Detailed proposal for IR sensor modulation
Posted by dc42
|
Re: Detailed proposal for IR sensor modulation January 14, 2014 11:47AM |
After some discussion with RRP, I've implemented this. I used the scheme described in my original document attached to the first post in this thread, with a few modifications:
1. I used 3K3 as the load resistor instead of 4K3 because I found the phototransistor could still saturate when the sun was shining.
2. I changed the firmware to keep track of the last 8 readings and maintain the sums of the last 4 odd and even readings, so that instead of only having a new value every 5 polls, a new value is returned on each poll. Polling currently occurs at intervals of about 6ms. So the old method gave a reading averaged over 30ms which could be up to 30ms out of date, so on average it's the value 30ms ago. The new method gives a reading averaged over 48ms which may be up to 6ms out of date, so on average it's the value 27ms ago and with less jitter.
3. When modulation is being used, the G31 command returns the calculated value and the average high-value. I don't bother reporting the average low-value, it isn't needed and can be calculated from the other two anyway. The high-value is needed to make sure the phototransistor isn't going into saturation. So a response to G31 of "400 (450)" means the calculated average is 400, and the average of the high values only is 450 (therefore the reading when the IR emitter is off averages 50 in this case).
Here is a photo of the modified sensor board. I cut the track leading to the IR emitter, soldered a BC817 above the cut track with the base pin bent upwards, and added a 1K resistor from the base to a header pin epoxied to the top of the board. In fact the header pin needs to point up and right, I had to bend it to clear the z-nut trap.
To test it, I calibrated the z-homing at 1mm nozzle height, giving me a setting of G31 Z1.0 P475. I then did a series of G30 bed probe tests, and determined that the head usually overshoots a little, so that the final G31 reading is about 483. If I then turn on a 60W incandescent lamp about 30 cm from the bed, the peak reading goes up to about 730. The calculated reading also rises, but only to about 500. So the effect of the ambient light on the reading has decreased from about 247 to about 17. It's still a bit sensitive to ambient light, but the sensitivity has been reduced by a factor of 14.5. The practical upshot of this is that when I do a bed probe or z-axis home, the effect of having that light on is make the nozzle just over 0.1mm too high. Without modulation, having the same light on makes the nozzle 2.5mm too high.
I can see that the G31 reading is more variable with the light on than off. I think this is due to 50Hz modulation in the light. If I can find the time, I'll look at running the ADC on a timer interrupt, doing conversions for the z-probe and the thermistors in the background.
Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod
Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].
1. I used 3K3 as the load resistor instead of 4K3 because I found the phototransistor could still saturate when the sun was shining.
2. I changed the firmware to keep track of the last 8 readings and maintain the sums of the last 4 odd and even readings, so that instead of only having a new value every 5 polls, a new value is returned on each poll. Polling currently occurs at intervals of about 6ms. So the old method gave a reading averaged over 30ms which could be up to 30ms out of date, so on average it's the value 30ms ago. The new method gives a reading averaged over 48ms which may be up to 6ms out of date, so on average it's the value 27ms ago and with less jitter.
3. When modulation is being used, the G31 command returns the calculated value and the average high-value. I don't bother reporting the average low-value, it isn't needed and can be calculated from the other two anyway. The high-value is needed to make sure the phototransistor isn't going into saturation. So a response to G31 of "400 (450)" means the calculated average is 400, and the average of the high values only is 450 (therefore the reading when the IR emitter is off averages 50 in this case).
Here is a photo of the modified sensor board. I cut the track leading to the IR emitter, soldered a BC817 above the cut track with the base pin bent upwards, and added a 1K resistor from the base to a header pin epoxied to the top of the board. In fact the header pin needs to point up and right, I had to bend it to clear the z-nut trap.
To test it, I calibrated the z-homing at 1mm nozzle height, giving me a setting of G31 Z1.0 P475. I then did a series of G30 bed probe tests, and determined that the head usually overshoots a little, so that the final G31 reading is about 483. If I then turn on a 60W incandescent lamp about 30 cm from the bed, the peak reading goes up to about 730. The calculated reading also rises, but only to about 500. So the effect of the ambient light on the reading has decreased from about 247 to about 17. It's still a bit sensitive to ambient light, but the sensitivity has been reduced by a factor of 14.5. The practical upshot of this is that when I do a bed probe or z-axis home, the effect of having that light on is make the nozzle just over 0.1mm too high. Without modulation, having the same light on makes the nozzle 2.5mm too high.
