TMAG5170: Integrating off axis encoder with BLDC

Michael,

Thank you for reaching out on E2E with your question.  TMAG5170 can be used in an off-axis alignment to measure the position of the motor.  It is likely that the field components measured from your spinning magnet with have unequal amplitude, so you will need to do some amplitude correction to help linearize your angle result.

I would recommend trying to achieve as high of a SNR as possible on your measurement.  This can be done by reducing air gap between the sensor and magnet, using a strong magnetic material, and by enabling averaging within the device.  Averaging will effectively reduce the observed noise, but does add sample delay of which you should be aware.  At high speeds this delay will influence timing of your system control. 

Thanks,

Scott

Michael,

The amplitude correction is done using the GAIN_SELECTION and GAIN_VALUE register settings.  One axis may be selected to adjust gain by a scalar value ranging from 0 to 2.  If your manufacturing process is repeatable, you may be able to characterize the best typical value for your system, but optimal performance would require measuring each system as part of an end of line calibration. There is no delay associated with this correction to the final measurement.

The update rate is shown in this table

It is difficult for me to quantify what is too fast, however, since this largely depends on how much phase delay you can accept for your needs and the level of averaging selected.  The fastest sampling option would be to set the device to capture 2 axes of data with averaging set to 1x (13.3 ksps).  You should start here and consider how much the magnet will rotate at the speeds you are considering over the course of one sample phase.  

Thanks,

Scott

Michael,

TMAG5170 will produce an output which can be used to track the X,Y, and Z components of the magnetic flux density vector.  The output varies linearly as the field changes.  Using a configuration that produces sine/cosine inputs allows the device to calculate the absolute angle created by any two field vectors.

By comparison the TMAG5115 is a latch type device.  It can be used to track rotation of a magnet and will toggle the output between high and low states as opposing magnetic poles are presented to the sensor.  Using a standard dipole magnet, you would only be able to detect when transitions happen with 180 degrees separation.  As the pole count increases, you would have more resolution.  This might be helpful in tracking speed or detecting an exact stopping point. If you happen to be at the midpoint between switching locations, you would have trouble detecting which direction the shaft turns until the next transition point occurs.

We have an application note here where we calibrated actual angle measurements for TMAG5170 in a low speed setting:

https://www.ti.com/lit/an/sbaa463a/sbaa463a.pdf

and here is a document which describes various calibration methods:

https://www.ti.com/lit/an/sbaa539a/sbaa539a.pdf

I think 2400 RPM should be within a reasonable speed for TMAG5170. In this case each revolution requires about 0.025 seconds.  At 32x averaging it should be possible to configure the device for about 600 ksps, or about 0.00166 seconds per sample.  You would be able to capture about 15 measurements per revolution at this rate.  Since you do not require high precision control, you could probably set averaging at the default 1x and then have new data available every 75 us.  

Thanks,

Scott

Michael,

The plot you are referring to is in the second application note referenced above.  What this plot is trying to show is how much the magnet will rotate over the sample conversion period of the sensor.  So at 5000 rpm, the magnet turns at 30,000 degrees per second.  For 1x the conversion time is 75 us, so the magnet will rotate about 2.25 degrees during this time.  At 32x the conversion takes about 1.625 ms and the magnet will rotate 48.75 degrees.

You are correct that this indicates the lag of the measurement to the rotor position, and this is common for any system.  The best practice is to always evaluate this delay, and then apply a forward calculation based on the rotor speed to determine the actual position.

All of our 3D sensor have a similar sample rate, so you would be faced with the same limitation.  We do have a 2D AMR based sensor which is faster, but this device (TMAG6180) is meant for on-axis applications.  If you can place the sensor accurately, there should be a location where you can orient the sensor to capture two field components with equal magnitude.

Thanks,

Scott

Yes, if you are running high RPM, you will need to forward calculate and this should be a workable solution.

With TMAG5170 you have more flexibility on your angle measurement because we are able to capture all three field components.  An example of what I was referring to is shown here:

This simulation was done using this tool:

https://www.ti.com/tool/download/MAGNETIC-SENSE-ENHANCED-PROXIMITY/

and these settings:

In this case the Y and Z components are well matched at this spacing from the magnet.  The sensor is almost co-planar to the magnet bottom, but not exactly.  Using TMAG6180 in this position would require rotating the sensor to detect the Y and Z components. Placing TMAG5170 here would minimize the effort on correcting gain, and you would just configure the device to report angle using these axes.

Thanks,

Scott

Michael,

The tool does not account for stray fields.  Ring magnets a certainly possible though, and you are able to increase the pole count if desired.

I created another setup more closely matched to your magnet requirements.  In this case you might use the +/- 25 mT range on the A1 variant:

Thanks,

Scott

TMAG5170 can sense the field from either diametric or multi-pole magnets.  When using a multi-pole magnet, you lose the ability resolve the absolute angle of the mechanism.  That is, you now have n points which can map to your measured angle value, where n is the number of pole pairs.  Multi-pole ring magnets do tend to be more expensive, but you should be able to work with one if desired.

TMAG5110 or TMAG5111 are able to detect two components of the magnetic field, and you should wind up being able to produce quadrature data.  For a standard diametric magnet, there are 4 output states per revolution, and thus you can determine which quadrant you are in.  If you use a multi-pole ring magnet, then you will get 4 states per pair, so your resolution improves.  The multi-pole ring magnet tends to have a weaker field, so you do need to make sure you have enough input to always trigger BOP  and BRP.

DRV5055 can also work here, but two sensors placed 90 degrees apart are needed to measure angle.

