Encoder Selection: Incremental vs Absolute vs Glass Scale for CNC Axes
Key Takeaway
Encoder selection for cnc axes comes down to what the feedback device has to see. An encoder on the motor reports rotor position and cannot see screw wear, bearing play or structural deflection. A glass scale on the table reports the real position and sees all of them. The cost difference between the two is small compared with the cost of not knowing where the table is.
Encoder Selection for CNC Axes
INCREMENTAL
Counts relative motion
Loses position on power loss
Needs a home switch
1000 to 5000 lines typical
ABSOLUTE
Holds position with no power
No homing cycle needed
17 to 21 bit typical
Single-turn or multi-turn
GLASS SCALE
Measures actual table motion
Sees screw and bearing error
0.1 to 1 micron resolution
Needs clean, dry mount
RESOLUTION
Arcsec = 3600 x 3600 / ppr
10000 ppr = 0.13 arcsec
Counts per rev x 4 = quad
Servo feedback at motor
Encoder resolution in arcseconds
10000 ppr gives 0.13 arcsec
Scale sees errors the motor cannot
Absolute encoder skips homing
Figure 1: The three feedback strategies for a CNC axis and what each one can actually measure
Table of Contents
1. What the Device Actually Measures
What the feedback device measures decides what your machine can actually hold, and almost every position error on a retrofitted CNC comes from a gap between what the controller is told and what the table is doing. This guide separates the three feedback strategies, gives the resolution arithmetic in arcseconds, and explains what each device can and cannot see. The sections end with a selection matrix and the four tests that verify the feedback on the real machine.
The first thing to get right about encoder selection for cnc axes is that a feedback device measures something specific, and different devices measure different things. This is the distinction that determines whether a machine can hold a tolerance under load.
The motor encoder
An encoder on the servo motor shaft reports the rotor angle. The controller multiplies it by the gear ratio and the screw lead to infer the table position. Everything between the rotor and the table is inferred, not measured: the coupling, the screw, the nut, the bearings, the screw supports, the way, and the structure of the machine itself.
This is acceptable when the mechanics are good. It is not acceptable when they are not, and the whole point of a retrofit is that the mechanics are often not good, because the machine is old.
The glass scale
A glass scale is a graduated strip mounted rigidly to the machine structure, with a read head mounted on the moving carriage. The read head measures the relative position of the carriage with respect to the machine structure, directly, with no drivetrain in the signal path.
Everything that happens between the screw and the table is therefore visible to the scale: screw wear, backlash, ball wear, nut wear, bearing play, gib looseness, and structural deflection under cutting load. The scale does not see the motor’s own error, so the best systems use both: the motor encoder for the servo loop, and the scale for the position loop.
The external drive feedback
Some servo drives close the position loop themselves and report the position to the controller over a serial bus. This is functionally equivalent to a motor encoder and carries the same limitation.
Section summary. The three devices in encoder selection for cnc measure three different things. A motor encoder reports the rotor angle and infers the table position through the coupling, the screw, the nut, the bearings and the structure. A glass scale measures the carriage against the machine structure directly. An external drive reporting position over a serial bus is functionally a motor encoder. The choice in encoder selection for cnc is really a choice about what you need to be able to see.
2. The Numbers: Resolution, Accuracy and Repeatability
Encoder specifications are quoted in a way that is easy to confuse. Four separate quantities matter, and they are not the same number.
Resolution
Resolution is the smallest position change the encoder can report. For a rotary encoder with P pulses per revolution and 4x quadrature decoding, the counts per revolution are 4P, and the resolution in arcseconds is:
Resolution (arcsec) = 3600 x 3600 / (counts per revolution) = 12,960,000 / (4P)
Worked examples:
| Encoder | Pulses per rev | Counts per rev (4x) | Resolution (arcsec) |
|---|---|---|---|
| Standard 500 ppr | 500 | 2000 | 6.48 |
| Common 1000 ppr | 1000 | 4000 | 3.24 |
| Precision 2500 ppr | 2500 | 10000 | 1.30 |
| High resolution 5000 ppr | 5000 | 20000 | 0.65 |
| High resolution 10000 ppr | 10000 | 40000 | 0.32 |
Accuracy versus resolution
Resolution is what the device can count. Accuracy is how close that count is to the truth. A 10000 ppr encoder with 0.32 arcsecond resolution might have an accuracy of plus or minus 60 arcseconds, which is nearly 200 times worse than its resolution. Resolution is the theoretical limit; accuracy is the real one, and it is the one that shows up in the finished part.
