TPS92643-Q1: overshoot

Hello Hanno,

The first thing I would look at is the inductor current.  That will tell you if the inductor current is being controlled correctly.  The output current will not be the same as the inductor current if there are parasitics affecting the circuit.  If you have two current probes that would be good to see.  I don't quite understand the second waveform.  Can you explain what you are doing during the second waveform?  I would also like to see a zoom of the waveforms the initial startup.  Can you explain your setup too.  Do you have long cables between the output and the load?  Are the waveforms using the lower  Vf, IR LEDs (old) or is this with newer IR LEDs with Vf of 2.8-3V?  What is your max current setpoint?  Is it 650mA or do you plan to go higher?  What are you using for the inductor?  Did you go through the calculations for min sensed ripple or correct value for 800kHz switching frequency?  I believe the inductor value on the EVM is 22uH.  Please confirm.

You will have to change the bootstrap capacitor (C7) to something smaller if you plan to use this at low PWM frequency of 37 Hz.  I recommend  0.033 uF.   I am concerned you are hitting LS ILIM during startup.  Having the inductor current and SW node current would help determine this, please max the waveforms zoomed in so I can see what is going on.  

-fhoude

Hello fhoude,

today I am in home office and can not perform any measurement. We do have multiple current probes and I will provide some detailed measurement with inductor current tommorow.

The second waveform shows the startup behaviour of the output current (yellow trace) and the switchnode (blue trace). It is showing the startup behaviour of the device.

The setup is using approximatly 50cm cables to the LED, comparable to our costumers spec. An external analog voltage should be used to change the current for the LED (which is working fantastic). Another point is a quite low frequency PWM (37Hz) which is provided by a frequencygenerator.

650mA is the maximum current setpoint. There are no plans to use higher output currents.

The waveforms for the old IR-LED(Vf=1,6V) and the new one (Vf=3V) are almost identical. The behaviour concerning the overshot are neglegiable. I will provide measurement screenshot tomorrow.

I will change the bootstrap capacitor as suggested and repeat the measurements. I also doublecheck the calculation of the minimum on-time and inductor value and provide detailed answers to this topic as well

Best regards

Alexander Geißler

Hello Alexander,

You need to put a current loop in series with the inductor to monitor the inductor current (usually I lift the Vout side of the inductor and NOT the SW side of the inductor to put in the current loop).  Then also have a current probe on the output current to the load.  I would include the SW node voltage and output voltage on when capturing the waveform.  Let me know what you get.  

-fhoude

Hi fhoude,

I double checked my caclulations for minimum ON-time with both LED types. With a LED forward voltage V_f=3V and a V_in=24V the device meets the required minimum on-time of 111ns. The minimum on-time is violated when using the LED with the lower  forward voltage of 1,6V @ f_sw=800kHz. The switching frequency needs to be adjusted(lowered) to meet the requirements. Never the less the device is still working with the lower forward voltage and I can not see any varying of the pulse width at the switch node voltage which would show instability.

I confirm that the 22µH inductor is populated on the PCB. The minimum sensed ripple should be 24mV. A minimum of 8mV is needed. So this should be OK. The calculated and measured inductor ripple match(~150mA). The minimum t_ON with a V_f = 2.8V is 145ns @ a f_sw=800kHz. This is higher then the minimum ON time of 111nS in the datasheet and should be fine aswell.

All screenshots are done with the LED with the higher V_f

The first picture shows a screenshot as suggested. It shows the startup of the device. For all measurements the probe setup is as follows:

• CH1: Switch Node / yellow trace
• CH2: Inductor current / blue trace
• CH3: Output current / violet trace
• CH4: Output voltage / green trace

Startup behaviour / no PWM

A zoom of the start and end of the ramping is shown in the next two pictures:

Beginning of ramping

End of startup

PWM with 37 Hz/~550mA

The following picture show the TPS92643-EVM with.

• IADJ = 1,2V
• PWM with 37Hz and DC of 10%
• R1 = 0,15Ω

The overshot is short and settles after 16µs. For the design implmenetation though it is an unintended behaviour because the current is continously monitored and when reaching a certain level the device is deactivated.

The measured overshot is captured in the following screenshot and is roughly 140mA

Multiple PWM cycles can be seen in the following screenshot:

PWM with 37Hz / ~150mA

The following picture show the TPS92643-EVM with.

• IADJ = 0,3V
• PWM with 37Hz and DC of 10%
• R1 = 0,15Ω

With reduced output current the overshot behaviour is getting better. To get a more detailed view I changed the vertical resolution in the next pictures for the current probes to 50mA.

The overshot reduces to ~20mA.

Let me know if you need more measurements/information

Best regards

Alexander Geißler

Hello Alexander,

I think you can see that the inductor current is pretty well controlled and doesn't overshoot much.  That means the control loop is doing it's job and reacting quickly with minimal overshoot.  The rest of the output current overshoot is due to the parasitics on the output like the harness inductance and the output capacitance.  What does your test setup look like and can you give me more details about your requirements for the overshoot in current?  

Hello fhoude,

I try to sum things up and give you a brief overview of the spec.

The constant current source in conjunction with the LED is used for medical treatment. For a security reason we must provide a overcurrent protection. In case of a malfunction/overcurrent the LED shall be disabled. Therefore the current must be monitored continously. At bootup a selftest is performed to confirm that an overcurrent can be detected.

