As some of you know , i took a 8846A and a screwdriver a couple of days ago based on some threads running on the forum here.
And here's what i found ...
We're all familiar with this one. Big fat graphics display , rubber keypad and a USB host port. Bananajacks on the left. Nothing special.
Made in the USA. No internal user serviceable parts ... right... where's me solderin iron... ?
The front panel uses a CIG ( Chip in glass ) VFD display from noritake-itron.
You can see the massive amount of traces leaving the internal chips in the above image.
Data is fed over a parallel bus and all display scan electronics is inside the vacuum glass. 11 filaments are used to emit electrons so the pixels can be turned on.
The frontpanel is a simple PCB that only has the USB connector and the pads for the membrane keyboard. nothing really interesting to see.
The connection from display board to main board is over a right angled 2mm pitch dual row pin header.
The silvery backside of the display reflects the board pattern. There is a metal shield (removed in this picture) over this whole assembly as the high voltages used in the display, and the scanning no doubtedly causes interference in the sensitive electronics of the meter.
The main circuitboard holds all electronics:
There is a clear split between the inguard and outguard portion of the system. Inguard is a term used to denote the portion of the meter electronics that has direct galvanic contact with the meter inputs and probes. The outguard is the chassis-earth reference electronics. there is a galvanic barrier between them ( clearly seen as the large empty zone runnong diagonally over the bord from top left to bottom right.
In the above image anything bottom left of the barrier is the inguard.
Here anthing bottom left of the barrier is the outguard.
Since we need communication between two worlds we need a bridge that preserves galvanic isolation. Here Fluke has gone the opto way. But, instead of using off the shelf optocouplers, they went all out using discrete IR leds and optotransistors under a shield.
The pair on the left left is the channel fron outguard to inguard while the one on the right does inguard to outguard.
The entire system has ample supply voltages almost all made using LM317 / LM337's. There is a power line sync circuit just below the yellow wire on the connector. Here they sync onto the mains frequency and trigger on the zero cross. This signal is used by the meter to trigger the actual measurement.
NPLC: Near- Power Line Cross. NPLC can also stand for 'Number of Power Line Cycles'. Both are correct and there's two things at play:
- For DC measurements the machine will execute the sampling of the input signal at the exact moment of the power line cross. The electrical field emanating from nearby power cables is then minimal. For sensitive measurements this drops the noise floor ( improves Signal/noise).
- For AC measurements we integrate the result post-rms converter over multiple power line cycles. Here we count a number of power line cycles and we integrate the measurement to even push noise further away. You can use both mechanisms to gain at least an extra digit of resolution.
Before we dig in to the succulent world of the inguard , let's take a look at the outguard area.
We find an Altera Cyclone 1C12 fpga. Not exactly a new part since we are at Cyclone IV now ... a big fat SDram sits in the back and booting is done from a Spansion Flash chip on the right. The flash chip holds the FPGA configuration and implements a NIOS softcore running a Linux-based operating system. An ICE + trace connector sits top right. Kind of strange to have this still on the board as the flash chip is clearly pre-programmed before soldering ( the 2662989 and 64916 strings on the spansion chip are lasered AFTER the chip is programmed ) . There's also a little 'oopsie' wire visible around the flash chip. Looks like someone didn't run DRC when they did the board...
This circuitry is complemented by a NXP / philips ISP7164 USB host controller and a National Semi DP8384 ethernet Phy ( the MAC no doubtedly sit in the FPGA. )
an LTC1385 does old-skool RS232 translation.
Left of the flash chips we find the GPIB bus transceivers: a 75160 and 75161 pair. The GPIB bus controller is no doubtedly integrated in the FPGA as well, as the classic bus controllers all have gone the way of the wind.
TI's TMS9914 / ML9914, Intel 8291/8292, NEC upD7210, Motorola 68488, Philips NXP HCF4788 are all long obsolete. The ML9914 is still in production but you will only find that one in Agilent equipment. It is a mask lithography gate array with the 9914 logic embedded. Pin and function compatible with the TMS9914. Only the 9914 / upd7210 were popular for GPIB.
You may encounter the 68488 in older hp equipment that is based around an 6802, 6809 or 68000 cpu. As for the philips chip.. i've only ever seen that in philips test equipment.
