Receivers for low-bandwidth optical (through-the-air) communications


When designing a receiver that is intended to be used for optical through-the-air communications, there are some things to be considered that are very different from what might be required in most other situations where detection (and demodulation) of an optical signal is to be done:
 What this means is that techniques for detecting higher-bandwidth signals (IR remote controls, fiber optic receivers, etc.) are not particularly well-suited for the detection of weak, through-the-air signals.

Additional comments about this page:
 Different types of detectors:

Vacuum tube detectors:

Among the older types of detectors, one is the phototube, while another is its far more sensitive cousin, the photomultiplier tube (PMT).  Still used today, the photomultiplier tube is unmatched in its ability to detect extremely low levels of light (especially at shorter wavelengths) provided that the tube's response is well-matched to the wavelength of light of interest.

A quick examination of photomultiplier specifications will indicate that most devices are fairly insensitive toward the "red" end of the spectrum:  The so-called red-sensitive "S-1" types have fairly miserable (0.1%) quantum efficiency in the red wavelengths while more common S-4 types have quantum efficiencies around one percent in the 600-650 nanometer area.  While there are some good, red-sensitive devices available (such as the Burle C31034 series with a GaAs photocathodes, or some of the Hamamatsu multialkalai devices such as those in the R669 or R7400 series) these tend to be fairly expensive when bought as new devices and rarely show up on the surplus market.  Even the garden variety photomultiplier tubes that are available on the surplus market (such as the venerable 931A types) tend to be much more expensive than a simple photodiode.

Photomultiplier tubes are also somewhat awkward to use:  They typically require 800-1500 volts for operation (depending on the tube and application) and its large target area makes it slightly more awkward to optimally illuminate with very simple optics.  Most awkward of all is that they are extremely fragile, both physically - because they have fragile glass envelopes and, especially, optically:  A good phototube can be wrecked by even a brief exposure to daylight or strong light sources.  Depending on the particular tube, the power supply, and the intensity of light, this effect may only be temporary, with the tube returning to "normal" in a few hours or days, or permanent damage may have resulted in the diminution of the tube's ultimate sensitivity.  Whether the effects are temporary or not, a simple mishap could easily end (or delay) experiments that are underway.

Note:  In the near future, I hope to run some "A/B" comparison tests between the PIN diode optical receivers and readily-available photomultipliers, such as the 931A and a more modern multialkalai PMT.

Photoresistors:

Photoresistors (such as the Cadmium Sulfide, or CdS types) are photoresistive - that is, their resistance drops upon exposure to light.  While these can be fairly sensitive, they are quite slow in comparison to most other solid-state and vacuum tube devices - on the order of minutes if one is looking at their specifications for ultimate sensitivity in near-total darkness.  It is this extremely slow response that makes them generally unsuitable for optical through-the-air communications work, although they have been used with limited success in short-range voice-bandwidth communications systems.  Another important factor is that the sensitivity of these types of cells is mostly in the green visual wavelength - a distinct disadvantage if one anticipates using red or infrared wavelengths to minimize atmospheric effects.

Phototransistors:

Phototransistors are convenient to use in that they can handle fairly high signals and are inherently self-amplifying, but they are, ultimately, not very sensitive when compared to photodiodes.  The ultimate sensitivity of phototransistors is limited by their high intrinsic noise - much of which is a result of collector-base leakage currents:  It is these noise currents that tend swamp out the much weaker, photon-induced currents at very low light levels.

Photovoltaic cells:

Also called "Solar Cells" these are designed to produce electricity when exposed to light.  As detectors, however, they have a fairly slow response and fairly high leakage current and capacitance - all being distinct disadvantages when trying to use them to detect weak, modulated signals.

Photodiodes:

Photodiodes are essentially very small photovoltaic ("solar") cells, but are typically much smaller in area to minimize the capacitance and they are optimized in their manufacture to minimize leakage currents, intrinsic noise and to provide consistency amongst devices.  When photons hit the surface of a photodiode , electrons are mobilized, generating currents that are generally proportional to the amount of light hitting them.  Photodiodes can also be operated in a photoresistive mode in which the impingement of photons results in current to flow through a reverse-biased photodiode.

Photodiodes are available in a large variety of sizes, from the so-called "Small Area" types to the "Large Area" types.  As the name implies, the primary difference between these is the actual size of the silicon substrate and the larger the substrate, the higher the device capacitance.  The so-called Small Area photodiodes (around one square millimeter or smaller) are most often used in high-speed receivers:  Their low capacitance (often well below 10pF) allows better frequency response to be obtained, but their small surface area can limit their intrinsic sensitivity.  Large Area photodiodes (over 10mm square) have the opposite problem:  They can "capture" more photons, but their response time slowed by their much higher capacitance (in the 100's or 1000's of pF) so they are often used where excellent sensitivity is required without the use of additional optics, but frequency response is not as important.

The optical response of silicon photodiodes is best in the near-infrared around 850-900 nanometers for common photodiodes, but it is still pretty good into the red portion of the spectrum, falling off rapidly at shorter wavelengths.  Fortunately, their response is a reasonable match for optical through-the-air communications involving red and/or infrared wavelengths - the very wavelengths of interest in through-the-air optical communications.  The BPW34, for example, has a peak quantum efficiency of 0.9 electrons/photon at 850 nm, but it drops to about 0.6 photons/electron at 630nm (the approximate wavelength of Luxeon Red LEDs), to 0.4 photons/electron at 530nm (green) and down to around 0.25 photons/electron at 460nm (blue.)  Some diodes have wavelength-specific packaging to limit their response - such as the black (or darkly tinted) "infrared-only" versions that are commonly used for infrared remote controls.

Another type of the photodiode is the Avalanche Photodiode or APD.  This device is somewhat analogous to the Photomultiplier tube in that it has intrinsic amplification and can replace the photomultiplier tube in many applications - but the old photomultiplier technology still wins when it comes to the ultimate in photon sensitivity in many cases.  Like the photomultiplier, the APD requires a high voltage supply, typically in the region of 100-200 volts - but their use is somewhat complicated by the fact that several precautions need to be taken in the design of the voltage source to assure proper performance over varying operating conditions, arguably making them more difficult to use than photomultiplier tubes.  At present, these devices are rather specialized and are rather expensive when purchased new, are difficult to find on the surplus market, and are often packaged as complete detector units that include the power supply and amplifier.  APDs are most useful where fairly low light levels are encountered but high speed is needed.  Recently, testing was done using APDs - some of the details of which are mentioned below.

