About six months ago, I re-designed my EEG pre-amplifier board to have more channels at about the same size. When the boards came in, they looked very nice and would test very well with simulated signals, but would often exhibit wild signal transients and intermittent wide-band noise when hooked up to live subjects. To say the least, the past half-year has been hectic as I tried to figure out what kind of ghost was causing the problems, which were odd, as it seemed that the new pre-amp design was just like the old working design, but with more channels.

milmax-hs-sized.jpg

( full size )

First, some background. As can be seen from the schematic, there are input channels and output channels. The input channels are of three groups: depthA-F, screwA-D and occipital peg (OP). The depth electrodes are bipolar, activated iridium-oxide microelectrodes (150um diameter per pole) while the screw electrodes (including OP) are 1mm diameter stainless steel contact electrodes. For the purposes of this, we can ignore the depth electrodes. The oversimplified idea is that when enough neurons that are spatially close and in the same orientation fire an action potential, the movement of ions due to the action potential is great enough that potential shifts (in the uV range) can be seen on the scalp. Due to electrode dynamics, these screw electrodes can be modeled as voltage sources with an output impedance of 1-10kOhm (at 30Hz.) Another point of interest is the grounding scheme. The recording system has a small enough foot print that it made sense to link the signal ground to the subject by a 10kOhm resistor to the OP, so that over a finite time scale, the recording device and the subject would be at the same potential, thereby reducing common mode. Since the potential difference was assumed to be small anyway (500-100mV), negligible current was assumed to flow over R33. This scheme worked fine in the previous version of the pre-amp as screwA and B were treated as one differential channel (meaning, both were fed into the same instrumentation amp giving one output) and screwC and D were treated as another. The OP was there only for the purposes of providing an ohmic path between the subject and the electronics. This pre-amp also had only two differential depth channels, so a revision was made to increase the number of channels while maintaining similar footprints.

The first thing that was changed was the connector. We went to a MILMAX dual row connector with 50mil spacing, as compared to PLASTICS-ONE connectors, which gave us almost triple the contacts in the same space. The connectors are one of the biggest points of failure in an electronic system, but it seemed that the MILMAX connectors had a better design (spring mechanism) as compared to the PLASTICS-ONE, so this change was made without too much thought. Upgrading the number of depth channels was easy, as I simply took the design for the components around U1 and duplicated it two more times in U3 and U4. As for the screw electrodes, we were still going with the same montage of five, so the easiest idea was jut to put in four differential channels and have the negative input of all of these be OP (as shown in the schematic.) The PCB design went from two layer to four layer, but I was careful about routing to avoid any problems in the design.

After we got the boards back in, I got an uneasy feeling about the OP-differential configuration. One electrode was driving the input stages of four amplifiers and acting as a current path between the subject and signal ground. I shrugged it off and considered that the output impedance of the “large” electrode was more than enough to accommodate for the pA input bias current of the amplifiers and that the current across R33 would reach a steady state. All of this looked good on paper, and in the lab as the pre-amp passed the tests performed by various signal sources, however, these assumptions were in error and were overlooked long enough to cause frustration. The lesson here is one that we all learn early on: do not pass current through the “voltage” probe of your measurement system. The reason that it was overlooked for so long was probably because the previous version worked so well that I immediately assumed that I just doubled the channels from the previous design in a safe way and only after building a dozen of noisy pre-amps and staring at the schematics for days did the answer come to me: don’t pass current through OP if it is used as a reference measurement. The solution is to remove the resistor at R33 and sacrifice one of the other screw electrodes, say screwD by putting a 10kOhm resistor at R30, so it becomes the current return path. Another assumption that may have been wrong is that current across R33 is small enough that it can be dumped on signal ground (SGND). If this is wrong, the pre-amp will have to be re-worked further to either buffer SGND when connecting to the resistor or to run it to power ground. In either case, this is just a few of the modifications to go into Rev 1.2.

 
del.icio.us | digg