You will have some high speed considerations with DRV5055 when you add in the time for your ADC to convert the output.  Essentially you will also experience lag from that conversion process and need to allow time for it.  

Thanks,

Scott

Michael,

You are correct here.  If you do not use the X input, it is not important whether or not the input saturates.  In the real device, you do have the ability to set the axes sensing range independently, but we have coded the tool to expect them to match.

Additionally, we have no way of knowing in advance which two axes you plan to use, so we plot angle results for all three combinations.  You can ignore the two which use the saturated X output.  If helpful, you can click the legend entries and hide the lines you are not interested in.

The minimum input should match the maximum input.  The maximum input allows you to adjust the sensitivity range (i.e. +/- 25 mT, +/- 50 mT, or +/- 100 mT)

I think your approach is good.  We would typically want to have the SNR as large as possible to reduce effects of noise on your angle calculation.  Then, the next step is to increase averaging if you are able to manage the extra latency.

You probably notice there is slight non-linearity to your angle result, which is caused by the amplitude mismatch between Y and Z.  With the actual device you could adjust the sensitivity gain of either Y or Z and this will produce the best result..

Thanks,

Scott

Michael,

You are right again.  For multi-pole you end up adding ambiguity of which pole pair you are observing, so you lose absolute angle of the mechanism.  However, for each pole pair you can observe 360 degree of angle change for each pole pair.  This may improve angular resolution somewhat, but you wind up also sacrificing the peak input amplitude you observe.  Placing the sensor closer to a multipole magnet tends to result with some distortion from an ideal sine and cosine input, so you will have more angle non-linearity as a result. You may be able to improve resolution somewhat, but these magnets are often more expensive.

Here are the main options when considering a diametric magnet:

• On-Axis w/ TMAG5170 (SPI) or TMAG5173 (I2C) provides absolute position over full 360 degree rotation using a diametric magnet.  Conversion time and communication add to latency of measurement. On axis is the least compact arrangement.  TMAG5273 can be used as well, as a lower cost / lower performance device.
• Off-Axis  w/ same sensors.  This will be more compact in solution size, but you will need to do more correction for amplitude and mechanical errors.
• On-Axis w/ TMAG6180 or TMAG6181.  These AMR based sensors are analog output with differential sine and cosine.  Automatic gain control will adjust itself to maintain a fixed output magnitude, so this device is best positioned on-axis.  This device is high bandwidth so your measurement will mostly be limited by your ADC performing the conversion.
• DRV5055 or TMAG5253 offer ratiometric sensitivity which adjusts itself for changes in Vcc.  This is meant to keep ADC conversion results constant as battery voltage droops.  For full angle measurements it will require two of these devices installed 90 degrees separated from each other about the magnet.  With analog output, your system latency will also be mostly determined by your ADC sampling.
• TMAG5110 or TMAG5111 these can be placed in most positions around a diametric magnet (90 degree resolution) or off-axis with a multi-pole ring (no field at the center).  These will have a fast response and do not require ADC sampling, but you have limited resolution with 4 positions per pole pair and only relative angle accuracy.  That is, you can only track angle changes based on your start point at power up.

Since in your case we need to exclude the on-axis alignment, I would probably still opt for TMAG5170 or TMAG5173 for the highest accuracy.  There should be locations where you can match two components in amplitude naturally, and this will help minimize the amount of calibration needed.  By maximizing input amplitude and setting averaging at the lowest setting you can sample at the highest rate to reduce latency.

Thanks,

Scott

The sensitivity ranges do vary on TMAG5173 vs TMAG5170.  You can choose from +/-40 or +/-80 mT.  

The A1 and A2 variants do have somewhat different tolerances.  I would recommend you compare the spec table for both.  They are too large to comfortable copy and paste here.

I would also try for a stronger magnet and higher sensing range to improve SNR.

We have had customers use the device with motorize applications, but I don't know that I have any specific anecdotal RPM values from their final product, and I haven't worked with anybody that reported problems based on the field from the motor altering their measurements.

Thanks,

Scott

Michael,

The field behavior can be a bit odd for the diametric ring depending on the geometry when the sensor is placed close to the magnet.  It is not uncommon to see some distortion to the magnetic field.  I do agree that it's probably best to place on the OD and to give yourself a bit of range to the magnet to allow for the field to appear more circular.

The A2 variant does have higher input ranges (+/- 75, +/- 150, and +/- 300 mT).  

In general, as large as possible applies so long as the input does not become distorted.  Once there is distortion on the input (say from putting the sensor too close to a ring magnet) then you will have an increase in angle non-linearity as well.

Thanks,

Scott

Michael,

I mean a non circular field, but saturating the input of the sensor will also produce a kind of distortion, and this would also have a negative impact on linearity of the measurement.

Comparing the +/-25 mT range to the +/-75 mT range, you should see a 3x reduction in the output code when comparing from A1 to A2 if all other variables are the same.  

Input referred RMS noise for XY axes at room temp using that A1 variant and no averaging is shown here

And the same for the A2 variant

While the max RMS value does increase when going from A1 to A2, if you are able to increase the signal by a more significant amount you should have a measurement which less susceptible to error from noise, but non-circular input or input saturation will not be as easily managed.  Noise can be reduced through over-sampling (averaging), and then accounting for the additional delay.  Angle non-linearity will require calibrated corrections in the MCU to resolve the angle error.

Thanks,

Scott

For either sensitivity option you should try to use as much of the FS range without saturating the input.  A2 has slightly higher RMS noise than A1, but you can provide a larger total input signal for A2 in order to reach a better SNR.  

Thanks,

Scott