Comments on appropriately sizing photodiodes:

When used without optics, a larger photodiode will intercept more photons than a smaller photodiode because there is simply a greater "capture area" to be struck by the photons.  When used with external optics, however, the size of the photodiode is less-important in terms of ultimate sensitivity because the light-gathering aperture is no longer just the surface area of the photodiode itself, but the capture area of the optics being used with the photodiode.

If you are using external optics to focus light onto a photodiode's active surface, the size of the photodiode is less important in terms of sensitivity.  What is most important is that not only is the photodiode placed at the focus of the optics being used, but that the area of the photodiode is somewhat larger than the "blur circle" of the lens being used at optimal focus so that all light from the intended source will, in fact, strike the active area of the photodiode.  If too small a photodiode is used, some of the received light may be "wasted" - that is, spill out around the photodiode and not having its photons do the intended job -  that is, making electrons move about!

If a photodiode is used that has a much larger active area than the light focused onto it, several things happen:
What size of photodiode should be used?  This question is one that can be answered appropriately by knowing the characteristics of your optics.  Very high-quality glass lenses should be capable of resolving a distant point of light and focusing it onto a very small area, making the use of a small-area photodiode quite practical.  More imprecise optics - such as Fresnel Lenses or less-precise plastic or glass "conventional" lenses will have a larger "blur circle."

One important fact to recognize is practicality in actual use:  While extremely precise, finely-focused optics may offer the best match for small-area photodiodes, aiming them in the field will be correspondingly more difficult.  Unless it is the highest possible speed that you are after, it may, in fact, be more convenient to use somewhat larger-area photodiodes than those that might be optimally-matched to the size of the blur circle of your lens (and possibly de-focusing slightly) to increase the "spot size" on the diode.  If this is done, the ultimate sensitivity will suffer minimally (provided that the larger photodiode's noise and capacitance characteristics aren't the limiting factor) as all of the intercepted light is still impinging on a photoactive surface, but aiming tolerances may be relaxed somewhat, simplifying setup and potentially improving longer-term system stability.

For more detailed information on photodiodes, read the application note included on the "Modulated Light DX Receiver Circuitry" page.


A good starting point - the K3PGP receiver:
Figure 1:
This is a very sensitive optical receiver designed by K3PGP.  While extremely sensitive, it has rather limited bandwidth.  The version shown is suitable only for nighttime use.
Click on the image for the same-sized version.
Schematic of K3PGP detector

Let us first discuss one of the simplest possible optical detectors - the so-called K3PGP receiver, a typical schematic being shown in Figure 1.  While this receiver is simple, its operation is deceptively complex.

One of the most striking aspects of this receiver is the connection between the gate of the MPF102 and the photodiode:  If ideal component models were used, this would simply be a floating junction and as the photodiode reached full potential, charging would simply stop and the circuit would not function - but real-world physical effects come into play.

In this circuit, at very low light levels, the photodiode is operating as a photovoltaic cell:  As photons strike the photodiode, electrons are mobilized and a voltage appears at the gate of the MPF102 JFET.  Because both the JFET and photodiode exhibit some leakage, this charge will drain away, thus the voltage on the gate of the JFET will eventually reach equilibrium and be more-or-less proportional to the amount of light hitting the photodiode.

Assuming that a typical "medium area" photodiode is used (like a BPW34 - a fairly good, but inexpensive device) the capacitance of the photodiode will be in the general area of 70-80pF.  While this may not sound like much capacitance, this is, in fact, enough to severely limit the frequency response to only a few hundred Hz.  After building and testing this circuit, I observed that the -6dB rolloff point, using a BPW34 diode, was around 200 Hz under very low light conditions.

This rolloff is due largely to the capacitance of the photodiode (about 75 pF) being paralleled by a high a resistance which is intrinsic to the photodiode and the JFET, largely in the form of leakage currents.  If one does some simple math, it can be seen that this leakage resistance could be modeled by paralleling an ideal JFET and photodiode (e.g. those with no leakage currents of their own) with a resistor in the range of 10 Megohms or so to simulate low-light conditions.  It should be pointed out that this is a very incomplete analysis as other factors should be considered (e.g. Miller effect of the JFET, photoconductive effects of the photodiode - parameters that depend heavily on the amount of light, etc.) but this very simple model will suffice for the illustration of the frequency response limitation.

The rest of the circuit is fairly straightforward:  The MPF102 JFET forms a common-source amplifier providing significant gain, while the following common-emitter bipolar stage provides even more gain.  This circuit cannot tolerate very much ambient light before the photodiode will achieve its maximum open-circuit voltage and/or the JFET stage will saturate, so it is only useful for very low light conditions.

Comment:  I constructed the K3PGP circuit using a 2N5457 for the JFET and a 2N5089 transistor for the bipolar stage instead of the MPF102 and 2N5088/2N4124 originally suggested.  Both of these devices are generally better characterized in terms of noise performance and other operating parameters than the original devices.

For a more-detailed discussion of this circuit, see the "Modulated Light DX Receiver Circuitry" page.

The VK7MJ optical receiver:
Figure 2:
The well-proven VK7MJ Optical receiver.  Negative feedback allows this to operate as a transimpedance amplifier and improve bandwidth - but at the expense of sensitivity.
Click on the image for a larger version.
Schematic of the VK7MJ optical receiver

Figure 2 shows the VK7MJ Optical receiver.  This well-proven design was devised by Mike Groth, VK7MJ, having been adapted from circuits used to detect low-level emissions in nuclear medicine.

Before we get to the photodiode portion, let's examine the amplifying portion of this circuit:  The input JFET, a 2N5457, is wired with the BC179 as a cascode amplifier.  This configuration greatly reduces the Miller Effect (where the gate-drain capacitance is multiplied due to the voltage swing of the drain) by having BC179 respond to the varying drain current of the JFET - and it provides a significant amount of gain as well.  The output of the BC179 is buffered by the BC109, wired as a high-impedance bootstrap circuit, which is further buffered prior to the output, by a source-follower circuit using an MPF102.

