[time-nuts] Homebrew frequency counter, need help
kb8tq at n1k.org
Fri Nov 28 12:41:04 EST 2014
> On Nov 28, 2014, at 10:14 AM, Li Ang <lllaaa at gmail.com> wrote:
>> 1) What frequency is the crystal at? (can you drive the chip from an OCXO?)
> I'm clking this chip with the refclk/2=5Mhz, which is done by cpld.
> something like always @(posedge refclk) tdc_clk <= ~tdc_clk; Not sure if
> it's noisy.
If the clock source into the CPLD is clean, the output should not be to bad.
> 2) Is there more bypassing on the circuit than shown? (If not, add some
> Since there is only few power supply pins, I just put 100uF + 100nF there.
> I can try put more to see if helps. And I'm going to use dedicated LDO for
> digital part and analog part next time to make a new pcb.
I suspect that a few more 100nF caps might be a good idea. They may not help, but it would rule out the supply as an issue.
> 3) How confident are you of your input signal? (can you check it with a
>> “known good” counter?)
> The best counter I get is the RACAL DANA 1992. That is not good enough. I
> think the accurarcy is not a problem, as if the reference is good enough,
That’s correct. I was hoping you might be able to borrow a SR620 or 53132 for a few minutes to check your input signal.
> it's only the matter of math. I prefer to enhance the resolution and reduce
If the noise is on your test signal, it can be very frustrating to chase it with a lot of software work.
> I'm using Trimble GPSDO(NTPX26AB) as the signal source
The GPSDO should be fairly quiet, but it has an ADEV that’s a bit high. The outputs also can have noise on them.
> and FE5650 Rb
The FE Rb’s tend to have a lot of spurs on the output. In some cases that can get you in trouble.
> as the ref. I've tried Rb+MV89A
If the MV89A is working properly, it should be a pretty good source. Based on some of the prices I’ve seen on the internal China market, you might get a couple of them as sources.
> and Rb + SMY01 signal generator, same
> performance. The 2nd-hand 53132A,PM6690,SR620 about 600~800$ here in China.
> I'm trying not to get one unless necessary. If I have get one to compare
> the performance, what's your suggested model?
The SR620 is a good counter, so is the 53132. They both can have problems. The 53132 display wears out and it’s power supply can fail. The 620 can run a bit hot, which kills a variety of parts in it. It’s better to pay a bit more for one you can actually check out before you buy than to get one shipped in.
>> 4) Have you tried jumping the 10 ohm resistor on the regulator output? (it
>> may not be helping things …)
> I'll try to remove that tomorrow
It may be allowing the supply to drop a bit when the chip goes into some sort of computation. Often these things happen at just the wrong time …
If you have a CPLD and a MV89:
1) With a 10 MHz sine wave out of the OCXO, you need to convert it to logic first. A biased input is a pretty good way to do this.
2) Generate 200 or 400 ns wide pulses out of the CPLD for testing. That will eliminate any issues from the 5 MHz crystal in the MV-89.
3) Keep the PCB as simple as you can. You need at least a double sided board (one side ground plane). If you can get a cheap 4 layer board, go for it. A full internal ground plane is a good thing.
4) Route the high speed signals (like the OCXO output) through solid connections. Flying wire leads are not a good idea. Mounting a MV89 direct to the PCB is a good way to do things. SMA connectors are also good.
5) If you have an oscilloscope or can borrow one, take a look at the signals on your board. Even a quick check can tell you a lot about signals that are not what they should be.
6) Be careful of ground loops and power supply issues. I’ve spent a *lot* of time on breadboards that didn’t work because I had power line noise running around.
Good luck !!
> All of which is *plenty* good enough to make a decent counter. That
>> assumes that they are talking about accuracy (even 1 sigma) rather than
>> just the resolution of the LSB. Specs are often confusing on parts like
> I guess the accuracy is not important in the interpolator scenario. It has
> a feature to output the result of delta_time / ref_cycle_time(it will
> measure the 2cycle_time - 1cycle_time after the delta_time measurement and
> do the float calculation). All I need is this part(it's the fraction part
> of refcnt). So if I can reach 90ps resolution they claim, the counter can
> tell 1/1000 of one reference cycle. That's 3 digits.
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
More information about the time-nuts