Originally I was looking around a well known auction house for a digital Spectrum/Network analyzer from one of the older Tektronix 2715 or HP8566 series to extend my measurement rack to higher frequencies, but they are hard to find for a reasonable price in relation to the risk of buying a device in unknown condition. Still, I wanted one of the older models, because of their excellent design and repairability. Maybe without the exception for some very special, custom parts – but we’ll just hope that those don’t break.
Also, in my understanding a slightly older system from the top series at the time still outperforms most more expensive, modern, all-digital-hey-we-compensate-all-the-errors devices. That is not to say that digital processing and compensation of systematic errors is bogus, of course! But at the same time, any weak measurement hardware can be made to appear top-class by taking several thousand complete sequences and averaging. Getting it right on a single try is an art for itself, and designing a combination of precise hardware and just the right amount of post-processing is the reason for the price. Or maybe I’m just a sucker for retro tech, with all its edges, heavy metal and shiny parts.
So, I finally got a fair deal on this 8565A unit. I might have wanted to choose its bigger brother 8569B instead, which has a wider external mixer span of up to 115 GHz and a digital control interface, but the LED readout certainly adds a special flavor to the set. The seller had informed me that the device would be uncalibrated and the sweep time selector didn’t work anymore. Usually such estimates contain some tolerance, so I already expected some other things that maybe nobody noticed. Since my original plan was to recalibrate whichever device I got anyway and fix all the problems over time, that would be okay. The fixes will be documented here.
The first and most important thing, which you also would not get right away for modern devices, is the service manual. In this case there are two variations:
The first is the standard operating and calibration manual while the second also contains the schematics, block diagrams, parts list and engineering comments. The sheer amount of info inside is incredible, down to detailed explanations of the exact functionality of each module, why it was designed like this and how it is supposed to work in combination with all the others. I used the shipping time to study the circuits up front.
When the box arrived (it was shipped in the original flightcase, padded with thick foam), I was shocked by the weight of 31 kilograms. It survived the transport in pretty good shape, in contrast to my worst nightmares. The only immediately visible damage done was a broken handle, probably because the case protects everything by soft foam EXCEPT the handles. Both handles are removable options and I want rack-mounting options anyway, so maybe I’ll fix the grip for cosmetic purposes at some later time. Should be possible since both are solid cast aluminum and have plenty of “flesh” for some screws. Actually, the whole device could be a cast block, wouldn’t change the weight at all. Its frame is made of thick and sturdy cast aluminum and steel bars, made to last for decades.
Prior to the first turn-on, I inspected the device more carefully. Took off the top and bottom covers and looked for mainly mechanical damage.
The first and most obvious problem was the fact that the CRT tube was sitting loosely inside its shell, rattling around on movement of the case. Yikes! Usually, the tube should be fixed in some kind of a shock mount which allows it to move a little but not to strike any hard surfaces, preventing mechanical impact. The 8565A tube was mounted like that before, but the foam inside the front part of the protective shield had failed due to age, and it seemed like just enough had remained to protect the tube this one last time. It survived without any apparent damage! Pretty mad if you think about a thin, heavy piece of highly evacuated glass sitting in a super-clunky metal box, being hurled around in a truck. Phew!
While poking around inside the case, I next found a small metallic finger-like part:
This is obviously a contact wiper out of some kind of rotary switch assembly – could possibly explain the broken sweep time selector, since that is indeed a rotary switch. Unfortunately, the selector usually contains TWO of these, and for the switch to be dead both need to be gone. Wait a second…I know that it can only have left the case if someone opened the lid because the openings are too small, and that means if one was inside, the second one probably also still is. The big question is: Where? From my experience, in such cases the answer usually goes like “wedged in between the live and neutral mains” or “sticking in the HV supply”, or anywhere along those lines. Never expect loose parts to stay harmless, ever – and if you decide to anyway, try to stay away from flying fragments ;-)
In the end I couldn’t find the damned thing during inspection, which is partly due to the whole case being packed full to the brim with RF components, or maybe it really did get out somehow. Since the most crucial and harmful places were clear, I risked turning the analyzer on in the end, and it proved safe.
