Saturday, February 11, 2017

A novel APD-based speech bandwidth optical receiver

In a previous posting I wrote about a novel application of a JFET (Read about that in the article "Gate current in a JFET - The development of a very sensitive, speech-frequency optical receiver" - link) in which the flow of gate current was integral to the operation of a photodiode-based optical detector.  In testing this circuit, which included tests using an indoor "photon range" and out in the field, it was observed that the sensitivity of this circuit was, at "audio" frequencies, on the order of 8-20 dB better in terms of signal/noise ratio than any of the more conventional "TIA" (TransImpedance Amplifier - read about that circuit here - link) circuits that had been tried.

In the analysis of this circuit it was determined that several factors contribute to the ultimate sensitivity, including:
  • The intrinsic noise of the JFET.  This can be minimized by hand-selection of the device itself for the lowest-possible noise as well as selecting a device that can operate at a higher drain current to reduce the "bulk noise" - or even the use of several JFETs in parallel.
  • The contribution of noise by other circuitry.  In the design this was minimized through the use of a cascode circuit topology as well as the use of a low noise, high impedance current source to supply the bulk of the drain current and the complete avoidance of other components being connected to the photodiode-JFET circuit junction.
  • The capacitance of parasitic circuit elements, such as capacitance (including the Miller effect) that reduces the amplitude of the signals from the photodiode, particularly as the frequency increases, effectively reducing the signal-noise ratio.
  • The noise contribution of the photodiode itself.
Of these factors, the majority of the noise would appear to be due to the JFET itself, particularly above the low audio frequencies frequencies (e.g. below 100Hz or so) where 1/F noise would dominate. One of the possible approaches to get better noise performance is to cool the circuitry, but this is fraught with difficulties related to condensation which would require that the device itself be sealed in an atmosphere (e.g. dry nitrogen) in a manner similar to that used to cool CCD imagers for astronomy.
Figure 1:
The outside view of the completed APD-based optical receiver.  Because
of its extreme sensitivity it must be well shielded to minimize the pick-up
of stray fields such as those from AC mains and radio transmitters/phones.
Click on the image for a larger version.

What else may be done to improve the performance?

Perhaps counter-intuitively, the use of a smaller photodiode can help a bit and provide at least as much signal output as a larger one, provided that the optics can focus the given amount of light from the distant source of light efficiently onto its active area:  A smaller device will have lower self-capacitance shunting a smaller amount of the AC currents being produced in response to the impinging, modulated light in addition to having a lower intrinsic noise contribution.  In the case of an optical receiver the active area of the device is less important than in some other applications as lenses and mirrors may be used to concentrate the light from the distant source onto the photoactive area.

When reducing the size of the device one must assure that the optics themselves will resolve the distant spot of light to an area that is not larger than the active area of the device as well as taking into account additional constraints with respect to the accuracy and stability of the aiming and pointing mechanisms.  For example, using reasonable-quality molded Fresnel lenses of common focal lengths (e.g. an f/D ratio of approximately unity) one can expect only to resolve a spot with a "blur circle" of approximately around 0.2mm at best while high-quality glass optics should be able to reduce this by an order of magnitude or better assuming a suitably-distant source, a corresponding small subtended angle and proper paraxial alignment and focus.  If the resolved spot of light is much larger than the active area of the device - perhaps due to the device being too small for the optics' ability to resolve or due to the quality of and/or misalignment of the lens(es) - there may be an additional loss of available optical energy and signal-noise ratio as some of the light from the distant source is being "wasted" when it spills beyond the active area of the photodetector.
For more information on "spot sizes" using inexpensive, molded plastic Fresnel lenses see the article "Fresnel Lens Comparison:  A Comparison of inexpensive, molded plastic lenses and their relative 'accuracy' and ability to produce collimated beams" - link.
Aside from the reduction of the size of the photodiode or cooling, where else may one eke out greater performance from this circuit topology?

The Avalanche Photodiode:

The Avalanche Photodiode (APD) is a type of photodiode that contains an internal mechanism for amplification.  Simply put, instead of a single photon having a given probability of mobilizing a single electron when it impinges the active area of a standard PIN photodiode.  In an APD, what might have been a single electron being loosed in a normal PIN diode that same single-electron event can cause the mobilization of many electrons via an "Avalanche" effect, providing amplification of the optical signal and hence the name.  The result of this intrinsic amplification is that the output signal from this diode from a given photon flux can be much higher than that of a standard PIN photodiode.

