About this project:

Fairly early on in my work with high-power LEDs such the Luxeon 1 and 3 watt devices, I have considered that PWM (Pulse Width Modulation) techniques would be an interesting means of modulation the LED.  In the strictest sense, this could be capable of producing the lowest distortion modulation on an LED because the linearity of the LED Current versus Luminous Output curve would be irrelevant:  The apparent brightness would be integrated by the receiver to produce a voltage proportional to the duty cycle.

For actual audio from the very modulator described and pictured, listen to the audio clips found here.

What is PWM?

Figure 1:
Graphics showing how PWM works.  These graphs are from the Luxeon link on the Modulated Light web page by Chris, VK3AML and Mike, VK7MJ.

Pulse Width Modulation is widely used nowadays in low-powered audio amplifiers and the so-called "1 bit" D/A converters and the operation is simple:

• For a "steady state" output (that is, no output) a 50% duty cycle square wave is generated at a frequency several times higher than the highest-frequency component in the audio being reproduced.  While this frequency could theoretically be as low as just twice the highest audio frequency, it is usually several times higher than that to simplify lowpass filtering to cut costs.
• To increase the output voltage, the duty cycle of this square wave is increased, with 100% being "full on."  Conversely, to decrease the voltage, the duty cycle would be decrease, down to 0% being completely off.  In reality, most PWM circuits avoid getting too close to either 0% or 100% as either extreme would produce objectionable "hard" clipping.
• The PWM output is filtered to average out the square wave, the ultimate result being a voltage that is directly proportional to the duty cycle of the original square wave.

As it turns out, the linearity of a PWM generator could, in theory, be absolutely perfect:  The duty cycle is simply timed precisely using digital counters - which are absolutely precise.  What is necessary, however, is that there be enough timer resolution in order to provide the needed resolution of duty cycle.

Take, for example, a 10 bit PWM converter.  Because 10 bits represents 1024 steps, it would be necessary that the original timing clock be 1024 times that of the sampling rate.  If, for example, our original clock were 20 MHz, one 1024th of that would be 19.53125 kHz.

In practical terms, the frequency response of the circuits in a typical optical receiver may not be able to respond at the PWM frequency, with the result being a voltage that is very close to the original analog signal applied to the modulator.  This technique usually works, as the more sensitive optical receivers (and/or their following amplifier stages) don't have the frequency response characteristics necessary to reproduce the original PWM waveform.

Some caution should be exercised, however:  If the optical receiver does have the bandwidth to recover the PWM signal, or if there is a reduced (but still sufficient) response of the audio chain at the PWM frequency, this could play havoc with downstream audio devices in several ways:

• The audio amplifier may be capable of amplifying the PWM signal, robbing power from the audio-frequency components.  In this situation, the audio amplifier is putting out its normal power, but some of it may be wasted at the PWM frequency and be inaudible to human hearing.  In this case, the audio amplifier may overload at lower-than-normal volume levels.
• Aliasing artifacts on digital audio devices.  Computer sound cards and digital audio recorders may not be able to sufficiently filter the PWM frequency from their inputs and this may result in odd aliasing artifacts, which may include noise, distortion, or odd mixing effects.

While the above are possibilities, I have not experienced these effects when using my Version 3 optical receiver (with the lowpass filter switched out) with digital audio devices - but the fact that this receiver starts to roll off severely above about 7 kHz is no doubt a mitigating factor.  The caution here is that all equipment should be tried out before going out into the field to verify that there is compatibility.

A PIC-based Pulse Width Modulator:

Having done some DSP programming using the Microchip PIC microcontrollers over the years, I knew that it already possessed the hardware to make a nice Pulse Width Modulator for LEDs.  I chose the PIC16F88 for this task as it had some useful onboard peripherals:

• A 10 bit PWM generator.  With a 20 MHz crystal, it could generate a PWM signal with 10 bits of resolution at 19.53125 kHz - a frequency sufficiently high enough for voice bandwidth communications without having to use elaborate anti-aliasing filters.
• A 10 bit A/D converter.  Again, this is a useful feature if you want to take an analog signal and do anything with it, and the ability to use multiple analog inputs allows several analog voltages to be digitized.  Another useful feature was that the A/D converter could be configured to use an external voltage reference.
  
