Today's experiment is a 10 MHz low pass filter to clean up a square wave and turn it into a sine wave.  This comes from the realization that the reference output from my Racal Dana 1992 universal counter is very unclean square wave instead of a pure sine wave.  Because square waves are actually composed of numerous harmonically-related sine waves, a good filter can extract just the fundamental tone of interest.  Cool, huh?

Why is this important?  Let's delve in a bit further . . .

The Racal counter has an OCXO timebase (ovenized crystal oscillator) which is extremely accurate.  I've had my GPSDO (GPS-Disciplined Oscillator) prototype working several times which gives me an atomic clock level of frequency accuracy.  Comparing that to the Racal, I found that the Racal is accurate to within about 0.4 Hz which is about 40 parts per billion.  Pretty incredible for a $75 eBay find!

But the Racal outputs a square wave and it isn't even a clean square wave.  It has many intermod distortion products and transients.  Ignoring that, for a moment, a square wave isn't the best reference for sensitive lab equipment.  It may be OK for some devices, depending on how they process the incoming reference, but a clean, pure sine wave is much better for eliminating noise from measurements.

Indeed, I have been finding 10 MHz noise in the lab, both radiated and conducted, as well as some harmonics.  It's slight, but noticeable.  So now I want to clean up that signal so I can keep relying on it while I continue building my ultimate frequency reference -- the GPS Disciplined Oscillator.

Background info:  For a terrific explanation of why a square wave is actually a composite of many sine waves (counterintuitive, yes?), watch this excellent YouTube video by Alan Wolke, W2AEW.

I've done some experimentation in the past with a Minicircuits PBP-10.7+ bandpass filter.  That is really intended as an IF filter, but its frequency range works for this project.  It does a lot to clean up noise, but leaves some to be desired.

Frequency Counter
Frequency Counters
(click for a large view)
Minicircuits Passband Filter
Minicircuits Passband Filter
(click for a large view)

So let's try a simple passive LC low pass filter.  I used RFsim99 to design the filter (I have to run it in a Windows XP virtual machine since it doesn't work in Windows 7).  The design parameters are 5th order Chebychev with 0.1 dB allowable passband ripple and a 10.9 MHz cutoff frequency.  Why these parameters?  Because it let me use parts I already had on hand such as 1μH inductors.

After soldering it up on plated perfboard with BNC connectors, it works!  Heck, it works better than the Minicircuits!  Well, mostly because the Minicircuits' purpose is really a bit different.

Check out this comparison.  The yellow trace is the Minicircuits bandpass filter.  The purple trace is my low pass filter.  We only care that 10 MHz gets through -- anything above that, especially multiples of 10 MHz (harmonics) need to be suppressed:

Notice that at the second harmonic (20 MHz), the Minicircuits filter only suppresses the signal by 11.28 dB.  But the low pass filter has 31.55 dB of suppression.  And it just gets better at the higher frequencies.

Filter comparison
Filter Comparison
(click for a large view)
The Assembled Filter
The Assembled Filter
(click for a large view)

The filter also has a very good impedance match.  I built it with trimmer capacitors to fine tune the circuit, but they weren't all that necessary after all.  The output trimmer is a little flakey so the output match isn't as good as the input, but it's still great.  Input impedance is 50.2Ω with very little reactance for an SWR of 1.06:1.  The output is 52Ω and has 1.1:1 SWR.

Finally, here's a quick comparison showing harmonic suppression:

DevicePower @ 10 MHzInsertion
Total Harmonic
Distortion (THD)
No filter 7.72 dBm (n/a) 31.90%
Minicircuits PBP-10.7+ 2.95 dBm 4.77 dB 5.43%
Chebychev 10 MHz LPF 7.19 dBm 0.53 dB 0.37%


Add comment

Security code