The biggest difference between this and the original K3PGP circuit is the addition of a negative feedback path from the output to the input.  The addition of this path creates a Transimpedance amplifier, that is, the amplifier to responds more to the current being output by the photodiode rather than the voltage and in doing this, the swamping effects of the capacitance on a changing voltage are effectively reduced:  Any current from the photodiode is amplified and countered by a sample of the inverted output fed back into the photodiode-JFET gate junction through Rf, the feedback resistor.  To maintain amplifier stability, a small amount of feedback capacitance, Cf, is added to counter the photodiode's own capacitance.

By virtue of this (mostly) "current-only" response the frequency response of this circuit can be much better than the K3PGP circuit in Figure 1.  This improvement in frequency response is not without costs, however:  The addition of the feedback circuit and the effectively paralleled resistances decrease the sensitivity of this receiver as compared with the K3PGP circuit, not only by the addition of noise sources from the added components, but by a reduction of the amount of signal from the photodiode itself, further dropping already-weak photon-induced currents farther down into the noise floor of the active and passive components.

An additional feature of the VK7MJ circuit is the application of reverse bias on the photodiode.  In this circuit, about 5 volts of reverse bias is applied, effectively reducing the photodiode's capacitance from around 75pF (for an unbiased diode) to something in the area of 20-30pF.  This has the expected effect of improving the bandwidth as well, thus reducing the required amount of negative feedback that would be required to accomplish the same amount of bandwidth improvement, thereby improving the amplifier's low-noise performance.  One caveat of the addition of reverse bias is that it has the potential to increase Shot noise - but this is a rather minor penalty at voice frequencies, as it turns out, and only seems to be a significant factor at very low audio (<200 Hz) frequencies.

Note that noise performance may be improved by increasing the value of Rf (consisting of R3 and R4 on the schematic) the feedback resistor - at the cost of a reduction of bandwidth.  This circuit shown does not have sufficient gain to allow effective use of a feedback resistor more than 50-60 megohms or so, so further increases in this resistance will not necessarily improve performance.  Note that below the "gain limit" imposed by the maximum value of Rf (and the noise floor of the devices) that S/N will increase.  For example, as noted by Yves, F1AVY, increasing Rf from 10 Megohms to 40 Meg will cause a fourfold increase in signal but only a doubling of the noise, resulting in a net doubling of the signal-noise voltage ratio, but all of this is at the expense of reducing the bandwidth.

As can be seen in Figure 2 there another variation of the circuit, notably the "daylight" alternative circuit.  This modification allows the circuit to operate under higher ambient light conditions by capacitively isolating the DC response of the photodiode from the rest of the circuit.  Were this AC coupling not done, a combination of the increasing photoconductivity of the photodiode (in response to the higher light levels) and the higher photovoltaic output would saturate the amplifier stages fairly easily, causing them to slam to a power supply rail.  The addition of this "daylight" circuit does have inferior nighttime performance as compared to the DC-coupled circuit, mostly owing to the addition of another 10 Megohm resistor across the photodiode:  It should be noted that this resistor causes further attenuation of the photodiode's output (dropping it further into the JFET's intrinsic noise level ) and is, itself, a potential source of thermal noise.

Comment:  I constructed a version of this receiver using a 2N5457 for the JFET, a 2N5087 in lieu of the BC179, a 2N5089 for the BC109, and an MPF102 as the source follower:  All of these devices have equal or better performance specifications than the ones suggested on the schematic:  It is this circuit that I use as my "standard" reference.

For a more-detailed discussion of this circuit, see the "Modulated Light DX Receiver Circuitry" page.

Improving on the VK7MJ receiver circuit:
Figure 3:
Top:  Schematic of the improved transimpedance optical receiver, version 2.02.  Bottom:  As-built prototype of the circuit wired in "PIF" configuration.
Click on either image for a larger version.
Schematic of version 2.02 optical receiver
As-built prototype of the version 2.02 circuit in PIF configuration

While the VK7MJ receiver is a well-proven and solid design, it occurred to me that there were several things that could be done to eke a bit more noise performance out of it - as well as make it a bit more versatile:
The results of these modifications may be seen in the circuit shown in Figure 3.

As can be seen, the cascode arrangement is maintained with Q1 and Q2, but a significant difference is the addition of Q3, a bipolar current source.  Q3 is used to maintain a fairly constant, high level bias current in Q1, the JFET, to minimize its noise contribution, but because the Q3 current source operates at a fairly high impedance, it can supply current to the JFET without incurring signal losses that would otherwise occur were a low ohmic value of resistance used to supply a similar amount of current.  A further advantage of the current source is its ability to work over a wide range of supply voltages without the need of significant readjustment.

The bipolar (Q2) section of the cascode arrangement operates at a comparatively low current and high impedance and by doing so it can operate at fairly high gain without requiring particularly high supply voltages.  This cascode circuit is somewhat unusual in that it is self-biasing:  Because the drain voltage of Q1 can vary depending on differing conditions and with different devices, it is necessary to have Q2 maintain a more constant current contribution under all conditions - but it is also essential that the AC base impedance be quite low in order for it to have high signal gain, hence the R6/R7/C3 arrangement which allows Q2 to "track" Q1's DC properties.

The remainder of the circuit consists of a simple noninverting op amp gain stage:  There is nothing particularly special about this amplifier, except that it should be of fairly a low noise type, but exotic amplifiers need not be used.  In this case, it is wired to provide a voltage gain of about 32 - enough to provide enough source signal for a feedback circuit, but the gain could easily be made variable by substituting a potentiometer for R13.

