This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TPS3852-Q1: Minimum pulse width

Part Number: TPS3852-Q1
Other Parts Discussed in Thread: TPS3852, TPS3703, TPS3702, LM4041-N, LM2903-N

I have a question about the WDI minimum pulse duration spec.

Is it referring to 't' or 'd' in the picture below:

  • The 50nsec refers to what you tagged as "d", the width of that WDI pulse. What is the reason for asking this question?

  • Thanks,

    We have a clock with a small duty cycle so wanted to confirm if it will work with the part.

  • Is there a minimum value for that pulse width? Datasheet only has typical.

  • Will check if there is any data on this and report here, if found. What is the smallest pulse in the minimum duty cycle of the clock? what is the clock frequency? what is the allowable jitter you are checking for?

  • looking for a minimum pulse width of 50 ns to 100 ns. Clock frequency is 1 Hz. +/- 0.4 ms would work. The TPS3852-Q1 can't do that. Any other parts that are tighter?

  • If I understand correctly, the min pulse width can be as wide as 100ns and the WDT timeout should be 1sec. Does the +/-0.4ms mean you are looking for that as the "window"? That is, if the next pulse arrives earlier than 996ms or later than 1.004sec it is a fault?

  • Pulse width can be 100 ns max but I was able to invert in the circuit to give a high to low pulse width of close to 1 s. I assume this minimum WDI pulse duration is for glitch filtering? I would still like to know what the max value is though. Yes, the window should be 1 s +/- 0.4 ms. Obviously it can be looser than that but something tighter than the TPS3852 would be better.

  • The min max window are set up for what's commonly seen as variability in processor code - 25%, 50%, 75% variation. On clocks it's different. The tightness of each window is around 10-15% due to the current source variation (ICWD), charging voltage sensing accuracy (VCWD) and the external capacitance (on CWD) accuracy. I am not aware of any other WDT product choices at TI or other suppliers that are any better in the timing accuracy. You be able to convert the time to voltage domain and use our supervisors which have better than 1% sensing accuracy of the threshold voltage. Could something like that work?

  • If you can suggest some parts to do the clock to voltage conversion that would be great. Thanks for the suggestion.

    Please answer: " I assume this minimum WDI pulse duration is for glitch filtering? I would still like to know what the max value is though."

  • When I wrote that I was thinking of integrating with capacitors but I realize that has the same inaccuracy issues. I have been looking at alternatives for measuring such small 0.4% jitter. Most of what I am finding are clock and timing devices or instruments and not devices.

    Sorry to miss answering your other question - The reason for min pulse width is not intentional glitch fltering; that is done only for the voltage monitoring to avoid noise and spurs on the voltage line from being considered a voltage out of regulation. The min pulse width is due to the ability of internal detection circuit. If the pulse width can be as wide as 100ns, I sense this is not really the issue.

    As I had mentioned the window option on the TSP3852 are 4x, 2x or 1.25x (25%, 50%, 75%) ratio between lower window and upper window. As an example, in the snippet you had from the datasheet you will notice 800ms lower window typical with upper window at 1600ms (1.6sec), closest to the 1sec typical frequency of your clock. But the accuracy of both the upper and lower window yields clock period detected as the smallest period could be as small as 680ms and as the widest period as big as 1.84sec which is much bigger tolerance than you wish. Most of the similar watch dog timer devices on the market are similar in accuracy.

    So such devices are probably not an answer to your problem. Let me research some more and get back to you. I will need some more time for this. what is your design timeline and what alternatives have you come across so far? Thanks for your patience

  • We have some time to look at alternative solutions.

  • OK give me until a few days; I will come back to you - please don't post a reply till then. Thanks,

  • I have looked into this further. I can't find any product that by itself provides the solution.  I can explain my thinking and suggest a discrete approach which may work. The challenge is that 0.4% window (4msec around 1sec) is a very tight accuracy window. As I mentioned most of our watch dog supervisor products have timing accuracy around 30-50%. This will improve in future to around 10% but that is still quite far away from your need.

    First diagram below shows the overall concept. Let's stay the integrator using a current source and a COG (zero temp variation, accurate) capacitor is restarted on every pulse. The voltage rises linearly and we can sense the window where to look for a valid pulse. If the voltage is below (undervoltage of UV) and the pulse arrives OR the voltage is above (OV) [and the pulse did not yet arrive since the integrator did not yet get reset], we can declare it as an error:

    The second diagram is a zoom in view of the window time. The accuracy of detection depends on the accuracy of the current source, the capacitor and the detector of UV/OV. My first thinking was a discrete integrator (as I will describe below) but to use one of our supervisors as the UV/OV. However, our best voltage supervisor products that can detect the voltage threshold have accuracy around 0.7-0.9% (TPS3702 or TPS3703 for example).

    For the current source, the current references in TI's portfolio are 1-3% but you can check out this app note (the LM4041-N A grade has 0.1% initial accuracy):

    https://www.ti.com/lit/an/snoaa46/snoaa46.pdf

    Voltage reference: https://www.ti.com/product/LM4041-N

    That same voltage reference, LM4041-N, paired up with a dual comparator like the LM2903-N can product the needed UV/OV.

    The LM2903-N small offset voltage, quick response should be adequate but this product is not in our portfolio; I can link you to another person who could help.

    Let me know if you wish to head in this direction of a discrete implementation or have found another solution. In either case, I would be curious to know your application and how the problem gets solved. Wish you the best of luck and let us know how we can help you.