Keithley 148 Nanovoltmeter (a brief teardown)

[manganin]

Two Keithley 148 Nanovoltmeters saw the daylight when I was re-arranging my storage room.




The specifications are quite respectable, even nowadays:

10 nV full scale to 100 mV full scale
noise 0.2 nV rms (most sensitive range)
drift <10 nV per 24 hours non-cumulative (in practice ~1 nV for several hours)
mains common mode rejection >160 dB


Top cover removed.




Bottom cover removed.




Both are the European 50 Hz version.




The sensitive front end components are mounted inside a double wall metal box. The outer layer is steel and the inner seems to be mu-metal. Everything from the input through the chopper to the transformer windings is made of pure copper with crimped only connections.




Thermal sinks. The crimped copper lugs are clamped together using a brass bolt. The base of the whole assembly is a thick copper plate which keeps all the joints in the same temperature. The lugs are mounted on a copper washer insulated from the copper plate only by a thin high thermal conductivity mica sheet.




Even the feedback switch (connected to the transformer center tap) is made of pure copper.

[Vgkid]

Thanks for the teardown. I like looking into these micro/nano-voltmeters.

[Gyro]

I'd never thought of using mica washers for close thermal coupling of circuit connections (ie. anything other than transistor heatsinking). Obvious really. 

[VintageNut]

Nice pictures. Have you fired them up and tested them?

I have 2 x 147 and 1 x 148. Both 147 units work great. The 148 requires some restoration.

I used the 147 extensively and was able to characterize some very low resistance devices. The experience was an education about thermal voltages and magnetic interference. If you get the nanovoltmeter too close to a strong transformer, you will see some oscillation of the meter needle. After you assemble a circuit with a very low resistance, you have to wait quite a while for all of the stray microvolts to settle.

It would be great if there was a hack to make one of these into an AC noise meter.

[zhtoor]

looks like there is some confusion about the input stage in 148,
does it use the CK512AX audio pentode or some JFET?

it certainly looks like the OP's model uses a tube.

regards.

-zia

[dietert1]

Old thread, yet new answers and questions.

That nanovoltmeter existed in two versions, one with tubes, one with FETs. On electrotanya i found a nice description of the version with tubes. The FET version also includes one tube, with the heater used as a PTC in the 94 Hz oscillator.

My instrument has three tubes and it includes the 1.2 V power supply for the heaters. Since those are direct heaters (heater used as cathode), how can those tubes be lower in noise than FETs? This is my question.

More doubts:
Resistor R232 takes about 5.2 V in my instrument and gets extremely hot (large black area inside top cover).
"Battery charging" lamp DS202 lights when? The other lamp DS201 gets about 5 V in my instrument and according to the parts list they are the same part.

Regards, Dieter

[Kirill V.]

The use of direct-heated tubes in highly sensitive instruments was standard practice, but I can't give you the exact reasons. Dieter, you can refer to the literature of that era about measuring devices to find the answer.
As far as I know, the cathode of indirect heating gives a little more flicker noise than direct heating. But if this meter uses a chopper any flicker noise is effectively suppressed.

[Martin.M]

thank you for posting,
I will look to finde one

[dietert1]

I was asking for the noise, because i understood that the heater power supply noise enters the signal chain near its input with a direct heater used as cathode. Maybe the bipolar transistors used for generating the 1.2 V heater supply (Q43/Q44) are lower in noise anyway.

Regards, Dieter

[Kleinstein]

Tubes are not very good or DC signals anyway. The AC part of the heater voltage can be filtered.

[dietert1]

Meanwhile i got our 148 into operation. In the low thermal box the burnt 111R "Daven" resistor got replace by two UPW50 resistors. Then i found a short to ground in one of the copper terminals. I unmounted all of them to clean the copper lugs. The tubes got 1N4004 diodes parallel to their heaters as over voltage protection. A lot of parts are new. The instrument received a Li-Ion battery pack and the charger got a small heat sink. It's a 2N1535 Ge transistor and it ran hot. The range switch needed cleaning.
The instrument runs well. It's native temperature offset is about 30 nV (with zero suppress "Off"). Will use it to check some low thermal multiplexers.

Regards, Dieter

[Dave Wise]

The tube description at Electrotanya is "keithley_148_measurement-of-nanovolts_nanovolt-mero_sm.pdf" .

