SM2000 Part 5 – PA Filter and Self Resonance

I now have a prototype 1W PA working (based on a RD01MUS2 FET) and need to design a filter to remove harmonics. With a transmit signal around 150MHz, this filter needs to attenuate harmonics at 300 and 450MHz.

My first attempt was a Q=5 “Pi” filter (56pF, 33nH, 56pF), which can be viewed as two L networks back to back with a virtual 5 ohm resistance in the middle. However when tested it had poor stop band rejection, just 30dB maximum at 280MHz (yellow):

After a day of messing about, reading, and LTSpice simulations, I traced the issue to stray inductance in the capacitor leads. By reducing the lead length for the capacitors, stop band performance improved by over 20dB (purple)!!

Here is a photo of a 4.7pF cap with roughly the lead lengths I started with:

Not much but combined with 50pF or more of capacitance the inductance of these leads can have a big effect. To reduce the lead inductance I soldered the capacitors across the back of the SMA connectors, and clipped the leads very short:

A few mm of lead can make a big difference, around 1nH per mm. This becomes quite significant at UHF, e.g. 10mm is 10nH which is 31 ohms at 500MHz.

Here are two simulations with long (green) and short (blue) capacitor leads, modeled as 8nH and 2nH series inductance. The longer 8nH leads have a series resonant frequency f = 1/2*pi*sqrt(LC) = 259MHz, before our 2nd harmonic. This makes the initial slope steep and produces a notch in the frequency response, however after resonance the capacitor is now inductive, and the stop band attenuation starts to get worse. The notch is not visible on the real world sweep, perhaps due to finite Q of the real world capacitor. However the same 30dB stop band “floor” can be seen in the simulation.

Also modeled is a guess of the 33nH conductor series resistance and parasitic parallel capacitance.

This sensitivity to component lead lengths was a nasty surprise! Although I’m developing a VHF radio, the behavior of components at UHF needs to be taken into account. Self resonance and parasitic effects is one of those things I’ve “sort of known” for a while – but the experience of a screw up and having to solve it really drives the lesson home! So: best to use small value, surface mount capacitors and ensure the self resonant frequency is above 500MHz. Quite amazing what 5mm of component lead can do at 500MHz!

5th order Chebychev

Armed with this graphical lesson in UHF construction I set about building a 0.5dB ripple, N=5 Chebyshev filter, with a 3dB cut off of 180MHz, using the tables in RF Circuit Design (by Chris Bowick).

Here is the circuit:

If you zoom in on the photo you just see a 33pF SM capacitor soldered across the back of the SMA connectors. I used a 47pF through hole cap, however I soldered the entire 5mm lead of one end to ground and had virtually no lead at the hot end. I used air core inductors for high Q, mounted at right angles to minimise coupling. They are 4 turns loosely spaced on a 5mm ID drill bit. I adjusted them with my tracking generator/spec-an to series resonate at 98MHz with a 47pF capacitor.

Here is the sweep of the filter:

I am very proud of this sweep! At least 70dB stop-band attenuation all the way out to 1.5GHz Yayyyyyyyyy.

Here is the output spectrum of my prototype 1W PA after being cleaned up by the filter:

With an output power of 1W (+30dBm) the 2nd harmonic is 53dB down. This exceeds the ACMA/ITU Amateur Radio Spec of 43dB + 10log10(P) by 10dB. For comparison the 2nd harmonic of my FT-817 with 27dBm (0.5W) output is 56dB down. My Baofeng UV-5R on low power (+32dBm) has several rather interesting VHF spurious emissions, the worst being just 42dB down at 180MHz.

I’m no Yaesu (my DV system is better), but these results are not bad for a VHF/UHF noob.

Reading Further

I stumbled across Construction Techniques for LC Highpass and Lowpass Filters used in the 1 MHz to 1 GHz Frequency Range. This is a really thorough treatment of how parasitic effects upset filters, has lots of of experimental results, and tips to improve real world filters.