• Onboard timing.  This processor has several onboard hardware times, allowing very precise generation of clock periods - something that would be useful for generating audio tones for testing.
• An onboard 4-bit R-2R D/A converter.  This peripheral, originally intended as a voltage reference for the onboard comparators, would prove to be very useful when used in a unique manner.

Description of the hardware:

Figure 2:
Schematic of the LED PWM circuit.
Click on the image for a larger version.

Signal input stage:

Audio input can be one of three sources:  A built-in electret microphone, an external microphone via J1, or an external line input via J2:  S1, an SPDT switch, selects which source is to be used and if a "center off" switch, that position can serve as a "mute" setting.  Note that J1 is a "disconnecting" type of jack and is wired to disable the internal microphone when an external one is plugged in, such as a desk-type computer microphone or a microphone-headset combination.

In experimentation, it has been noted that computer-type microphones are wired in one of two ways:  While the audio is always on the "tip", some connectors apply bias to the tip and leave the ring disconnected while others apply the bias voltage only to the ring.  The circuit shown accomodates both wiring schemes.  Note that this microphone input is not suitable for dynamic or crystal microphones - only electret microphones should be used.

Note also that J2's is wired such that the two resistors will sum (and attenuate) a line-level stereo input to a monaural signal.  The resistors (R3 and R4) are necessary in many audio amplifiers because it has been noted that with many audio devices, simply shorting the left and right channels together often results in distortion.

The input signal is buffered by Q1, an emitter-follower and the source impedance is set with R9.  Q2, a JFET, is used as a variable resistor, controlled by the microprocessor, U2, to reduce gain of the input stage.  U1B is a non-inverting amplifier used to boost the low-level audio input signals (from the microphone, for example) to a level suitable for the A/D converter on the microprocessor.  U1C, along with R18 and C8 form a 3-pole 3.5kHz anti-aliasing lowpass filter used to limit the frequency response to below the Nyquist limit of the microprocessor's sampling rate of 19.53125 kHz.

LED Current Driver:

U1A is wired as a precision current sink:  With a closed feedback loop, the drive on the gate of Q1, an N-channel power MOSFET, is adjusted by U1A as necessary to obtain a voltage drop across R30, a 1 ohm resistor, that is equal to the voltage on pin 3 of U1A, the non-inverting input.  In this way, the current through R30 - and thus through LED1, the high-power LED - is exactly proportional to that voltage applied to pin 3.  Because the maximum, peak current through a 3 watt red Luxeon LED is 2 amps, R28 is adjusted to provide precisely 2 volts at the peak of the PWM waveform (from U2) when R29 is all of the way up:  When adjusted this way, with a 50% duty cycle, the average LED current is thus 1 amp.

For current monitoring, R31 and C20 provide a filtered voltage reference where 1 volt equals 1 amp of average current.  For audio monitoring, a pair of headphones may be used with R32 being used to adjust the audio level, R33 limiting the drive to the headphones to a safe level with C21 blocking DC and C22/R33 filtering most of the PWM switching frequency out of the monitor point.

In order to "mute" the LED drive without powering down the circuit (and avoiding the wait for the circuit to re-stabilize if it is powered up) S3 simply disconnects the LED from its voltage source.

Modifications to minimize voltage drop:

In testing, it has been noted that the LM324 used will properly operate down below even 10's of millivolts and because of this, it should be possible to reduce the value of R30 down to at least 0.1 ohms.  If a low on-resistance MOSFET is used for Q1, it should be possible to construct a circuit that will fully modulate the LED with less than 0.5 volts of additional voltage drop.  What this means is that with these lower resistances it is possible to run a single Luxeon III from a 6 volt supply, or up to three Luxeon III's in series from a 12 volt battery supply!  (Note that if you were to use a 6 volt supply, you would have to assure that the +5 volt regulated supply could maintain accurate regulation!)

Comment:  I plan to revisit the audio monitor and install a buffer amplifier so that it will be insensitive to the load of the LED:  As shown, the audio output from J3 decreases when R29, the current setting is decreased - or just goes away when S3 is opened.

