Photovoltaic vs Photoconductive Operation in Pulse Oximeters

[MildInductor]

Hello, I am trying to figure out if pulse oximeters typically use the photovoltaic or photoconductive mode of operation. I have't been able to gain much insight into this besides looking at some example circuits online, from which it seems like the photoconductive mode is preferred. Although I have only looked at one circuit. I originally thought that the photovoltaic mode would be better because it has a lower dark current. However, apparently it is 'slower'. I am not sure if 'slower' is actually too slow for the application though.

[Kleinstein]

In many cases one uses a photodidoe with a TIA and virtually zero voltage at the detector. So just in between photovoltaic and photoconductive.
There is little extra effort to add a reverse voltage. The speed advantage can be quite a bit, but 2 x to 10 x due to a reduced capacitance. This is not just speed, but also a reduced noise in some frequency band.
Sensors with modulated light usually don't work with a really high frequency, so chances are they may still use zero bias.

[electron_plumber]

Not familiar with pulse oximeters specifically but in my experience with optics, applying a reverse bias to the diode decreases junction capacitance which can reduce noise and increase bandwidth. To determine whether this is advantageous, build a little diode model and simulate it against your transimpedance amplifier/integrator-- there are cases where (for example) feedback parasitics in the amplifier dominate bandwidth, and therefore diode junction capacitance reductions are less useful.

While this likely isn't your case, there are also scenarios where diode internal resistance is super low (e.g. low-bandgap semiconductors tuned to niche wavelengths can present as low as 30 ohms). In those cases, any applied bias voltage creates more DC current than it's worth.

Edit: check out the "bias" slide of this deck https://e2e.ti.com/cfs-file/__key/communityserver-discussions-components-files/14/Transimpedance-Circuit-Design-Considerations_5F00_TI_5F00_Precsion-Amps.pdf

[MildInductor]

Thank you both. That information helps a lot. I will have a play with the various configurations and see how it goes.

[CatalinaWOW]

Over multiple decades working in the infrared field, photovoltaic operation was a property of the detector.  A photovoltaic detector has a junction and detection is based on boosting electrons over the bandgap of the junction.  Photoconductive detectors relied on carrier generation (electron-hole pairs) which reduces the impedance of the detector when radiation impinges on it.  Because these detectors involve movement of electrons and holes they effectively have twice the number of charge carries and thus a root two noise penalty.

That said, I believe the detectors used in these devices are junction diodes and thus photovoltaic.  The comments above on biasing are appropriate for that mode.  Photoconductors cannot detect at zero bias.

[daisizhou]

https://www.mydigit.cn/forum.php?mod=viewthread&tid=471827&highlight=%E8%A1%80%E6%B0%A7

The link above is a successful case.

But I am interested in the spo2 simulator,If you are also interested, please see the attached file

[MildInductor]

Ok, I think the plan is to reverse bias the diode with some voltage and then use a TIA. I know the capacitance of the diode decreases with increased bias voltage. However, I am wondering what is stopping me from biasing the diode with something like 50 V or so as compared to around 5 V which the data sheet seems to reference.

This project is pushing out of my knowledge scope so I’m sure I will have more questions like this to come

Edit: The datasheet claims a maximum reverse voltage of 20 V is acceptable so let’s say 20 V rather than 50 V.

[Kleinstein]

With a very high reverse bias the leakage current goes up and adds additional noise. The curve of capacitance vs voltage gets increasingly flatter to higher voltage. So most of the gain is from the first few volts of revese bias. It is also extra effort to create the higher voltage.

I don't think the photodiode is the really critical part with a Pulse Oximeters. The more important part can be the optics to get the same path for both wavelengths and to get a stable intensity from the LEDs (or has a 2nd detector to measure). There is no need for high speed, it is more about amplitude accuracy.

[MildInductor]

Ok, got it. I hadn't considered the leakage current increasing from an increased reverse bias voltage.

I haven't even begun to look at the LED cluster circuit yet 

[MildInductor]

How does the op-amp bandwidth play into a TIA photodiode circuit? I'm not understanding where bandwidth comes into play at all with photodiodes really. I have read a TI application note (https://www.ti.com/jp/lit/an/snoa942a/snoa942a.pdf) but I am still unsure.

[Kleinstein]

The TIA usually needs a small capacitor in parallel to the feedback capacitor to avoid oscillation / ringing.  How much is needed depends on the speed of the OP-amp and capacitance at the input.
The linked application node is more about really high speed. For a Pulse Oximeter one is more in a range with lower current and lower frequency (e.g. ~ 1 khZ) and bandwidth.

[MildInductor]

Ok I see, so I should be looking for an op-amp with a bandwidth of approximately 1 kHz or greater.