Comments:

Adjusting for the proper amount of feedback:

As can be seen from bottom of the diagram, several configurations are offered - and we want to apply the "standard" configuration of using a high-value feedback resistor.  One of the ways that this circuit differs from the VK7MJ circuit is the way in which feedback is applied:  There are provisions to vary the amount of signal (using R10) being put into the "hot" side of the feedback resistor and thus provide the exact amount of feedback to obtain a flat frequency response.  This adjustment is approximately thus:
Potentiometer R11 is used to set the proper amount of gate bias.  For good operation, this is typically set at a voltage that is roughly equal to (or slightly below) Q1's drain voltage  and this will often vary from one JFET to another so the best voltage for lowest-noise operation (particularly with feedback resistors above 50 Megohms) will have to be determined by experiment.  Perhaps the easiest method is to make the adjustment in total darkness, but with a weak (very dim) optical signal, adjusting R11 from one extreme where the receiver works properly to the other, and then setting the potentiometer in the middle of that range.  Note that the bias voltage can be tweaked somewhat to improve performance under conditions of high ambient light.

It should be noted that with the addition of R10, the "feedback adj" that the "Cf" (feedback capacitor) noted in Figure 2 may not be required if R10 is adjusted properly, with existing capacitance of the feedback resistor and other components being adequate.  It has been suggested that slight improvements in performance may be possible with the addition of a small amount of additional of feedback capacitance (about 0.5pF to 2pF) and a reduced amount of feedback, but I did not note any obvious performance advantage in doing so.  With the addition of this small capacitor, however, lower levels of feedback (which means higher value of feedback resistor) may be used, possibly providing potential benefit in terms of gain/noise performance for a given bandwidth.  If the circuit tends to oscillate or is excessively "peaky" in terms of frequency response and adjustment of R10 doesn't seem to help, try a larger amount of capacitance for Cf - but it is unlikely that much more than 5pF would ever be required.

Improved ambient light tolerance:

One of the benefits of this circuit as compared to the original VK7MJ circuit is that it is quite resistant to ambient light, being able to tolerate wide variations without saturating:  This property makes this receiver a good candidate for "general" use in a wide variety of conditions - from total darkness to an urban light-pollution setting - and maybe even for "attenuated" daylight experimentations.  Further testing is required to see how the ambient light tolerance of this circuit compares with that of the "daylight" version of the VK7MJ circuit.

Operation over a wider supply voltage range:

Another advantage of this circuit design is that it operates well over a wide voltage range - from 7 to over 14 volts, drawing from 7 to 15 milliamps, depending on the voltage.  A caveat here:  At differing supply voltages, not only does the gain of the Q1/Q2/Q3 circuit change somewhat, but so does the amount of reverse bias on the photodiode and these two factors will affect the optimal setting of R10, the feedback adjustment as well as Cf, if it is used.  In practice, one would set R10 at the voltage at which operation was expected, but good performance could still be expected (albeit with a somewhat different frequency response) at different voltages.  If a Zener diode (9 volts or so) is installed from the base of Q4 to ground, this problem can be avoided if the power supply is above 10-11 volts and regulation is occurring.

It is important to be aware that your choice of op amp may also be a limiting factor in how low of an operating voltage will still yield good performance.  I found that the circuit still performed well (if R2's value was reduced as mentioned below) at about 6 volts - even though this was below the published supply voltage specification of the TL082 op amp that I used.  At these low voltages, the gain of the JFET/Bipolar circuit drops off noticeably and the reduction of the photodiode's reverse bias causes frequency response to suffer due to increased capacitance, both being factors that require a readjustment of the feedback.

Comment:  Resistor R2 in Figure 2 could be reduced to 100 ohms if it was determined that a zero or slightly positive gate bias was appropriate for the JFET used.  Lowering the gate bias would also allow for a commensurate increase in the reverse bias of the D1, the photodiode as well as permit the circuit to operate at a lower operating voltage.

Note about the "PIF" configuration:

In Figure 3 (both on the schematic and in the caption) there is mention of a "PIF" (Photodiode In Feedback) configuration.  This configuration is mentioned in the paper "Low-Noise Photodiode Amplifier Circuit" by Hyyppa and Ericson (IEEE Journal of Solid-State Circuits, Vol. 29 No. 3, March 1994, pp. 362-365) where a small amount of negative feedback was applied to the "cold" end (e.g. the non-signal side) of the photodiode.  The claimed advantage of this circuit is that it eliminates the need to establish a feedback path at the junction of the photodiode and the gate of the JFET - a potential noise source.  A copy of this article may be found in the "Files" section of the Optical DX Yahoo Group - note that membership in this group is required to access this file.

As noted at the bottom of the schematic shown in Figure 3 there is a mention of a circuit configuration to provide the "PIF" circuit.  When tested, it was observed that the PIF circuit did, in fact, have a performance advantage over the conventional "feedback" type circuit - but only below the "knee" frequency - that is, the frequency at which the capacitance of the photodiode was causing a 6dB/octave rolloff to occur:  At frequencies above this "knee" frequency it offered no advantage.

What this means is that in experiments using a "medium area" photodiode like the BPW34, there was no performance advantage above about 200 Hz.  This has to do with the fact that at the higher frequencies, the parallel capacitance of the photodiode essentially bypasses the feedback, negating the beneficial effects of the feedback.  It is worth mentioning that the photodiode mentioned in the article (the Siemens SFH229) has a much smaller area than the BPW34 and the "knee" frequency would be quite a bit higher than with the BPW34 owing to the lower capacitance of that device.  Although experimentation with the "PIF" circuit didn't prove any significant advantages for voice-bandwidth circuits, further experimentation may be warranted.

Test setup to determine relative circuit performance:

In order to provide a means of evaluating relative circuit performance, I constructed a "Photon Range" in a basement room without any windows.  This test setup consists simply of a red diffuse lens LED attached to the ceiling while the circuit under test is placed on the floor (about 7 feet, or about 2.1 meters) directly below it:  In no case did the LED or the receiver's photodiode have any optics.  The LED and receiver are connected via wire to an adjacent room, and using a function generator, the LED is driven with a square wave and the current is set to just a few 10's of microamps - just enough to be able to tell that the LED is illuminated at a distance of several meters.  The use of the generator allows the LED's modulation frequency to be varied from less than 1 Hz to several megahertz, although a frequencies above 10 kHz were not routinely used as the computer's sound card's input frequency range was the limiting factor.

