The following document was authored by Ed Breya and originally posted on the TekScopes Yahoo forum.
In the 7S14 design, the sampling diode pair is reverse biased by two 1.35 V Hg cells that float along with the sampled signal at the input to the high-z input amplifier, in a megohm impedance environment. Mercury batteries have not been available for quite some time, so a good substitute is needed. The main problem is that Hg cells are very stable, more than any other type, and they have a voltage different from other types, so without knowing the design requirements of the circuitry from about 40 years ago, it's hard to know how precise they actually need to be, and what other voltages may work.
My initial thought was to just use 1.5 V alkaline or silver oxide cells, and increase the strobe drive level to maintain sampling efficiency, but I figured that they would tend to drift around a lot, and still need regular changing - perhaps annually - and eventually leak and make a mess if forgotten for a few years. This still may be a good enough solution, but I didn't pursue it further.
The only other options I could see were to redesign the input amplifier so the biasing didn't need floating sources (not very appealing), or to make the sources with optical power. Since the circuit is high impedance after the sampling gate, it turns out that not much current is needed, so small optical devices seemed possible. The objective was to first get enough power, then regulate it to ±1.35 V, all in a space no larger than the original cells.
Optocouplers seemed like a good start, but you need two (or more) Si junctions in series to get into the 1.35 V neighborhood. Using conventional LED/transistor types, you're stuck with six-pin packages (to access the base), with at least four for each channel, plus any additional circuitry.
It seemed like a few μA was fairly easy to attain, so next I looked for regulation parts. I found some shunt regulators from Maxim that can run on 1 μA (!) but of course only at a few standard voltages, including 1.2, 2.5 and 3 V, so it would still be necessary to use resistive dividers or more complex circuitry to produce the exact voltages needed, but it is possible.
After pondering trying to cram all this stuff in there, I decided to experiment with discrete LEDs, to see if I could get enough voltage using fewer devices with bigger band gaps. I found that high intensity blue and "white" ones can make 1.5 to 1.6 V open circuit, so that looked promising, but none would produce even 1 μA at 1.35 V, so they couldn't even run the shunt regulator alone.
After much experimenting, I settled on a design using one high intensity red LED, driven by one high intensity "white" LED, to substitute for each Hg cell, making 1.35 V, using the intrinsic diode characteristic as the shunt regulator. The only other parts are a bypass capacitor across each red LED, and the resistors etc. to set the driving LED currents.
In Part 2 to follow, I will explain the part selection process, the details of construction, and some experiments on circuit performance.
Design and construction
The forward voltage of red LEDs happens to be in the right range, at low current, at room temperature. I looked at a number of them on a curve tracer and figured the knee region was reasonably flat enough past about 1 μA. The trick was then to get enough optical power into them, and land at the right voltage range. The only ones that seemed to have enough output were high intensity types. The plan was to have the generating LED be nose to nose with the driving LED for maximum power transfer. I found that although using the same type of LED for the driver was potentially more efficient, it was very difficult to align. The best drivers seemed to be "white" LEDs - the type that are actually blue, with a phosphor to generate the whitish spectrum. I think the phosphor causes the light to be more diffuse, yet still very bright, so alignment is not as critical.
The easiest way to pick the generators is to measure them with a DVM that has at least 10 MΩ input resistance, while illuminated with one of those LED flashlights - the kind with a cluster of LEDs. It should be possible to find some that will run at about 1.35 to 1.4 V into a 10 MΩ load. I used one of my good old Fluke 8400As, which has virtually infinite input R in the 10 V range, and 10 MΩ in the 100 V range, to easily see the loading effect.
Now for the construction. This may not be suitable for some, since surgery is involved. Before attempting this, it would be best to wait until Part 3 (to be added since this part is getting too long already) which describes my findings and recommendations.
On the sampler board there are (approx 1/4 inch diameter) holes at the centers of the Hg cell locations. I removed the sampler assembly and drilled 4 holes through the aluminum box for the drivers - one exactly aligned with each board hole, 2 per channel. The LEDs are standard size type about 3/16 inch diameter, so the holes should match, and be a snug fit. I made a small one-sided vector board piece to attach to the bottom of the sampler, with the LEDs passing through it and the box. The LED leads and series resistors then attach to pads on the board for support and interconnections. There is just enough room for this between the bottom of the sampler box and the delay line bracket below it. Alternatively, the LEDs could be glued into the holes, and air-wired to the circuitry elsewhere.
When the sampler boards are remounted, there will be a driver LED sticking up through the box under each of the holes. Then a generator LED is mounted upside down over each one, attached to the pads where the Hg cells were. It is a tight squeeze, but there is just enough total height available to clear the LEDs standing up as long as the leads are bent right in close.
