Hewlett Packard 740B DC Standard Digital Voltmeter (and 740A)

[Dave Wise]

Here are pictures of my DIY input and output cables.  Hope this works, preview seems broken.

[Dave Wise]

Here's 740B #2 with original jacks for comparison.

And here is my cable pinout.

Oh that didn't edit right, what am I doing?

[Dave Wise]

And a text version of my notes.

000LN was omitted from the CODE LIST OF MANUFACTURERS, but it is

Deutsch
MDR07-7P-090 (input), NSN 5935-00-169-9672
MDR07-7PB-090 (output), NSN 5935-00-410-7256

Mounting hole is 0.685" .

AMP CPC Series 2 shell 11 just fits, flange behind panel,
secured with countersunk flat-head 6-32 screws.
Plug P/N is 206434-1.

TODO: Get one Series 2 Keying Plug to prevent input/output cross-mating.

NEW CABLE PINOUT

SHELL CONTACT NUMBERING looking at contact side of plug

 1 2
3 4 5
6 7 8

INPUT
1  2  3  4  5  6  7  8
G  I- C  O  O  X  O  I+

G = Guard
I = Input
C = Chassis
O = Signal Guard.  Continuous version of O+, i.e. 5 wire between
    A11 and K1A.1 and C4.  Plug has these contacts populated but
    not connected to anything.  This shields pin 8 from the rest.
X = No contact

Cable is triax.
Center       = I+
Inner shield = I-
Outer shield = G
Chassis is a separate wire, bundled to the triax with spiral wrap.

OUTPUT
1  2  3  4  5  6  7  8
G  S- C  J  S+ O- J  O+

G = Guard
S = Sense
C = Chassis
J = Jumper between pins 4 and 7 in plug, energizes output relay in jack.
O = Output

Cables are triax.
CABLE 1
Center       = S+
Inner shield = S-
Outer shield = G
CABLE 2
Center       = O+
Inner shield = O-
Outer shield = G
Chassis is a separate wire, bundled to the triax with spiral wrap.

Cables are strain-relieved at boxes by the common plastic strain relief
used on zip cord, with the projection ground down to accommodate the
larger size of our cable without kinking it.  Smaller size relief on
input (1-triax) cable, larger size relief on output (2-triax) cable.

WARNING: Unless you fit the INPUT jack pin 6 with a keying plug
to prevent cross-mating, if the OUTPUT cable is inadvertently
plugged to the INPUT jack, amplifier output voltage will appear
at the OUTPUT box's SENSE+ terminal even when output is disabled.

[Dave Wise]

I can't remember where I got the triax.  Some eBay seller I suppose.  The outside jacket is smooth and slippery-feeling like teflon.  The center conductor insulation is thin but I'm sure it was designed to take high voltage because it is covered by a conductive coating.  At the plug and box it's fiddly but essential to scrape it off without nicking the insulation.  Using my Fluke 895A as a high-R ohmmeter, I can't detect any leakage at all.  Virtually all of the instrument's 10+ gigohm input resistance is the connectors or inside the instrument.  The latter is partly the A9 input/range board and the S1 and S2 MODE and RANGE switches, partly the amplifier input current.

[beanflying]

Thanks Dave 

[Dave Wise]

I have detailed notes any time you want.

Having accidentally blown up the Germanium Q1 and Q2 years ago, 740B #1 is retrofitted with Silicon, plus mods* to accommodate its different characteristics.  (Currently KSA1010 driving MJ2955 but I want to try TIP42C driving BD912.  I also found some 2N6331's.  I'll drive that one with an MJ2955, similar to HP's own choice.)

* A 1nF/330R in series from collector to base, or remove A7C8; and an RCD "tracking voltage clamp" across T4 primary to tame the leakage-inductance spike.  That's 10uF, 1K, UF4003.

It's a good idea to put a heat sink on A7Q3 if main switch Q1 is Germanium.  A7Q3 has its work cut out for it discharging that big heavy base, and it's the hottest transistor in the instrument.  (With Silicon Q1, A7Q3 barely conducts at all.)
Make sure A7Q2 is working; if it goes open, A7Q3/R7/R9 will fry.

In late production, HP added a freewheel diode across Q1 to prevent transformer ringing from driving the collector positive.  They seem to have used a garden variety rectifier but I discovered that a fast-recovery part like the MUR115 or UF4003 is way better.