I can see that the G31 reading is more variable with the light on than off. I think this is due to 50Hz modulation in the light. If I can find the time, I'll look at running the ADC on a timer interrupt, doing conversions for the z-probe and the thermistors in the background.
Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod
Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].
|
Re: Detailed proposal for IR sensor modulation January 18, 2014 07:02AM |
I did a bit more investigation. The expansion connector has only one ground pin and it is connected to totally the wrong place to use as the ground return for analog inputs. The expansion connector really needs a separate analog ground pin near the other end. However, the ground connections along the bottom of the board go fairly directly to the processor, so the ground here should be fairly clean and good for the ground side of an analog input, provided you do not run anything off any of these ground pins that injects noise. The only things running off these grounds are the IR sensor board and the fan. The stepper drivers use a different ground return path.
I've also found that if I have the 12V supply on and all the steppers and fan connected, but JP9 out (so 5V power is supplied from US
, then the noise on the analog reading goes away. So I now suspect that the problem is noise from the switching regulator getting on to the 3.3V rail. I recall that someone previously reported that he observed on a scope that noise on the 5V rail seemed to be passed through to and amplified on the 3.3V rail.
I'll attach a scope and carry on with the investigation.
Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod
Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].
I've also found that if I have the 12V supply on and all the steppers and fan connected, but JP9 out (so 5V power is supplied from US
, then the noise on the analog reading goes away. So I now suspect that the problem is noise from the switching regulator getting on to the 3.3V rail. I recall that someone previously reported that he observed on a scope that noise on the 5V rail seemed to be passed through to and amplified on the 3.3V rail.I'll attach a scope and carry on with the investigation.
Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod
Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].
|
Re: Detailed proposal for IR sensor modulation January 18, 2014 10:13AM |
@dc
What about ferrite beads on each of the IR Sensors wires individually, something like this: RS: 725-9032
It has a 1mm inner dimension and 3.6mm outer and is 4.85mm long and are very cheap.
We have used these in the past to reduce noise going down wires between devices like keyboards and wired devices etc basically any data lines that was getting noise etc.
In some cases we have had to fit these very close to both of the ends to stop the cable becoming an antenna to noise like wireless devices and mobile phones etc.
Paul
RS Ormerod No 436
What about ferrite beads on each of the IR Sensors wires individually, something like this: RS: 725-9032
It has a 1mm inner dimension and 3.6mm outer and is 4.85mm long and are very cheap.
We have used these in the past to reduce noise going down wires between devices like keyboards and wired devices etc basically any data lines that was getting noise etc.
In some cases we have had to fit these very close to both of the ends to stop the cable becoming an antenna to noise like wireless devices and mobile phones etc.
Paul
RS Ormerod No 436
|
Re: Detailed proposal for IR sensor modulation January 18, 2014 07:08PM |
Well I looked at the waveforms on the +5V and +3.3V rails with a scope. First thing I saw was a lot of ringing at about 200MHz, several hundred mV peak to peak. As my scope only has a bandwidth of 100MHz, I reasoned it must actually be somewhat greater than I could see. I was a little suspicious of the input and output capacitors on the LM1117. The datasheet recommends 10uF tantalum at the output, but the board has 10uF ceramic and 100uf aluminium. Ceramic capacitors of that value are made with dielectrics that lose most of their specified capacitance as their rated voltage is approached, and as the parts list on KiCad doesn't give the capacitor voltage, I wondered whether a 6.3V capacitor had been used. So I tried adding 10uF tantalum right at the regulator leads. No change. I also tried a 47uF input capacitor between the regulator input and ground leads, again with no effect.
Eventually I found that I could see the same 200MHz the ringing on the ground terminal of the regulator. I realised that the ringing was probably caused by resonance of the inductance of the loop formed by my scope probe and earth clip, together with the input capacitance of the probe. This loop was picking up the sharp changes in magnetic field as the switching regulator operates.
I then turned my attention to the ripple I could see at 1MHz. The amplitude was 10 to 30mV p-p depending on where I looked. I realised that once again the loop formed by my scope probe and earth lead was picking up the magnetic field generated by the switching regulator, so it was hard to say how much ripple there really was. Measured at the sensor cable 3.3V and ground connections with the sensor disconnected, the ripple appears to be about 18mV.