Microcontroller:

U2 is a PIC16F88 microcontroller with a built-in 10 bit A/D converter, 10 bit PWM generator (used as a D/A output) as well as other peripherals.  The 5 volt PWM signal is applied to the U1A current sink via R28/R29.  U2 also contains a 4 bit voltage reference originally intended to be used with onboard comparators, but it is used in this case as another D/A channel, being filtered by R23/C16 and amplified by U1D:  A divided-down version of this voltage is also used to provide a centerline reference for the filtered and amplified A/D input.

Also connected to U2 are two potentiometers, R19 and R20.  These produce variable voltages that are read by the A/D converter and used by the software to change settings such as gain or tones.

Description of the software:

Figure 3:
Top:  Front panel view of the modulator, which is connected to a Luxeon 3-watt emitter module with a current limiter.  (No, I haven't gotten around to properly labeling things just yet...)
Middle:  The interior of the modulator showing the component side of the circuit board.
Bottom:  Another view of the interior of the modulator, showing the backside of the board and front panel.
Click on an image for a larger version.

Timing:

U2 is clocked by a 20 MHz crystal and this is used to providing timing, including the 19.53125 kHz PWM frequency and the corresponding 19.53125 kHz sampling rate done in an interrupt service routine.  Because of the sample rate, the maximum allowable audio frequency that may be sampled without distortion is just under 10 kHz:  The 3-pole lowpass filter attenuates input audio by about 20dB at the Nyquist frequency, effectively preventing such problems by rolling of frequencies above 3 kHz at a rate of 18dB per octave.

Operation in audio mode:

Automatic gain control (AGC):

When in "audio" mode - that is, when audio from a microphone or line input is being modulated, the voltage at input AN4 is digitized.  When doing A/D conversions, however, it is important to keep the audio level near the maximum input limit of the A/D converter, yet not overdrive it and cause distortion.  To prevent this, the software looks at the incoming digitized audio and if the sample voltage is too close to minimum or maximum, gain is reduced to prevent overdriving the A/D and thus the PWM generator.  Each time a sample of high audio is detected, RB1 ("Gain Control Pulses")  is set high, but it is kept low at all other times.

Every time RB1 goes high, charge is added to C5 via R13 (through D2) and R14.  As the voltage on C5 rises, Q2 begins to conduct, shunting away some of the signal being applied to U1B and reducing the audio being input to the A/D converter and resulting in fewer "high audio" conditions that cause a gain control pulse.  If the audio has not exceeded the "gain reduction" threshold, RB1 is kept low and C10 is discharged more slowly via R14.  With this feedback system, the gain is balanced, keeping the audio from getting too high, too often.

Another output, RB7, is used to indicate when the CPU detects that either the A/D input and/or PWM output exceeds a high amplitude (slightly higher than the gain reduction threshold) by outputting a pulse that turns on Q3 and causes LED2 to flash, indicating to the user something about the amount of audio present.  Note that it is normal for this LED to flash once in a while and that occasional audio peaks that get "hard clipped" by the input A/D won't cause objectionable distortion on speech.

The time constants of C10, R13 and R14 are chosen to be fairly fast in order to track speech.  The effect of this is that this audio AGC acts very much like a compressor-type speech processor and can help maintain a tight peak to average ratio - something that can greatly improve intelligibility of speech under conditions of poor signal/noise ratio.

Manual gain control:

Another means of gain control is via R19, a potentiometer:  The voltage from this pot is digitized and internally converted to a value that is applied to the "Comparator Voltage Reference" (Cvref) output on pin 1.  Although this is a 4-bit D/A converter, 8 bits of resolution are obtained via software dithering and using U1D, the Cvref output is filtered with C16/R23, amplified, and applied to pin 2, Vref+, the voltage reference input of the A/D converter while a sample of half this voltage is applied to the audio input through R27 to provide a "mid-scale" voltage reference for the A/D converter.  By lowering the reference voltage, the gain of the A/D converter is effectively increased.