740B #1 also has a DIY HV transformer wound on ferrite instead of iron because the original overheated when operated full-throttle overnight.  Its relatively low saturation ceiling forced me to develop a simple mod to the A7 board current limiter which is compatible with the original parts.

I refitted both 740B #1 and #2 with H11F-based choppers, and I recently realized I had matched the modulator parts for the wrong characteristic.  They need to be matched for temperature coefficient.  If this results in a large net offset, it's easy to trim out, by simply inserting a resistor at one or the other end of the front panel ZERO control.  I matched the parts in a toaster oven, 20C-60C, using a Keithley 147 Nanovolt Null Detector to measure the offset.  Out of nine same-lot H11F2's, I found two good matches with less than 2uV drift over 20C-60C.  The mod-demod efficiency increases and it's desirable to reduce AC amplifier gain to avoid overload on the millivolt ranges.  Adjust A3R19 not A3R9, I don't know what HP were thinking there.

While I was in there, I reduced A4 noise by replacing transistors.  I used KSC900.
Beware of the 2N3391 data sheets.  Central Semi, National Semi, Fairchild, and New Jersey Semi call the 2N3391 "general-purpose", but GE (HP's vendor for 1854-0033) is "Ultra High Beta, Low Noise".

The original neon-and-photocell input protection circuit is trouble - the HV cell HP used to disconnect the input generates 10-100uV of offset.  (This is how I discovered the oven was oscillating - offset ramped up and down as oven load affected the light level on the photocell.)  Even if it didn't, the circuit is not usable with H11F's with their 30V absolute max.  I engineered a new circuit based on the Supertex LND150 high-voltage depletion-mode MOSFET which is optimized for current limiting, and a TVS.  There's also an SSR to keep input current out of the KVD in STD mode.  This circuit works better than the original did when it was new.

Years ago I discovered that 740B #1's oven had run away and cooked the PCB inside so I rebuilt it with the two separately-floating circuits on separate boards to avoid leakage.  It's important to keep the reference zener thermally connected to the inner can.  The thermistor was cooked too so I replaced it with a modern part, discovering in the process that control stability can be improved by attaching the thermistor to oven wires instead of the can, to introduce some phase lead.  I still had to reduce control loop gain to keep it from oscillating.  I discovered that 740B #2's oven was oscillating too and reduced its loop gain until it stopped.  The frequency is in the milliHertz; I graphed it with a storage scope while perturbing it with an HP 203A oscillator.

Speaking of leakage, I discovered that T2, the transformer powering the in-guard system, has a lot of leakage between the +34V and -12V secondaries, which messes up low-level readings.  A simple mod reduces the voltage between them and hence the leakage.

When I was doing hi-pot testing, 500V between MINUS and GUARD in VM mode leaked into A2C17 and punctured it.  This proves HP never did that particular test.  Symptom is no reading in DVM mode.
I added a TVS across the new cap.  SMAJ30CA since I was already using that to protect the H11F's.  It can be almost anything, normally the cap has less than a volt across it.

(I've said this one before, I'm repeating it here to put it all in one place.)  Battery leakage messes up circuit operation and must be cleaned.  The battery on A10 can be anything you like.  I used a lithium coin cell with tabs.  It will last for its shelf life.  If you use some kind of power supply instead, it must float 1000V with no leakage.  The battery on A11 should be relocated onto the GAIN CHECK switch.  It can't be 3V lithium as that amount of offset will overload the chopper amp.  I used a silver oxide button with tabs.  It will last for its shelf life.  If it's not 1.35V you need to adjust the GAIN CHECK bogie value.  If you do away with the photocell choppers, GAIN CHECK becomes a pointless function anyhow since the H11F's will last forever.

PROCEDURAL NOTE: When doing INTERNAL CAL, wait several minutes after you switch between 1-9 and 10-12 to allow dielectric absorption in the switches to relax.

[Cubdriver]

Dave, thanks for the very thorough writeup!  I will definitely be going through that again when I eventually get the beast onto the bench.

Much appreciated!

