Convert CMOS-level signal to microphone-level (0..3V --> -1..+1V)?

Question

I'm building a microcontroller circuit that feeds its output into a microphone input of an audio recorder. The output is a square wave. (Think old-style 3.5mm Square credit card readers).

What can I use to covert a signal in the range of 0..3V to -1..+1V? I'm looking for something like MAX232, but with lower voltage levels.

I could just use a voltage divider and feed 0..+1V to the audio, but I understand that many audio input circuits use AC coupling and the recorded waveform can have unpredictable DC bias. This is undesirable, since the waveform encodes information in zero-crossings.

The question really is about a circuit that can generate negative voltages without a separate negative supply, but for completeness, I can add a few more details. The application is a kind of old-style telemetry format. The data rate is fairly low (~2,000 zero-crossings/second, or 2kbps). The client audio circuit is usually (though not always) a digital recorder. The client records my input and (usually) saves it as a digital sound file. Clients' expectations are that third-party programs can extract original digital information from these files. Sometimes, instead of recording my signal directly, the client feeds my signal into an audio-frequency radio TX (think walkie-talkie) and records the signal from the RX.

Answer 1

I could just use a voltage divider and feed 0..+1V to the audio, but I understand that many audio input circuits use AC coupling and the recorded waveform can have unpredictable DC bias. This is undesirable, since the waveform encodes information in zero-cr ossings.

From the comments:

If it helps, the waveform is similar to Manchester coding.

That's a fairly important detail. For one thing it means that the average DC value will be 1.5 V from the micro and that on your audio side the mean DC level will be zero. (This wouldn't be the case in the bitstream example I gave in the comments.) This in turn means that if you could give an adequate lead-in of a string of 1s or 0s that a simple decoupling capacitor would work because any initial DC offset would fade away after a few cycles.

I have never attempted anything like this but something along the following lines might work.

^{simulate this circuit – Schematic created using CircuitLab}

Figure 1. On the first positive or negative pulse the output will be clamped at about +/- 1.5 to 1.7 V. Thereafter it should swing fairly symmetrically around 0 V.

The C1 value is the default in CircuitLab. I didn't do any calculations.

Figure 2. Simulation of C-R coupling and Zener clamping.

TTL

• The TTL signal could be dormant low or dormant high. The simulation starts (1) with a high.

OUT_R

• The right side of C2 starts at 0 V (2).
• On the first negative-going edge of the TTL signal OUT_R is 'kicked' to -3 V (3). This may overload your audio circuit.
• As time progresses R1 removes the initial DC bias and, with the values chosen, the signal is centred around 0 V by time (4).

OUT_ZD

• This also starts of at 0 V (5) due to the capacitor.
• As in the OUT_R circuit the first negative-going edge of the TTL signal tries to drive the right side of C1 to -3 V. This time, however, D2 will breakdown in reverse mode and D1 will be forward biased. The combination of the 1 V Zener breakdown of D2 and the 0.7 V or so forward voltage of D1 will clamp OUT_ZD to about -1.5 to -1.7 V as shown at (6). At this stage their work is almost done.
• Subsequent changes of polarity will change OUT_ZD by 3 V from a starting point of 1.5 V away from 0 V so the diodes will hardly have to do any further work (7).

The back to back Zeners just means that the pair work symmetrically on positive or negative pulses. Because Zener's work as a regular diode in forward mode we need to reduce the Zener voltage by the forward voltage drop. You're looking for ±1.5 V and V_f is about 0.6 to 0.7 V so I went for 1 V Zeners.

Answer 2

I'd say that you don't need a level shifter if what you'll receive in the end is a digital audio file. Whatever biases present in that can easily be fixed with a simple equalizer (FFT filter). Microphone inputs don't usually care if the signal is not symmetrical around 0 (speakers and power amps are a different story).

I know this, because I've done something similar. I have a small transducer that converts light into sound, it's really a photodiode + preamp + a resistor divider so that I get a 0..1V signal, just as you suggested. If my input signal is 50% square wave (e.g. I'm probing the backlight PWM pattern on an LCD screen), the audio file will have a proper square wave, symmetric around 0, because the microphone input's decoupling capacitor acts to remove the "always positive" signal bias. The only gripe is that it can be sometimes slow to do so, as its time constant is typically 0.1-1s, so if you start with a steady state (0V) and then have a digital signal (0-1V), it will begin as a positive waveform, that will slowly sink to the symmetric, with the time constant set by the decoupling cap. If you need a faster reaction to such changes, you can postprocess the audio file with some editing software (e.g. using Audacity - use the Equalization effect and select it to reject all frequencies below your some suitable threshold, e.g. 400 Hz).