HP437B with a new microcontroller system

[diodak]

I have an HP437B meter. I don't have a probe for it. So I decided to modify it to be able to use an external probe on AD8307 (and similar). The assumptions behind the modifications are as follows:

• possibility of easy return to the original version
• use of the original keyboard and display
• support for the analog part to use the mode as in the original probe
• no HPIB support but data output to serial port
• additional OLED display in place of the tilt indicator (which in my case is damaged)

I also plan to make an equivalent of the HP848x series probes (based on a diode detector). I'm not interested in the range beyond 1GHz (at least for now). The board connecting the digital and analog parts has already been designed and I am waiting for delivery. For convenient connections, it is made of 4-layer laminate. By default, for the new control system to work, you need to reconnect the keyboard and display connectors. The rest on the IDC34 connector may be transparent. The system will consist of two uPs - AVR64EA32 and ATTiny414 for keyboard operation.

[diodak]

For the purposes of adaptation and checking the meter itself, I made a circuit corresponding to the HP8484A probe. I read some documentation about this probe and measurements performed in a similar way (Marconi 6290). There are many indications that crosstalk (symmetry) of the 220Hz signal driving the FET gates will be a problem.
I have also seen projects using repurposed probes as calibrators. As I already know, most problems will be at the ends of the range (from the background/noise side). But I'm really curious what the meter will show if I connect such a probe instead of the original one.
Will it pass autozero and what will it show for a 0dBm signal (with attenuator)? I think this will be instructive before I start implementing the new controls.

[diodak]

My first tests are behind me. I assembled the HP8484A probe simulator. The effect is that the HP437B shows a level of approximately -46 dBm. Everything indicates that there is some voltage on diode D1. Shorting C3 or TP1 to ground (signal) causes the meter to read approximately -65 dBm. What could be the reason? It doesn't seem to be about thermoelectric effects. Initial tests of the HP437B show that there is no damage and the analog path works.
There are many indications that the problem lies in the Q1-Q2 system, i.e. the circuit that converts direct voltage into alternating voltage. Or maybe in such a system (built this way) it is impossible to go below -50dBm?

[ch_scr]

Have you tried it with a known level input? I assume you've used not the original HP diode - so the forward voltage may be different. Not sure what the consequences of that would be.

[iMo]

It depends on the driving voltage levels for the Q1 and Q2 - there is the gate-source diode which may leak when forward biased and you will get a voltage on that input diode. Btw the 1N5711 is one of the best diode for the input detector up to 1GHz (HP5082-2800 eq. afaik), imho.

[Kleinstein]

The electrolytic capacitors C4 and C5 could be an issue. They have quite some dielectric absorbtion and this way can hide some charge for a time much longer than the visible RC time constant suggests. There is also a possibility that diodes react to light and produce current this way.

[diodak]

Generally, in this version, the meter will not perform zeroing or calibration. An input short circuit, i.e. a level of -65dBm, allows the meter to perform zeroing. Therefore, the indications differ significantly from the input power.
I checked for a possibly light-sensitive diode - it was one of the first things that came to my mind. No effect.
There may actually be something wrong with the electrolytic capacitors. They are all of the tantalum type. However, one of them at the input (4.7uF  - black in the PCB photo) is new but with an old production date. And that will be something I will weld.
However, a version with a diode connector in the FET transistor is also possible. And I guess we will have to use a different transistor (currently it is according to the MMBFJ111 diagram) to see if it will change anything.

[Kleinstein]

MMBF111 are a bit strange choice. They require quite some voltage to turn off and that voltage can also scatter quite a bit. Some may be to high in the threashold. I would normally expect lower threshold JFETs, more like MMBFJ113 or maybe MMBFJ309 for lower capacitance.

I don't fully understant the target performance of the circuit, but the circuit looks odd in some places. E.g. C4 and C5 as electrolytic at the input that would make the response of the probe slow. The chopper part looks really low impedance as if going for extremely low noise. The tendency is for low resistance choppers to also produce more bias current and in this case this would mean a shifted zero.

[diodak]

Capacitors C4 + C5 are equivalent to 8.2uF from the HP8484A probe diagram. In Marconi 6920 there is 4.7uF but the chopping frequency is 925Hz and the amplitude is from 0 to -5V. Removing one does not change the background level. Removing both results in the message "please zero" and, after execution, "can not zero". It is similar with only the added ceramic 100nF.

