<TL>/<DR> If you just want to know the solution scroll down.

Problem (short)

I needed a (even dirty) source of reference frequency to calibrate a PFM3000. 

Epic

So it all started when I was searching the internet on how to calibrate a Frequency Counter, after I acquired a second hand PFM3000. Not knowing if the previous owner played with the calibration pot or not, that was a must.

In my sailing through the rough seas of the world wide web, I found a paper describing the different methods and one of them was using a GPS module. Turns out GPS modules have the possibility of outputting a frequency that is as exact as any reference one.

Great! I have a couple of GPS module for my

RC airplanes. Can I use those? More sailing.

 I decide to open the shield on one and discover it uses the M8030-KT (U-Blox). Searching through datasheets, yes, it has a configurable TIMEPULSE (whatever that is). But what pin is it connected to??? U-Blox don't provide pinout for their M8030-KT for the likes of us (hobbyists), and after more digging in the dark corners of the internet, I discover the  chinese clone/equivalent chip AT6558R and it's datasheet which explains the pinout: https://datasheet.lcsc.com/szlcsc/1901021833_ZHONGKEWEI-AT6558R_C350653.pdf. 

Playing with the multimeter I confirmed it that it is similar. But Still couldn't figure out which pin was the TIMEPULSE. Next I stumbled upon the Youtube video:

Doh.. I didn't have to waste so much time.. it's the Red Led pin!!

OK.. What next.. installing the U-Center from U-Blox and trying to configure the BN-220 through the nice UI.. (I wired the BN-220 through an Arduino UNO, and shorted the RESET pin to GND, yes you can do that!) I got the NMEA messages, but the configuration just wouldn't work.. no ACK came back. A few hours later I gave up on them.  

Second day. I was still thinking that something is missing, since in the iNav it would auto-configure the GPS to another baud rate, refresh rate and use UBX. So why can't I use it in the UI??

I get my arduino Zero, find the library from Sparkfun https://github.com/sparkfun/SparkFun_u-blox_GNSS_Arduino_Library for M8 GPS chips and start hacking away.

2 hours later (after lots of stupid errors and bad documentation) I had the GPS module outputting the frequency of my liking. They even have a sample that does exactly that. Time for a nice cup of tea.

P. S. My PFM3000 was 1.4Hz off on 1MHz. Well within specs (2PPM).

Solution

You can use any U-Blox M8 module (or newer, I believe). Beitian BN220 was on hand so I just used that one. Need to check the protocol version that it's >=18 (the first lines when powering the GPS module).

Arduino Zero is the weapon of my choice because it's not using a serial chip so you have a dedicated serial for USB communication and another one for peripherals (also lots of horsepower that weren't used in this project).

Check the Instructions for wiring and sketch.

Results

Yellow is the signal entering the resistor, green is the one exiting the resistor and entering the LED.

A pretty good square wave for 1KHz:

At 10KHz it's still good:

Things just start getting slightly messy at 100KHz, occasional spikes, and the waveform starts deforming:

At 1MHz we see a totally different picture:

Next we have the 10MHz and that's where things go wrong, as the waveform is all over the place because of PLL trying to divide the 48MHz to get a 10MHz. There is no integer number to divide by. That's also where I stop trusting my Hantek and probes as to vertical accuracy.

My PFM3000 still managed to get an accurate reading, with just 14Hz error (within 2PPM).

12MHz - there might be some spikes that my Hantek can't catch, so I wouldn't trust it too much, but the main thing is the wave is very uniform, unlike the 10MHz:

And because I'm considering using 24MHz in a project, here's the wave:

Ignore the horizontal lines in the last two screens as I hurried to get the screenshot before the gps signal was lost (in the house), and did not wait for those to expire.

Also ignore the Freq reading, as it's way off. Hantek isn't known for it's accuracy.

Afterthoughts

This project did not intend to get a clean output frequency/oscillation. But this can be done, and in the referenced video it was done. I did configure my code to output several frequencies, including 10MHz and 12MHz. 10MHz was jumping all over the place, but 12MHz was stable, exactly as explained in the video - the source is a 48MHz clock, and it can be divided by exactly 4 to get 12MHz.

If you want better signal, maybe unsoldering the LED and resistor would give better output - I don't know, comment if you do. I still need my modules in working condition so I won't do the desoldering as I'm pretty sure I am not that good at removing a 0603 LED without destroying it.

I am playing with another circuit for HF radio, where it uses a 74ACT00 to get a nice 5V square wave, so that can be a solution if you want to do something more serious.