1
\$\begingroup\$

I am trying to design a 4th-Order bandpass filter with the following specifications; centre frequency 30 MHz, bandwidth 5 MHz and of course minimising loss.

I present my current design below which I determined using the ELSIE simulation design program. The problem is I am not very confident in electronics and I am not convinced about my circuit diagram.

schematic

simulate this circuit – Schematic created using CircuitLab If anyone has any advice or can spot any mistakes I would appreciate any thoughts!

Cheers!

Quick Edit: I have attached the output schematic of my ELSIE simulation in case it is useful to anyone: enter image description here

\$\endgroup\$
9
  • \$\begingroup\$ Try out this tool for a better design ti.com/lsds/ti/analog/webench/… \$\endgroup\$ – user3219492 Nov 25 '16 at 17:21
  • \$\begingroup\$ @user3219492 thanks I'll take a look at that! \$\endgroup\$ – Q.P. Nov 25 '16 at 17:26
  • 2
    \$\begingroup\$ What you have labelled as the output is the ground that's common to both input and output. The output is taken from the junction of C6 and C4. \$\endgroup\$ – Neil_UK Nov 25 '16 at 17:40
  • \$\begingroup\$ @Neil_UK thanks for that spot! That was actually one of the things that I was confused about. \$\endgroup\$ – Q.P. Nov 25 '16 at 17:42
  • \$\begingroup\$ @user3219492 I tried out the tool you recommended - unfortunately it is restricted to a centre frequency of 10 MHz. Thanks for the suggestion though \$\endgroup\$ – Q.P. Nov 25 '16 at 17:53
1
\$\begingroup\$

You may be constrained by Network Analyzer impedance of 50 ohms. But your filter need not be so constrained. End-matching networks allow you to optimize these goals:

(1) Choose filter components whose unloaded Q is high (at 30 Mhz).
(2) Choose filter components whose nominal value has high accuracy.
(3) Choose filter component values whose reactance dominate circuit parasitic reactances.
(4) Minimize passband attenuation.
One approach (perhaps Elsie allows this) chooses source and load resistance that gives optimized component values. Usually, both source and load are equal (and for your filter type, larger than 50 ohms). Then end-matching networks are added to the design to accommodate 50 ohm source & load Z. This does have the disadvantage of adding (at minimum) two extra components. Familiarity with available commercial components is of paramount importance in choosing an optimum filter impedance. And estimating circuit parasitic reactances is important too.
You will likely find that a filter that can meet design tolerances will need to be tune-able - because fixed, standard component values that Elsie spits out are not available, and because their manufactured accuracy is too loose. If production volume is high enough, then you might move to a monolithic surface-acoustic-wave (SAW) filter and avoid tuning. This solution is chosen by TV manufacturers.


Here is an example 3rd-order approach to illustrate:
Filter design chooses an arbitrary inductor value of 265nH, unloaded Q=100. This (in my opinion) is an "optimum" component value, that also yields "optimum" capacitor values. Centre frequency=30Mhz., BW=4Mhz. Source and load resistance turns out to be 405 ohms:

schematic

simulate this circuit – Schematic created using CircuitLab Hopefully, Elsie allows you to specify 405 ohm source/load Z. Choose source/load Z that yields realizable, optimized component values spewed by Elsie. You'll soon see that higher Z yields large inductors & small capacitors, while lower Z yields small inductors & large capacitors. This filter is then adapted to your desired source & load impedance of the network analyser (50 ohms) with end-matching networks:

schematic

simulate this circuit Note that with all top-coupling capacitors, the attenuation outside the passband is not symmetric...attenuation is poor above the passband. Choosing end-matching to be inductive can improve this situation. You can transpose a filter too:

schematic

simulate this circuit After 50 years, "Handbook of filter synthesis" (Anatol Zverev) is the reference for lumped-component filters (in your frequency range), but its not for mouse-jockeys.

\$\endgroup\$
9
  • \$\begingroup\$ thanks for this answer - it's great! I will look into the things you have suggested. I actually got it working, making a very crude first attempt. The span and centre value were pretty much on target but the loss was around 13 dB... \$\endgroup\$ – Q.P. Nov 25 '16 at 22:27
  • \$\begingroup\$ А useful companion to Zverev's "Handbook of filter synthesis" a.k.a "Red Bible" is Zverev and Blinchikoff's "Filtering in the Time and Frequency Domains". \$\endgroup\$ – My Other Head Nov 25 '16 at 23:16
  • \$\begingroup\$ @MyOtherHead Thanks, I will definitely take a look at this! \$\endgroup\$ – Q.P. Nov 26 '16 at 0:13
  • \$\begingroup\$ @glen_geek If I could ask another question about your first design where you have 405 Ohms source and load resistance. Should these be physically components on the circuit board, or are they just a representation? \$\endgroup\$ – Q.P. Nov 26 '16 at 14:39
  • \$\begingroup\$ @QuantumPenguin , Every filter is designed for real source & load resistances. Your network analyser very carefully ensures that it provides 50 ohm source & load - they are physical resistors. In a test setup, those 405 ohms parallel R's would become 355 ohm series R's to your network analyser. In actual use, 355 end R's would cause too much attenuation. Inductor Q was 100, which says there are missing R's in parallel with each ideal 265nH inductor (4995 ohms) - you should add these to a simulation. \$\endgroup\$ – glen_geek Nov 26 '16 at 15:47

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.