24th Jun 2020, 4:07 pm | #81 |
Pentode
Join Date: Apr 2018
Location: Canterbury, Kent, UK.
Posts: 189
|
Re: Replacement for HP8640B signal generator
Ah the 8662 has a "digital sweep" function. See p114 of this PDF
http://ftb.ko4bb.com/manuals/94.15.3...8662_90069.pdf |
24th Jun 2020, 7:03 pm | #82 | |
Dekatron
Join Date: Sep 2010
Location: Cheltenham, Gloucestershire, UK.
Posts: 3,077
|
Re: Replacement for HP8640B signal generator
Quote:
It's a bit too hot to run any test gear for very long this evening but I did have a quick go at sending a very simple sweep routine to the 2019 via GPIB. This stepped 100 frequency points using the (delta) UP command after I programmed the step sizes to 100Hz. The result was a bit disappointing. It took about 6 seconds to complete the sweep and I had to slow it down a bit more to get reliable results. One alternative would be to inject a high power noise source into the filter and then look at the filter output on the analyser. I have done this a few times over the years and the analyser should show the filter shape filled with noise.
__________________
Regards, Jeremy G0HZU |
|
29th Jun 2020, 8:52 am | #83 | |
Pentode
Join Date: Apr 2018
Location: Canterbury, Kent, UK.
Posts: 189
|
Re: Replacement for HP8640B signal generator
Quote:
https://www.hpl.hp.com/hpjournal/pdf...Fs/1979-08.pdf Depending on the sweep width the first LO may be locked only at the start of a scan or more often (called by the quaint name "lock and roll"). This, and comments on the HP group, suggests that a HP8444A tracking generator modified to avoid drift on its own internal oscillators, will give a stable trace on the HP8568B. Or do I misunderstand (again)? Alan |
|
29th Jun 2020, 2:47 pm | #84 |
Dekatron
Join Date: Sep 2010
Location: Cheltenham, Gloucestershire, UK.
Posts: 3,077
|
Re: Replacement for HP8640B signal generator
I'm not a member of HP groups.io so I can't see what their expert view on this is.
I'm very rusty on this stuff but the HP8568B contains a free running cavity oscillator for LO2 although this probably isn't obvious to many people. Therefore, I can see why someone would claim that (on narrow spans) LO1 + LO2 etc is phase locked (and drift free) to the 10MHz reference and some people might think that all of the local oscillators are locked to the 10MHz ocxo reference because this analyser shows a rock steady response on the CRT display with low phase noise. But I think it is only the overall synthesiser loop that is corrected like this on narrow spans. This means that LO1 will have drift forced upon it by the correction system to counter the drift in the free running LO2. Can you ask the HP groups if they realise that LO2 is a free running cavity oscillator and can you ask them how they think the HP8568B is able to correct for the drift and noise of this LO2 oscillator? I don't think there is any PLL error correction applied directly at LO2. It just runs freely and drifts very slowly within the overall PLL system. I believe that the answer is that the system 'corrects' LO1 instead and this means LO1 will drift in a similar way to LO2. It might slowly drift 10kHz in 10 minutes for example. A normal user of the HP8568B will be blissfully unaware of this because the two drifts cancel out in the overall PLL system and this gives the impression that all of the LOs are drift free.
__________________
Regards, Jeremy G0HZU Last edited by G0HZU_JMR; 29th Jun 2020 at 2:55 pm. |
29th Jun 2020, 3:23 pm | #85 |
Moderator
Join Date: Mar 2012
Location: Fife, Scotland, UK.
Posts: 22,867
|
Re: Replacement for HP8640B signal generator
There IS a free running cavity oscillator AND the first LO is locked. However, the first LO is locked to a frequency assembled using that cavity oscillator.