Further tests remain towards a sampling system but with other N-JFETs as keys. I ordered a few types for testing and we'll see.

[ffranzini]

Comparing the original 8484A schematic seems you have direct connection between guard (J) and connector body (ground), this may explain the behaviour of your probe.

Best

Francesco

[Bud]

I'd check the output stage gain. T3 is a half of the amplifier stage, the other half is in the meter. Marconi manual says gain of this composite stage has to be 1000. Incorrect gain may cause zeroing problems.

[diodak]

Comparing the original 8484A schematic seems you have direct connection between guard (J) and connector body (ground), this may explain the behaviour of your probe.

Best

Francesco

The metal housing of the M16 connector is connected in the probe to the entire probe body. And point J is connected to the M16 plug housing through the cable shield. The signal ground against which the voltage is sampled is a separate signal (point E). I think this is how it is in the original layout. Only guard connections are made on pins J, M, F and in the original, as far as I can see, they are connected to the guard paths around the Q1/Q2 system. In my simulator I abandoned this circuit. I don't have detailed photos to see what these paths protect. That's why I left them out. In general, the entire circuit Q1/2, R3, R5, R2 and R4 is a hybrid. So that could also have an impact on the whole thing.

I'd check the output stage gain. T3 is a half of the amplifier stage, the other half is in the meter. Marconi manual says gain of this composite stage has to be 1000. Incorrect gain may cause zeroing problems.

In HP8484A it is even worse and this circuit has a gain of about 1700. And here is the question: is the gain of Q3 (h21) important in this circuit or are the resistors in this circuit responsible for the entire gain?

The new JFETs have arrived, so I will test how they affect the zero level.

[ffranzini]

still unclear to me, comparing, with the original schematic J and F are connected together on the probe side, they are used to guard the sampling section of the probe, this guard must be floating and connected to ground only on one point to avoid any ground loop, in your schematic J is tied with E (signal ground) and with F trough C2.
If i remember right the magnitude of DC signal at -50 dB is around 9 uV then any issue on loops, layout, etc. will reflect in a severe degradation of the performance.



 
 

[diodak]

still unclear to me, comparing, with the original schematic J and F are connected together on the probe side, they are used to guard the sampling section of the probe, this guard must be floating and connected to ground only on one point to avoid any ground loop, in your schematic J is tied with E (signal ground) and with F trough C2.
If i remember right the magnitude of DC signal at -50 dB is around 9 uV then any issue on loops, layout, etc. will reflect in a severe degradation of the performance.

It's true, there is an error in the marking. Currently, the cable shield is connected to J. From the analysis of the PCB photos in the early versions of the HP8484A, there is probably a capacitor between J, F, M and the probe housing. Similarly, A is connected to J, F, M. In newer diagrams this connection is no longer present. In any case, in the HP437B socket, pins A, J, M and F are fastened together with a cable and connected to the housing screen.

But there is also new good news Replacing Q1/2 with MMBFJ113 significantly improved the situation. The background level is -60.5dBm. However, when we choose measurement range 5, i.e. -50dBm, the reading is approximately -51. In auto mode, the reading does not always go to a lower range (6 or -60dBm). Zeroing is still not possible. Similarly, calibration for -30dBm (correction of the reference signal to -27dBm already allows calibration). But the measurements are already as follows for signal/reading:
for a -50dBm signal, the reading is -48.5
-40dBm gives -42.5
-30dBm gives -33.1
-20dBm gives -23

I may experiment with MMBFJ309 and MMBFJ201 when I assemble the second PCB.

[Kleinstein]

The circuit looks like a relatively low impedance chopper amplifier. So probably no need to try the MMBFJ201 - that one is low capacitance higher resistance and would likely be a bit too noisy.
When using a J309 one would likely need smaller capacitors for C6 and C7.  C7 is anyway a bit strange, maybe a bit large. The capacitance could be a point to trim the input bias and this way the zero point.

Another point maybe worth trying could be using MLCCs instead of electrolytics for C3 and C4 and maybe also C10.

The capacitor type and value for C11 can also be an issue. The value may well effect the gain and X7R are known to drift over time and temperature.