-Pat

[Dave Wise]

Don't rush into removing all the chopper photocells even if your gain is below spec.  Try modernizing just the demodulator side first and see if that gives you enough boost in efficiency.  I think the main challenge will be syncing the demodulator drive to the modulator, unless you punt and drive both from a new circuit.  (BTW I have PIC source code for that.)  Don't try for a final design, just breadboard something to see if it's enough.  (I'm speaking from the hassle I went through matching pairs of H11F.)   If you do essay to install H11F's, the main amp has to be H11F1 or H11F2 for the 30V breakdown.  (The main amp - even when gain = 1 - has to generate 30V to get the expected slew rate out of A4.)  The meter amp can use any H11F including the 15V H11F3.
Be careful with the mod-demod phasing - Figure 3 in the manual is wrong.  On the demod section of A17, V4 is actually lit by DS1 and V3 by DS2.  The chopper amp overall (2 wire to 926 wire) should invert.

Do be suspicious of A9V2.  Mine was generating on the order of 100uV photovoltaic offset.  I've replaced it with an SSR; Ixys CPC1981Y works fine and leakage is negligible.  Standex SMP-1A40 would also be great.  You can wire in the LED in place of A9DS2 if you add series resistance.  I'm sure HP would have used one of these SSRs if they existed in 1965.

[Dave Wise]

I'd like to add one thing.  The H11F1 characteristic that needs to be matched is not offset - by manipulating resistors A10R6/R7/R13, you can null out at least 1000uV.  The important thing is drift.  You want a pair of H11F1's that drift in the same direction, the same amount.  This is even more tedious than offset matching, but I was able to get several satisfactory pairs out of about 20 candidates using my toaster oven as a cheap environmental chamber.

[Dave Wise]

NOTE: Everything in this thread is applicable to the 740A, 741A, and 741B, as well as the 740B.  It may be useful for the 419A Null Detector as long as power consumption stays low.

NOTE: With respect to chopper phasing, the 740B schematics are a confused
muddle, including my well-intentioned annotations.
For now, my definitions: Both Main and Meter.  Physically, cells are
numbered top looking down V1 thru V4 counterclockwise.  V1 is mod series,
V2 mod shunt, V3 demod shunt, V4 demod series.  V2/V3 lit by neon DS2 or
LEDs DS2/DS3, V1/V4 lit by neon DS1 or LEDs DS1/DS4.  DS3 and DS4 are
internal to the H11F1 demodulator switches.

I have new findings.  I am grading my photocells by scoping them against a current source, adjusting illumination so R-lit is 20K, and recording the time it takes the cell to recover to 100K.

Recovery is a straight line at normal illumination, which means that T10 (HP's figure of merit, see AN-919) is meaningless unless you also adjust illumination so all cells have the same R-lit.

Of my sixteen cells (two 740B's), eight look very promising.  I think I can use photocell modulators along with H11F1 demodulators.  The latter's near-ideal behavior lets us do a very important trick.

FINDINGS

1. Asymmetric Demodulation Timing
We must turn on series switch V4 at V1 on and keep it on throughout V2, to route the entire half-cycle waveform to the A4 integrator.  This boosts efficiency by around 30%.
Use a microcontroller.
Four states.  V1+V4, V4, V2+V3, none.

2. Customized Modulator Drive
We MUST tune the light intensity on each cell, which rules out neon.  I suggest poking LEDs down the empty V3/V4 wells.
You should insulate and guard.
With custom drive currents, modulator pulse width is not critical, just set 50% dark ratio (beta) - on both channels - and forget it.

Doing this gives me another big boost in efficiency.  My results are comparable to my estimate of original new-instrument performance.  Don't go overboard, remember HP put lower limits on V1-V2 average series resistance.  (160K on A17 - main, 350K on A16 - meter.)

3. Asymmetric Modulation Timing (partially true)
Bright short pulse V2, dim wide pulse V1.
E.g. A17#1V1 treated as V2.  100uA 5ms is about 10K valley and
it hits 100K about 1.8ms after turnoff.  300uA 500us is also
10K valley, but it hits 100K about 1.6ms after turnoff.
dR/dt is almost constant for a given Rlit, slowing down as the cell is driven harder.
So if you maintain a constant Rlit, you will have essentially constant dR/dt
regardless of pulse width.
On the other hand, a shorter pulse makes room for more dead time,
so it is still useful: brighten V2 and shorten V2/V3, stopping short of
degrading cap charge, and lengthen V1/V4, stopping short
of degrading Rin.

[Dave Wise]

UPDATE.

My fastest photocell is also nearly blind; using a 24V Luxeon 3535HV LED, 3mA is required to get the cell down to 20K.  Using a 3V Cree JB2835, I expected 24mA, but it actually requires 60mA, a factor of 20 instead of 8.