Repeatability
Repeatability is how tightly the encoder returns to the same reading from the same physical position on the same direction of approach. For a CNC, this matters more than absolute accuracy, because the controller can often correct a systematic offset but cannot correct a wandering one. A good servo encoder has repeatability of 1 to 3 arcseconds; a cheap one has 10 or worse.
Converting to machine position
Encoder counts become machine microns through the gear ratio and the screw lead. For a 10 mm lead screw driven directly (1:1):
Counts per mm = counts per revolution / lead in mm
With 10000 counts per revolution and a 10 mm lead: 10000 / 10 = 1000 counts per mm, so one count is 0.001 mm, or 1 micron. That is a useful sanity check: a retrofit with a 10000 count encoder on a 10 mm lead screw has a theoretical position quantum of 1 micron. Any measured accuracy much worse than that is a mechanical problem, not an encoder problem.
With a 5 mm lead, the same encoder gives 2000 counts per mm, or 0.5 micron per count. With a 20 mm lead, 500 counts per mm, or 2 microns per count. The lead therefore sets the resolution limit of the whole axis, which is why a coarse-lead screw on a precision axis is a mistake even with an excellent encoder.
Section summary. Resolution, accuracy and repeatability are three different numbers and confusing them is the most common error in encoder selection for cnc. Resolution is 12,960,000 divided by the counts per revolution, so a 10000 count encoder gives 0.32 arcseconds. Accuracy can be 200 times worse than that. Repeatability matters most on a CNC, because the controller can correct a systematic offset but cannot correct a wandering one.
3. Incremental Encoders
An incremental encoder outputs pulses as the shaft rotates. It knows the relative motion and it counts. It does not know where it is after a power cycle.
The consequences
- A homing cycle is mandatory. Every power-up requires the machine to find a reference. This is the most common operational complaint about open systems, and it is the strongest argument for absolute encoders.
- The position is lost on a power interruption. A power cut mid-cycle means the machine has to be re-homed before it can run again, and if the E-stop was active the axes may have moved.
- It is cheap. A 1000 ppr incremental encoder for a servo motor is around Rs 4,000 to Rs 8,000, versus Rs 18,000 to Rs 30,000 for an absolute equivalent.
When incremental is the right answer
For a retrofit where the machine will be attended and homed every shift, incremental encoders are entirely adequate, and they are the default for a reason. They are also more robust to contamination and cable damage, which matters in a shop environment.
The one situation where incremental is clearly wrong is a machine that runs unattended overnight. If the machine has to be able to power up and run without an operator, the absolute encoder is worth every rupee of its extra cost.
Section summary. Incremental encoders are cheap, robust and entirely adequate for an attended machine, and the reason they are the default is that they work. The cost of incremental feedback in encoder selection for cnc is a mandatory homing cycle at every power-up, and the position is lost on any power interruption. The one situation where incremental is clearly the wrong answer is a machine that has to run unattended overnight.
4. Absolute Encoders
An absolute encoder outputs a binary or serial code that identifies the shaft position within one revolution, without any reference to a previous position. It knows where it is the instant power is applied.
Single-turn versus multi-turn
- Single-turn absolute encoders report position within one revolution and cannot count revolutions. For a CNC axis, that is useless on its own for position, because the controller still has to know how many revolutions the shaft has turned. In practice a single-turn absolute encoder is used on a servo with a gearhead where the absolute gear is a separate multi-turn device.
- Multi-turn absolute encoders count revolutions as well, up to a limit of typically 32 or 64 turns. For a screw with a 10 mm lead, 32 turns is 320 mm of travel, which covers most axis travel in a small machine.
Resolution in bits
Absolute encoders are quoted in bits per revolution. 17 bits is 131,072 positions per revolution, giving a resolution of 360 / 131072 = 0.00275 degrees = 9.9 arcseconds. 21 bits is 2,097,152 positions, giving 0.000172 degrees = 0.62 arcseconds. 23 bits gives 0.155 arcseconds.
For comparison, the 10000 count incremental encoder from the table above gave 0.32 arcseconds of resolution. So a 21-bit absolute encoder is roughly half the resolution of a 10000 ppr incremental encoder, at three times the price. A 23-bit absolute encoder is about the same resolution as a 2500 ppr incremental.
Where absolute encoders are worth it
- Unattended operation. The machine powers up and runs without a homing cycle. This is the single strongest reason.
- Vertical axes. An absolute encoder tells the controller where the Z is after a power cut, so the machine does not drop the tool onto the workpiece.
- Tool changer and turret axes that must not be re-homed, because homing moves the tool.
- Removing a homing routine from the operator’s list. In an Indian shop where the operator may skip steps, a machine that cannot be left un-homed is a machine that will be run in an unknown state.
