of 6,000 rpm is what constrains you.

What is the difference between singleturn and multiturn absolute encoders?

A singleturn device knows the angle within one revolution and starts over after a full circle. A multiturn device also counts the revolutions themselves. On the Sick AFM60A-S4NB018x12 that is 18 bit per revolution plus 12 bit of revolutions, which is 4,096 revolutions of absolute travel.

Does an encoder connect straight to a variable frequency drive?

It depends on the drive input. Closed-loop inputs on VFDs normally take incremental TTL or HTL, and then only through an add-on card. EnDat, SSI or a resolver needs an evaluation module — in SINAMICS, parts 6SL3055-0AA00-5BA3, 5CA2 and the rest of the table above.

NPN or PNP: which should I take?

Whichever your input understands. In the E6B2-C series, the CWZ6C version gives NPN open collector and a 100 kHz maximum response frequency, while the CWZ5B gives PNP and 50 kHz. So the output type affects not only wiring but the speed ceiling too, and flipping polarity on site with a resistor does not solve it.

Where to start

Answer one question: can the machine safely find zero after power-up. Yes — take incremental and work out the speed from the datasheet formula. No — take absolute, and read the error limit rather than the bit count. Then check what your drive accepts, and only at the end look at resolution.

All nine parts are listed in the encoder section. If you are unsure about your drive's interface, send us the drive model and the part number of its expansion card and we will check the datasheets ourselves. Our details are on the contact page.