Before I knew this, I sketched my tube-type 147.  The filament power supply is five transistors: a common-emitter differential pair driving a single-ended common-emitter stage driving a common-collector darlington pair.  Its reference is +12V, and it outputs 10% of that, powering V1 then V2, in series, to ground.  V1 G2 and P are at +12V, with the 200H inductor parallel-tuned to 94Hz by a 0.0143uF 1% capacitor.  (The trap is the same as the FET version.). V1 G1 is at ground, so V1 sees -0.62V to -1.2V bias.  V2 plate load is 100K, G2 drop is 100K and bypass cap, and G1 return is 4.7Meg same as FET, giving V2 only 0V to -0.62V bias.

As far as I can tell, the 147 and 148 are identical except at three points.
1: The 147 adds an input current cancel circuit, just a 100Meg resistor to an adjustable voltage.  Adjust for minimum shift as source resistance changes.
2: The 148 adds a sensitivity range.  It's just one more step increase in the feedback resistor network, the amplifier topology stays the same.
3. Different part number for the input transformer.  Different ratio?

The manuals state different values of input resistance, and different time constants for step response.  The loop gain must be set higher in the 148.

I encountered two interesting problems in my 147.

SYMPTOM: Erratic reading plus high offset
CAUSE: Crimp failure.  Every connecting wire in the compartment measured at least an ohm from end to end, some read 10's of ohms and noisy.  The crimped component leads were okay, it was only the point-to-point wires.  Must have been a bad batch of copper.
FIX: Zap.  I used 60mF at 6V.  I will monitor over time to see if the welds keep working.

Now it stays on scale, even at 0.03uV range.

SYMPTOM: Zero Suppress inoperative on mV function.
CAUSE: Assembly error at the factory?  Look at the bottom half of the S101 (FUNC) wafer outside the compartment.  9/4 wire (ZS source current) should be on S101 rotor contact, but rotor contact was N/C and 9/4 was soldered to 9/8 (uV target).  So uV target is energized on uV and mV, and 9/6 wire (mV target) is never energized.

Anyone who has a 147 open, please compare.  Thanks.

Regards,
Dave Wise

PS: The supposed electron tube filament, R322, is actually a lamp.  T1-3/4 envelope and wire leads.  I don't know the volts/amps yet but cold resistance is around 27 ohms.

[Dave Wise]

The instrument received a Li-Ion battery pack

Dieter, in your 148, what is the minimum battery voltage for good 12V regulator operation?  In my 147, below 6.2V, the -12V regulator drops out.  Since the original battery was 6.0V, either something's wrong or Keithley goofed.  My money's on the former, but the inverter transistor bias and the reservoir cap - the main suspects - are okay.  What's your battery discharge and charge current?  Did you tune the charger to stop prior to over-voltage, or is your pack protected internally?  What is your inverter frequency?

Dave Wise

[dietert1]

When i put in that Li-Ion pack shown in the photo, i kept watching during the first charge-up and adjusted the charger voltage limit to about 8 V. That battery pack has 2x2 cells. so the cells will get 4 V each. The discharge is very slow. I know that i can run the instrument two consecutive 8 hour shifts without reloading. In the manual they explain that the intended mode of operation is charge-up during night and run the instrument from battery during daytime for about 8 hours. I just made sure the new battery pack has more capacity than the nominal capacity of the original one.
The 148 power supply is strange insofar as during normal mains operation the low voltage supply is about 50 % higher than during battery operation, which results in up to 22 V output of the isolation converter  = input to 12V regulator. That regulator transistor runs quite hot during mains operation.

When running on battery our 148 has a sigma of about 1 nV with a low thermal short. With a low thermal short at the end of the 1.5 m measurement cable i made that increases to about 2 nV. This is with a plastic cover over the input connector.

Regards, Dieter

[Dave Wise]

I looked closer at my -12V regulator.  As far as I can tell, Keithley goofed.

It won't matter after I install the new 7.2V battery, but it still bothers me.

Q35, Germanium, TO-8, RCA, 2N1183, is the pass transistor.

(148 manual says RCA 40139 but that's a typo, 147 manual says 40319.
That's 2N4036 family.  My Q35 is marked RCA 36340 same as charger driver Q26,
148 manual calls it TI 2N1372 - TO-5 - but 147 manual says it's RCA 2N1183.)