The protocol matters
On a Mesa-based LinuxCNC system, absolute encoders usually connect over the smart-serial protocol: a differential data link carrying data and power on a single small connector, up to 10 m. This is why the 7i76E and the 7i80 have dedicated serial encoder channels, and the controller needs the right encoder configured for that channel. Mixing an incremental encoder on a serial channel, or a serial encoder on a quadrature channel, produces no data and no error message, which is a confusing fault to diagnose. The board assignments are in the LinuxCNC hardware guide.
Section summary. Absolute encoders remove the homing cycle, and that is worth more than the resolution difference suggests. A 21-bit absolute encoder is roughly half the resolution of a 10000 count incremental encoder, at three times the price, so the argument for it is not precision but availability of position at power-up. It matters most on vertical axes, on turret and tool changer axes, and on any machine an operator can leave un-homed.
5. Glass Scale
A glass scale is the only option that measures the actual machine position rather than inferring it.
How it works
A graduated glass strip is bonded to the machine structure. A read head, carried on the moving carriage, senses the graduations and outputs a sine or cosine signal whose phase encodes the position. The controller interpolates the signal to a fraction of a graduation, which is how a 20 micron graduation gives sub-micron resolution.
Typical resolutions: 1 micron with 20 micron graduations, 0.1 micron with 5 micron graduations, and 0.01 micron with 1 micron graduations. The accuracy is typically plus or minus 1 to 3 microns on a good scale, plus or minus 0.1 degrees of mounting alignment error over the length of the scale.
The mounting rule that matters
A scale measures the distance between the read head and the scale strip. If the strip is not rigidly attached to the same structure that carries the carriage, the scale measures a mixture of the carriage position and the frame flex, which is worse than not having a scale at all. The strip must be bonded to the bed casting with an epoxy that does not creep, or bolted along its whole length, and the read head must be carried rigidly by the same carriage that carries the load.
The environment
- Cleanliness. A scale is the most contamination-sensitive feedback device on a machine. A chip of aluminium on the strip is a scale error. Use an air purge or a wipers cover, and keep the machine’s chip management working.
- Moisture and coolant. Most scales are rated for a limited coolant exposure. Mount them out of the direct spray path where the machine allows.
- Vibration. A scale is stiff, so it is not sensitive to the structural resonance that makes a motor encoder loop unstable. That is one of its less obvious benefits: adding a scale often lets you raise the servo gains.
How much accuracy does a scale buy
This is worth quantifying, because the answer is often smaller than expected and it keeps the decision honest.
Take a retrofit with a 10 mm lead ball screw. The screw’s own pitch error over 500 mm of travel is typically 0.01 to 0.02 mm for a good screw and 0.05 mm for a worn one. Add ball screw backlash of 0.01 to 0.05 mm if not compensated. Add structural deflection of 0.005 to 0.02 mm under a 1500 N cutting load on a bed with 0.02 mm of deflection per 1000 N. Now compare with a scale: a 1 micron resolution scale with 1 to 3 micron accuracy and 0.1 degrees of alignment error over 500 mm, which is 500 x tan(0.1 degrees) = 0.87 mm… that is the worst case of the alignment tolerance, so the alignment has to be controlled much more tightly than that in practice; a good installation holds the alignment error to about 20 arcseconds, which is 500 x 20/206265 = 0.048 mm.
So a well-installed scale on a machine with a worn screw typically improves the achievable positioning accuracy from about 0.05 to 0.08 mm down to about 0.01 to 0.02 mm. That is a real improvement, and on a precision job it is the difference between a marginal machine and a good one. On general job work with a 0.05 mm tolerance, it is not worth the money.
Section summary. A glass scale is the only device that reports the real machine position, and the honesty check is worth doing: on a machine with a worn screw, adding a scale typically improves the achievable positioning accuracy from about 0.05 to 0.08 mm down to about 0.01 to 0.02 mm. On general job work at 0.05 mm tolerance it is not worth the money, and encoder selection for cnc should say so rather than overselling the option.
6. Verifying the Feedback on the Real Machine
Whatever you install, verify it. The procedure is quick and it catches the majority of feedback faults.
The hand-push test
- Enable the servo on one axis and put the drive in position mode with the motor at zero speed, so the axis is holding position against the servo loop.
- Read the position display in the controller and note it.
- Push the carriage by hand, slowly, in one direction, by about 0.5 mm. The drive will resist and report a following error, and the position display should hold or return to the same value once you release it.
- Now, with the machine in a mode that reports the raw feedback count rather than the commanded position, push the carriage again. With a motor encoder, the count does not change, because the rotor has not turned: the backlash and the structural flex were absorbed by the coupling and the screw. With a scale, the count changes by the amount you pushed.
That difference is the scale’s entire value, and the test proves whether you have one.
