[time-nuts] Re: Working on a low-phase-noise frequency multiplier

[Richard (Rick) Karlquist]

On 8/25/2023 10:39 AM, Bill Ezell via time-nuts wrote:
> I have some old frequency standards that provide 1, 2.5, or 5 Mhz 

> It seems that doing an integer-N PLL with a VCXO would be the way to go. 
> The capture range can be quite narrow since the input will be something 
> that's within a 1 Hz range and a long loop constant might be good. Plus, 
> with a good-quality VCXO, I would expect the noise figure to match the 
> VCXO's pretty closely.

This only makes sense if you just happen to have a high quality VCXO in
your "junkbox".

I designed a frequency multiplier chain for the 5071A that had 5 
doublers in cascade to go from 10 MHz to 320 MHz.  Here are some
take aways from that:

1.  IIRC, I used Mini Circuits ASK-1 mixers where I connected the LO 
input in series with the RF input.  This works better than connecting 
them in parallel and is simpler than using a power splitter.  The "IF" 
was then the doubler output. Note:  it is not at all clear from the data 
sheet that this is even possible.  However, it turns out that these two 
inputs are floating primaries on the respective transformers in the 
ASK-1.  There is nothing unique about the ASK-1; I'm sure lots of other 
mixers would work as long as you can connect the LO and RF in series, 
which isn't always the case.
Also, the ASK has an IF bandwidth of 600 MHz.

2.  This is very important:  the input to the ASK-1 must be a pure sine 
wave.  This is not intuitively obvious at all, but if the input contains 
any harmonics of the drive frequency, it will degrade the output 
spectrum.  IIRC, I used a resonant impedance matching circuit that had a 
fairly high loaded Q to accomplish this.  It took something like +12 dBm 
of drive.

3.  The output filter was not some "obvious" Butterworth low pass ladder 
filter.  Rather it was a BANDPASS filter at the output frequency, such 
as 20 MHz if the input was 10 MHz.  A bandpass is necessary to reject 
subharmonics, but the "obvious" top coupled resonator design was not 
used.  Instead, the filter consisted of series resonant tank circuits 
and parallel resonant tanks, arranged in a ladder network alternating 
between parallel resonant tanks in shunt, and series resonant tanks in 
series.  No one tank had a particularly high loaded Q, but given enough 
of them, there was a lot of selectivity.   It was all built with SMT 5% 
inductors and 1% capacitors.  Using only E12 values, combinations of 
these were used to tune each tank to the exact value needed.  After the 
first PC board came back, each and EVERY L and C was then hand tweaked 
to compensate for PC board parasitics.  There were no adjustable 
components.  The board was released to production and it just worked.  I 
never heard anything about it from the production engineers.

Perhaps this is overkill for the original poster's question, but I 
thought I would share it with the group anyway.

Rick N6RK



> 
> Thanks, Bill
>