Because the signal from the Avalanche photodiode itself is amplified internally it is more likely to be able to overcome the effects of the capacitance on frequency response as well as the noise intrinsic to the JFET amplifier, support circuitry and components, providing the potential of producing a greater signal/noise ratio for a given signal. Typically an Avalanche photodiode is incorporated into a TIA (TransImpedance Amplifier) with good effect, but what about its use in the previously-described "Version 3" photodiode receiver circuit that utilizes JFET gate current?

The basic design:

From the previous article (link) one can see the basic topology of the "Version 3" circuit using a "normal" PIN photodiode depicted in Figure 2, below.
Figure 2:
A diagram of the "Version 3" optical detector that utilizes JFET gate current.  In this circuit Q1 and Q2 comprise a cascode
circuit with Q3 providing the majority of Q1's drain current while U1b is configured as a differentiator to compensate
for the low-pass effects of the intrinsic capacitance of D1, the photodiode and Q1.  Resistors R1 and R2 along with
C1 provide a filtered reverse bias for D1 which not only decreases its capacitance, but it also biases Q1 to
its operating state where it is drawing maximum drain current.  In this circuit the connection between the Photodiode (D1)
and the gate of the JFET is made in air and not on a circuit board to minimize capacitance, stray signal pickup and
most importantly a source of leakage currents and related noise.
Click on the image for a larger version.
In this design PIN photodiode D1, a BPW34, is reverse-biased via R1 and R2.  One of the main benefits of doing this is that the capacitance of D1 decreases from approximately 70pF at zero volts to around 20pF at the operational voltage, reducing the degree to which high frequency signal are attenuated by this capacitance.  A somewhat less tangible benefit of this is that in addition to photovoltaic currents produced by the impinging light, the bias also allows photoconductive currents to flow from the bias source, through the photodiode and into the gate of the JFET.  As noted in the original article it is the presence of the gate-source junction of the JFET (Q1) and its conduction that limits the gate-source differential to around 0.4-0.6 volts, permitting D1's reverse bias to become established without the need of any additional noise-generating or lossy components.  In this configuration the drain current of the JFET is still proportional to the gate-source voltage (but with an offset of drain current greater than the "zero bias" drain current) and like a bipolar transistor's base voltage and current, the relationship between gate voltage and gate current is logarithmic.

A question now comes to mind:  What about replacing D1 with an avalanche photodiode?

Testing with an Avalanche Photodiode:

Like its more-sensitive distant cousin, the Photomultiplier tube, the avalanche photodiode requires a rather high bias voltage in order to function at maximum gain.  Rather than requiring a kilovolt or so as is needed for a photomultiplier, typical photodiodes may operate with up to "just" a few hundred volts.  Like the photomultiplier, the current required for "dark" operation is minuscule - a few hundred microamps in "dark" conditions is more than enough.

In perusing the various component catalogs I noted that Mouser Electronics carried some avalanche photodiodes - but as expected, there was a price - literally:  Around US$150 at the time for just one APD.  In a compromise between size, availability and cost I chose the AD1100-8-TO52-S1 by First Sensor  (previously known as "Pacific Silicon Sensor") - a device with a round, 1mm2 (1.128mm diameter) active area - a reasonable compromise between cost, size, and the practical limit of the Fresnel-based optics.  This device, which came with its own test sheet, indicated a maximum gain ("M" factor) of approximately 1000 occurring at 134 volts at a temperature of 25C.

In most ways using an APD is just like using a reverse-biased PIN photodiode - except that the reverse bias voltage will be much higher.  Perusing the literature and manufacturer's specifications one will note that many designs depict a temperature-compensated bias voltage supply, but further investigation reveals that this is necessary only if the device is being used at/near maximum gain (and maximum voltage) and/or if it is necessary to precisely maintain a precise gain over a wide temperature range.  For our application we don't really care if the gain changes with temperature, so an arbitrarily adjustable high voltage supply is fine - and actually preferred.

In my initial research I noted that the internal action of any APD suffers an inevitable, but expected, effect:  As the gain goes up with increasing bias voltage, the intrinsic noise of the device itself increases at a faster rate than the gain.  What this means was that there is going to be a point above which a further increase of device gain will cause the signal to noise ratio to decrease even though the actual signal level continues to increase with bias voltage.  With this in mind, the question is "At what voltage might this happen, and would this 'crossover' point occur at a point where we can expect the overall 'gain+noise' to offer a net advantage over a PIN photodiode?"

