|
Progress!
First revision boards are ordered now.
I found some other prescaler part number datasheets that mention the spurious count thing and suggest a 10K resistor to VCC from the unused channel for a couple mV of bias (and the resulting decrease in sensitivity).
So I have made a provision for a resistor and also a capacitor for good measure (extra pads to play with 'just in case').
Once the boards arrive, I'll try this and that and have a play around to see what happens.
I'll need to get a function gen that is good for a bit more than 3GHz (my gear currently tops out just a bit above 1GHz with a Tek SG504) so that's on the 'gear acquisition list' now too.
I think I should be able to use my 8481A/E4418B power meter to roughly characterise the input response at different bias voltages too.
Jared.
|
Ok, so I whipped up a test board tonight as I got the PCB's delivered yesterday.
One small hiccup... I accidentally ordered an older revision of the PCB by mistake (because I'm an idiot haha) so I had to get creative to get some of the parts to fit.....
The good news is that it is now working! :D
So far, I have only tested it to a tiny bit over 1GHz from my Tek SG504 levelled Sinewave Generator, but using that and my SG503 levelled sinewave generator for the lower frequencies, I have verified that it currently works from 15MHz up to 1.05GHz over the entire input voltage range of 0.5V to 5.5V peak-to-peak. Not bad so far...
It does exhibit the expected spurious readout when there is no input applied, but playing with the bias on the prescaler chip inputs is a job for tomorrow. It's late, time to sleep happy with tonight's success. :)
Jared.
|
So, a little testing was done. I think I need to do some alterations to get this thing in spec.
I had a play with a simple trimpot from Vcc to the invert input, but it severely reduced the sensitivity. The best I could get was about 765MHz to 2430MHz. I think there's some magic in the ... I'll have to figure it out (Unless anyone has some insights?)
I also used the tracking generator on my Siglent VNA that is good to 3.2GHz as a function generator to test the input sensitivity without the trimpot or any input bias. (Set VNA span to 0 and the center frequency becomes your frequency output setting).
I hooked it all up with my E4418B power meter and 8481A power sensor to see what dBm level the counter display became unstable and made this graph:
I'm thinking that I might need to add in a third stage of amplification to boost the signal to the prescaler at the higher frequencies (I think the prescaler gets a bit deaf up there)?
Unfortunately the highest bandwidth scope I have is a TDS794D at 2GHz, so it's a little hard to scope the input to the prescaler directly to see what's going on. Any ideas on how to see the signal level there?
Jared
|
I will try to find some time to revisit my board and make a graph of its sensitivity vs the Agilent and we can compare.
I am using the MC12079 prescaler and two Mini Circuits amplifiers so it is a bit different.
Never had a real prescaler board from HP/Agilent to compare so not sure if the OEM board has the input problem, but I would guess not.
I had added a comparator and 74LVC1G17 buffer after the prescaler chip and that may be part of my problem.
The 53131 does not need the 5V signal, the output of the prescaler chip should be enough, so I will jumper those out and see what happens.
ed
|
I've added another amplification stage to my design, and replicated the HP bias network too to see if it helps with the random counting with no signal.
It's not specifically designed for my chip, but at least it'll give me a start to getting the design figured out.
I just wish I could find an app note or something on this bias network in this application...
I'll send off an order to the PCB fab house soon and see how the performance improves.
I'd love to get my hands on a genuine HP Opt 03 board, but they cost quite a bit, and then once I have it, well I don't need to make my own... :D
If one does come along, I'd love to make a bunch of measurements to see how it really performs across the frequency range.
Jared
|
I swear, every time I publicly say I can't find something, I immediately go ahead and find it. :D
Two app notes on suppressing prescaler self-oscillation by biasing the inputs...
I'll have more of a poke at things tonight.
Jared
|
Thanks for the links.
I use a similar bias scheme but I do a pull up on one input and pull down on the other with a resistor between biasing the inputs around 2.5V point.
I now see that I may not be able to arbitrarily bias about the half way point.
I will have to check the input pin voltage without the resistors and set the bias around that point.
ed
|
Hmm, I'm getting to the point I'm going to give up on this now. I've spent too much time chasing my tail and I'm just not smart enough to figure it out.
