Bruel and Kjaer type 3053 audio DAQ module repair

[hanakp]

My workplace has an audio lab which (among other stuff) includes 18-channel microphone beamforming array that works as "acoustic camera." The microphones are connected to Bruel and Kjaer type 3053 input DAQ modules, one 12-channel and one 6-channel.

https://www.bksv.com/-/media/literature/Product-Data/bn1467.ashx

https://www.bksv.com/en/products/data-acquisition-systems-and-hardware/LAN-XI-data-acquisition-hardware/modules/type-3053

But one day, the 12-input module suddenly stopped working, all analog channels failed all their built-in self-tests. A new DAQ module costs around 10,000 EUR and the official service costs start at whopping 2400 EUR, regardless of actual failure. So my colleagues asked me to look into it first. Of course, there was absolutely nothing about the 3053 modules online, but it wasn't as hopeless as it may seem: the digital circuits worked fine, but all analog channels were dead. In other words, there was some "big" systematic problem that affected all the analog circuits simultaneously, which shouldn't be that hard to find, even without schematic. And it wasn't.

The self-test can be run from the B&K PULSE Front-end software, it performs 13 tests on each analog channel. The results can be then saved as CSV file, I'm pasting the tests' names here so they will get indexed into search engines:

Noise in 10 V Range
Noise in 1 V Range
Distortion in 10 V Range
Distortion in 1 V Range
Voltage at 1024 Hz 10 V Range
Voltage at 12800 Hz 10 V Range
Voltage at 25600 Hz 10 V Range
Voltage at 1024 Hz 1 V Range
Voltage at 12800 Hz 1 V Range
Voltage at 25600 Hz 1 V Range
High Pass Filter (0.1 Hz)
High Pass Filter (22.4 Hz)
High Pass Filter (Intensity)

It also runs digital self-tests, but curiously, I found no way to save their results. You can view digital errors (if any) only on tiny display on the front panel of the module.

Cracking the module open is pretty easy, you need to remove 4 countersunk screws in the bottom cover. There are three ADS1274 simultaneous-sampling ADCs, which are connected to two Actel AGLP060V5-CSG201I CPLDs. These ICs were covered by soft thermal pads, but more about that later. At first, I checked the voltage test points near the edge of the PCB, but all were OK. Next, I suspected one of the CPLDs may have developed a solder joint crack, some BGAs are rather prone to it. I weighed them down with 3 kg load and run the self-test again, but nothing changed. Then I moved to the ADCs and sure enough, their reference voltage (VREFP, pin 56) measured 0.7 V instead of expected 2.5 V. I didn't want to remove the PCB from the top cover because of thermal pads (more about that later), so I soldered a thin wire to the nearby VREFP decoupling capacitor and carefully forced 2.5 V from an external power supply. It drew around 40 mA which seemed safe to me. It helped, but only partially - the noise tests passed, all others failed; it was clear the reference circuits are more complex. So I had to remove the PCB from the top cover after all. I soon found the reference circuits on the other side of the PCB, in unused areas between the ADCs. There is LT1019AIS8-2.5#PBF precision reference in SO8 package and its output is routed to (at least) two buffers with LT6233 opamps in SOT23-6 package. The first buffer feeds all three ADCs and the other feeds the complex input signal conditioning circuits. The LT1019 reference was dead; it was powered from +12.7 V supply, but it's output (pin 6) fluctuated around 0.7 V. When I soldered another wire and forced 2.5 V on its output, the 3053 module passed all tests. Obviously, I replaced the reference for the same part number.

Like I mentioned, the module uses numerous soft thermal pads to transfer heat from ICs to the aluminium covers. There are 4 different pad thicknesses: 0.5, 1, 2 and 3 mm. Most of them returned to their original shape after a day or so, but not the 0.5 and 2 mm ones. The covers have rough grainy texture inside which became permanently imprinted on them. I wasn't sure if they'd have good thermal contact and I didn't want to take any chances, so I replaced them. The 0.5 mm one is used only once, it covers Altera EP2C20F484I8N FPGA on the black digital board. I used these part numbers from Bergquist, they seem to be the same as originals:

GPVOUS-0.020-01-0816
GPVOUS-0.080-01-0816

In the end, the pads were the most expensive part of the repair (about 70 EUR), but of course that's still nothing compared to the official service cost. But if you know where I could buy them in smaller quantities, please let me know.

I must say the faulty reference IC baffled me. Sure, I encountered one or two dead tuning voltage references in CRT TVs and VCRs, but never such expensive industrial one such as LT1019AIS8-2.5#PBF. It doesn't even run hot, I measured about 42 deg. C with an IR camera. Curiously, the ADCs use 2.7 V reference instead of recommended 2.5 V. Their buffer is connected as noninverting amplifier with 1k and 12.7k resistors, so its gain is 1.0787. This ideally gives 2.5 x 1.0787 = 2.698 V. The other buffer near LT1019 has unity gain, but there is divider from 1.96k, 1.58k and 140k resistors which creates 1.1 V from the 2.5 V.

Hopefully, someone will find this info useful.