What Is an Encoder and Decoder in CNC Machine Control?

• 1
  What an encoder and decoder in CNC machine feedback actually do
• 2
  Incremental, absolute, and where each one belongs
• 3
  Resolution, accuracy, and repeatability are not the same thing
• 4
  How the feedback chain fails, and what it looks like on the part
• 5
  What this means for the parts you send out for machining
• 6
  Which feedback setup fits which axis

Say the controller commands an axis to move to 152.734 mm. The servo drives the ball screw, and the axis moves. Something has to confirm it landed there and not at 152.731 mm. That is the encoder's job. It is a transducer bolted to the motor shaft or to the slide itself, and it outputs a pulse train whose count is proportional to travel.

The decoder is the second half of the pair. Encoder output is not a number yet. It is a pair of square waves, usually called A and B, plus a once-per-revolution index mark called Z. The decoder counts edges, reads which channel leads to establish direction, multiplies the count by an interpolation factor, and hands the controller a clean position value and a velocity value.

That loop closes thousands of times per second. If the commanded position and the decoded position disagree by more than a set window, the drive faults out before the tool ever touches metal. This is why the feedback chain, not the ball screw alone, sets the practical floor on accuracy.

One consequence engineers should keep in mind: the decoder only knows what the encoder tells it. Backlash, thermal growth of the screw, and tilt of the column sit outside that loop unless a second scale is fitted on the axis itself.

• 1
  EncoderMeasures actual position and speed; outputs A, B and Z signals.
• 2
  DecoderCounts edges, resolves direction, produces position and velocity values.
• 3
  ControllerCompares command against feedback and corrects the drive.

Incremental encoders count from a reference point. They have no memory of absolute position. After a power loss the machine must return to a home or reference mark, using the Z pulse, to re-establish where it is. Any motion during the outage is simply lost information. On a three-axis mill this homing routine costs a few seconds and is rarely a problem.

Absolute encoders report a unique position value at every point in travel, so the machine knows where it is the moment power returns. No homing cycle, no lost position after an e-stop. The trade is cost and wiring complexity. On a five-axis machine with a tilting head, that trade usually pays for itself because re-homing a rotary axis mid-setup is slow and easy to get wrong.

Linear encoders sit on the slide rather than the motor. They see the table itself, so they capture screw pitch error, thermal growth, and backlash directly. Rotary encoders on the motor shaft see only motor rotation and trust the mechanical train in between. For work at ±0.005 mm, a linear scale on the critical axis is often the difference between holding tolerance all shift and chasing drift.

Magnetic and optical versions cover the same ground with different constraints. Optical scales give finer resolution but dislike oil mist and chips on the read head. Magnetic scales tolerate contamination better and are common on lathes and on machines cutting graphite or cast iron.

• 1
  IncrementalLow cost, needs homing after power loss.
• 2
  AbsolutePosition known at power-up, better for rotary and 5-axis.
• 3
  Linear scaleReads the slide, captures screw and thermal error.

A spec sheet that says 0.1 μm resolution does not mean the machine holds 0.1 μm. Resolution is the smallest step the decoder can report. Accuracy is how close the reported position is to the true position. Repeatability is how consistently the axis returns to the same point. Buyers mix these up constantly, and the mix-up leads to unrealistic purchase specs.

Consider a common setup: a rotary encoder with 2,500 lines per revolution, decoded with four-fold quadrature, on a 10 mm pitch ball screw. That gives roughly 1 μm of reported resolution per step. The machine may still only hold ±0.01 mm because of screw lead error, thermal drift, and servo tuning. Resolution sets the floor, not the result.

Interpolation inside the decoder multiplies the raw count. A 1 Vpp sine encoder with 1,024 periods, interpolated 4,096-fold, reports very fine numbers indeed. But interpolation amplifies signal noise along with signal. If the scale is dirty or the head alignment is off, a high interpolation factor makes the position reading jittery rather than more precise.

The practical question is not which encoder has the biggest number. It is whether the feedback chain is finer than the tolerance you need to hold, with margin left for everything downstream.

• 1
  ResolutionSmallest reported step from the decoder.
• 2
  AccuracyCloseness of reported position to true position.
• 3
  RepeatabilitySpread of returns to the same commanded point.

Contamination is the most common cause. Oil mist, fine chips, or coolant film on an optical read head scatters the light and drops pulses. The decoder sees fewer counts than the axis actually travelled, so the controller thinks it is short of target and keeps driving. The visible result is a part that measures oversize on one feature and on size on the next, with no consistent pattern.

Cable and connector problems show up as intermittent faults that appear only at certain axis positions. A partially broken signal wire inside a drag chain can pass continuity checks on a stationary machine and fail the moment the axis moves. If a drive alarms only in one region of travel, check the cable before replacing the encoder.

Electrical noise from a spindle drive or a welding cell nearby can corrupt the A and B channels. Shielding, separate routing, and proper grounding fix most of it. A differential line driver output instead of a single-ended one is the other standard fix.

Mechanical issues look different again. A loose coupling or a slipping belt between motor and screw means the encoder reports motor rotation that never reached the slide. The controller is satisfied, the part is not. This is exactly the case a linear scale would catch and a motor-mounted encoder cannot.

• 1
  Symptom: oversize featuresDirty optical scale dropping pulses.
• 2
  Symptom: alarm at one positionBroken wire inside the drag chain.
• 3
  Symptom: erratic size, no patternNoise on the signal channels.

When a shop quotes ±0.005 mm, it is quoting what its machines and its inspection can actually hold. That number comes from the whole chain: encoder, decoder, screw, structure, thermal control, and metrology. The feedback system is one link, and it is usually a well-understood one on modern machines.

For most machined parts, the tolerance driver is not the encoder. It is fixturing, tool wear, and heat. A block of 6061 aluminium that warms 5 °C during a long roughing pass moves more than the feedback resolution ever will. That is why in-process measurement and a cool-down before finishing matter more than the encoder spec on the datasheet.

Where feedback does become the limit is on long axes, on rotary axes, and on parts with tight true-position callouts across several features. On a 4,000 mm travel machine, screw thermal growth over a shift can exceed the tolerance band. A linear scale on that axis is not a luxury.

So the useful question when you review a supplier is not which encoder brand they run. It is whether they can explain how they hold the tolerance on your specific feature, and whether they inspect it before the part ships.

• 1
  Short axesFeedback rarely the limiting factor.
• 2
  Long axesThermal and lead error dominate; linear scales help.
• 3
  Rotary axesAbsolute feedback avoids re-homing errors.