A less-blind cell that needs 1mA on 24V, needs 10mA at 3V, a factor of 10 instead of 8.  This points to a steep drop in LED efficiency as current goes up.

This means I cannot run 3V LEDs on 5V, I must use 24V LEDs driven by transistor switches.  The extra space taken precludes through-hole hand-assembly of a prototype, plus it's not usable in a 419 Null Detector with its 24V supply.  You can probably manage with 9V LEDs, which should find themselves on the left side of the efficiency knee.  (12V-18V parts are expensive for some reason.  The cheap ones are 3/6/9/24/48.)

I'm working on a circuit board layout with surface-mount parts including in-circuit-programmed uC.  There's enough space for everything.  The board will mount on the demodulator side of the old chopper housing, sticking LEDs down the V3/V4 wells.  (Remove the old photocells first.)  The back side of the board is copper fill connected to Guard.  Cover it with a few layers of Kapton or polypropylene tape to insulate from the grounded chopper housing.  The H11F1's mount on edge, with the LED pins in through-hole pads above the V3/V4 wells.  Air-wire the signal pins to the old terminals.

I'm a total noob with KiCad and OSHPark.  Maybe someone can look over my shoulder and stop me from making dumb mistakes I can't back out of. HP-Chopper-Photocell-Driver-first-draft.pdf (17.29 kB - downloaded 29 times.)

[Dave Wise]

I think we have a winner.  Wonder what OSHPark will say.
Now I get to lay out the LED discs that go into the photocell wells.

I tried to make it friendly to hand-solder, with 0805 and 1206 parts.

[Kleinstein]

For the driver circuit it would help to also show the schematics.  For the layout the design rule checks and PCB tools usually do a good job ckecking for major errors.

From what it looks like the optocouplers ouput side is not connected, not even to some connectors, which is odd.


The origianl neon bulb circuit has one hidden function, that may not be obvious: it alternates between the 2 neons and has a little pause in between.


It is kind of normal that the fast LDRs are less sensitive, but there is also a point when the sensitivity gets too low. With very high power from the LEDs one gets increasing levels of thermal effects that can cause thermal EMF. This would not be real drift, but cause a rather slow settling for the meter to warm up.

[Dave Wise]

Thanks, Kleinstein.  I added the schematic to post number 61.

Everything on the board is at or near Guard potential, which may be up to 500V away from the signal pins.  I will solder in only the LED side of the optocouplers.  I'll connect the signal pins to the module terminals with point-to-point wire, to avoid leakage.  I wasn't willing to put in the work of creating a custom footprint for half of an H11F1, so in KiCad I just pretended they were hanging off the edge, with the signal pins set as no-connect.  (The only DRC warnings were "Silkscreen outside board outline".)

The microcontroller program will include dead time.  The original 740B driver has dead time on the Main chopper but not the Meter chopper.  Testing showed that it's of critical benefit to both.

I listened to my blind, brightly-lit photocell with my General Radio 1232-A Tuned Amplifier, and didn't hear a thing.

I worried about those four clearance holes.  The teflon insulators the holes are meant to clear extend right to the edge.  A drill bit would want to bend.  I made them recesses in the outline, so OSHPark can mill them instead of drilling.  I figured out how to make KiCad do it.  It's tedious but I did it.  String together lines and arcs.  It's finicky about the outline making a closed shape, you have to drag line ends together at extreme zoom.

Back copper is a fill zone, connected to Guard.  Have to insulate from the chopper housing.  Kapton tape, polypropylene tape (Scotch "Magic" tape)... how about a phone screen protector?

See any show-stoppers?

[Kleinstein]

Partially mounting of the OK makes absolute sense. The drive level is quite high and this can cause some thermal offsets, that may increse the warm up drift (would not make is slower but more amplitude).

I would assume the LEDs to drive LDRs are for the input modulator and the H11F1 is for the demodulator. The demodulator part is after amplification and thus way less critical. 

I don't hink the H11F1 would need that much current for the demodulator - the signal current is still relatively low.
From the datasheet there is rather little gain from using more than some 5 mA and even 2 mA should be enough.
Anyway the drive current could come from the 5 V supply - no need to start from 24 or 40 V to drive a 1.2 V IR LED. Of cause this does not help with the 5 V derived from the 40 V with a linear regulator.
some 10 mA for the H11F1 would be quite some power from the 40 V rail !
For the 5 V regulation it may with adding more filter capacitance - for just the µC and low power the simple zener may be enough.
Similar the 40 V supply for the LEDs should have a filter capacitor (e.g. , so the extrernal supply would not see much of the modulated current.

