Filter for saw modulated with triangular

I have saw signal that I measure time it needs to reach 200mV from 0V that rises linearly fastest within 1ms and slowest 4s. This signal is amplitude modulated with another triangular signal that ranges in period from 120µs to 80µs and has amplitude of 15mV p-p. Problem is how to filter out as much as possible of that triangular signal and not introduce time delay error in measurement of period of saw signal. I can have pauses between saw slopes as much is needed for filter to "come down" after input drops to 0V, but still don't know how to design filter that is sharp enough to let trough fastest saw and stop as much as possible of slowest triangular signal without introducing additional time error. I was playing with National Semiconductor's "Webench" but could not produce filter with any improvement to time measurement. Any suggestion is welcome, thank you!

Presumably you have a "trigger" signal that initiates the saw signal. Can you pass it through an identical filter, and measure the time between the trigger filter's output and the output of the saw filter?

deltamodulated saw signal ?

10kangstroms said:

Presumably you have a "trigger" signal that initiates the saw signal. Can you pass it through an identical filter, and measure the time between the trigger filter's output and the output of the saw filter?

Click to expand...

Yes, I do have trigger signal, but slope of saw signal is not known, so I can't pass it through identical filter since I don't have clean saw to compare it width. Delay for itself is not a problem, but delay variation between different slopes is. FIlter distorts slope to a curve that varies with the slope.
Without any filter time error window will be ~3%, and filter that is sharp enough to supress unwanted triangular wave adds error that is by far bigger than that 3%. Selecting filter with least ammount of phase variation from webench seems that does not help much. My knowledge is very thin in this area, so I am not sure if this is possible at all?

Added after 6 minutes:

artem said:

deltamodulated saw signal ?

Click to expand...

Saw rises from 0-200mV and drops to 0 with selectable pauses in between and triangular wave with 15mVpp rides on top of this saw signal(added). I am measuring time its needed for saw to reach this 200mV and to reduce error,I would like to suppress that triangular wave.

may it be help you to isolate triangle signal first by HPF (as its freq variation is less than signal you try to measure) and substract it from original signal and then work with that resulting?

artem said:

may it be help you to isolate triangle signal first by HPF (as its freq variation is less than signal you try to measure) and substract it from original signal and then work with that resulting?

Click to expand...

Interesting idea, I will try it out. Thanks!

I don't think you understood my idea. Attached is a block diagram. I simulated this with a 1ms saw and 120usec triangle (which is the toughest case for the filter), and the delay error was only about 3usec.
The delay should not be a function of the saw slope.
The bandpass idea is good, but my understanding was that you have no control of the triangle frequency, so how can you select the bandpass frequency? Also, you will still have harmonics remaining on your saw.

10kangstroms said:

I don't think you understood my idea. Attached is a block diagram. I simulated this with a 1ms saw and 120usec triangle (which is the toughest case for the filter), and the delay error was only about 3usec.
The delay should not be a function of the saw slope.
The bandpass idea is good, but my understanding was that you have no control of the triangle frequency, so how can you select the bandpass frequency? Also, you will still have harmonics remaining on your saw.

Click to expand...

I understood connection topology the first time, but in simulations I have done in webench it showed me that there would be delay variation to reach 200mV dependant on the slope of the saw signal. Delay for trigger pulse would be constant, so its easy to compensate that bias delay in processor even without that additional filter. I will try once again Bessel filter in webench to double check for this delay variation and if you are right, all I need is one filter. Also, bandpass filter was never mentioned, Artem mentioned isolating triangular signal with HPF and substracting it from main signal. Although I think this idea would have great deal of problems, it pointed my stupidity to me as I have clean triangular signal available. All I have to do is substract it from modulated signal. I still have to look into this option, if my clean signal is actually in phase with modulated signal, but it looks promissing. For now I have to finish design without any filter, but I made provisions on board for upgrade to process signal.

Thank you very much to you both for your help and effort! If you do remember some more ideas, please post it!

May be this idea can also help you :
if signal will be oversampled digitaly and decimation is done at period of triangular wave frequency - triangle should be eliminated after decimation in resulted digital form. It is just imho.

Unfortunately, I just measure time to reach this 200mV.

Interesting...