The intel 8291/8292 pair is a slave controller / host controller. HP agilent uses these in quite a few designs as well. Although only the 8291 will be found. You only need the 8292 ( which is essentially a preprogrammed 8048) if you want to make a controller. As test instruments are typically slaves you will not find that thing. The upd7210 is a NEC clone of the 8291.
National instruments have their TNT4888 asic and some agilent and Rohde/schwarz equipment uses those. These machines are typically PC-based as the tnt4888 is an ISA bus compatible part. The 4888 is a slave device only. There is a 4888s is a master device and you need to sign an NDA to get that one. Royalty payment is involved whenever you wish to make a bus master.... slaves are royalty free. Although the status right now is unclear. Last i heard the patents finally expired while other sources claim not until 2015 for the 488.2 extentions...
Enough sidetracking.
The analog section is well protected using the classical bussman fuses found in many a Fluke instrument.
The input section uses proper dual coil relays, as not to influence incoming signals with the magnetic field of the coils.
Fusable resistors , voltage spark-gap capacitors (black vertical devices with slits in em ) , ptc's (black drums left) and mov's (blue drum's right) galore
Current sensing is done properly using a kelvin resistor
The whole macine uses precision laser trimmed hybrid devices:
These are not your run-of-the-mill open hybrids but they are vacuum sealed with a glass cover. ( probably quartz ) and are trimmed after sealing. Pretty expensive stuff.
Range mulitplexing is done using high precision analog switches from Maxim (wide body ic) and plenty of guard traces are available around input pins of both the chips and the precision resistors.
You can clearly see al the guard-loops on the backside of the board. some of these lie on ground potential but some are connected to output signals of opamps.
The True-RMS section has no surprises either.
The classic circuit using the AD637 and precision capacitors . Look like polystyrene SMD versions.
The entire inguard system is controlled by its own FPGA: another Altera Cylcone 1c12 , tied to an Alliance semiconductor Static ram. this FPGA boots from a serial configuration memory. There is a trace port available, as opposed to a simple JTAG port, which leads me to believe there may be another NIOS processor mapped in this thing running a piece of code.
This processor probably gets its firmware over the optical link during the system start. This allows for easy update as the entire meter firmware is held by the outguard flash chip.
The entire inguard system derives its long term stability and accuracy from a linear technologies thermally controlled buried zener ( much like the LM399 )
Now, here's a bit i didn't like at all : the input switch.
The front panel buttons are a wobbly bit of cheap plastic that are simply clamped on the pcb. they move push-rod that controls an even crappier construction at the back of the machine
A plastic block with some simple stamped wiping contacts being dragged over pcb traces to swicth between front and back. Simply snapped on to the pcb....
This doesn't fit a 6 1/2 digit multimeter. For a little 3 1/2 digit handheld ... ok.. But for this kind of machine ? It just feels wrong...
I'm going to go out on a limb here , and most of this is speculation on my part, but i think this machine was designed by committee.
- The analog engineers went all out creating a really precise system using expensive hybrids , precision electronics under control of an fpga .
- The marketing guys absolutely needed a very expensive graphics ddisplay , ethernet capability and other 'eye-candy'. They had to sacrifice so much money on the dual fpga , flash , rams, usb and ethernet stuff that the development cycle would have been too long and the cost of software development and maintenance high ( including the LXI connectivity over LAN) . Time to market would be impaired.
By the time everything was assembled the price point of the hardware was so high that there was no money left to make this thing profitable...
So firmware development money was cut by slapping in a linux kernel and used some code-monkeys to write an application as opposed of doing real firmware on an asic.
There was no money for a decent multipole switch ( those are pricey too ) so they went the wobbly wonky style pcb switch way.
It feels like a true fluke on the analog design part. High precision , High quality , and when that was done the panic kicked in... we, need to save money ! we are late !... but we will not scrap the eye-candy or the lan connectivity , so we will sacrifice some more money on a second fpga and then cut corners wherever else we can ... like using a cr2032 battery to store data in the inguard ? Why ? Ever heard of eeprom ? or Fram if you are worried about multi write cycles... it's not user replacable since you break the calibration seals by opening the machine and that's a no-no in a lab environment.
It just doesn't 'feel' right... but that's my opinion. As of running a full blown 2 megabyte operating system for a mulitmeter ? That's shooting mosquito's with an entire platoon armed with bazooka's...