The performance of the optical detector was measured by using a laptop computer running the Spectran program at a bandwidth of 1.3 Hz.  The signal-noise ratio was checked using the same "standard" VK7MJ receiver - and the same unit was used for all tests.  For each test session, the first and last readings were done with this "standard" receiver not only to verify that the equipment was configured the same as with previous tests, but also to provide a basis for comparison for the other tests and to make sure that the amount of light emitted on the "Photon Range" was consistent during the entire testing session.

In order to provide the best measurements, it was usually necessary to operate the laptop computer from battery to minimize introduction of coupled AC line currents into the receiver.  Because these tests were done indoors, the circuit under test was placed in a recessed, grounded metal box (one with a top open to the LED mounted on the ceiling) to provide shielding from stray AC fields that would have otherwise ingressed the receiver, making measurements difficult.

Figure 4:
Weak signal comparisons of the circuit of Figure 2 and that of Figure 3.
Click on the image for a larger version.
Comparison of VK7MJ circuit and the circuit in Figure 3
Comparisons of receiver circuits:

Notes:
 With the described test range, I was able to quantify performance differences between the various circuits, so I decided to test the sensitivity of the VK7MJ circuit as compared to the circuit shown in Figure 3.

The first readings were done using the circuit in Figure 2 (Version 2.02) as a basis of comparison.  For those tests, I used a feedback resistor (Rf) with a value of 22 Megohms.  I then checked the optical receiver shown in Figure 3, also using a 22 Megohm feedback resistor, and found the readings to be with a few 10ths of a dB - too close to call.  In each case, the signal-noise ratio was 19.5-20.5dB, depending on frequency:  A typical result may be seen on the bottom row of Figure 6.

I then changed Rf to 54 Megohms in both receivers, making modifications/adjustments to the feedback circuits as necessary, and then re-ran the tests:  The results are shown in Figure 4.  As can be seen, the two circuits perform similarly - about 3dB better than with the 22 Meg feedback resistors - but the Figure 3 circuit has a slight performance advantage over the original VK7MJ circuit.  This slight improvement is likely a result of somewhat improved noise performance of the JFET input stage's bias and amplification circuit as well as a slightly lower noise contribution of the feedback circuit, but it is also likely that some of it is due to normal variations in the active devices being used.

After the test with 54 Megohms of feedback resistance, I changed the Figure 3 circuit to use a 148 Megohm feedback resistor.  (The VK7MJ circuit did not have sufficient gain to permit resistors higher than about 60 Meg to work properly.)  I noted nearly identical performance with the 148 Meg resistor as obtained with the 54 Meg resistor in terms of signal/noise ratio, indicating that the sensitivity was likely being limited by the performance of the photodiode and/or the JFET.  I did note, however, that with the 148 Meg feedback resistor, the noise performance was more strongly affected by the setting of R11, the bias resistor, than it was with a 54 Meg feedback resistor, and that there seemed to be a wider degree of component-related performance variation:  This has the implication that with the careful selection of the lowest-noise components and the optimal setting of R11, better performance may be obtained with the 148 Meg resistance than with Rf=54 Meg.

Comment:  With a 54 Megohm feedback resistor, the -1dB bandwidth of each receiver was about 30 kHz, but the receiver in Figure 3 dropped to about 8 kHz or so when Rf was increased to 148 Meg.  Given the results of these tests, there is likely to be little benefit of using a feedback resistor of higher than 50-60 Meg unless very careful component selection is made.

Evolution of another receiver circuit:

Having done these tests, I though back again to the K3PGP circuit (Figure 1) and wondered about its performance.  In my original tests, I noted that while this circuit provided excellent sensitivity, its frequency response was, by itself, somewhat unusable for speech owing to a 6dB/octave R/C rolloff that began at 200 Hz or so.  Despite the rolloff, I wondered what the signal/noise ratio would be for a signal detected with this circuit and, more importantly, if it would be better or worse (and by how much) than that of the VK7MJ circuit in Figure 2.

Digging up my original K3PGP prototype, I did some tests which yielded some interesting results, so repeated these same tests using the Version 2.02 circuit in Figure 3 which I reconfigured to be without feedback or reverse bias on the photodiode - essentially converting it into the K3PGP circuit in that the photodiode was without bias or feedback of any kind - and found that it slightly outperformed the original K3PGP design, likely owing to the higher FET current, if not normal component variations.  In these tests I noted that while the signal output dropped off by 6dB per octave (above the "knee" frequency of 200 Hz or so) the noise dropped off at nearly the same rate!  In other words, the signal/noise ratio decreased at a slower rate versus frequency than the amplitude did.  At this point I decided to apply reverse bias to the photodiode and noted that higher frequency (>200 Hz) S/N and gain performance improved markedly.
Figure 5:
Top:  Optical receiver without feedback, version 3.02.  Top center:  Interior of enclosure with version 3.02 circuit.  Bottom Center:  Exterior of enclosure.  A strip of felt was used along the lid to prevent light ingress between it and the body of the enclosure.  Bottom:  The prototype of the Version 3.01 circuit (e.g. no lowpass filter) as mounted in the "cheap enclosure."  Despite the lack of significant shielding, the circuit has not proven to be particularly susceptible to AC or RF fields as compared to those circuits using feedback.
Click on an image for a larger version.
Schematic of version 3.02
Interior of as-built version 3.02 circuit
Exterior of as-build version 3.02 circuit
Prototype the Version 3.01 circuit as mounted in the "cheap" enclosure.
With these promising results I constructed another prototype, adding to it an op-amp differentiator to compensate for the 6dB/octave rolloff caused by the photodiode's capacitance and the lack of any feedback:  The ultimate result was the circuit shown in Figure 5.  Because of the Q2/Q3 current source/cascode sections, the circuit is able to operate properly even if the JFET (Q1) is conducting heavily.  It is because of this that the circuit still works even when the photodiode is reverse-biased, causing the gate-source voltage to become positive and even causing the gate-source junction to conduct.

