Replacement for HP8640B signal generator

[Alan_G3XAQ]

Yes, digging into the 8444A manual I see they added a 1.55GHz free running cavity oscillator as well as the OPT-059 500MHz free running L/C oscillator (to replace the LO output from the earlier model SAs) to get it running to 1.5GHz with the 8658B.

So the TG will be a bit noisier that the base 8658B but, worse, it is bound to drift a lot. There are comments about needing to sweep crystal filters in single shot mode ("max hold") to get something to actually look at.

Ah well, I see the 8444A just as a freebie thrown in to already good deal on the 8658B.

73, Alan

[G0HZU_JMR]

I suspect there will be other minor niggles as well. I don't think the LO1 source from the 8568B will be very accurate in frequency when used with the tracking generator.

I'm very rusty on this stuff but the overall LO system in the HP8568B contains a free running cavity oscillator and this is used for LO2. Therefore, to achieve the clean phase noise and accurate frequency shown in my earlier 'system' phase noise plot I'd expect LO1 to be deliberately coloured with similar noise and drift to the LO2 cavity because the overall system needs to cancel and correct for the noise and drift caused by the free running LO2 cavity. Normal users of the analyser (i.e. those not exporting LO1 to a tracking generator) don't need to worry about any of this because the overall system phase noise will be cleaned up regardless of any drift or noise at the LO2 cavity oscillator.

In other words, I think the overall frequency uncertainty of the HP8568B + tracking generator system will be affected by the drift and noise of two cavity oscillators, not just one. There is the cavity oscillator used for LO2 in the 8568B and the cavity oscillator in the tracking generator.

[G0HZU_JMR]

To clean up all this stuff in the HP8568B and still maintain an extremely impressive close to carrier noise performance the HP8568B has quite a wide PLL loop bandwidth as drawn in my earlier image.
I'm at lunch here at home and I have powered up my HP8568B and see below for a plot of the noise pedestal. This particular HP8568B seems a bit cleaner than the 'typical' plots in the HP manual. Mine achieves about -113dBc/Hz within the loop bandwidth.

It may look a bit unsightly but it brings the benefit of 'lots' of noise cleanup within the loop bandwidth so it can easily wipe away the noise contribution from the LO2 cavity as an error signal within the system loop. You can see the noise is getting slightly cleaner as you look closer to the carrier. This allows the analyser to examine the close in phase noise of HF/VHF/UHF synthesisers on very narrow spans.

For an example, see the plot next to it that shows the HP8568B measuring a Marconi 2019 at 262MHz on a 5kHz span. This is the pink trace. The green trace below it is a much cleaner signal source being fed to the HP8568B and you can see how much lower the HP8568B system phase noise is compared to the noise of the Marconi 2019. I'd expect your HP8640B to have similar noise as the pink (Marconi 2019) trace when set to 262MHz so you could do a health check on the HP8640B and your Marconi 2019A up at these frequencies.

[Radio Wrangler]

The 8444A was intended as a tracker for the 1250MHz free-running spectrum analyser plug ins for the 140 scope mainframe, so free running wasn't seen as a problem.
THere are plenty of them around from those systems so you could buy a cheapie to raid for a spare output amp/alc.

Jeremy's left phase noise picture shows a classic PLL phase noise response with a loop bandwidth of about 60kHz art work. From 100kHz offset outwards, you see the raw, unaffected noise sidebands of the final oscillators. Within 50kHz offset you see the phase noise of the reference oscillators, the jitter of divides and phase detectors, and the LF noise of loop amplifiers all scaled up by the frequency multiplication factors.

Between these two limits, the noise behaviour does a segue form one to the other. The art is picking and planning a change-over characteristic. There will be a bump in the noise, but you want to minimise it without spoiling other factors.

The synth in a spectrum analyser has to follow the sweep reasonably accurately, and retrace in reasonable time.

David

[Alan_G3XAQ]

Yes, I can probably wait and pick up a non-opt059 8444A and raid it for its output amp. But given the drift problems is it worth much effort? Someone on the HP group suggested injecting a 2050MHz signal into the 8444A and avoiding using both the internal free-running oscillators.