Building a prototype receiver similar to that depicted in Figure 2 I substituted an APD for D1 using a string of sixteen 9 volt batteries and a 1 megohm potentiometer with a 100k resistor in series with the wiper (and some bypass capacitors to ground on the "hot" side of the diode) in lieu of R1 to set the bias voltage.  Placing this prototype in my "Photon Range" - a windowless room in my house where there is an LED mounted to the ceiling that may be modulated - I compared the sensitivity of this prototype to both my "standard" TIA receiver (the VK7MJ design) and an operational exemplar of my "Version 3" design.

Varying the voltage from 10 volts to around 140 volts I noted that at a bias voltage comparable to the reverse bias applied in Figure 2 (approx. 8 volts) the apparent sensitivity was roughly on par with that of the Version 3 unit using a normal PIN photodiode after the signal levels were corrected to compensate for the smaller area of the APD as compared with the BPW34 (e.g. 1mm2 of the APD versus 7mm2 of the BPW34 - the larger size gathering proportionally more light in this lens-less system).  At around 130-135 volts, the output of the APD-based prototype was very high, but the weak, optical signals from the test LED were lost in the noise.  In the area of 35-45 volts I observed that while the overall signal levels, while significantly higher than they were at 8-10 volts, they were a fraction of what they were at 130 volts but the signal/noise ratio was roughly 6-10dB higher than it was at the lowest voltage when the differences in active area of the APD versus the photodiodes in the test receivers were taken into account.

  • The test receivers used BPW34 PIN photodiodes with an active area of 7mm2 while the APD has an active area of just 1mm2.  Because there are no optics in front of the photodiodes there will be 7 times as many of the LED's photons hitting the larger device, resulting in an approximate 8.5 dB difference in signal/noise - assuming all other parameters being equal.  It is when using the device in this "lens-less" configuration that this factor must be accommodated.
  • While it is theoretically possible to use a photomultiplier tube (PMT) in lieu of an APD, there are several practical concerns.  Even though an "S-1" type of photocathode has a peak in the red-NIR area, its low quantum efficiency makes it a rather poor performer overall.  The "931A" PMT - widely available surplus - has a more typical blue/violet peak response (type "S-4") in which the longer red wavelengths suffer greatly in terms of quantum efficiency.  Field testing of these devices by British amateur radio operators has shown that they offered no obvious advantage over the "Version 3" PIN photodiode design for "red" wavelengths.  As of the time of this writing the use of PMTs with more exotic photocathodes (such as multialkalai and GaAs) that are better suited for "red" wavelengths (but much more difficult to find surplus!) have not been field-evaluated.
A practical design:  The high voltage APD bias supply:

First, a few weasel words:
Even though the currents are very low, there is some risk of injury with the voltages involved (e.g. several hundred volts) and it is up to you to educate yourself about high voltage safety!
If you wish to construct these circuits, be aware of possible hazards and always assume that any capacitors are charged, even after power is removed.

You have been warned!
Because it is not convenient to carry around a lot of 9 volt batteries, a simple high voltage converter was designed to provide the  microamp-level current required for the APD bias supply and it is depicted below in Figure 3.
Figure 3:
High voltage supply for the APD receiver.  U101a is an oscillator that drives Q101 to produce a high-voltage, low-current bias for the APD.  The output is regulated via U101b and associated components to the voltage set by potentiometer R111.  R109 is used to set the highest voltage that may be obtained when R111 is adjusted for "maximum".  R109 is shown with the two "ends" grounded only because it was convenient to wire it this way when the prototype was built.
Click on the image for a larger version.
This design is a simple "boost" type switching converter using a high voltage transistor and an inductor to produce the needed bias.  In this circuit U101A forms an oscillator that drives the high voltage transistor Q101, and when Q101 switches off, the magnetic field of L101 collapses, producing a high voltage spike that is rectified by D101 and filtered and stored by C102, R106 and C103.  To regulate this high voltage a sample is divided-down by R108 and R109 and compared with a 5-volt reference from U102 that is made variable with R111:  If the output voltage is too high, U101b turns on Q102 to pinch off the drive for Q101.  Because I used an "ordinary" op amp with an output that could not go all of the way to the negative supply rail, LED101 was put in series with the transistor's base to provide a drop of around 2 volts to assure that Q101 could be shut completely off.
Figure 4:
Inside the high voltage (bias) supply for the APD receiver.  Potentiometer
R111 and the indicator, LED101, are mounted in the front of the
case.  Both the high voltage generator and the receiver itself are powered
from a single 9 volt battery.  The typical combined current consumption
for the both sets of circuits is less than 35 milliamps.
Click on the image for a larger version.