Essentially the 8568 contains a duplicate receiver strip running in reverse like a tracking generator. The yig is tuned to bring the pseudo tracker into lock at the wanted frequency of centre span and the tuning error voltage is frozen. The yig is then analoguely tuned to the start frequency and then ramped across the screen. As it passes the centre, the sweep voltage passes through zero and the frozen error correction voltage should be good to put it on the centre frequency accurately for the instant it passes the centre. The sweep continues to do the other half of the scan. As the cavity osc drifts, the system shifts the YIG to compensate. THis is peculiar to the 8568 The 8566 has a multiplied-up 100MHz crystal oscillator not a cavity and doesn't have the dual strip structure. The 100MHz xtal is used to make all the other LOs needed and is itself locked to a 10811A ovened ref osc (1 in 10^8 grade) THe YIG is mixed down using a harmonic gen/sampler and is phase locked to the wanted freq for the centre and a correction voltage is frozen in a S/H. It then unlocks and does an analogue sweep with the S/H fholding the correction voltage to get the centre right. So the analyser spends a proportion of time lurking at the centre frequency settling its PLL and not measuring. Then it does an analogue sweep and then it's PLL time again. 8566 uses more expensive stuff, but it can use YIG osc harmonic mixer and YIG filter to go to 22GHz and beyond (100GHz with external mixers) Multiplied Xtal osc beats the cavity on stability and phase noise. As you go for narrower and narrower spans the deviation from the centre frequency gets less and your frequency accuracy gets better and better. So if you download picture data from an analyser, don't take the freq dimension as phase locked gospel. There will be drift of freq from when lock was on to the time of the middle of the scan, and there will be analogue scaling error of the distance from the centre. Select zero span to get real time lock for a true spot measurement. Or select marker count. Sweep stops on a marker, locks up and runs the counter on the IF. David
__________________
Can't afford the volcanic island yet, but the plans for my monorail and the goons' uniforms are done |
29th Jun 2020, 3:31 pm | #86 | |
Dekatron
Join Date: Sep 2010
Location: Cheltenham, Gloucestershire, UK.
Posts: 3,077
|
Re: Replacement for HP8640B signal generator
Quote:
Just spotted David's reply above and he seems to agree about LO1 and LO2 in the HP8568B. On very narrow spans (<1MHz) I recall that the HP8568B system is fully locked over the whole sweep but the drift problem with LO2 still has to be accounted for so I think LO1 will still drift in sympathy with LO2 to keep the whole PLL system locked. The drift of LO2 can be seen as a slowly wayward error that the system can easily track and correct for.
__________________
Regards, Jeremy G0HZU Last edited by G0HZU_JMR; 29th Jun 2020 at 3:42 pm. |
|
29th Jun 2020, 3:48 pm | #87 | |
Pentode
Join Date: Apr 2018
Location: Canterbury, Kent, UK.
Posts: 189
|
Re: Replacement for HP8640B signal generator
Quote:
OK, so the two of you agree that LO1 will drift. The demonstration would be for Jeremy to plug LO1 (as fed to the TG) into his expensive freq counter and quote us directly measured drift numbers. People say the 8568B should be left running for hours and hours before it becomes stable but if, after the paint I've been watching has fully dried, LO1 drifts 1KHz/min then the tracking generator approach is not for me. Back to the idea of a high power noise source methinks... Alan |
|
29th Jun 2020, 8:02 pm | #88 | |
Dekatron
Join Date: Sep 2010
Location: Cheltenham, Gloucestershire, UK.
Posts: 3,077
|
Re: Replacement for HP8640B signal generator
Quote:
To get the very best from the OCXO it should be left running 24/7 but I really don't think that is necessary. The drift in LO2 is cancelled by the internal PLL so users of a healthy HP8568B should see a very stable analyser even on a 1kHz span. I think you could sit there all day and not see the response change assuming the 10MHz OCXO has had a few minutes to warm up. If the whole analyser is started from cold then there will be drift issues for a few minutes because the 10MHz OCXO will be many Hz away from 10MHz with a cold oven. But this applies to all RF gear that uses a similar oscillator. My Anritsu frequency counter has a huge OCXO that takes about 50 minutes to settle from cold and that is a real pain.
__________________
Regards, Jeremy G0HZU |
|
29th Jun 2020, 9:23 pm | #89 | |
Moderator
Join Date: Mar 2012
Location: Fife, Scotland, UK.
Posts: 22,867
|
Re: Replacement for HP8640B signal generator
Quote:
Ho ho ho ho ho! A cheapy plain crystal oscillator has an AT cut crystal which has a temp/freq curve like an uld U-bend sink trap. You ask for +/-10ppm over 0-70C and they adjust cutting angles to fit the curve in that box. Turn it on and it comes up somewhere in the box and moves slightly from slight warming. so 10ppm off. A TCXO is the same but with a thermistor, varactor and a bit of curve shaping to compensate the worst of the temp change out. They'll get you to +/- 1ppm over the temp range, maybe half that. Turn them on and they're not far off their final settled frequency. An ovened oscillator runs hot... typ +80C because heating is easy, cooling is painful so it's one-sided aircon for them. For best performance they use a different cut of quartz, designed to have a parabolic freq/temp curve around 80C. So it's very flat at the temp it runs at. So any ovem variations have less effect. It takes 10-20 mins to heat up from cold and to get settled. That parabolic curve is a long way from the design frequency when it's down at 22C. About 1000ppm off So turn on and your super duper kilobucks ovened oscillator is about 100x worse than a cheap crappy crystal. The 8668 and lots of other premium stuff is all absolutely lousy at frequency accuracy as turned on. Much worse than the versions without the posh reference option. Joe Bloggs never expects this! So the settling time for the 8568 isn't really the cavity osc it's the master reference. The PLL and lock and roll work from the get go, but what they lock onto is feeling a bit poorly until it gets some heat into it. By the way, some of the YIG spheres are on little PTC heater mountings to stabilise their temperature. They don't take as long. David
__________________
Can't afford the volcanic island yet, but the plans for my monorail and the goons' uniforms are done |
|
29th Jun 2020, 10:10 pm | #90 |
Pentode
Join Date: Apr 2018
Location: Canterbury, Kent, UK.