For the LEDs the focussing of the LEDs can also have quite some effect. The LEDs for illumination use tend to be wide angle.  The viewing angle can also be different for the 3 V LED.
The Cree LED is spedied for 150 mA, so at 60 mA the efficiency should still be good. Another point could be the color: the white LEDs have quite some red and blue. The red light part may not have any effect and the blue part could be already less effective. The color temperature make quite some difference to the spectrum.  The question would be if a high efficiency green or even orange LEDs could work better. 

[Dave Wise]

Thank you!

You have it exactly right, I'm using a pair of H11F1's on the demodulator side - where offset can be ignored - while keeping the modulator photocells and driving them with LEDs.

With a 3V LED, my observation did not agree with the datasheet; Cree JB2835 needed 60mA to excite the photocell the same as my 24V LED at 3mA.
Or maybe the spectrum changes when you drive it hard, to something the photocell doesn't like.
Anyway, I abandoned that design path.

The LEDs I tested are supposed to emit 120-150 degrees which covers the whole photocell.

During "cruise", the demodulator only carries a few uA, but when slewing, the shunt switch has to charge 1uF to 20V in a couple milliseconds.  The meter loop series switch peaks around 200uA.

Are we looking at the same datasheet?  OnSemi Rev 2 (2021) Figure 2 Output Characteristics shows almost equal-size increases in saturated current for drive at 2-6-10-14-18mA, with 16mA yielding about 600uA.  I gravitated to 16mA because it's the test current for several specs.

Using the -42V rail, current transients are not an issue since the instrument's PWM output pulls amps not milliamps.
That said, if 9V LEDs work, I want to ensure the board can be used on the 419A, so I am looking at the lower voltage supplies, like +16.5 or -16.5 .  And since the board has moved from the main chassis to the (grounded) chopper housing, there's an exciting possibility of running it off one of the in-guard rails like +22 or -22 which would simplify the insulation.  Like -42V, these power high-level signals that don't need to be isolated from transients.  The 419A expects a quiet battery so I will see if I can squeeze an RC filter onto the board.

UPDATE.
9V LEDs are usable!

My blind photocell that needs 3mA on Lumileds 24V and 60mA on 3V Cree JB2835BWT-G-H30GA0000-N0000001, needs
* 12mA with 9V 2700K 80CRI Lumileds L128-2780EC35000B1
* 20mA with 9V 3000K 80CRI Cree JK2835BWT-W-H40EC0000-N0000001

[Kleinstein]

For the datasheet i have lloked at the 2002 Fairchild version. The main difference is whether looking at the on resistance or the the maximum current (e.g. some 2 V). For the current a higher drive seems to help, for the on resistance mot very much beyond some 6 mA. I don't expect the slew rate to be that relevant. The analog meter and input filter are slow anyway. I don't see a need to be faster than some 100 ms to come out of saturation.

If the H11F1 really needs some 10 mA, I would looks for a different solution that does not get as hot. Photomos may be option, though a litte on the slow side (e.g. 0.2 to 2 ms).

Even it the circuit may tolerate some ripple in the supply, it is still worth having supply filtering. The LED current part (especially with unequal current) is in sync with the chopper signal and can thus effect the result. Mains ripple would be far less critical. Filtering is easy (just a resistor (e.g. some 100 Ohms)  and a electrolytic cap (e.g. 470 µF). There is anyway pelnty of voltage to loose on the series resistor.

The spectrum can change a little with intensity, but not very much. The color temperature of the white LEDs can make a much biger difference. From looking at LDR datasheets it looks like they are most sensitive for green to orange light, just the part of the spectrum where most white LED are rather week. So both a green and an orange LED could be an alternative even though the brightness to the human eye may be lower. The extra red and blue has mainly a negative effect in heating up the LDR and possibly causing thermal EMF problems (more warm up drift). The other point with high power is heating up the circuit and this way indirectly causing possible drift from the LED intensity changing with temperature.

[Dave Wise]

Looks like I could fit in a 6.3mm through-hole cap; that's either 47uF/50V or 100uF/35V.   There's not enough room for 470uF.  I don't think 100uF is big enough to make a difference.