When R19 is fully-counterclockwise (minimum gain) the Vref+ voltage is set to about 5 volts, corresponding with a full-scale range of 0-5 volts of analog input voltage.  When R19 is but fully-clockwise, the Vref+ voltage drops to about 0.3 volts:  Those familiar with this chip will note that the minimum specified A/D converter Vref+ voltage is, in fact, 2 volts, but this is only true if full 10 bits of A/D resolution is required.  The penalty of using a voltage lower than this is simply that the lower-order A/D conversion bits will start to be lost in the noise.  At the lowest voltage, the A/D converter resolution is roughly equivalent to 6 or 7 bits,  but since the lower-order bits contain what sounds like white noise, the overall effect is that as A/D conversion gain goes up with the lowering of Vref+, so does the noise level, which manifests itself as a background "hiss."

While this added noise is noticeable, it is not really objectionable and on any sort of weak optical path - the condition where the gain might be run up to maximum to more-heavily compress the audio for improved intelligibility - it probably wouldn't be noticed at all amongst the other noise sources.  Note also that under normal conditions, the "manual gain" control would not be operated at maximum, anyway.  In normal operation, it has been noted that the gain is sufficient to pick up the voice of anyone within several feet of the modulator when its internal microphone is used.

Note also R11, another gain control.  This sets the maximum gain of U1B, the first amplifying stage.  On the prototype, I'd used a 100k potentiometer, but noted that I had to set it nearly at maximum gain (minimum resistance) so a lower value of potentiometer (20k-50k) would probably be more appropriate.  Note that setting R11 for too high of gain can result in instability of U1B and/or excess noise.

Operation in "tone" mode:

Another feature of the software is tone generation:  Using DDS techniques, low-distortion sine waves can be generated at practically any audio frequency below the Nyquist limit with a resolution of 0.298 Hz..  Having this capability allows several tone generation modes:

• Continuously variable frequency.  Using R20, the audio frequency can be adjusted from 20 Hz to about 2.5 kHz (2457 Hz, actually.)  In this mode, the rotation of R20 is "de-linearized" to make it easy to adjust the tone frequency over a wide range.
• Selection of fixed frequencies.  When in this mode, one of 8 precisely fixed tone frequencies may be selected using R20 as noted below.
  
• Ascending or descending tone sequence.  The tone sequence consists of four dissonant tones that are very easy to pick out of the noise.  R20 is used to adjust the sequencing rate.
• Activitation of a pilot carrier.  In this mode, a 4 kHz tone (12 dB below 100% modulation) is digitally mixed in software with the microphone (or line input) audio.  The software takes the presence of this pilot carrier into account and prevents overmodulation of the LED with the combined audio sources.  At the receive end, this pilot tone can be filtered out and is available for analysis of scintillation or used for peaking the receiver.
  

The selection of tone and audio modes is done by grounding RB3, RB4 and/or RB6 using diodes D3-D9 and a rotary switch to generate a binary code as follows:

A - Adjustable tone:  RB3, RB4 grounded, RB6 open
B - Selection of 8 fixed tones:  RB3 grounded, RB4 and RB6 open - see below for a list of the audio tones
C - Ascending tone sequence:  RB4 grounded, RB3 and RB6 open - see below for the list of tones used in the sequence
D - Descending tone sequence:  RB3, RB4, and RB6 open - see below for the list of tones used in the sequence
E - Audio with pilot tone:  RB4, RB6 grounded and RB3 open
F - Audio with no pilot tone:  RB6 grounded, RB3, RB4 open

The 8 fixed audio tones available in Mode B are:

1 - Musical note B0 (actual freq. = 30.9944 Hz)
2 - Musical note E1 (actual freq. = 41.1295 Hz Hz)
3 - Musical note C4 - middle C (actual freq. = 261.6674 Hz)
4 - Musical note F4-sharp (actual freq. = 369.8468 Hz)
5 - Musical note A5-sharp (actual freq. = 932.26912 Hz)
6 - Musical note - E6 (actual freq. = 1318.52896 Hz)
7 - 440 Hz - Musical note A4 (actual freq. = 439.907 Hz)
8 - 1kHz tone (actual freq. = 999.9242 Hz)

Note:  The ascending sequence (Mode C) consists of tones are #'s3, 4, 5 and 6 (in that order) while the descending tone sequence (Mode D) are the same tones in reverse order.