The circuit operation is nearly identical to that of the Version 2.02 circuit in Figure 3 - at least until U1A.  U1A is simply a unity-gain follower, used to establish a low-impedance source for U1C, which is a differentiator.  The differentiator has a classic 6dB/octave emphasis curve that nicely cancels out most of the R/C rolloff of the photodiode.  As in the Version 2.02 circuit, the op amp, U1, should be fairly low noise:  I tried a TL074, TL084, and LF347 with equally good results - all much quieter than the Q1/Q2/Q3 amplifier, but an LM324 was noticeably noisy, decreasing the the receiver's ultimate sensitivity.

Added to this circuit is a 3.5 kHz lowpass filter that may be switched in and out with S2 to remove some of the "hiss" coming from the photodiode amplifier - the high frequency components of which could cause "ear fatigue" when trying to dig out signals with poor signal/noise ratios.  The lowpass filter also has the advantage that if an optical signal is being received that is generated using PWM techniques, the majority of the PWM switching components are removed - an important consideration if you plan to record the audio to a digital or magnetic take recorder or computer,  not to mention preventing a normal audio amplifier from distorting from the PWM frequency components.  Note that the lowpass filter adds about 7 dB of audio gain.  Also added is a gain switch (S1) - just in case one is trying to detect a weak signal and one needs as much audio as possible.

Comments:
 How it works:

It is worth mentioning the similarities and differences between this circuit, the K3PGP circuit, and a one using a transconductance amplifier - like the VK7MJ circuit:
 Perhaps the most unique aspect of this circuit is the fact that the JFET's gate-source junction is, in fact, conducting!  From what I can tell, there are few (if any) other published circuits in which the gate of the JFET being biased into conduction is an essential aspect of their operation.  Furthermore, there is surprisingly little information to be found in the literature describing how JFETs operate under conditions where gate current is flowing.  In my experimentation, I have observed that the drain current of most depletion mode JFETs will continue to increase even after the gate-source junction begins to conduct - even to current levels well in excess of the saturation current specified in the device's datasheet.  As you might expect, the gate-source voltage begins to follow the classic voltage/current diode curve once gate-source conduction occurs.

Concerning this circuit configuration, some interesting things happen:
Performance of the Version 3.02 circuit

Even before I did more scientific, comparative testing in my "photon range," I could tell, by ear, that this circuit easily outperformed any others that I had tried:  The results of comparative performance testing may be seen in Figure 6.

Along the bottom row is the performance of the standard test receiver, the VK7MJ circuit shown in Figure 2.  The top row of Figure 6 shows the performance of the circuit in Figure 5 when operated from an 11 volt supply - a configuration that results in about 8.5 volts of reverse bias across a BPW34  photodiode.  As can be clearly seen, the signal/noise ratio at 1250 Hz is about 14dB better than the original VK7MJ circuit - and 8-9 dB better than the Version 2.02 circuit in Figure 3.  As expected, performance degrades with higher frequency, but even at 5 kHz (the highest frequency that I could test with my laptop) it was still outperforming any other circuit that I had tried.

This circuit isn't without its drawbacks, though as its high frequency response does have a distinct limit.  If you wish to have a frequency response that extends above 15kHz or so, it is likely that a transimpedance amplifier such as the VK7MJ circuit in Figure 2 or the version 2.02 circuit in Figure 3 will easily outperform it (in terms of signal/noise ratio) at these higher frequencies when using a medium-area photodiode like the BPW34 - but it is worth remembering for all optical detectors, high-frequency performance gains may be obtained (at the possible expense of some additional optics) with the use of small-area photodiodes.
Figure 6:
Performance comparisons of the VK7MJ receiver shown in Figure 2 and the version 3.02 receiver shown in Figure 5.
Click on the image for a larger version.
Performance comparisons of the Version 3.02 receiver and others

Another aspect of this circuit is that it is fairly limited in its ability to work with good sensitivity in ambient light when compared with the circuit shown in Figure 3 - but it does tolerate ambient light much better than the VK7MJ circuit, owing to its inherent self-biasing.

Notes about frequency response with the "Version 3.02" circuit:

One peculiar quirk of this circuit at higher levels of ambient light is the fact that the frequency response becomes skewed and the audio will begin to sound distinctly tinny.  The reason for this is that at higher levels of ambient light, the photodiode becomes more and more conductive, and as this happens the capacitance of the photodiode - the primary limitation of high frequency response - is increasingly shunted by the lower effective resistance, thereby improving high frequency response.  Additionally, as the photodiode current (along with the gate current) increases, the gate impedance also drops, further reducing the effects of the photodiode's capacitance.  Under very low-light conditions, the "knee" frequency is around 150-250 Hz for the BPW34 (depending on the photodiode's capacitance and the amount of reverse bias) and is generally unnoticeable (except as a slight lack of "bass" response) but with higher levels of light this "knee" moves well into the middle of the audio range where the post-emphasis effects of the differentiator become quite obvious in the audio.

Comment:  It would be a fairly easy matter to provide an extra control to adjust the "knee" frequency of the differentiator to manually compensate for frequency response differences as well as an extra resistor and capacitor to recover the "lost" bass response - but the low-frequency rolloff may be an advantage because of its tendency to attenuate 100/120Hz hum from AC-powered light sources.

With the increase of ambient light comes a dramatic increase in Shot noise as well - both from the photodiode itself (and possibly the JFET) and the source of ambient light.  While this effect is, for all practical purposes, negligible in the voice frequency range at very low light levels, it will eventually become a roar of noise at much higher light levels - those at and beyond the point where the audio becomes "tinny."  For this reason, if operating under conditions of high ambient light or where the transmitting station is backgrounded by some light, improved performance may result from adding a bit of optical attenuation to the receiver!

Remember:  The receiver tested had about 14dB better intrinsic sensitivity than the original VK7MJ receiver, so you may very well be able to tolerate a bit of "optical attenuation" on this circuit and still have performance that is on par with the "non daylight" version of the VK7MJ circuit.  It is also worth noting that almost all (ambient) light sources contain large amounts of noise, so it may be that the distant signal source may simply be drowned in a sea of optical noise from these other sources, anyway.