I set up the drivers so each channel has two LEDs in series, then a resistor, and a reverse protection diode, so I could get about 2 to 8 mA through each one, and hooked up to drive from an external supply so I could experiment with it. When powered up, the generators should all be illuminated, and it is then possible to bend them around a little, to maximize and match their outputs. This can be done on the bench - the sampler doesn't need to be installed or powered up.
I added a large bypass capacitors across each generator to make sure there would be no extra effects on the loop dynamics, but I'm not sure how much is actually needed. I had some 4.7 μF high-k chip caps, so in they went. It takes a few seconds to charge them up, but that's less than the CRT filament warmup time, so OK. Since the LEDs are much smaller than the Hg cells, I am sure that this setup has much less stray capacitance than the original circuit.
When all was said and done, I could run the 7S14 in a 7704, with the driver power supplied from a variable external supply, and I had already roughly characterized the generator voltages during alignment, so I could remotely set them from 1.3 to 1.4 V - it was all set for experimenting.
Findings and recommendations
I fired it up and applied a fast pulse input, and had the supply set to make 1.35 V on the generators. Everything seemed to be fine, so of course I had to mess with it. Running from about 1.3 to 1.4 V, it had no apparent effect on gain or risetime, and a small effect on offset - it would shift up about a minor div at typical settings. At maximum sensitivity 2 mV per div, it obviously would move up several divisions as the generators went from 1.25 to 1.4, and you could clearly sense the "knee" of the diode shunting effect - the change in voltage, hence offset, with increasing drive power, was very rapid and then compressed as the regulation flattened out.
Then I tried it with NO drive, and was surprised to see that it still gave a pretty good representation of the signal. The sampling diodes were rectifying the input signal and charging the caps up most of the way anyway! All it did was clip the overshoot and square it off a bit. Increasing the drive, you could clearly see that it started working again as the generators got back up to about 1.2 V - this was with input level about 25 percent above the p-p spec input range. I found the same effects on the other channel.
Conclusion: There is nothing magic about the 1.35 V or its precision. If that voltage changes a bit, the display will move a little, but the effect is probably less than the normal drift of the system anyway. The front panel offset pots have a huge range compared to the drift caused by changes in the bias voltage, and the samplers probably also have a bigger effect. I could not find any spec about this in the manual, so I assume it's either perfect (unlikely) or very loose. The main thing is that the bias voltage shouldn't change too rapidly, or way too far, so anything that hits close enough and is reasonably stable, should work just fine.
Knowing what I know now, if I was to "fix" another 7S14, I would not worry too much about the exact voltage and its regulation. So, I would back up a few steps to my original idea of using conventional LED/transistor optocouplers, but with no fancy regulation schemes needed.
The design would require five commonly available optocouplers (such as 4N26) per channel, with their LEDs in series and a current set resistor from one of the supplies, and outputs (B to C&E) all in series. When activated by around 5 to 15 mA through the LEDs, it should produce about 2.7 to 3 V. This voltage is just right and stable enough, but since it's an odd number of devices, it has to be split by a divider of say 47 k ohm resistors across the middle coupler of the string, to provide the center point. Add a couple of bypass caps to make sure the AC impedance is low, some pigtail leads to connect and support it, and that's it. The assembly could be built on the bench with the couplers lined up end to end, and just chained together on each side (if you think about the pinout you'll see what I mean) with air wiring, and then fitted into the sampler assembly. Three leads would connect to the original cell pads, and two leads would need to pick up power from + or − 15 V and ground. Since my LED version works just fine, I have no need to build this type, but I am virtually certain it will work.
There are also available photovoltaic MOSFET gate drive couplers that easily supply enough voltage from a single device, which would be much more compact, but then it still needs to be regulated back down, and split, adding a few more components. They are readily available new in the US$ 3-5 range. I have a sample one from many years ago, but couldn't find it when I started this experiment, so didn't look to see if it would work, but I think that any that can deliver a few uA or more at 3 V should be enough to operate some kind of shunt regulator junction devices or IC plus a resistive divider. I prefer to use good old LED/Q types (at least for low V, anyway) because they are cheap and plentiful, and pack a pretty good punch - tens of μA, although only about 0.6 V max.
For those who want to stick with batteries, I would recommend using a single 3 V Li cell such as CR2032 (220 mA-hr), which is available with solder tabs, thus solving the attachment problem. It just needs to be split with a high resistance divider (500 k to 1 M) range (and bypassed with caps). The Li chemistry has a very low self-discharge rate, so life would be limited mostly by the divider current, so it may be worth experimenting to see how high the resistance can go and still work. At 1 MΩ, I think the life should be in the six to seven year range for a fresh cell. Also, the battery can be checked without removing the cover by having small holes drilled through it in the right places (don't hit any of the runs), to poke test leads through to to access the ends or other test points.
So, whether you've "seen the light" or want to stick with batteries, there are a number of ways to solve the problem.