In the 740B, +/-22V don't have enough filter cap; we're stuck with out-guard supplies.  Insulate the board to 500V.  Use -42V or string it across +/-16.5V; these tolerate noise.  Either connection can use 24V or 9V LEDs.

The 741B doesn't have Guard.  Use -34V (24V LED) or +25V (9V LED).  These tolerate noise.

In the 419A, disable the neon driver (disconnect XA2 pins 9 and 11) to save on battery.  Use 9V LEDs, and string the new board across +/-13V behind an external RC filter.  Insulate the board to 500V.

In all cases, select the zener series resistor and the LED dropping resistors to get the appropriate current with your supply voltage.  In the 419A, pick a supply voltage, tune pulse widths on an external power supply, then adjust the R in the RC filter to get that voltage.

I tested two eBay H11F3's, one marked GE, the other Harris.  Both far exceeded H11F1 datasheet Figure 2 Output Characteristics current, and U3 would not need 16mA drive.  Put them on a curve tracer, with LED on Base and FET on Collector, and see how much you need to make the switch conduct 0.5mA .

I'll put all this in an Installation Guide, if there are enough people in the world interested enough to use it.

[Dave Wise]

Boards ordered from OSHPark.

[Dave Wise]

Assembling one to see if I need to fine-tune any mechanical interference issues.  The A16 meter-loop chopper assembly is near some HV points on the A11 board.

[Dave Wise]

Board v1.0 has some trouble.  It's fine in the spacious and quiet A17 position, but in A16 you have to bend the connector pins 45 degrees to avoid nearby board A11.  New v1.2 layout ordered.  This one has connectors up high to avoid mechanical interference.  It's also EMI-hardened because A11 emits high-voltage spikes.

[Dave Wise]

Pictures of v1.0 board and how it's installed in the chopper module.  I'm using a scrap of 0.01-inch clear PET film as the insulator.  It was the packaging for some random trinket at the store.  My Fluke 895A reads > 2e12 ohms at 1000V.  If Henkel doesn't warn me off it, I will secure the film to the chopper housing with Loctite "Stik-N-Seal" Outdoor Adhesive, then secure the board to the film with dabs of the same filling the voids around the PTFE turret insulators.

[Dave Wise]

Henkel says, they don't test Stik-N-Seal for electrical, therefore, don't use it for electrical.

I tested it and it's fine.  It's a synthetic rubber in a volatile solvent.  With dielectric absorption settled and about 50% relative humidity, I read about 1e14 ohms at 1000V sustained for an hour.  Understand, I'm using it because I have it, and it's cheap and abundant.  Normally, Guard and In-minus are at the same potential, but the spec allows them to be 500V apart and I don't want to be the weak link.

A scope probe bypassed by 10K in the A16 space picks up about 200mV peak-to-peak from A11 (home of HV rectifier A11CR1) when the instrument is idling.  This will not interfere with calibration.  (I see about 1V at 1000V out.)  Anyway, you can calibrate the module outside the instrument - unless your photocells are marginal (in which case why are you doing this?), just select and trim the frequency and leave the pulse timing at default.

I made neon-less mounting plates out of polycarbonate sheet.  You could also cut up a CD/DVD/BD.  I'll include a dimensioned sketch in the writeup.

Since 740B#2 contains my input protection re-design, I've replaced A9DS1 with 30V TVS diodes across A17V1 and V2.  This will protect them from surges in case the operator accidentally switches range or mode with high voltage present.  As soon as I've verified all the program's calibration functions, I'll build a V1.2 board, set its LED drive for A16's photocells, pull out the KVD module, and install A16 and A17.

The attached picture shows the A17 powered on the bench.  V2 is brighter because that photocell is blind.  It doesn't show in the picture but I had to replace the old capacitor because (a) it's too bulky and (b) one lead is too short.  I threw together a cal harness, it reuses the 5-pin ICSP jack.

[Dave Wise]

I have installed my hybrid modems in 740B number two.  It's looking good.  They easily exceed efficiency and resistance requirements.  There's less drift, and less input current.  (Compared to previous quad-photofet modems.)  After I've recorded some observations, I'll modify 740B number one.  I'm continuing to test as time and energy permits, and I'm working on a writeup.

[Dave Wise]

Here it is.  For 740, 741, 3420, and 419, with possibility of repurpose elsewhere.
Writeup, MPASM source code and hex, and KiCad schematic and board layout.  Please look it over and comment.