Other circuit comments:

As can be seen from the schematic in Figure 5 there are provisions to add an external bias voltage.  Under zero and low-light conditions, the leakage currents of the photodiode are in the nanoamp range - far lower than the leakage of the bypass capacitors, so one or two 9 volt transistor radio batteries wired in series with the main supply may be used to provide this voltage if it is desired - and an "off" switch for the bias supply is likely unnecessary.  Under brighter conditions, the photodiode will conduct more heavily - eventually being current limited by R1 and R2, but under normal conditions, the amount of leakage experienced is likely to be a small fraction of the self-discharge of the batteries that you might use.  As noted in Figure 6, under low-light conditions, however, a higher bias voltage (up to 30 volts) can allow for further improvement in the signal-noise ratio at higher audio frequencies (e.g. above 2 kHz or so.)

Figure 6 also shows some tests using a Hamamatsu S1223-01 photodiode - a larger, lower-leakage photodiode than the BPW34.  At lower frequencies, this device performs better - partly because of its larger surface area (13mm^2 for the S1223-01 versus 7.5mm^2 for the BPW34 ) allowing it to accumulate more light in the absence of optics - but at higher frequencies, its higher capacitance begins to degrade performance.  Figure 6 nicely illustrating the limits of the efficacy of this circuit at higher frequencies while providing a dramatic demonstration of the improvement obtained by the lower junction capacitance associated with higher reverse bias.

It is worth noting that the very low (below 200 Hz) frequency performance may be hindered by the application of reverse bias.  In the case of Figure 6, this is shown by a slight degradation at 150 Hz when using the S1223-01 photodiode - but this effect is more pronounced at still-lower frequencies where the noise due to the reverse bias leakage current has more impact.  What this means is that for very low frequencies (in the 10's of Hz) it is likely best to follow K3PGP's advice and to not apply reverse bias.

Theoretically, the S1223-01 should, when no optics are used, have about 4.7dB better sensitivity than the BPW34 simply because of its larger surface area - but this does not take into account the fact that more surface area also means more capacitance and more photodiode junction material to contribute noise (e.g. a higher "NEP") - nor does it necessarily take into account the noise from the rest of the amplifier.  When used with external optics, the size of the photodiode is likely to be dictated more by how the distant light source is focused onto the photosensitive material:  In this case, it is best to use as small a photodiode as possible - provided that the photodiode is at least as large as the "blur circle" of the lens system being used.  Doing so minimizes photodiode capacitance and leakage current - both of which improve the signal/noise ratio.  Additionally, a smaller-sized photodetector can be used to reduce the beamwidth of the optical receiver - something that can further improve the signal-noise ratio by virtue of reducing the response to off-axis light sources.

Simplified version of the "Version 3" optical receiver
Figure 7:
Simplified version of the "Version 3" optical receiver - see text.
Click on the image for a larger version.
Simplified version of the Version 3 optical receiver

Figure 7 shows a simplified version of the "Version 3" optical receiver.  The performance of this circuit is the same as that shown in Figure 5 (above) but a bit of "minimizing" has been done - most notably the removal of the power-supply filter (Q4) and the low-pass filter (U1D.)  Retained is the high/low gain switch (S1) and the reverse-polarity protection (D6 and TH1) as these items were considered to be very important.  In this circuit, R4 is lowered to 10 ohms to reduce the voltage drop and allow operation from lower supply voltage, and it is used to measure the current through Q1, but this may be optionally bypassed with J1 after measurements are completed.

This circuit is intended to be operated from its own, single 9-volt battery - which is one of the reasons why the reverse-polarity protection is present:  It is extremely easy to momentarily connect a 9-volt battery backwards while fumbling in the dark - something that could instantly destroy U1.

While the use of an LM833 has been shown, practically any low-noise dual op-amp may be used.  Note that operating an LM833 from a single 9-volt battery pushes the low voltage limit of this device which is 10 volts:  Testing has indicated that the LM833 seems to operate reasonably down to at least 7 volts, but this is not a guaranteed specification!  If you are constructing this, keep in mind that there are many other op-amps that offer good performance but can operate from much lower supply voltages, such as the National LM4562 or the LMC6482.

Suitable enclosures and shielding:

The two center pictures in Figure 5 show the enclosure, constructed of double-sided copper-clad circuit board material, containing the as-built version 3.02 circuit.  It should be noted that all signal and power leads are passed into and out of the enclosure through solder-type feedthrough capacitors in order to avoid the ingress of RF energy.  A careful observer will also note that in the center of the enclosure, set back from the hole, one can see the photodiode and Q1, the JFET:  Note that the photodiode-to-gate connection is done in midair to avoid any possible leakage paths that might occur on circuit board material.  Also, the photodiode is set back from the hole by several millimeters to permit the enclosure itself to provide some shielding of the photodiode from any e-field energy that might be present.  Finally, note that the top of the inside of the enclosure is painted black to minimize reflections from off-axis light sources that might affect the sensitivity of the receiver.

As mentioned before, this circuit is less-susceptible to the effects of stray AC and RF fields than either the VK7MJ or the Version 2.02 circuit for one simple reason:  The most sensitive junction (that of the JFET's gate and photodiode) has nothing else connected to it.  In the case of the other circuits, a feedback resistor is connected at this most-sensitive junction and will more-readily pick up any stray fields that may be present.  Despite its relative immunity, it is still quite sensitive to AC fields, so one should still employ good construction practices when building this circuit.

Thoughts on further performance enhancements:

The sensitivity performance of the Version 3.x circuit is not to likely to be increased too much, although some minor gains (a dB here and a dB there) may be had from things like:
Other Comments:

JFETS
 As mentioned above, I chose to use a 2N5457 instead of an MPF102.  While the MPF102 is a pretty good device, a quick glance at the spec sheets will show that it is broadly characterized - that is, given a hundred devices from different manufacturers made at different times, you'd see that the measured parameters were all over the place.  The 2N5457 is a much more consistent device and one is likely to be more similar to another than MPF102s are to each other.  Having said that, it is still reasonable to obtain many more devices than you need and sort through them, using only those that have the best performance.  If all you have is a bunch of MPF102s, it may be worth going through several of them, finding the one(s) that have the lowest noise - something that is also likely to be related to the highest zero-bias drain current.

Finally, it is worth mentioning that some JFETs mayl NOT operate in a useful way when a positive gate voltage is present.  It seems as though many common devices like the MPF102 and 2N5457 continue to provide lower channel resistance even if the gate goes into conduction, as some "pinching room" of the channel still seems to be available.