The two main parameters are:

1. current set point (the current the LED is working with)
2. safety current ( if this threshold is exceeded an overcurrent is detected and the LED shall be deactivated)

As an example:

• The current set point is defined to be 200mA
• The safety current shall be set 10% higher than the current set point. In this example 220mA

Between the current source (TPS92643) and and the LED there is a switch to break the connection in case of an failure. The goal has been to use the switch just in case of an overcurrent to deactivate the LED. The TPS92463 is providing functionalities like a PWM as current source.

The main issue is that with the current overshoot the overcurrent detection will be triggered. For sure there are multiple options to overcome this issue (make the overcurrent detection slower e.g.) but even though in the measurements above the overshoot settles in about 10µs I am not able to tell if the higher emmited intensity could possibly harm the patient. I'd like to go for the save option from the hardware point of view: Minimize overshoot, in best case no overshoot.

Due to the overshoots and no possibility to reduce the overshoot it seems to me that this chosen path is not the right one to go. (Setup as described above).

Another possible setup I have not tested so far would be to use an external load switch (discrete or a TPS22919 for example) and control the current ramp with the help of the switch. Do you have any ideas or suggestions for a possible solutions. I am open to any suggestion.

Tomorrow I am back in the lab and I'll provide a picture of the setup.

Thanks for your support so far

Best regards

Alex

Hello Alex,

10% threshold for safety current is really tight!  The part has a tolerance of +/-4% in a dc state.  How do you currently implement this protection?  It seems to blank out that overcurrent for a set time or filter it the detection threshold such that the fast transients are ignored.  Given the PWM dimming frequency is so slow, maybe you could set the IADJ pin to a lower setpoint then increase it a bit slower so that we have a small step changes towards the end where you get to final setpoint.  How is IADJ created?  DAC?  DC rectified PWM signal?  

-Francis Houde

Hello Francis,

today I achieved some improvements which seem to lead in the right direction.

But first some answers to some questions you previously asked:

The following picture shows the lab setup I use for measurements. I use a frequency generator to produce the PWM signal. The frequency generator is also capable of generating a dc-voltage for the IADJ pin. The IR-LED is in a box.

A blockdiagram shows the concept of the first protoype. The PCB also does have linear regulated current sources/sinks which basically work by the same principle (works pretty well). During the bring up phase of the prototype we ran into trouble with the smps current source (by a competitor) because we had to deal with overshoots up to 1A.

We use a DAC (DAC63004) to generate the control voltage for IADJ and the current threshold voltage.Highside current sensing is used for a contious current monitoring. The PWM is generated by a controller. During an overcurrent event the safety logic opens the power switch. As a possible solution we thought to blank out the overshoots but in the current prototyp the safety logic is deactivated.

A good hint is the +/- 4% tolerance in dc state.

• Do you know out of the box a device with a better dc performance?
• What would be a current threshold you would recommend ?

I usually do some tests to make things worse or better and started by adding more capacitance to the CSP Pin (Added 1µF). The overshoot got way worse, as I would have expacted. So I added more capacitance to CSN (4,7µF). At first I would have expacted to get the same result as on CSP. To my surprise the opposite happend.

The overshoot was reduced by 100mA @ 600mA output current. R1, C1 and the added 4,7µF seem to form a delay network ? The results seem promising.

At 150mA output current the overshoot could be reduced by ~100%. From 21mA to 10mA.

So far I have not tested other values. This is how far I could get for today and now TGIF (thank god it's friday)

Have a nice weekend and thanks for all your input

Best regards

Alex

Hello Alex,

Thanks for putting the effort in answering my questions and collecting data.  

• A tolerance of +/-4% at full scale (max reference voltage) is actually a really good tolerance.  Keep in mind the lower the current setpoint the worst the tolerance is.  This is due to the fact at light loads we have smaller reference voltage and the offset voltage error (which doesn't change with operation), line most any amplifier would have, is a larger percentage of the setpoint.  The lower the VIADJ the larger the percentage error due to the offset voltage.  Here is diagram of what I am talking about.

This also the reason we tell people to select their current sense resistor such that is close to the full scale voltage (max current), that makes sure the current accuracy is best at the highest current.   

• Increasing the output capacitance  (CSN pin) is taking some of that current that would be seen as overshoot and putting it into charging the capacitance.   You can play try and manipulate both Cout (output capacitance) and Ccomp (compensation capacitance) to optimize the transient response (bandwidth of control loop) and the filtering of the load.  
• The Ccomp controls the soft start on the initial pulse during PWM dimming.  The next pulse is regulated based on where COMP was while it was regulating previously.  This can be thought of as a sample and hold method method which makes for fast transient response, but in this case having to fast a transient response might cause the overshoot.  Increasing Ccomp reduces the bandwidth and slows down the response.  Thus the reason to see changes here might help with overshoot.    
• I did have a questions of where the current sense element is going to be in your system.  I also want to know if this eventually going to be all on one board?  All these long cables that have inductance and adverse affect performance.  It might affect where you put decoupling capacitors, for example do we want it at the output of the driver or at the load.  
• Probably a tolerance of 15% is viable (4% for load reg and 10% for transient), but this assumes no effects of parasitics due to cables. 
• If you have control of the VIADJ with a DAC you could snychronize the PWM high with a slightly lower setpoint then raise it after a short period such that the control loop doesn't have to slew so fast and it won't overshoot.  This quite a bit more complicated but could work if we can't tune the response to what you need it.

Looking forward to what you find next week.  I can do a virtual meeting if that works better for you too.  Let me know.

-fhoude

Glad to hear that good progress was made.  Is there anything else I can do to help?  If not, can you confirm that this post resolved your issues?  Thanks.

-fhoude