Posts: 189
|
Re: Replacement for HP8640B signal generator
A chap on the HP/Agilent group has done a proper job, taking all three LOs from the 8568B to a gutted 8445A chassis and mixed them to give the TG signal. He used a lot of expensive packaged parts but admits it could be done more cheaply. Either way it's still more than I would want to bite off. I think you can read the thread and look at the pictures without joining the group
https://groups.io/g/HP-Agilent-Keysi...28157646#92614 Alan G3XAQ |
29th Jun 2020, 10:20 pm | #91 |
Dekatron
Join Date: Sep 2010
Location: Cheltenham, Gloucestershire, UK.
Posts: 3,077
|
Re: Replacement for HP8640B signal generator
I just did a few quick tests after letting the HP8568B run in standby for a while.
When I turned it on I looked at the 10MHz OCXO on a counter and it drifted up very slightly for a few seconds but settled at 10MHz -0.06Hz. I've only adjusted this oscillator once in 15 years and that was probably a couple of years ago. So this is a good result I think. I let the analyser run for maybe 5 minutes and looked at LO1 with the analyser set to 10MHz CF and a 5kHz span. I also fed an accurate 10MHz from a decent sig gen into the analyser. This sig gen has a 10MHz OCXO reference and is very accurate. No drift was seen on the analyser display but the centre of the LO1 sweep was slowly drifting upwards. I also looked at the drift of LO2 with an EMC probe. This showed the same drift rate as LO1. With the analyser still warming up the drift rate for LO1 and LO2 was 1kHz every 17 seconds. The 10MHz reference inside the HP8568B didn't change and stayed at 10MHz -0.06Hz.
__________________
Regards, Jeremy G0HZU |
29th Jun 2020, 11:44 pm | #92 |
Dekatron
Join Date: Sep 2010
Location: Cheltenham, Gloucestershire, UK.
Posts: 3,077
|
Re: Replacement for HP8640B signal generator
I just retested both LO1 and LO2 and they have now both drifted up just over 100kHz. This is probably over a timeframe of about 100 minutes so my analyser does seem to have an LO2 cavity oscillator that drifts about 1kHz per minute.
If this was a VFO running at 1.750MHz then the equivalent drift would be 1 Hz per minute which would be extremely impressive so I'm not sure my LO2 cavity oscillator is poorly. Not all HP8568B LO2 oscillators will have the same drift pattern so you might get lucky. In normal use none of this matters because the PLL system cancels out the drift of LO2.
__________________
Regards, Jeremy G0HZU |
29th Jun 2020, 11:53 pm | #93 |
Dekatron
Join Date: Sep 2010
Location: Cheltenham, Gloucestershire, UK.
Posts: 3,077
|
Re: Replacement for HP8640B signal generator
I just measured the current rate of drift for LO2 (and LO1) over just 1 minute and it is down to 500Hz/minute so the drift rate is slowing down. The 10MHz OCXO currently reads 10MHz -0.05Hz so it has hardly moved at all.
__________________
Regards, Jeremy G0HZU |
29th Jun 2020, 11:56 pm | #94 |
Moderator
Join Date: Mar 2012
Location: Fife, Scotland, UK.
Posts: 22,867
|
Re: Replacement for HP8640B signal generator
Yep, that all sounds as it should be.
10811 and its 10544 take a long long time to cool, so you may not have seen the bulk of it. You don't have to buy many packaged bits from Minicircuits M/Acom etc before you've blown the price of an old network analyser, and you get phase! You don't even need the network analyser to go as high as the spectrum analyser. Broadband noise source trickery gives you little dynamic range. Alan, you know you really want a 3577A....... David
__________________
Can't afford the volcanic island yet, but the plans for my monorail and the goons' uniforms are done |
30th Jun 2020, 11:07 am | #95 |
Octode
Join Date: May 2010
Location: Grantham, Lincolnshire, UK.
