Hi,

I am designing a Gaussian shaping amplifier based on Cremat-200 series shaping amplifier module. I always have a fixed 1-1.2 us duration voltage spike looking signal at the input and I want to transform that signal into roughly 10 us duration Gaussian shape signal at the output. I am using 2 stage Sallen-key filters for that (all resistors are 10k, all capacitors are 220 pF). Problem arises when after each stage I see a stronger pole in my signal (check attachments) that I have to somehow eliminate. I tried to make a PZ circuit just like they suggest in the Cremat-200 module datasheet, even tried to different pot values (10k and 100k) and capacitors (100 pF to 1 nF) but that made no difference. You can see all the timings and the amplitudes of the pole in the attachments. Can you please give me exact guidelines how do I make the PZ circuit? I did my research online, but all this PZ theory only looks good in terms of math. It is really difficult to adapt all that complex math to real life scenario RC network. Thank you in advance.

 

Hi,

I am designing a Gaussian shaping amplifier based on Cremat-200 series shaping amplifier module. I always have a fixed 1-1.2 us duration voltage spike looking signal at the input and I want to transform that signal into roughly 10 us duration Gaussian shape signal at the output. I am using 2 stage Sallen-key filters for that (all resistors are 10k, all capacitors are 220 pF). Problem arises when after each stage I see a stronger pole in my signal (check attachments) that I have to somehow eliminate. I tried to make a PZ circuit just like they suggest in the Cremat-200 module datasheet, even tried to different pot values (10k and 100k) and capacitors (100 pF to 1 nF) but that made no difference. You can see all the timings and the amplitudes of the pole in the attachments. Can you please give me exact guidelines how do I make the PZ circuit? I did my research online, but all this PZ theory only looks good in terms of math. It is really difficult to adapt all that complex math to real life scenario RC network. Thank you in advance.

Hi,

I did not look over all your data yet but you might do some math experiments to check out exactly how well pole zero cancellation works. The zero has to be very very close to the pole or it won't cancel enough. This is harder to get to work in practice, and the problem is if any of the components shift in value and do not track, there is no longer any 'cancellation'.
If done in pure math this isn't a problem, but you can investigate by changing some values a little manually just to see what would happen.

 

So you chose 10k for all the filter resistors and 220pF for the capacitors. That gives each second-order section a Chebyshev response with a Q of 1.0. That has an underdamped pulse response. Two in series will give an even more underdamped (I.e. oscillatory) response.

If you want a Gaussian pulse output, why are you using a Chebyshev filter, not a Gaussian?
You will find tables of coefficients for a Guassian response here.

 

TI has a FilterPro program that will calculate the component values for various active filters, including gaussian.
You might try that.

 

TI has a FilterPro program that will calculate the component values for various active filters, including gaussian.
You might try that.

I’d forgotten about Filter-Pro and it seems that so has TI.
Their online filter design tool is nowhere near as comprehensive, and doesn’t do Gaussian beyond 3rd order.

 

So you chose 10k for all the filter resistors and 220pF for the capacitors. That gives each second-order section a Chebyshev response with a Q of 1.0. That has an underdamped pulse response. Two in series will give an even more underdamped (I.e. oscillatory) response.

If you want a Gaussian pulse output, why are you using a Chebyshev filter, not a Gaussian?
You will find tables of coefficients for a Guassian response here.

Please provide a bit more details how did you calculate it (without any inductor value). What exact number should I be aiming for? What is the best topology for such circuit? Just RC components in series or high/low pass filter like? Maybe parallel RC network in series? Thank you in advance?

 

Please provide a bit more details how did you calculate it (without any inductor value). What exact number should I be aiming for? What is the best topology for such circuit? Just RC components in series or high/low pass filter like? Maybe parallel RC network in series? Thank you in advance?

I analysed your existing circuit using this.
You will find tables of coefficients for a Gaussian response here. It would seem that you need Qs of greater than 0.5, so an RC filter isn't going to manage. Only an active or an LC filter will do the job.
You can use Sallen & Key or MFB topology for your filter, provided that you make sure it has enough gain-bandwidth product and slew-rate to cope.
Microchip explains how much of each you need here.