LED101 also provides two other features:  It functions as a "power on" indicator, and since it is in series with Q101's base drive it is modulated at approximately 6.5 kHz (determined by experiment to be the frequency at which Q101 and L101 produced the highest voltage with the best efficiency) and can be used as an optical signal source to verify that the receiver is working.  Worth noting is that R112 is placed across the "hot" end and the wiper of R111 to "stretch" the high voltage end of the linear potentiometer's adjustment range a bit to compensate somewhat for the fact that near the maximum voltage, the gain goes up exponentially with the bias voltage, making fine adjustments at this setting easier.

The APD (optical) receiver:

The actual optical receiver section is depicted in Figure 5, below:
Figure 5:
The optical receiver which works in a manner very similar to that depicted in Figure 3.  In this implementation
the high voltage bias is applied to the cathode of D201, the APD, which has its anode connected to the gate of the JFET,
Q201.  Q201 and Q203 comprise a self-biasing, AC-coupled cascode amplifier while Q202 provides the a high-
impedance source for the bulk of Q201's drain current.  The components in the sections marked "HV Filter"
and "LV Filter" are used to keep the residual switching frequency energy from being conducted into these circuits.
As with other circuits of this type, the connection from the photodiode to the JFET's gate is made in air and not via a
circuit board trace - this, to minimize capacitance, leakage currents and noise.
Click on the image for a larger version.

Not surprisingly this circuit looks very similar to the "Version 3" optical receiver of Figure 2.  Notable features include an R/C filter consisting of R201, R202, C201 and C202 to remove traces of the 6.5 kHz power supply ripple from the high voltage supply while L201, C211, R215 and C212 do the same for the 9 volt supply that the receiver circuitry shares with the high voltage generator.  The two sections - high voltage supply and optical receiver sections - are separate, connected by a 3 foot (1 meter) umbilical cable, both to provide isolation of the extremely sensitive optical receiver from the electrostatic and electromagnetic fields of the high voltage converter and also to remotely locate the controls on the high voltage supply away from the lens assembly on which the receiver portion is mounted so that adjustments can be made without disturbing it.
Figure 6:
Inside the receiver portion of the APD receiver.  This section is physically
separated from the high voltage converter to prevent the switching energy
from getting into these extremely sensitive circuits.  In the center is
a small sub-board with the APD and JFET that is mounted on short pieces
of 18AWG wire to allow its position to be adjusted in all three dimensions
to provide both paraxial alignment and focus.
Click on the image for a larger version.

The APD itself is mounted on a small sub-board along with Q201 (the JFET) and the other capacitors noted in the box in Figure 5.  Most of Q201's drain current is provided by Q202's circuit, a current source, that operates at high impedance while Q203 is the rest of a cascode amplifier circuit that is designed to be self-biasing at DC and to provide gain mainly to AC signals.

The output of the cascode amplifier is passed to U201b, a unity gain follower amplifier.  This signal then passes to the circuit of U201a, a differentiator circuit that is designed to provide a 6dB/octave boost to higher frequencies to compensate for the similar R/C low-pass roll-off intrinsic to the APD and JFET itself:  Without this circuit, higher frequency audio components of speech would be excessively rolled off, reducing intelligibility.  By design the frequency range of the differentiator and its surrounding circuitry is intentionally limited so that low frequencies (below several hundred Hz) are strongly rolled off to prevent AC mains related hum from urban lighting from turning into a roar as are very high frequencies - above 5-7 kHz - which would otherwise become an ear-fatiguing "hiss" were the differentiation allowed to continue to frequencies much higher than this.

An interesting property of the photodiode circuit is that the "knee" related to this 6dB/octave roll-off occurs varies somewhat with the bias voltage and thus amount of device capacitance and, to a certain degree, its gain.  Because of this the frequency response of the APD/JFET circuit and the differentiator don't match under all operating conditions but experience has shown that it is better to have a bit of extra "treble boost" than not when it comes to making out words when the distant voice is immersed in a sea of noise.

