I design a 7-order linear phase low-pass filter (which is exactly the same anti-aliasing low-pass filter used in Agilent 33220A), which is:

Agilent 33220A LPF

This filter is used as an anti-aliasing filter to filter the current output of AD9744 DAC (digital to analog converter). Plus the passive filter itself, it also include some load resistors such as R1, C1, R3, R2 which are shown in the following picture. I simulate this circuit in LTSpice:


However, the magnitude and phase response is NOT flat, as can be seen in the following picture (V(ap) is the voltage at the ap node, V(bp) is at the bp node):

LTSpice simulation

Where the voltage at the ap node in NOT flat, which seems like to be fluctuated, and the response at bp is also NOT flat. So what is wrong with my circuit?

The followings are my simulations using NuHertz:

enter image description here

And the simulation in NuHertz using the values in Agilent 33220A:

enter image description here

  • 1
    \$\begingroup\$ So you didn't really design it - you just copied it and now you are relying on it having a characteristic that somehow suits your expectations of what you thought that the Agilent filter did. Do you have any evidence of what the Agilent filter does (other than your simulation data)? \$\endgroup\$
    – Andy aka
    Apr 10, 2017 at 13:15
  • \$\begingroup\$ Yes, I just copied the circuit from Agilent reference manual. I would it to have 12.5MHz filter passband, and ~46dB stopband attenuation. I simulated in nuHertz and it works as expected. \$\endgroup\$
    – liubenyuan
    Apr 10, 2017 at 13:24
  • 1
    \$\begingroup\$ Huh, your component values and layout in NuHertz are completely different from your LTspice simulation. \$\endgroup\$
    – pipe
    Apr 10, 2017 at 13:25
  • \$\begingroup\$ @pipe I want to design the filter parameters as closed to as Agilent, these values are the closest ones so far. I add the simulation result in NuHertz using the same values as in Agilent 33220A. \$\endgroup\$
    – liubenyuan
    Apr 10, 2017 at 13:30

1 Answer 1


The Agilent design uses inductor and capacitor values that you can actually buy. The NuHertz design (at least the first one) uses values you'd have to have custom wound and carefully trimmed.

Of course component values also vary from part to part and when the operating temperature changes. The Agilent designer likely determined that given the requirements of the application, the off-the-shelf values are close enough to give adequate performance. For your design, you should simulate the response with different values for the components depending on the precision of parts you're going to use, and verify that performance is adequate over the expected range of variation of all the components.

  • \$\begingroup\$ I know that Agilent choose the components that are available on market. What puzzled me is that, I want to filter the output of a current source (from DAC), and the voltage swing at the (ap) node is not flat. I do not know the reasons and wonder whether this design suits my needs. Thanks. \$\endgroup\$
    – liubenyuan
    Apr 11, 2017 at 0:14
  • \$\begingroup\$ @liubenyuan, try what I suggested about simulating with variations in the circuit values. You'll find it's not possible to make a 7th order passive filter that remains flat when there are component variations. So you must decide, how flat is flat enough, and design for that. Since this is a 20 MHz frequency generator application, you don't expect to see several frequencies present at the same time, so some ripple, especially above 20 MHz, should be tolerable. \$\endgroup\$
    – The Photon
    Apr 11, 2017 at 0:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.