On the first look, everything seemed to be in order. The tube came on, all readouts worked – except for the center frequency. The oscillator certainly did sweep a frequency range around DC according to the screen output, so the voltage controlled YIG oscillator had likely survived. Still, the center indicator showed a steady 0.000 GHz. Judging from the manual, this should be an error in the driving circuit. The indicator is an actual voltmeter that shows the selector voltage, which means that there is no way a missing oscillator output will force the display to zero. The first LO is then driven by a circuit that translates this voltage into the necessary tuning currents. This translation is necessary because the LO is usually not identical with the received RF frequency for filtering reasons. On a hunch I opened the top lid again and pulled all the modular boards again after cutting power. The connectors were clean and shiny, but something seems to have been loosened during transport, because after replugging of all cards and reachable plugs the center selector worked just fine across all bands.
So much for the past, let’s check what works right now and what doesn’t:
The sweep time selector is obviously dead, probably for mechanical reasons. Fortunately, HP designed the device in such a way that if all contact wipers in the switch are lost, the timing circuit will switch to AUTO mode as failsafe. Could be worse. AUTO means that the device selects the ideal sweep timing from a table, which is usually ok. In some cases it becomes a nuisance, but this fix has a somewhat lower priority.
Another part that seemed to have become stuck is the RF attenuator. If -10dB is selected, no relay clicks on inside and the real attenuation is actually 0 dB, which can become dangerous for the first mixer. Also, all odd attenuation levels that employ the the -10dB-segment are also wrong. This may be due to a contact problem because the attenuator selector is built the same way as the sweep time selector, and needs to be fixed first.
In conjunction with the attenuator, something is wrong with the REF LEVEL readout. This is given on the dial itself by a very clever mechanism, and also on the screen bezel via LED modules. The two values differ by 10dB and the dial shows wrong values if the highest attenuation and lowest ref level is selected, so this is just a misalignment of the dial.
The REF LEVEL setting also shows weird behaviour. Each step of the dial should increase or decrease the trace amplification by a fixed 10 dB. For the first +50, amplifiers are inserted in the IF section while for +60 and above an offset is applied to the logarithmic amplifier. This is handled internally according to noise figures. The gain stages are composed of one +10 dB and two +20 dB circuits where the second +20 is enabled only for gains of +40 and above.
Have a look at Fig. 5. To take this picture, I enabled persistence display, noise averaging and applied no input signal, which results in a horizontal line. I then stepped through the REF LEVEL settings in WRITE mode and switched over to STORE mode. The 0 dB gain line is just below the visible screen area, and you can clearly see the wrong distance between the +30 and +40 dB lines. This happens only for this one specific step, which hints at a misalignment of the second +20 dB step gain amp.
Finally, one of the two 50 Ohms termination plugs for the mixer and 1st LO BNC connectors was missing. Can be bought, no big deal.
With these exceptions, the condition of the device is pretty good for its age. It probably needs a complete recalibration soon, which depends on when I can get access to the necessary precision references, and how I get them close to this mother of a beast. For now I will concentrate on making it fully useable.
An extra that I want to try and add in the future is a 10 MHz reference input. The 8565A has an analog PLL that can lock the first LO to an internal 1 MHz crystal reference to stabilize it for precision measurements. This crystal takes a long time (~30 minutes) after turn-on, so the lock will drift some for the first ~30 minutes. Since I own a Trimble Resolution T frequency standard module, I want to be able to at least lock the 1 MHz oscillator to the external 10 MHz reference as it is standard for most modern units. At this point, a modern digitally controlled PLL really would come in handy, but that brings its own problems. The whole issue with the 8565A is even a little more complicated than I can explain right now because of the technique used to achieve the locking, but it should be possible. Mostly without extensive modifications to all original circuits I hope, since I really don’t like to mess around with such a beautifully clean design. Fortunately, two of the internal HF mounting spaces near the back of the case seem to be unoccupied, so I could design my own plug-in module. Or I manage to get my hands on the 300/100 Hz IF filter option modules that are supposed to plug in there.