His idea is to use a stable low noise signal source... from... er... a high quality signal generator. Which I suppose you just go out and buy. Hmm, I doubt something lashed up on a scrap of Veroboard would cut the mustard.

Alan

[G0HZU_JMR]

Even if you fitted a perfect 2050MHz oscillator to the 8444A it would only be half an answer because there will still be the drift effect of the free running LO2 cavity oscillator inside the HP8568B. I think the LO2 oscillator will 'paint' its drift and noise into the HP8568B PLL system as an error signal. The HP8568B can correct all this stuff internally but I think it will export LO1 with this error correction superimposed on top of it. So the LO1 signal from the HP8568B will show drift and some noise and this is because of the free running cavity oscillator used for LO2.

Note that this is what I think will happen not what I know will happen because I've never tried to used the 8444A with an HP8568B.

[Radio Wrangler]

Yep!

to make a real tracker for the 8568 you need all the oscillators extracting.

Alternatively you trust the 8568 to look after itself and use software to control an 8568 and an 8662 to step along tracking it. Slower, but low noise. Also allows offset tracking for mixer measurements, even frequency inversion for the other mixer sideband.

David

[Alan_G3XAQ]

It all looks more and more impractical. I'm contrasting this with my TF2370 with built-in TG, which does narrow scans of crystal filters perfectly adequately. For example see the attached photo. It answers the question: "Why does the passband on my 80m lockdown receiver sound lumpy". Looking at the 455KHz CW filter you can see why. The display is 100Hz/div by 10dB/div and it *is* lumpy because the filter is broken! The scan is perfectly stable for tens of minutes until the PLL drops out of lock. Thankfully I have another one out of an old TS930 coming in the post.

I know, it's just the price I have to pay for GHz frequency coverage.

Some might say I should keep both the old SA and still get the new one. But I daren't even think that within the house.

73, Alan G3XAQ

[G0HZU_JMR]

Presumably the 8444A has a manual rotary 'tracking' control on it and to try and maintain accurate tracking. I think this control would need fairly regular tweaking to maintain adequate tracking on narrow spans.

I'll leave my 8568B on for a while and monitor the drift of the ~1750MHz LO2 cavity oscillator. I can just pick it up using an H field probe wedged between the RF and the display sections of the HP8568B. At the moment the drift rate is fairly reasonable at 1kHz per minute but it might get better over time.

[Radio Wrangler]

Plan C: Just get a 3577. 200MHz network analyser. Perfect for crystal filters and very narrow scans.

David

[Alan_G3XAQ]

Originally Posted by Radio Wrangler

Plan C: Just get a 3577. 200MHz network analyser. Perfect for crystal filters and very narrow scans.

Get thee behind me, Satan.

Alan

[G0HZU_JMR]

His idea is to use a stable low noise signal source... from... er... a high quality signal generator. Which I suppose you just go out and buy. Hmm, I doubt something lashed up on a scrap of Veroboard would cut the mustard.

I'm not sure a lab sig gen would be needed. A fairly simple printed resonator based VCO at 2050MHz and an ADF4106 PLL chip referenced to a 10MHz VCXO could give a phase noise performance similar to the simulation plot below. A real prototype version would be unlikely to be better than this but it could be quite close to this level of performance. The whole circuit would go on a single PCB not much bigger than a fag packet. The rotary tracking control would be used to steer the 10MHz VCXO reference.

However, the overall system stability would still be limited by the drift of the LO1 signal extracted from the HP8568B.

The drift rate of the LO2 cavity oscillator in my HP8568B didn't improve much but it was very hot here this afternoon. Maybe it would have been better with normal room temperatures. It probably got as low as 800Hz/minute. I think this would be OK for most users. I'm not sure I could live with a system that drifted like this but there must be lots of happy HP8568B + 8444A users out there.

[Alan_G3XAQ]

Or would a ADF4351 eval kit do it all, so long as a TCXO tweaker was added? The Chinese have copied the eval board and stuck it in a box, for example e*ay item 124024524139. The AD datasheet suggests -100dBc/Hz at 10KHz spacing is possible at 2GHz with the VCO divided by 2 (figure 9) but who knows what the Chinese version is like?