I can get it to respond from 20 MHz up to 3.2 GHz without any input bias and it operates fine (if a bit deaf compared to the HP design, as per my previous messages), but then it has that free-running thing with no input signal.
If I use any sort of bias on the input (22K bias in the following example, the internal resistance from the input pins to Vcc is 7.5K) the random counts stop but the maximum frequency the prescaler operates to is only 2.4-ish GHz (minimum is still less than 100 MHz, so that's fine), BUT it seems there is some weird behaviour from the prescaler chip.
At a number of frequency ranges (shown below) it outputs roughly double the frequency that it should.... (changing the amount of bias just shifts the problematic frequencies up or down)
190 MHz - 195 MHz
224 MHz - 240 MHz
275 MHz - 305 MHz
350 MHz - 415 MHz
450 MHz - 495 MHz
I just don't know enough to figure this stuff out, so I'm quickly coming to the point of not continuing to waste time and dropping this project.
|
Are the two input chips identical?
Is it reasonable to expect the two chips to behave similarly if different?
Could it be that adding the resistors somehow adds enough capacitance to mess up the frequency response?
You're using surface mount resistors, right?
resonant circuits anywhere?
How is the prescaler driven?? Could that make a difference?
I am NOT an RF type, but there might be some validity here.
Harvey
toggle quoted message
Show quoted text
On 10/31/2023 12:19 PM, Jared Cabot via groups.io wrote: Hmm, I'm getting to the point I'm going to give up on this now. I've spent too much time chasing my tail and I'm just not smart enough to figure it out.
I can get it to respond from 20 MHz up to 3.2 GHz without any input bias and it operates fine (if a bit deaf compared to the HP design, as per my previous messages), but then it has that free-running thing with no input signal.
If I use any sort of bias on the input (22K bias in the following example, the internal resistance from the input pins to Vcc is 7.5K) the random counts stop but the maximum frequency the prescaler operates to is only 2.4-ish GHz (minimum is still less than 100 MHz, so that's fine), BUT it seems there is some weird behaviour from the prescaler chip.
At a number of frequency ranges (shown below) it outputs roughly double the frequency that it should.... (changing the amount of bias just shifts the problematic frequencies up or down)
190 MHz - 195 MHz
224 MHz - 240 MHz
275 MHz - 305 MHz
350 MHz - 415 MHz
450 MHz - 495 MHz
I just don't know enough to figure this stuff out, so I'm quickly coming to the point of not continuing to waste time and dropping this project.
|
Am 29.10.2023 um 16:15 schrieb Jared Cabot via groups.io: Unfortunately the highest bandwidth scope I have is a TDS794D at 2GHz, so it's a little hard to scope the input to the prescaler directly to see what's going on. Any ideas on how to see the signal level there? Use a down converter. If you have a -hp- 5257A Transfer Oscillator or similar, you can use that to downsample the HF to below 100MHz. Bernd
|
On Sun, Oct 29, 2023 at 04:15 PM, Jared Cabot wrote:
Unfortunately the highest bandwidth scope I have is a TDS794D at 2GHz, so it's a little hard to scope the input to the prescaler directly to see what's going on. Any ideas on how to see the signal level there?
I guess you don't have a wider spectrum analyzer? Raymond
|
My spectrum analyser is a Siglent SVA1032X, good for 3.2 GHz.
I'll have to get a down converter for my scope...
It's all SMD (I even mounted the resistors upside down to improve performance) and the entire signal path is 'coplanar waveguide with ground' at almost exactly 50ohms.
I haven't completely given up just yet, but I took the cheater's way out... I just ordered some MB510-PF prescalers from China, the same part that HP used. :D
I'll replicate the HP bias network, with the only main differences in my design being the 5V LDO regulator rather than a zener regulator, and the RF amps/diodes being substituted too as they are obsolete and my selections all seem to work fine.
I have just re-laid out the PCB for the new prescaler, so once I've slept overnight and looked at it again tomorrow with fresh eyes and fixed any stupid mistakes, I'll send the gerbers off to get the PCB's made and wait for the prescaler chips to come from China.
At least if I get a closer replica working to HP specs with a HP specified prescaler, I can use that as a baseline for further tweaking if needed.
One other question. I'm not sure if I'm doing it right, but I tried measuring the SWR of the input using the SWR function of my VNA for fun. Does this look correct/reasonable/etc or does SWR not matter for this application? I'm still learning what I'm looking at here..