Q36, Germanium, TO-5, TI 2N1381, drives Q35.

Note: In my early instrument, D206 is two 1N645 rectifiers in series
instead of a zener.

Output is -12.13V .
Q35.E is -12.284V (R223 measures 3.2R so output current is 48 mA, 110mV p-p)
Q35.B is -12.517V
Q36.B is -12.730V
JCT R224-R225 is -13.35V

My -12V regulator drops out when the battery drops below 6.2V .
There are three stages of dropout.

1. Q36 base current supply.  When the drop across R224 falls to less than
about a volt, there's not enough current to keep Q36 on.  If we elevate
R224 with an external supply or wire up a booster, we can continue operating
down to 6.0V .  Then we drop out at stage 2.

2. Q36 saturation.  When Q36.C drops below Q36.B, Q36 saturates, and further
drop will reduce the output voltage .  If we elevate Q36.C, we can eke out
another couple tenths and operate down to 5.7V .  Then we drop out at stage 3.

3. Q35 saturation.  When Q35.C drops below Q35.B, Q35 saturates, and further
drop will reduce the output voltage.  There's no fix, we're already
using Germanium transistors and there is no more headroom.

The inverter drive is a square wave exactly like the manual shows.
The inverter transistors switch from open to a couple tenths
of a volt, just as they should.  T202 primary center tap is Vbatt minus F202.
Rectifiers D210-D211 and reservoir caps C208-C210-C211 are fine.

All I can figure is T202 simply has insufficient step-up ratio,
which is either a design defect or a manufacturing defect.

What is your dropout voltage?

[Dave Wise]

I installed two Lithium 18650 cells in place of the Nickel-Cadmiums.  They fit in the old Gould battery holder.  I secured them with a dab of hot glue.  They stand proud of the holder, so put some insulation on top.

Load current is about 250mA and my cell capacity is about 2800mAh, but the regulators drop out before the pack is fully discharged.
Looks like I will get about 10 hours running time from a full charge.

YOU MUST NOT OVERCHARGE OR OVER-DISCHARGE.  This mandates some kind of protection circuit.  Some cells have built-in protection, but many ads lie and say they have it when they really do not.
I used an HX-2S-JX20 resistive balancing BMS board.  My board has 39-ohm bleed resistors which turn on at 4.25V; this can bypass about 100mA of residual charger current.
Many ads for BMS boards lie and say they balance when they really do not.  Look for bleed resistors in the picture.
You can add a TL431 programmable zener if you don't trust the board.  It will need an external pass transistor due to heat.

 1. Change zener D214 (9V) to 12V
 2. See "POWER SUPPLY CHECKOUT AND CALIBRATION", section 4-7 in the 148 manual or 5-8 in the 147 manual
    With POWER SUPPLY set to AC and battery disconnected, adjust R228 for Test Point = -9.4V instead of -8.1V

Charge current starts at about 300mA, or about C/10.
At full charge, the BMS clamps the pack to 8.5V with average bypass current about 70mA, and DS202 is dark, with about 0.5V across it.

[Enginerding]

I picked up a 147 and 148.  Feb 1978 markings on the original battery of the 147.

Opened up, clean inside, no obvious issues.  Felt worthy to give it some line voltage w/ the thermal cam handy.  Left the batteries hooked up, because kinda curious how 42 year-old batteries would respond.  Set to "millivolt", "100" range, and turned zero suppress all off,  (not that it should matter).

Got a buzz and some needle movement.  No lights on the "AC connected" lamp (ds201) or "battery charging" lamp (ds202).  Same issue on both the 147 and 148.  (sidenote:  based on the manual, whenever the 147 is connected to AC with the switch in the "off position", the battery will charge if necessary.)  Buzz is coming from the mechanical chopper at the input.

Decided to start fixing the 147 first.  Disconnected the now bulging battery, ha.  After being on for 10 seconds or so, w/ the thermal cam I see large temp difference on the "PC-74" card, one of 2 upside-down plug-in cards, which seems to be the home of power supply circuits.  While the rest of the internals are at room temp more or less, R232 on PC-74 is quickly ~150F (top cover is dark in that area too, same as Dieter).  Taking the board out, Q27, a 2N1535, is also quite hot (to-41, germanium PNP).