There are other JFETs such as the Philips BF862:  This JFET is quite a remarkable device in that its designers seem to have achieved high transconductance and high current without inordinately high gate capacitance.  To be used with this circuit, however, modifications will be required as this FET's drain current is much higher than that of the 2N5457 - in the range of 15-25 milliamps.  In preliminary testing with this transistor, the source resistor was removed and the current source (Q3) was reworked to use a high-beta PNP device (although another current source topology - such as a Widlar or cascode current source may be more appropriate) to handle the current.  Initial testing shows that this device is a bit more "finicky" than the 2N5457 but that its performance it was at least equal:  It has yet to be determined if the much higher drain current capability of this device will provide any significant advantage.

PIN and avalanche photodiodes:

As mentioned before, when using external lenses, the size of the photodiode has less effect on the total amount of signal that will be recovered:  Envision a "spot" of light focused onto the photodiode, and as long as the photodiode is equal in size or larger than that "spot" the photodiode will not intercept more signal from the distant source.  The larger photodiode has a disadvantage, though:  It has higher capacitance that can swamp out the desired signal, and the larger diode has more semiconductor material that can contribute thermal noise and dilute the already-weak signal - not to mention the fact that a larger photodiode also implies a wider beamwidth and a decreased ability to reject off-axis, possibly interfering signals.  It has also been implied that if the chosen photodiode has a smaller active-surface area than the "spot" focused by the lens, some of the energy from the distant signal will be missed.

Additional testing was done with an Avalanche Photodiode (APD), a Silicon Sensor AD1100-8.  APDs, unlike standard PIN photodiodes, have an inbuilt gain mechanism.  Ideally, the gain of the APD itself would be able to overcome the intrinsic noise of the amplifier to which it is attached, but real-world devices contribute their own noise, thereby limiting the ultimate sensitivity.  In order to substitute an APD, several minor circuit modifications were made, referring to the schematic in Figure 5:
During testing, up to 130 volts (the rating of the photodiode that I was using) was applied with the receiver in the "photon" range and at these different voltages, the signal-noise ratio of the test signal was measured.  Also, the results were scaled to take into account that the active area of the APD being tested (1 mm^2) was much smaller than the PIN photodiode being used for comparison (7.5 mm^2):  Without any optics, the signal on the photon range received by the photodiode was proportional to the area of the photodiode, making such an adjustment necessary to provide meaningful comparisons.

At low voltages (<15 volts) the performance of the APD was comparable to that of a standard PIN photodiode - not surprising, as the APD operates as a standard photodiode at very low bias voltage.  At high bias voltages (>75 volts) the level of the test signal increased significantly - but the device's intrinsic noise floor increased even more:  While the gain increased by about 40dB, the noise increased by about 60dB, possibly causing the weak signal to become drowned in noise.

Optimal performance occurred in the 35-45 volt range:  While the intrinsic gain of the APD was much lower at these voltages, its noise contribution was also significantly reduced, improving the signal-noise ratio of the test signal.  While extensive testing has yet to be completed, initial results indicate a S/N improvement of at least 6-9dB over the best performance using a PIN photodiode.

It should be mentioned that using the circuit of Figure 5, which has no feedback mechanism, the raw frequency response of the APD-based circuit still has a 6dB/octave rolloff characteristic above the "knee" frequency - but this should not be surprising considering the intrinsic device capacitance and high impedance involved.  With the added intrinsic APD gain, however, the effective frequency response can be improved, as the higher-frequency components, which are decreasing in amplitude due to the capacitance, are also being amplified by the APD and do not drop into the noise floor as quickly.

Beware of microphonics and current loops!

It is also worth mentioning that, for a number of reasons, that all of the circuits shown on this page tend to be somewhat microphonic - that is, they will respond (in differing degrees) to mechanical vibrations.  It is very important that any loudspeakers used be located away from the optical receiver to avoid acoustic feedback!  This simple fact precludes the inclusion of a speaker contained within the same housing as the receiver itself.

Also note that it is best that the optical receivers NOT share the same power supplies as either the transmitter or speaker amplifier:  Doing so is inviting trouble, as circulating currents from these other devices tend to find their way into the (extremely!) sensitive receiver and will likely result in crosstalk and/or feedback!  It is for this simple reason that the optical receiver itself has been designed to operate from a single 9 volt battery!

Additional comments about high-sensitivity optical receivers in general:

Why not use low-noise op amps in the front end?

One might ask why discrete transistors were used instead of high-performance, low-noise op amps (like the LT1115, LMH6624, LMV751 to mention but a few) in the first stage of the optical detector?  The answer is that readily available op amps - even very good, low-noise ones - will not perform as well as a single JFET amplifier.  Why might this be?  As Bob Pease points out in his article on Transimpedance Amplifiers (see the article "What's All This Transimpedance Amplifier Stuff, Anyway" in the January 8, 2001 issue of Electronic Design) one has to add a JFET in front of an op amp in order to obtain the best possible noise performance for several reasons:
Of course, one could replace Q2 with an op amp to maintain the cascode configuration, but that would not likely offer any performance enhancements:  If you do, you must keep in mind that this stage should be self-biasing (like the Q2 circuit) to accommodate different voltage/current conditions present at the drain of Q1.

Also available are a number of devices that have integrated photodiodes and op amps contained within a transparent package, such as the TI (formerly Burr-Brown) OPT101, OPT201, OPT212 and similar.  While these components are useful in minimizing size and component count, experiments by others indicate that they offer little - if any - performance advantage over a less-expensive discrete photodiode coupled to a low-noise op-amp and have performance that is noticeably inferior to that of the VK7MJ circuit.


Acknowledgments:

Credit should be given to the fine work by K3PGP and VK7MJ for setting the groundwork for these experiments.  Also appreciated are comments by Yves, F1AVY on the Optical DX Yahoo group concerning various aspects of the operation of these circuits.

Related pages:

Return to the KA7OEI Optical communications Index page.

If you have questions or comments concerning the contents of this page, feel free to contact me using the information at this URL.
This page and contents copyright 2007-2008.  Last update:  20080714