The direction test
Move the axis in the positive direction, then back. Compare the position reading at the same physical point approached from each direction. The difference is the backlash plus the hysteresis of the feedback device. A motor encoder on a machine with 0.05 mm of backlash will show the full backlash, because the rotor has turned the same way both times. A scale will show only the hysteresis of the scale itself, typically under 0.002 mm.
The count ratio test
Command a move of exactly 100 mm. Check the position readout against the actual distance measured with a dial indicator or a test bar. If the counts per millimetre do not match the screw lead, the scaling is wrong, and the symptom is that the controller thinks the axis is somewhere it is not. This is a common fault after a retrofit where the encoder was set for a different lead.
The consistency test
Move the axis to the same point 20 times from the same direction and record the reading each time. The spread across the 20 readings is the short-term repeatability of the whole system, feedback and mechanics together. For a machine aiming at 10 micron accuracy, this spread should be under 5 microns. If it is 40 microns, the problem is mechanical or thermal, not the encoder.
Section summary. The four verification tests in encoder selection for cnc take twenty minutes and settle every argument about feedback. The hand-push test proves whether you have a scale or a motor encoder, the direction test separates backlash from hysteresis, the count ratio test catches a scaling error, and the consistency test separates a mechanical problem from an encoder problem. A machine that has not passed these has not had its feedback verified.
7. A Selection Matrix and the Decision
Putting it together:
| Situation | Motor encoder | Absolute motor encoder | Glass scale |
|---|---|---|---|
| General job shop, attended operation | Recommended | Optional | Not justified |
| Unattended or overnight running | Not recommended | Recommended | Recommended if precision work |
| Vertical axis on a machine with a brake | Acceptable | Recommended | Overkill |
| Precision work to 10 micron or better | Not sufficient | Not sufficient | Recommended |
| Retrofit of a machine with worn ball screws | Acceptable if backlash compensated | Same as incremental | Recommended |
| Budget under Rs 4 lakhs for the whole retrofit | Recommended | Stretch | Not applicable |
The practical recommendation for most Indian retrofits
Start with a good incremental encoder on the motor, 1000 ppr or better, from a brand whose cable and connector are replaceable. Add backlash compensation and get the mechanical side right. That combination delivers 90 percent of the achievable accuracy on general job work, and the money is better spent on the mechanical rebuild than on a 23-bit absolute encoder.
Then upgrade in this order, because the order is by value per rupee:
- Replace the couplings with rigid ones. A flexible coupling’s torsion is the largest single source of following error on a retrofit. This costs almost nothing.
- Fit a proper drag chain and shielded feedback cable. Cable movement and contamination cause intermittent encoder faults, and the intermittent fault is far more expensive than the cable.
- Move to absolute encoders if the machine runs unattended or has a vertical axis.
- Add a glass scale on the axis that determines the part’s critical dimension, if the tolerance is tight.
The verification procedure above is what turns the choice into an engineering decision rather than an opinion, and the errors it exposes are exactly the ones the troubleshooting guide then deals with. The parent retrofit guide covers where the feedback fits in the overall machine, and the cost calculator prices the difference.
Section summary. The matrix in encoder selection for cnc is the decision, and the practical recommendation for most Indian retrofits is a good incremental encoder on the motor, backlash compensation done properly, and the money spent on the mechanical rebuild rather than on a 23-bit absolute encoder. The upgrade order is by value per rupee: rigid couplings first, then a proper drag chain and shielded cable, then absolute encoders, and a scale only on the axis that sets the critical dimension.
Related Reading
- CNC Machine Retrofit: Complete Guide
- Servo Motor Selection for CNC Retrofit
- LinuxCNC Hardware Configuration
- CNC Retrofit Troubleshooting
- CNC Retrofit Cost Calculator
Sources
- ISO 10360, Test conditions for geometric accuracy of numerically controlled machines
- ISO 230-2, Accuracy of self-contained numerically controlled turning machines
- Mesa Electronics, 5i/7i Series Motion Hardware and THRS-21 Scale
- LinuxCNC Documentation, Hardware Compatibility
Key Takeaways
- Resolution in arcseconds is 3600 x 3600 divided by the counts per revolution, so 10000 counts per rev gives 0.13 arcsec.
- Quadrature decoding gives 4 counts per encoder pulse, so a 2500 ppr encoder gives 10000 counts per revolution.
- A motor encoder cannot see screw wear, bearing play or deflection; only a scale on the machine can.
- Absolute encoders hold position with no power and remove the homing cycle, which is why they are the first upgrade on a retrofit that will run unattended.
- Verify the feedback by deliberately moving the machine by hand under power and watching the position readout; a motor encoder will not see it and a scale will.