To my current understanding, the rate of change in the photocurrent entering the TIA due to a change in illuminance influences the op-amp bandwidth requirements. To think of an extreme case, I can imagine if, for instance, the op-amp bandwidth was 1 Hz, the output voltage of the TIA would not be able to keep up with the change in the photocurrent given a rapid change in illuminance on the photodiode, which would be bad. Is that more or less correct?

[Kleinstein]

The OP-amp still should be reasonable fast, so that it is not really limiting the speed, so more like 1-10 MHz GBW for the OP-amp. The current range determines the required resistor in the feedback and with a large resistor the TIA gets slower. The GBW of the amplifier can compensate somewhat.
To slow an OP-amp would not keep the input at a virtual ground and than the capacitance of the photodiode and amplifier input get more relevant.

For selecting the FB capacitor and looking at the required speed it can make sense to run a simulation (e.g. LTspice).

[dietert1]

In most pulse oximeters the photometer operates near zero detector voltage, using a transimpedance amplifier. This setup is preferred for best linearity.
As the oximeter method depends on AC only, the LEDs need a low noise current source but no long term stability. In general one uses a LED pulsing scheme with synchronous detection, similar to a lock-in circuit. Often there is a cable between photometer and TIA and speed will be limited by that cable. The reason is the need to disinfect finger clips using heat.
The photometer needs some protection from ambient light, e.g. direct sun light or AC modulated light from fluorescent lamps. Other optical parts aren`t required, once the distance between the red and IR emitter is small in comparison to finger dimensions. The photometer needs about 120 dB of noise free dynamic range in order to get a 1 % saturation result at 0.1 % perfusion. Lucky enough at low frequencies, so the usual way to go is oversampling. Clinical oximeters need to work properly in the presence of electro surgery devices. Nellcor finger clips had perforated metal shields around LEDs and photo detector.
The difficult part in making a medical pulse oximeter starts where the school books stop. Biosignal recovery is kind of a black art. E.g. the heart beat of a baby can be at the same frequency as the movements of an adult under drugs. Other problems is low arterial perfusion with cold hands. The saying is that Masimo spent tens of millions in research in order to come up with their signal extraction method (SET). Those patents aren´t for the faint hearted.

Regards, Dieter

[MildInductor]

In most pulse oximeters the photometer operates near zero detector voltage, using a transimpedance amplifier. This setup is preferred for best linearity.
As the oximeter method depends on AC only, the LEDs need a low noise current source but no long term stability. In general one uses a LED pulsing scheme with synchronous detection, similar to a lock-in circuit. Often there is a cable between photometer and TIA and speed will be limitid by that cable. The reason is the need to disinfect finger clips using heat.
The photometer needs some protection from ambient light, e.g. direct sun light or AC modulated light from fluorescent lamps. Other optical parts aren`t required, once the distance between the red and IR emitter is small in comparison to finger dimensions. The photometer needs about 120 dB of noise free dynamic range in order to get a 1 % saturation result at 0.1 % perfusion. Lucky enough at low frequencies, so the usual way to go is oversampling. Clinical oximeters need to work properly in the presence of electro surgery devices. Nellcor finger clips had perforated metal shields around LEDs and photo detector.
The difficult part in making a medical pulse oximeter starts where the school books stop. Biosignal recovery is kind of a black art. E.g. the heart beat of a baby can be at the same frequency as the movements of an adult under drugs. Other problems is low arterial perfusion with cold hands. The saying is that Masimo spent tens of millions in research in order to come up with their signal extraction method (SET). Those patents aren´t for the faint hearted.

Regards, Dieter

With regard to the low noise current source to drive the LEDs, do you have an example of the type of circuit that would be used? I presume driving the LEDs directly from a microcontroller GPIO pin is probably not the most ideal then. I totally agree that biosignal recovery is black magic

The OP-amp still should be reasonable fast, so that it is not really limiting the speed, so more like 1-10 MHz GBW for the OP-amp. The current range determines the required resistor in the feedback and with a large resistor the TIA gets slower. The GBW of the amplifier can compensate somewhat.
To slow an OP-amp would not keep the input at a virtual ground and than the capacitance of the photodiode and amplifier input get more relevant.

For selecting the FB capacitor and looking at the required speed it can make sense to run a simulation (e.g. LTspice).

Would one choose an op-amp first, then important the SPICE model, then finally play around with the feedback capacitor values? What type of current source would you hook up to the op-amp and what analysis method would you use? I am only really familiar with basic DC steady state analysis in LTspice. However, I am keen to learn.

[dietert1]

Our oximeters have a 22R shunt to ground as emitter resistor of a npn transistor, so the LEDs with their MUX would be operating near the 3.3 V positive supply. Base of the transistor is driven by a DAC pin of the MCU via a voltage divider. Both emitter and collector are connected to MCU ADC pins in order to monitor LED voltages and currents.
LED current pulses can be up to about 50 mA yet an average LED current is only about 2 or 3 mA.