Jared.
|
And just for funsies, here's the latest layout, using the MB510-PF chip and HP designed bias network as per the HP original design.
Let em know if there are any glaring errors. :)
Jared
|
Would it be a better layout if U1 and U2 were rotated to straighten out the traces?? No idea, but it's something that occurs just by looking at it.
Harvey
toggle quoted message
Show quoted text
On 10/31/2023 1:37 PM, Jared Cabot via groups.io wrote: And just for funsies, here's the latest layout, using the MB510-PF chip and HP designed bias network as per the HP original design.
Let em know if there are any glaring errors. :)
Jared
|
The MB510 should give more predictable results, having much more uniform frequency response than the UPB1506.
If tweaking bias and input frequency response tricks just don't cut it even with the MB506, and you must have no output (can't just ignore it) when there's no input signal, there is another option. This would entail having an RF signal level detector in the amplifier chain, and a comparator circuit that overrides the prescaler's bias so it can't respond unless the detector says there's enough input level to work right. The prescaler can then be run in stock mode, with maximum sensitivity, whenever it's enabled. This of course adds another set of considerations like what trip limits apply, and how accurately it all must be determined, and the added circuit complexity.
HP apparently got the original system to work adequately to avoid this nuisance response (right?), and meet some sort of specs - I'm not familiar with this model so don't know if these are true. Anyway, if they did, then you can too, but with the grief of having to use some different parts. I doubt that the original is extremely clean and flat and totally free of instabilities, but just enough for whatever the specs are.
Ed
|
Why two sets of diodes before the prescaler?
I would think that two diodes after the resistor should be enough.
The pair after the amplifier have no limiting.
Also agree with the comment about straightening out the traces on the amplifiers.
A question, are you using a 0.032 (0.8mm) thick board or other thickness than 0.062?
ed
|
The diodes are there because that's what HP did. I would also like to know the exact reason they did that too. :) It seems to work, so I'm sure there's a good reason for that layout.
The Coplanar waveguide specs are as follows:
PCB thickness = 1.6mm
Conductor width = 1.5mm
Conductor gap = 0.28mm
So that makes the trace Impedance = 50.1 ohms according to the Saturn PCB Design Toolkit.
Is there a benefit to going to a thinner board for signal integrity etc? I designed around 1.6mm thickness as it's the most common size (so cheapest, easiest, and quickest to get from China).
I'll have a go at straightening those traces too.
|
Jared, as I recall, the modern replacement diodes you're using are quite a bit more stout than the originals HP used. The only reasons I can picture for the doubled-up ones by HP is first, at the input, to handle some input RF overload protection level spec. Any ones deeper into the circuit could be limiting to protect the individual amplifier stage IC inputs at various points - I don't know what their specs are, but if they can output much more than the next stage's input rating, this could be an issue, requiring some diode clamping. The diodes can help to provide some degree of leveling too, but only where the level is already high enough to activate them. This would tend to flatten the overall frequency response, at least at high input levels.
Depending on your diode ratings versus the original, you can probably eliminate the double-ups, and the associated extra capacitance, which should speed things up.
Good luck. I know you'll make it work.
Ed
|
I don't know what HP's original layout looks like, but here's one possibility for the doubled-up diode clamps: Note that they're not directly in parallel -- they're separated by other stuff. If that other stuff is intentionally made to look like a segment of slightly higher impedance line (and therefore inductive), then the diode capacitance gets more or less absorbed into the T-line, rather than looking like a lumped capacitive load. That way, you can get substantially better protection without killing bandwidth or SWR.
Don't know if that was HP's reason, but I've used similar protection strategies in a lot of designs. It's hard to get good ESD withstand at mm-wave freqs without doing something like this.
-- Cheers
Tom
toggle quoted message
Show quoted text
On 11/1/2023 9:03 AM, Jared Cabot via groups.io wrote:
The diodes are there because that's what HP did. I would also like to know the exact reason they did that too. :) It seems to work, so I'm sure there's a good reason for that layout.
|
Good point Tom. The schematics don't necessarily indicate the fine details of circuit structure (especially at HF) and operation. I guess there is no picture of the original HP board layout? If one is available, it would help a lot.
Ed
|
Jared Cabot