Spectrum Analyser: just curious.
 

Here's one for Radio Wrangler! I used to have an HP8568b and was fascinated by it's clever architecture (the IF side chain and 'lock & roll'), I thought it was the work of a genius or possibly a madman.

Here's the question. The 8568 was developed in the late 70's and obviously used the technology available at the time, making it big, heavy and power-hungry.
If it was developed using the same ideas today, what tech would they use to achieve the same result?

I know there's probably no need to resort to such a complex and elegant solution these days what with DDS and heavy-duty digital techniques, but if they did?
Sorry about this daft question, but sometimes I get these spells....

Re: Spec Ans: just curious.
 

The 8568 has a rather clever architecture. The 22GHz 8566 is somewhat more ordinary architecture-wise.
8568 has a side chain which is pretty much a tracking generator and it is this which is locked.

HP lost the plot after these two monumental instruments. It made portable versions but never bettered them other than in improvements to the user interface and such fripperies as more powerful microprocessors for faster operation/interface.

They ran a series of R&D development projects for replacements for the 'Doomsday Box' analysers (R&D project name for them at the time) but none reached production. They all got cancelled, until one project fr an analyser with a DSP implemented IF made it to production. This one had a custom ADC with some special tricks, and a custom DSP chip. However it was found to miss signals which fell into 'sampling holes' and this is a grave problem with a spectrum analyser. R&S got ahead and stayed there once the 8566 and 8568 were no longer state of the art.

Lock and roll stabilises and gives you an accurate centre frequency, but the scan either side is analogue generated and can exhibit a growing error away from the centre. It's good enough for purpose, but you have to bear it in mind. Today, LOs are fully synthesised and give full accuracy at every point, and it saves the lock time lengthening the retrace time. I don't see future analysers ever going back to lock and roll.

However, the phase noise behaviour of a good YIG oscillator is hard to beat given the tuning range. A lot of work went into trying to develop varactor tuned oscillators as an alternative, but this was mainly for more 'cost effective' analysers where the aim was to dodge the staggering cost of a good YIG.

Today, a high performance spectrum analyser will still have a YIG oscillator but in a much more normal synthesiser arrangement. Frac-N synths have spurs or have noise sideband pedestals. Expect a low noise PLL with a lower significance Frac-N loop interpolating steps to get the wanted resolution. Future could see a bank of small tuning range VCOs switched to give the range.

DSP IFs with the ability to speed up narrow bandwidth sweeps by FFT are inevitable. The cost and care needed in those LC/crystal IFs is a strong incentive.

Also, the world has changed. The big money and turnover is on crappy-performance radios inside cellphones wifi bluetooth. There isn't the same market for a ball-crushing overperformer tha there used to be.

The regime within HP/Agilent had also changed. In the past, if you produced something with surprisingly good performance, you were a hero... it's what kept us ahead of the competitors. But then if you produced something better than the planned 'good enough' you got it in the neck, criticised under the assumption that you could have developed something worse, and assumedly therefore cheaper.

Spectrum analysers have also become commodities. Everyone and his father have brought them out. Few are excellent, but all are good enough for some purpose or another.

We're unlikely to see something as outstanding as the 8566/8568 again.

Having one and then not having one, seems like a daft thing to do. One of those boxes you'd always regret selling. Lost in a fire or stolen, or hit by a meteorite would be good excuses.

David

David

Re: Spec Ans: just curious.
 

Fantastic. Thanks very much for that, I'm always interested in stuff like this and could (and have!) spend many a happy hour with a pot of tea reading HP Journals and other write-ups.
Yes, I realise that the side-chain and 'lock & roll' was the answer to the question at the time and that we wouldn't need to resort to it now, I was just wondering how they would do it today if they still had to. It's a moot point I suppose, as the tech available now has made it unnecessary anyway. That and the 'crappy radios' that don't need testing with so much rigour.
Thanks again.

Re: Spec Ans: just curious.
 

There's a photo of a much younger and goofier-looking me in one of those HPJs, though I think I'm not going to divulge which one in public.

