Re: Win11 ??

Short of that kind of teasing apart…

I have to think about orders of magnitude here. The GPSD-OCXO has a short term (~30s) stability of around 1e-11. But it will drift longer term, and the GPS signal will discipline the OCXO to bring it back on frequency (10 MHz). Surely that discipline correction will be on the order of 1e-11.

Now the two signal generators (Agilent and Siglent) are both capable of 1e-12 tuning increments, so surely they aren’t going to lose their minds on a reference carrier shift of 1e-11. eh?

[ I can run another experiment where I feed the reference input to the Siglent using the Agilent, which itself will be held to the GPSD-OCXO. But then I can step the Agilent output frequency by increments of 10 micro-Hz and record what happens with the Siglent signal. Do this and a rate of a few steps per minute, and we can surely avoid the 76 minute glitch.]

But the central piece in the picture is an Icom IC-7700 Receiver. I already know that its internal DDS is accurate only to a granularity of a few mHz at 10 MHz. For example, the measured tuning offset at 15 MHz is -19.495±0.005 mHz, with a sample standard deviation (one look to another) of ±0.1 mHz.  

So, in other words, the Icom receiver can be trusted only to around 1e-9, or 1 ppb. That’s about 100x worse than all the other equipment.

In terms of DDS word-size, I would expect that the Agilent and Siglent are using 48+ bits of precision. The Icom RX is probably using a 32-bit counter.

So, if anything has trouble adjusting to a reference carrier change, I would expect it to be the Icom receiver.

That being the case, I should expect to see a regular series of glitches, at the same 76 min spacing, even when using the old trusty Agilent Generator.

… and I just saw my first glitch with Agilent after waiting about 50 minutes. We’ll see if it happens again in 76 minutes…

- DM


> On Sep 3, 2025, at 12:21, Yuri Davidovsky (as work at disclosure dot ie) <[email protected]> wrote:
> 
> 
>> On 3 Sep 2025, at 21:13, David McClain <[email protected]> wrote:
>> 
>> It lasts 6 ms. The separation of the peak signal on either side of the glitch is 16 ms, and the nominal peak-period should be close to 10 ms.
> 
> We are on the tail of the suspect. Let’s have a look what he looks like. To do that, we could subtract the original signal in its clean form from the distorted signal and we should end up with the clean view of the impulse. It can be tricky due to slight frequency fluctuations in the main signal but it looks like it should be doable.
> 
> <image0.jpeg>


_______________________________________________
Lisp Hug - the mailing list for LispWorks users
[email protected]
http://www.lispworks.com/support/lisp-hug.html