[diodak]

Thank you for your tips 

C6 and C7 are selected according to the service manual and may not be there, or even up to 10pF there may be one of them. I currently don't have these capacitors. Soldering a 4-14pF trimmer for C7 allows you to reduce the noise to -65dBm, but zeroing results in an error very quickly (twice as fast as without C7). C11 is actually currently 100nF X7R. Its short circuit also allows the noise to go below -70dBm.

Or maybe the gate current is important in this system? The currently tested JFETs had a max of 1nA and the MMBFJ201 has a max of -100pA.

[Kleinstein]

The gate leakage specs are more like test limits. The typical gate leakage is much lower, often less than 10 pA if the temperature is not high.
I would be more worried about the base current of the BJT Q3. The transistor choice looks odd. I would have expects a high gain audio part, not an RF part with relatively low gain.


If I understand the circuit right the 2 JFETs work as switches for a chopper amplifier with low input impedance. C11 and the 1.33 K in parallel provide quite some load to the rectified part.
This different from a normal voltage input chopper amplifier which wants high impedance and low capacitance after the chopper part.

[diodak]

Before further testing, I decided to change the PCB a bit. I added protection circuits around the gates. I also changed the shape to fit the Gainta G0123 aluminum housing.

Whether this will be better I don't know. With all this learning, an extra 5USD doesn't make a difference anymore and I hope it will bring something.

[diodak]

The next version of the probe has arrived, the housing is ready. It remains to test the modified system

[diodak]

Some time has passed, but the work has not stopped. The new, next version of the PCB did not offer much new information. The background level is approximately -49dBm.

Suspecting that maybe there was something wrong with the element values, I decided to simulate the system. This showed that the choice of NPN transistor is not really sensitive. However, in the sampling system, the best choice will be MMBFJ113 or MMBFJ309. Having already soldered the MMBFJ113, I decided to investigate the issue of grounding. And here I noticed great sensitivity. However, such conclusions only came to mind when I routed the 220Hz signal not via a cable to the probe, but completely separately. The cable currently used is 10x0.14mm^2 with a screen. I ordered a special cable with 3 pairs of 0.14mm^2 each in the screen and additionally 3x0.14mm^2. I'll check it out when I get it. I don't know what is in the original E9288A cable. Some diagrams suggest two beams in the shields, one for the amplifier input/output and the analog ground and zero level. The second one is for the 220Hz signal. I still have the option of testing the Ethernet cable. Perhaps running a 220Hz signal through a twisted pair will help.

In any case, everything from the measurements indicates that the problem is signal penetration from the chopper amplifier. Therefore, the background/noise level is too high and without an RF signal, actually the entire HP437 amplifier block is close to overdrive (at the most sensitive range - when the signal attenuators do not work).

[diodak]

I have the impression that this circuit cannot be copied to an acceptable level... I have completed further tests and replaced the cable with the one mentioned earlier. Another N-JFET tried. And the input noise is still too high. I present oscillograms from points TP15 and TP17 (the first amplifier and the amplifier with regulated gain, respectively).

The measurements concern the probe diagram, in which I have marked the elements not installed with a yellow frame (currently the N-JFETs are BSR57, capacitors C5, 6, 16, 17, 18 are not installed).  As you can see, shorting the C3/C4 capacitors eliminates the noise. So this signal is the limit to go below -50dBm. Behind the first amplifier there is a VGA with 101x gain, then a block of band amplifiers and the effect is that in TP4 (the output of the band filters, before the synchronous detector) we have a distorted signal. And this is without anything at the probe input.

Honestly, I have no idea what to do to make it better...

[diodak]

I did more tests. This time I was interested in the 220Hz generator circuit for controlling the probe - JFET switches. For some reason, this generator was designed very simply, and also slightly asymmetrical regarding the 0V pull-up resistor. Specifically, one branch has a 2.5k resistor and the other 2.613k. I wondered what would happen if I reduced this resistor. I added 4.7k to 2.613 in parallel. This resulted in a faster voltage rise to 0V and thus faster switching of the JFET. The result is a smaller signal on TP4.

To confirm this, I drove both JFET switches from the function generator. And suddenly everything calmed down and the noise disappeared... Just change the phase within a thousandth of a degree and the circuit overdrives again.