David

Re: Spectrum Analyser: just curious.
 

You could look at the block diagrams and service literature for the Agilent PSA and the more recent PXA to see how things are done for the replacements for the 8566/8.

However, one of the key features of the HP8568B was the very low phase noise close to carrier (at offsets from 300Hz to many kHz) when used across HF and into VHF. This was achieved with the complicated synthesiser with a 20MHz comb generator at the head of it all. This gave it an edge compared to the other analysers we had at work in the 1990s although I do remember that the 400MHz Marconi 2382 analyser was also very good in this respect.

The PSA and PXA vary the loop bandwidth of the LO synthesiser depending on the span set by the operator and this can give very low phase noise at offsets of a few kHz or so. The PXA can manage -130dBc/Hz here compared to maybe -112dBc/Hz for the HP8568B. But it could be argued that the PXA is cheating because it degrades the far out phase noise to get that improvement.

So maybe this is the modern method/solution to get close in phase noise about 20dB better than the HP8568B. It certainly looks good on the datasheet especially when the equivalent (joined up) phase noise plot is drawn that plots the best of both loop bandwidths. Impressive and misleading at the same time because this hides the fact that the loop bandwidth changes beyond a certain span. It is possible to fix the loop bandwidth in the menus and when this is done the overall phase noise plot for offsets from 10Hz to maybe 10MHz doesn't look as impressive.

At work we have owned or hired pretty much all of the flagship spectrum analysers of the 1980s through to the latest models and my favourite analyser (to use regularly) is probably the Agilent E4440A PSA analyser that dates back to about 2002.

However, the analyser that gives me the best 'tingle' when I switch it on is my old HP8568B because I know its history and I know what it meant to our company when it first arrived. It was treated like royalty from the directors right down through all the engineering ranks. I felt incredibly privileged to be allowed to use it on a daily basis about 30 years ago and best of all I was able to buy it from the company soon after the Agilent E4440A PSA analysers arrived in the RF labs at work. I have newer analysers here at home so I don't use it much these days. I really do have too many spectrum analysers here at home and some will need to go soon... The main role for my old HP8568B (and an old HP8714B VNA) last winter was as a supplementary fan heater when the central heating started playing up :)

Re: Spectrum Analyser: just curious.
 

One useful trick with the HP8568B is to set it to a span of 1MHz with a large test signal fed at the input. You will then see the wide noise pedestal set by the loop bandwidth of the synthesiser. This noise is typically about 70kHz either side of the carrier at about -112dBc/Hz and it can get in the way a bit.
Then key in a span of 1.001MHz and the analyser will go into lock and roll mode (for effectively the same span) and the close in phase noise will clean up a lot. This can be useful when looking for close to carrier spurious on a 1MHz span as the amount of noise reduction on the carrier is quite considerable when it goes into lock and roll mode. I think there is a front panel keypad combination that can toggle lock and roll mode but I'd have to look it up.

Re: Spectrum Analyser: just curious.
 

If you want to see some local oscillator phase noise performance comparisons between an HP8568B and something like the classic 26.5GHz Agilent E4440A PSA analyser I can show you if that helps?

I don't have the newer PXA here (I wish I did!) but I do have an old E4440A PSA that I could fire up and I still have my HP8568B. The Agilent PSA analyser range was the best we had at work 15 or so years ago. They are a bit long in the tooth now and most people prefer to use the PXA or the MXA or something from the Tek RTSA range. These modern analysers can capture and process >80MHz bandwidth and are suitable for modern digital signals. This old PSA only captures up to about 10MHz and it has limited capture memory.

It represents a crossover between the flagship 8566/8 swept analysers and today's modern signal analysers as it can work both as an ultra high performance swept spectrum analyser and also as a fairly restricted VSA (Vector Signal Analyser) with about 10MHz capture bandwidth.


https://www.keysight.com/en/pcx-x205...0&cc=GB&lc=eng

I have the 26.5GHz E4440A version in the link above and there should be loads of literature on the Keysight website that will include technical descriptions and block diagrams for the YIG based LO system.