Given that it's not going to make an 8568+8444A properly stable I find it hard to get enthused.

Alan

[G0HZU_JMR]

Yes, I think the phase noise would still be good enough with the ADF4351 solution. Doing it discretely with a printed VCO and the ADF4106 would be a lot cheaper but I agree that it isn't going to provide a completely stable solution. The 800Hz/min system drift would be enough to annoy me a lot so I can understand why you aren't that keen.

[Alan_G3XAQ]

Originally Posted by Radio Wrangler

Alternatively you trust the 8568 to look after itself and use software to control an 8568 and an 8662 to step along tracking it. Slower, but low noise. Also allows offset tracking for mixer measurements, even frequency inversion for the other mixer sideband.

I hadn't thought of that. I really like the idea of being able to look at mixers (or even whole receivers). The Mi 2019A I'm getting next week might have the GPIB option. Is that how you do it with the 8662: talk to both SA and sig gen over GPIB and send both "increment" or whatever commands? How fast will a pair of HP boxes step?

Alan

[woodchips]

Yes.

[Humber888]

This just posted on the HP/Agilent/Keysight Test Equipment Forum. The part about 8640B hybrids could be useful. I haven't checked further about postage to the UK, etc.

Sweeper Scales, 8640 spares and other hard to find HP goodies at stuff day...
From: walter shawlee
Date: Tue, 23 Jun 2020 14:45:23 PDT

We uncovered a lot of useful HP repair bits including 8620 and 8690 sweeper scales, 8640B power hybrids, 3400A spares, and some amazing counters, all went up on the stuff season page. There's an 10811A equipped and cal'd 5334A counter. and a fluke 8506A true RMS DMM. Lots of wonderful HP, Tek and Fluke goodies to browse through. see everything here:
https://www.sphere.bc.ca/test/stuffday.html

all the best,
walter (walter2 -at- sphere.bc.ca)
sphere research corp.

[G0HZU_JMR]

How fast will a pair of HP boxes step?

GPIB can be really fast but I think the bottleneck for the 2019A will be the loop BW of the main PLL. This is typically 150Hz. At a crude guess this will limit the retune/settle time to 20ms a step for smallish changes in frequency.

This would mean that a 100 step sweep would take 2 seconds. It might be possible to try for a faster speed if you accept that each tuning step won't be fully settled before it retunes again. Newer sig gens offer an internal sweep function that can be set up via the front panel but often the fastest tune time per step will be about 10ms.

[Alan_G3XAQ]

Originally Posted by Humber888

This just posted on the HP/Agilent/Keysight Test Equipment Forum. The part about 8640B hybrids could be useful. I haven't checked further about postage to the UK, etc.

Thaks for the link. Yes, the LSI output amplifiers are generally unavailable so s/hand at $75 they are a good buy. But post/VAT/RMhandling will make them dear.
I'm just back from collecting my third 8640B so I hope the scrapper will donate one of these spare output amplifiers.

Scanning the website displays lots of desirable stuff. If only I lived in that part of the world.

Regards, Alan

[Alan_G3XAQ]

Originally Posted by G0HZU_JMR

GPIB can be really fast but I think the bottleneck for the 2019A will be the loop BW of the main PLL. This is typically 150Hz. At a crude guess this will limit the retune/settle time to 20ms a step for smallish changes in frequency.

This would mean that a 100 step sweep would take 2 seconds. It might be possible to try for a faster speed if you accept that each tuning step won't be fully settled before it retunes again. Newer sig gens offer an internal sweep function that can be set up via the front panel but often the fastest tune time per step will be about 10ms.

Not sure if the HP8662 David mentioned will step any faster. 2 sec/scan is OK for stuff that just sits there but if I want to adjust pots/trimmers on the DUT it's not terribly convenient. I assume you'd scan/store the 2568B in discrete steps, to match the sig gen, and so get 100 discrete points on the screen? I don't have a feel for how it would be used in practice.

73, Alan