More to come.  Also, while looking around found some leaky electrolytics on the other plug-in card, PC-75, which is the 94Hz oscillator.  I changed them, but I don't think they were the source of any particular problem.  Out of circuit they tested pretty well, age considered, including leakage current.  If anyone's interested in that data, I'll post below.

Cap data:
Sprague "Verti-lytic" original caps;
replaced C304, C306, and C307 (15v 100uF), C305: (15v 10uF).
C304: 11V for 7.5 minutes, leakage of 995nA.  @100Hz 126uF, ESR:  0.5R, tan: -87.5 deg.  Replaced with Nippon KZE 50V 100uF (on hand, leakage was 200nA for comparison).
C305: 11V for 7.5 minutes, leakage of 1.9uA.  @100Hz 11.4uF, ESR:  4.1R, tan: -88.2 deg.  Replaced with a Nippon BT 25V 10uF (leakage 42 nA).

[Enginerding]

Next to get the battery permanently out. (Bottom cover?)

Then, look at the schematic some more.  Maybe try powering it from a DC source in place of the battery to bypass all the AC to DC conversion?

I suck at circuit analysis, so apologies if it's obvious or slow.

[Enginerding]

Took the battery out, which is as simple as removing the bottom cover, and taking out 6 screws that go directly into a bottom plate on the battery.  I noticed the inductor coil ("degaussing coil") around the mechanical chopper had fallen down, so I kapton taped it back up. 

Connected a linear power supply and gave it 6.3v.  Turning the main dial to "battery test" reads ~8.5; manual specifies a minimum of 8 for operation.  Interestingly, the enclosure is connected to positive (which I think is mentioned in the manual).  The K147 draws 0.30a in operation.  The PC-74 power supply card no longer gets hot, so that issue is bypassed for now.  The mechanical chopper has quieted down a lot, though still audible with the cover on.  I also noticed that dialing up and down the supply voltage noticeably changes the chopper frequency, so I'll have to check all that when I go through the alignment.  Nothing on the instrument is getting above ~85F after a few hours warm-up, so all seems well to play with a voltage reference.  I'll recap the main board later.

I'm going to compare it to the Keithley 260 nanovolt source and Keithley 181 nanovoltmeter (the next generation).  I'll also connect the analog output on the K148 (1v full-scale) to an HP 3457a.

This is the K147 with input shorted on the lowest range, 0.01 microvolt fs:   the noise is negligible, with short-term drifts around 6nv.
(the high pitched noise in the background the mechanical chopper).  The 147's output to an HP 3457a is at the end.



For comparison, here is the K181 with the same shorting plug:  short-term drift of about 15nv (filter off, damping off, "hi res" 6th digit on.  With filter on that circular drift is ~7nv, and with filter and damping ~4nv).

[Dave Wise]

Chopper frequency should be stable, the 12V regulators are dropping out when you turn down your battery eliminator.

The position of the chopper coil you taped back in place is critical, the manual has a procedure for aligning it.

[Enginerding]

Awesome.  I see you're a few steps ahead of me  Thanks Dave.

It'd be cool to do a project to replace the mechanical chopper.  Seems like they're unobtanium right now.

[Dave Wise]

Glad to help.

Keithley looked into different chopper types, it's in the manual.  At nanovolt levels, nothing beats a mechanical contact.

[doktor pyta]

At nanovolt levels, nothing beats a mechanical contact

Actually FETs perfom better as shown here:
http://www.emelectronics.co.uk/low-level-voltage-measuring-instruments/p13/

[Dave Wise]

How is it shown?  All I see is a product advertisement, FETs are not mentioned.

[Kleinstein]

For the nV preamplifier versions there are some infos around in the web and also in the forum here. They use JFETs as choppers a slightly special transformer and than a relatively normal AC amplifier and demodulator part. Chances are the very low noise version would use MOSFETs to get an on resistance well below 0.2 Ohms. Modern low voltage MOSFETs are pretty good, both as discrete or CMOS switches.

One has to find a balance from charge injection and residual noise / resistance. So the best input depends on the source resistance.  The modern design get quite good, so that over a rather large range (e.g. 1-3 decades) the Ohmic resistance of the source dominates over voltage and current noise.