Adjustment:

Maximum LED current:

• The LED connection is shorted out
• R28 is turned all of the way down (wiper grounded)
• R29 is turned all of the way up
• R28 is then adjusted for 1.1 amps of average current (when using a 3 watt red Luxeon) as measured at the LED Current Monitor point by observing 1.1 volts.

Maximum audio gain:

As mentioned before, R11 is adjusted to provide the maximum desired amount of microphone gain.  Care should be taken to avoid setting R11 to too low a value to prevent noise and/or instability if the U1B amplifier section.

Comment:  100% modulation is defined as modulation that goes all the way from zero up to twice the average (unmodulated) current as set by R29.

---

Important note:  It is strongly recommended that you never operate any modulator or LED without having current limiting on the LED.  This may take the form of a resistor, or a current limit circuit such as one using an LM317.  If an LM317-based limiter is used, you may need to install bypass capacitance to prevent distortion of the waveform due to the nonlinear nature of the PWM waveform.

Components:

• Diode D1 is a 3-6 amp, 50 volts diode or greater
• Diodes D2-D10 are small-signal diodes, such as 1N914 or 1N4148
• Q1 is an N-channel power MOSFET.  A recommended device is one that has a current rating of 10-20 amps at up to 100 volts.  Note that high voltage/high current devices have more gate capacitance and could make gate drive difficult.
• Q2 is an MPF102
• Q3 is a general-purpose NPN transistor.
• All potentiometers are linear taper.
• J1 is a disconnect-type 3-conductor (stereo) 1/8" (3.5mm) jack
• J2 and J3 are 3-conductor 1/8" jacks
• S1 is an SPDT switch.  A center-off switch is nice to have, but not necessary.
• S2 is a 6-position, non-shorting rotary switch.  I used a 6 position switch (from Radio Shack - P/N 275-1386.)  If necessary, several toggle switches could be used to set the various modes.
• S3 is an SPST switch
• U1 is an LM324 quad op amp:  DO NOT SUBSTITUTE!  The op amp used here must be capable of operating down to the negative supply rail. Other rail-to-rail op amps were tried, but did not work very well:  I need to look into this...
  
• U2 is an appropriately programmed PIC16F88 microcontroller.
• U3 is an 78L05 (or 7805) 5 volt regulator.
• R11 - Trimmer potentiometer, 20k-50k maximum.
  
• LED1 is a high-powered LED.  The use of a red (or red-orange) 3-watt Luxeon is assumed here, but other units may be used provided that R28 is adjusted for maximum safe current.  It is strongly recommended that the LED itself be equipped with a current limiter to protect the LEDs.
• LED2 is a normal LED, probably red.
• TH1 is a self-resetting, 3 amp "thermal" fuse.
  

Comments:

• It is normal for the "Overload" light to flash on occasional audio peaks.  With high input levels and/or excess audio gain, the light may flash much more frequently, causing some some minor clipping, but it may sound overly "compressed."  Under conditions of low signal-noise ratio, however, a heavily compressed audio signal may be more intelligible than one that isn't as compressed.
• S3 disconnects the LED to allow "muting" of the light output, but leaves the rest of the circuit powered up.  This keeps the circuit active, thus eliminating the need to wait for things to stabilize were the entire circuit powered down:  The current consumption with the LED off is about 40 milliamps.- Note that as the LED current is decreased, the audio output from J3 will also decrease.  When S3 is opened, the audio output will also go away.
• TH1 is a 3 amp self-resetting thermal fuse that is used to protect the circuit in the event of an internal power supply short, or in conjunction with D1 to provide power supply reversal protection.
• Some time after building this circuit, I joined the Optical DX Yahoo Group and noticed that David Smith, VK3HZ, had taken a similar PWM approach - it might be interesting to compare notes.
• This Pulse Width Modulator does not offer any power efficiency over a linear modulator because it still uses linear current regulation to limit the LED drive.
• I have also built a linear modulator that uses the same "Precision Current Sink" circuit but does not use a PIC to process the audio.  I did this circuit mainly to see how well it works, but having built both, I would recommend the "linear" version instead as it is somewhat simpler, and it does not have the audio frequency response limitation of this circuit.

Return to the KA7OEI Optical communications Index page.

If you have questions or comments concerning the contents of this page, or are interested in this circuit, feel free to contact me using the information at this URL.