Posts: 1,174
|
Re: Replacement for HP8640B signal generator
Just to add another element to the discussion, I have a couple of Tracor 900 VLF/LF receivers.
Bought many years ago, so from memory these phase lock to various government agency signals such as MSF, possibly 198kHz. There is a Rustrak chart recorder that ticks along giving the frequency error in centicycles. Again from memory, these are the real way to get a frequency standard without your own caesium unit. Weren't cheap, even got the manual for them but can't find it. Sure I got one or the other working. Any use? |
30th Jun 2020, 11:08 am | #96 |
Pentode
Join Date: Apr 2018
Location: Canterbury, Kent, UK.
Posts: 189
|
Re: Replacement for HP8640B signal generator
See my earlier comment about Satan...
|
30th Jun 2020, 2:02 pm | #97 |
Moderator
Join Date: Mar 2012
Location: Fife, Scotland, UK.
Posts: 22,867
|
Re: Replacement for HP8640B signal generator
If god is supposed to be omnipresent, then isn't the anti-god?
I'm not sure what to not believe in, these days. David
__________________
Can't afford the volcanic island yet, but the plans for my monorail and the goons' uniforms are done |
30th Jun 2020, 2:48 pm | #98 |
Pentode
Join Date: Apr 2018
Location: Canterbury, Kent, UK.
Posts: 189
|
Re: Replacement for HP8640B signal generator
Either way, I don't need your help with temptation.
I should be getting my Mi 2019 sig gen tomorrow. The 8568B is in its shipping crate in Cyprus awaiting collection by the shipper. I already have a NanoVNA so I don't really really have to have a proper one. So many toys, so little time... Alan |
30th Jun 2020, 11:11 pm | #99 |
Dekatron
Join Date: Sep 2010
Location: Cheltenham, Gloucestershire, UK.
Posts: 3,077
|
Re: Replacement for HP8640B signal generator
Wow, you have bought the HP8568B then!
I left my HP8568B unplugged from the mains for about 24 hours to let it cool off the RF unit and the OCXO within it and then powered it on immediately and logged the frequency of the 10MHz OCXO over about 20 minutes with a resolution of 0.01Hz and a 1 second gate time. The OVEN COLD warning should show on the CRT when this happens. The (cold) OCXO was about 350Hz high above 10MHz (10.00035000MHz) at the instant the HP8568B was powered up. See below for two graphs showing the warmup curve. The OVEN COLD warning turned off after about 8 minutes and this was very close to the point where the OCXO overshot 10.00000000MHz on the way down. This is shown in one of the graphs below. However, the OCXO has clearly not settled fully at this point as it goes on to overshoot by about 6Hz. By 11 minutes it is within about 1Hz and I would consider the analyser fit for casual use at this point. For more critical use it looks like it takes about 14 minutes to get within about 0.1Hz of the final settled frequency. By 18 minutes I think it is within about 0.05Hz of the final settled frequency. To get beyond this to a very settled state would take a few hours I think. I'm usually quite happy to use an HP8568B after about 10-15 minutes from a completely cold start.
__________________
Regards, Jeremy G0HZU Last edited by G0HZU_JMR; 30th Jun 2020 at 11:33 pm. |
1st Jul 2020, 4:05 pm | #100 |
Pentode
Join Date: Apr 2018
Location: Canterbury, Kent, UK.
Posts: 189
|
Re: Replacement for HP8640B signal generator
Yes. I can resist anything except temptation. I don't like to talk about money on here but in a PM I ran the price past David and he thought it was "friendly". I would have asked you as well but you only have a public persona.
I get the SA plus the flaky TG with broken LSI output stage, a Prologix GPIB/USB adapter and the usual cables. I'm unsure if paper manuals are in there but they are available online. I've ordered a Chinese noise source for next to nowt to tide me over while I think about TG solutions. As David says, the noise output on narrow RBW will be marginal but maybe I can band limit it to 30MHz and amplify it 20dB to better suit my needs. The Ed Breya chap on the HP forum says he's done a proper TG from the guts of an 8444A. He says LO2 is available at a coax socket test point and an LO3 signal can be bodged. The job is to "just" replace the 2050MHz (which the 8444A makes from its cavity and the external 500MHz signal) with LO2+LO3+21.4MHz. How hard can it be? He seems to have made a meal of reverse isolation on the LO samples, even adding 3 isolators more to the LO1 signal than HP originally used. But OTOH he says there are no visible spurs. In the absence of a junk box full of isolators an attenuator/amplifier approach would probably suffice. If I see an 8444A going for a song I might try it one day. My 2019A sig gen has travelled from Grantham to Hitchin but the planned handover today stalled so that will be rescheduled for the weekend or next week. Alan |