Eventually I realised that the source of the noise in the sensor readings was staring me in the face. The ground connection is at the bottom of the board, while the signal connection is right at the top. The loop formed by these cables and pcb tracks to the mcu will pick up the switching regulator magnetic field, just like my scope probe did. So the solution would be to eliminate this loop. To do this, I added a 3-way header strip at the bottom of the board, right below the existing pins used by the sensor.
I ran a wire from the expansion connector pin used by the sensor on the back of the board down to one of these pins, following first the pcb track from the expansion connector to the mcu, then the pcb track from the mcu ground plane area to the sensor ground pin at the bottom of the board. For good measure I routed the sensor modulation control signal similarly. I left the third header pin free.
With the sensor signal wire connected directly to the expansion connector, the reading I get from the sensor at 1mm nozzle height varies between 439 and 448. With the wire connected to the new header at the bottom of the board instead, it varies from 443 to 446. So this mod has reduced the noise by two thirds. This still isn't quote as good as I get when powering from USB only, but it is significantly better. I have some evidence that putting a 1mH inductor (a value I just happened to have lying around) in series with the 3.3V supply to the sensor reduces the noise a little more, but is isn't conclusive. I would like to try adding a 10nF capacitor between pins 74 and 78 of the mcu, but I don't have any capacitors with thin enough leads.
Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod
Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].
Eventually I found that I could see the same 200MHz the ringing on the ground terminal of the regulator. I realised that the ringing was probably caused by resonance of the inductance of the loop formed by my scope probe and earth clip, together with the input capacitance of the probe. This loop was picking up the sharp changes in magnetic field as the switching regulator operates.
I then turned my attention to the ripple I could see at 1MHz. The amplitude was 10 to 30mV p-p depending on where I looked. I realised that once again the loop formed by my scope probe and earth lead was picking up the magnetic field generated by the switching regulator, so it was hard to say how much ripple there really was. Measured at the sensor cable 3.3V and ground connections with the sensor disconnected, the ripple appears to be about 18mV.
Eventually I realised that the source of the noise in the sensor readings was staring me in the face. The ground connection is at the bottom of the board, while the signal connection is right at the top. The loop formed by these cables and pcb tracks to the mcu will pick up the switching regulator magnetic field, just like my scope probe did. So the solution would be to eliminate this loop. To do this, I added a 3-way header strip at the bottom of the board, right below the existing pins used by the sensor.
I ran a wire from the expansion connector pin used by the sensor on the back of the board down to one of these pins, following first the pcb track from the expansion connector to the mcu, then the pcb track from the mcu ground plane area to the sensor ground pin at the bottom of the board. For good measure I routed the sensor modulation control signal similarly. I left the third header pin free.
With the sensor signal wire connected directly to the expansion connector, the reading I get from the sensor at 1mm nozzle height varies between 439 and 448. With the wire connected to the new header at the bottom of the board instead, it varies from 443 to 446. So this mod has reduced the noise by two thirds. This still isn't quote as good as I get when powering from USB only, but it is significantly better. I have some evidence that putting a 1mH inductor (a value I just happened to have lying around) in series with the 3.3V supply to the sensor reduces the noise a little more, but is isn't conclusive. I would like to try adding a 10nF capacitor between pins 74 and 78 of the mcu, but I don't have any capacitors with thin enough leads.
Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod
Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].
|
Re: Detailed proposal for IR sensor modulation January 19, 2014 12:37AM |
It wonder if a digital solution would be best here. The MCU has numerous options for digital IO, but as a first go you could use PWM and convert the PWM output from the sensor to analog with an RC filter (EDIT: buffer wouldnt be necessary since the RC capitor should give low impedance and ideally buffer amp) in the lead at the Duet end Using (say) 100 kHz PWM, which after filtering with a time constant of (say) 1/50 th sec would still give a pretty smooth output. Alternatively a buffer amp at the sensor end with say 100 R output impedance,might be more robust and improve accuracy.
regards
Andy
Edited 1 time(s). Last edit at 01/19/2014 12:39AM by kwikius.
Ormerod #318
www.zoomworks.org - Free and Open Source Stuff
regards
Andy
Edited 1 time(s). Last edit at 01/19/2014 12:39AM by kwikius.
Ormerod #318
www.zoomworks.org - Free and Open Source Stuff
Sorry, only registered users may post in this forum.