Regards, Dieter

[Kleinstein]

For the drive side the question is if one can get away with just a stable current and temperature (e.g. via the LED voltage) compensation or wants an extra photodiode to measure the light at the other side of the finger too. Usually photodiodes are a lot more stable than LEDs. The extra photodiode would also reduce the demand on the current source.

[MildInductor]

Our oximeters have a 22R shunt to ground as emitter resistor of a npn transistor, so the LEDs with their MUX would be operating near the 3.3 V positive supply. Base of the transistor is driven by a DAC pin of the MCU via a voltage divider. Both emitter and collector are connected to MCU ADC pins in order to monitor LED voltages and currents.
LED current pulses can be up to about 50 mA yet an average LED current is only about 2 or 3 mA.

Regards, Dieter

That's really interesting to hear what some commercial products use. A few questions:
- What do you mean by 22R shunt to ground? Is the resistor acting as a current shunt for the current monitoring you mentioned?
- Is there anything to watch out for when picking a suitable npn transistor?

[dietert1]

Yes it's a current measurement shunt and gives 1.1 V at 50 mA. The remaining 2.2 V are for the LEDs, for the MUX resistances and to avoid transistor Uce saturation. The scheme works down to about 3.0 V. For the transistor you want one with high beta. Or substitute it by an opamp and a mosfet, thus taking away the base current from the shunt.
We used a feedback diode for the LED of an electronic patient simulator (part: IL300), in order to linearize LED response as function of current. Pulse oximetry finger probes don't have feedback diodes. It isn't necessary as the oximeter separates the signal of arterial pressure waves (AC) from the average transmission (DC). Pulse oximetry uses the same formula with thin and thick fingers and even painted fingernails don't disturb the measurement.
In the beginning we made our oximeters with variable LED currents and variable TIA gain. Later we simplified the circuit, using only variable LED currents. In battery driven monitors one can also save energy by adapting the LED modulation scheme for "easy" measurements with good perfusion and few motion artefacts.

Regards, Dieter

[MildInductor]

Ok I see, that's really interesting to hear.

When you say variable LED current, is that so you can dim the red+IR LEDs? If so, why would it be desirable to dim them?

Also I noticed the following on the Wikipedia page on pulse oximetry:

Continuous monitoring with pulse oximetry is generally considered safe for most patients for up to 8 hours. However, prolonged use in certain types of patients can cause burns due to the heat emitted by the infrared LED, which reaches up to 43°C

This doesn't seem correct. Surely a little red+IR LED diode cluster wouldn't reach a temperature that high?

[Kleinstein]

With an average current of a few mA one would not reach significant power and the light level should be no problem. With more current an IR LED can reach quite significant power, but that would be a different thing. If there would be a power problem the time constant is much less than 8 hours, more like 1 minute. The mechanical force to clamp the sensor would be annoying over a long time.

A variable LED intensity could be used to conserve power (for battery operation) and add some gain switching to compensate for different fingers. Switching the LED current may be easier than switching the TIA gain. If power saving is not needed one could as well have gain switching at a different position or a system with larger dynamic range.

[MildInductor]

That’s along the lines of what I was thinking re the temperature and the LED dimming. I’m sure I’ll be back with more questions shortly

[electron_plumber]

Would one choose an op-amp first, then important the SPICE model, then finally play around with the feedback capacitor values? What type of current source would you hook up to the op-amp and what analysis method would you use? I am only really familiar with basic DC steady state analysis in LTspice. However, I am keen to learn.

As I recall, there are generally two fundamental bandwidth limiters of a traditional transimpedance amplifier -- which one dominates depends on the circuit properties.

1. Single-pole response of feedback resistance and parallel capacitor: for high-gain applications in particular, this often limits amplifier bandwidth. Since it's a single pole response, it tends to be a smooth roll-off.
2. Double-pole response of input capacitance and effective input inductance of amplifier. This moves based on amplifier bandwidth product.

You might wonder where the "effective inductance" comes from -- afterall, there is no inductor in your circuit. It comes from the input impedance of the amplifier (i.e., impedance seen looking "into" the inverting input pin). The impedance looking into that pin will be near zero for frequencies in the closed loop bandwidth of the amplifier, but it will start increasing with frequency as your amplifier runs out of gain. So, in the frequency domain, this looks like an R-L network.

Lots of good resources about this online. Photodiode amplifiers by Jerald Graeme is an old book now but it explains this pretty well as I recall.

[dietert1]

A standard finger probe produces a differential photo signal. We used to have input inductors for the transimpedance amplifier inputs for reasons of EMI suppression. Later we replaced them by resistors (of 390 Ohm as far as i remember). The TIA resistance is about 1 or 2 MOhm. System bandwidth is about 25 Hz. For an ECG channel one wants about 400 Hz. Our oximeters do measure pulse transition time (PTT) with a standard deviation of about 1 or 2 msec.

Regards, Dieter