A sample of the output from U201b, before differentiation, is also passed to J20, the "Flat" output.  The audio taken from this point, lacking differentiation, will sound a bit muffled under normal low-light conditions and it is not subject to either the high or low pass effects of the U201a differentiator which means that it will pass both subsonic and ultrasonic components as detected by the APD amplifier itself.  On the low end, the sensitivity is limited by 1/F noise which becomes increasingly dominant below a few 10s of Hz while on the high end it is again the capacitance associated with the APD and JFET circuits.  In testing it was observed that at this "Flat" output it was possible to detect signals from an LED modulated up to several MHz, albeit with significantly reduced sensitivity.  The main purpose of this output is to provide a signal point suitable for both subsonic digital communications as well as ultrasonic for experimentation with low/medium rate data, FM carriers and SSB signals.

In this circuit the amount of drain current in the JFET will vary depending on the individual properties of the JFET itself, the bias voltage, and the amount of impinging light.  Under "dark" conditions the "standing" JFET current was set to approximately 7-10 milliamps by the current source and the drain-source voltage varied from around 0.21 volts when the APD bias was just 12 volt to around 0.155 volts when the APD was operating at its maximum rating of 135 volts.  The specified JFET, the BF862, is typically capable of handling more drain current than this - and to do so would likely reduce its noise contribution slightly - but it was set at this level (with R205) to moderate battery current consumption.

Circuit testing:

Although it may have risked component damage, the APD circuit was "torture tested" to check ruggedness.  In a completely dark room a xenon photo flash was set off just inches/centimeters away from the photodiode with the bias set at 135 volts.  While the receiver was deafened for a second or two, the time it took for the various circuits to recover (e.g. power supply, re-equalization of various capacitors, etc.), repeated tests like this did not do any detectable damage to the receiver sensitivity or its noise properties indicating that the APD and JFET were more than rugged enough to handle any conceivable event that might happen in the field, aside from directly focusing the sun on the photodiode!

This circuit has also been successfully used in broad daylight.  While the receiver worked, the background thermal noise from the sunlit landscape was the limiting factor for sensitivity and the recovered audio had quite apparent nonlinearity (distortion) with an altered frequency response (e.g. "tinny") because the ambient light and resulting photodiode conductivity effectively shunted the high voltage bias and device capacitance.  In short, in such high ambient light conditions this circuit has no advantage over other optical receiver topologies such as the original "Version 3" or even a more conventional TIA (TransImpedance Amplifier) but its ability to be useful under such conditions is indicative of its versatility.

The results of in-field testing:

This receiver was first field-tested on a 95+ mile (154km) optical path during the September 2012 segment of the ARRL "10 GHz and up" contest:  For detail on this communication, read the blog entry "Throwing One's Voice 95 Miles on a Lightbeam" - link
Figure 7:
My end of the 95+ mile optical path during the session where the APD-
based optical receivers were first field-tested.  As seen in the picture
the optical path passes over urban lighting which tends to slightly raise
the noise floor due to both Rayleigh and lens-related scattering
Click on the image for a larger version.

During this test the optical (voice) link was first established using the "Version 3" PIN Photodiode receiver depicted in Figure 2.

With the reasonably clear air and the moderately long path we noted that we could reduce the LED current to a tiny fraction of the maximum before significant signal/noise degradation was noted.  At this lower LED current each station at opposite ends of the path switched from the PIN photodiode to the APD receivers and after tweaking our pointing and reducing the LED current even more we observed what turned out to be between 6 and 10 dB improvement in the signal-noise ratio - about what was observed on the indoor "Photon Range" with the initial prototype circuit.  It is likely that the actual improvement in sensitivity was greater than this, but because our respective optical paths passed directly over populated areas (see Figure 7) our ultimate noise floor was degraded by light pollution which included a thermal "hiss" from the urban lighting and a low-level, harmonic-rich 120 Hz hum.

As was determined in the lab, the best signal-noise ratio in the field occurred with the APD biased in the 35-45 volt range where the "M" (amplification) factor was in the area of 3-10 (approximately 10-20dB gain).  At this rather modest bias voltage the "Gain+Noise" from the APD itself was sufficient to overcome much of the intrinsic noise of the JFET.  At higher voltages the gain continued to increase but the signal-noise ratio decreased at a faster rate until the APD's own avalanche noise drowned out the desired signal.

* * *

For more information about (speech bandwidth) free space optical communication, check out these links from my "Modulated Light" web site (link):

Be sure to check out the "" web site's other pages as well!


This page stolen from "". 

No comments:

Post a Comment


While I DO appreciate comments, those comments that are just vehicles to other web sites without substantial content in their own right WILL NOT be posted!

If you include a link in your comment that simply points to advertisements or a commercial web page, it WILL be rejected as SPAM!