Signal Conditioning

Blood pressure is one of the better known vital signs, measured by just about everyone, often with a non-invasive blood pressure cuff. But the cuff is inaccurate, yields only one measurement per cycle (approximately 60 seconds) and causes tissue damage if inflated too often, so for critically ill patients and in animal experiments blood pressure is recorded continually using an invasive arterial line. This is simply a catheter placed into an accessible artery, such as the radial artery, and connected via rigid plastic tubing (filled with sterile water) to a pressure transducer. These pressure transducers are very nice and small, laser-trimmed for superb accuracy, compatible with most fluids, and inexpensive enough for disposal. When used in our EPCS the transducer will be reusable as it won't come into contact with bodily fluids as it would when monitoring blood pressure, but the size, accuracy, power, and range are all ideal for pneumatic pressure monitoring. In additional every potential user of the EPCS stocks these transducers in bulk.

Of course, it would be too easy if we could wire these sensors directly to our system. In the interests of interoperability with many medical monitors and to minimize the complexity of the disposable components, the transducer requires two wires for excitation and outputs the pressure as a differential analog voltage on a second pair of wires. The gain (proportional to excitation voltage) is standardized at 5 μV/V/mmHg. So with a typical 5V excitation source, a pressure of 100mmHg would be 2.5 mV, differential. Both output lines carry a common-mode offset which is approximately mid-scale. Since our microcontroller has an integrated 2.5V precision bandgap reference, and that falls within acceptable limits on excitation voltage, we'll use that. Actually the reference output doesn't have good fan-out so an op-amp will be used in a unity gain configuration to buffer the voltage. Thus 100mmHg would be only 1.25 mV in our system.

The nice thing about using the same voltage for both transducer excitation and the analog-to-digital converter reference is that the actual voltage for full-scale cancels out. But we still have full-scale = ±200000mmHg. To obtain a 2mmHg precision with a 400000mmHg range like that, our ADC would need to produce over 17.6 bits/sample. On the other hand if we scale by a factor of 400 and shift by +20% of full-scale, then our range becomes (-100, 400) mmHg which is more than ample for the EPCS. Plus, the 12-bit ADC integrated in our microcontroller gives 4096 distinct codewords, or 1 LSB = 500mmHg/4096 or 8 LSB < 500mmHg. A 10-bit ADC would add no additional error to the ±1mmHg/±2% accuracy of the transducer, but 12 bits is pretty standard so we end up with a little bit of headroom.

So how do you scale and shift a low strength signal like this? An instrumentation amplifier is just the ticket. It combines unity-gain op-amps for buffering on each input, so the non-negligible output impedance of the transducer won't skew the feedback, followed by a third op-amp configured as a gain stage. Analog Devices doesn't make any fixed gain instrumentation amps near a factor of 400, so I've chosen a the AD8227, a variable gain (controlled by an external resistor) ultra-low power instrumentation amplifier which is also protected against ESD and overvoltage. Analog Devices is also kind enough to point out (on page 23 of the datasheet) that when this amplifier is used with a charge-sampling ADC, an R-C filter is needed. The example values given by Analog Devices provide a corner frequency of about 16 kHz which is more than is needed in the EPCS, by increasing the resistor by a factor of 100 we cut down the instantaneous current demands.

The AD8227 also handles the shift requirement because the output voltage is referenced to an offset input. Making the correct value could be a little tricky though, because there are no 4:1 ratio standard resistors. I know a way to get around that though: an array of four resistors lets us build a 2R:R/2 divider which is exactly 4:1, and because the resistors are in a single package the manufacturing inaccuracy and thermal effects will tend to be equal. Analog Devices again comes to the rescue with a helpful hint in the datasheet that the offset voltage needs to be buffered, the voltage divider shouldn't be used directly.

So the EPCS pressure signal conditioning circuit will look like this (subject to change, most likely the individual op amps will be combined with others and replaced with a single multi-channel IC):

P.S. I seem to use a lot of Analog Devices chips, anyone care to guess why?