Elliptic filter - implementation deviates from the design and simulation

Question

I have designed a 7th-order elliptic filter using this online tool.

Cut off frequency - 15 MHz
Pass band ripple - 0.1 dB
Stop band attenuation - 70 dB

I've asked it to use tge standard inductor value of 2.2 μH.

The designed filter is shown below:

The Kicad schematic:

The Kicad PCB:

The actual PCB:

This gives a stop band attenuation of around -65 dB which is good enough for my purpose.
I've also simulated this in LTspice which showed similar performance.

I implemented this filter on a homemade PCB. The main requirement of this filter is to cut a 50 MHz noise from an analog signal. In addition, I expected it to cut 2nd and 3rd harmonics of a signal which has a maximum fundamental frequency of around 10 MHz.

When the filter is tested however, the results I got were very different. The 50 MHz noise was still present considerably. So, I tested this filter using a function generator and oscilloscope. Simply, I connected the filter input to my AWG. Two channels of the oscilloscope were connected to the input and output of the filter. Then I changed the frequency from 1 MHz to 100 MHz gradually in the AWG. It showed good attenuation in 15 MHz range. However, after around 30 MHz the attenuation gradually reduced and stayed almost constant. I measured the peak value of the input and output sine from the filter and calculated the attenuation and it was only around -17 dB instead of -65 dB which is shown in the design.

I would like to understand possible causes for this discrepancy. A few that come to my mind are:

• Deviation of actual component values (bought from AliExpress), wrong approach in actual measurement/impact of oscilloscope probes/etc.

• Problems in PCB layout causes input to leak to output at high frequencies, issue in the design itself which is not evident in the simulation.

Answer 1

However, after around 30 MHz the attenuation gradually reduced and stayed almost constant

This measurement indicates that the inductors are likely beyond self-resonance. It results in a cascade of capacitive voltage dividers that have an attenuation that is independent of frequency.

Real passive elements have a tolerance that make targeting a specific frequency difficult. Use low tolerance parts <1% for good accuracy.

Use COG/NP0 ceramic capacitors. Other ceramics are very voltage dependent. Capacitors have a self resonant frequency above which they become inductive.

The inductors have a self-resonant frequency above which the become capacitive. They also have a series resistance that is usually significant.

When you buy components that have no data sheet, you must measure them yourself to characterize them within the bounds of your application. Even if there is a data sheet, measure them anyway to verify that they are consistent with the claims in the datasheet.

Rs is the source resistance not usually installed on the board. The AWG already has the 50 internally. Use a 50 ohm cable (coax) from the AWG to the board. R1 on the board should be 0 ohms.

RL is the load resistance not usually installed on the board. Use a 50 ohm cable (coax) from the board to the 50 ohm load. R2 on the board should be open. For testing, R2 of 50 ohms is fine. If you can set the scope to 50, that is better for loading.

However, a high value resistor (100k) can be used for R2 to ensure the capacitors are discharged when the board is not connected.

Attaching an oscilloscope to the circuit changes the circuit. A low capacitance probe still has 9 to 10pF. The ground clip lead can introduce 0.5nH.

Addition: Avoid using wire-wound resistors. They are inductive. Low inductance winding is available. Carbon film and metal film are better although they too can have significant inductance. I think metal foil is the best. Whatever, choose low inductance resistors and capacitors. end addition

Sorry for being long winded, but all these things got in my way when implementing and characterizing filters.

Answer 2

2 possibilities I can think of:

1. Please show the schematic of your test fixture too. Your filter was done assuming a 50 ohm source impedance and load impedance. Are you sure that's what you're placing when measuring your circuit?

EDIT:

I see you have added an schematic to your question now. Even though you have a placed a 50 ohm in series with your input, I'm not sure that's the impedance the your AWG will see. It should normally go to ground and the load should be high ohmic so that the level indicator in your AWG is accurate enough.

On top of that, you're adding a parallel RC at the input with the other probe to measure the input.

As a 1st step in your debugging process, I'd suggest comparing the measurement of your filter input to a measurement of your AWG connected directly to your oscilloscope. If the amplitudes differ, then that's one source of inaccuracy. You could also try to vary the frequency and see whether you see some frequency dependent effects when measuring your filter input, which you'll probably not see when measuring the AWG directly.

2. If your notches are not as steep as you expected, a common reason is poor performance of the inductors (poor Q, your frequency of interest is too close to the self-resonance of your inductor, etc).

If you're doing this for work, do yourself a favor and source components that have a proper SPICE model so that your simulation better approximates the real measurement, and make sure that the frequency of self-resonance is at around 1 order of magnitude higher than your highest frequency of interest.

Unless you have done a very poor job in your PCB design, a circuit track will mostly shift the notch of your frequency and probably steal a bit of depth of the notches you want due to their intrinsic parasitic impedance.

Answer 3

You should try this, made with "passive filter designer", microcap v12.