Editor’s Note:
Part 1 of this DI uses an electret mic to create infrasound. It starts with a basic equalization circuit validated with a DIY test fixture and simulations, and ends with a deeper analysis of the circuit’s real response.
Part 2 includes refinements to make the circuit more usable while extending its detectable spectrum with an additional technique that allows us to hear the infrasonic signals.
Although electret microphones are ubiquitous, they are more versatile than might be expected. With some extra equalization, their frequency responses can be made to range from earthquakes to bats. While this Design Idea (DI) ignores those furry mammals, it does show how to get a reasonably flat response down to way below 1 Hz.
Wow the engineering world with your unique design: Design Ideas Submission Guide
Electrets aren’t only used as audio pickups. For decades, they have been employed in security systems to detect unexpected changes of air pressure within rooms, while more recently they can be found in vapes as suck-sensors (or, more technically, “draw sensors”, according to Brian Dipert’s recent teardown).
An excellent description of their construction and use, complete with tear-down pictures, can be found here. The capsules I had to hand were very similar to those shown, being 10 mm in diameter by 6 mm high. Some experiments to check their frequency response—practical details later—showed a steady 6 dB/octave roll-off below about 15 Hz, implying that a filter with an inverse characteristic could flatten the response down to a fraction of a Hertz. And so it proved!
Building an equalization circuit
A basic but usable circuit capable of doing this is given in Figure 1.
Figure 1 Simple equalization can extend the low-frequency response of an electret microphone down to well under 1 Hz.
While this exposes some problems, which we’ll address later, it works and serves to show what’s going on. R1 is chosen to give about half the rail voltage across the mic, and A1 boosts the signal by ~21 dB. At very low frequencies, A2’s stage has a maximum gain of ~30 dB. This falls by 6 dB/octave from ~160 mHz upwards, reaching unity gain at ~4.8 Hz. C3/4 and R7/8 top and tail the response, and A3 boosts the level appropriately. (Not shown is a rail-splitter, defining the central, common rail.) The op-amps used were MCP6022s because of their low input offset voltage.
The low 3-dB point is largely determined by C1/R2. (Adjusting the values of R5, R6, and C2 and adding an extra resistor in series with C2 would, in principle, let us equalize a specific mic to give a flat response from a few hundred millihertz up to its upper limit.)
Figure 2 shows the overall response to changes in air pressure, with 3 dB points at about 500 mHz and 12 Hz. While this is an LTspice-derived trace, it closely matches real-world measurements.
Figure 2 The response of Figure 1’s circuit to air-pressure changes at different frequencies.
Validating the frequency response
That confidence about the actual response may raise some eyebrows, given the difficulty in getting decent bass performance in even the best of hi-fi systems. A custom test rig was called for, using a small speaker to produce pressure changes in a sealed chamber containing a mic-under-test. It’s shown in Figure 3.
Figure 3 Two views of a test rig allowing sub-Hz measurements of a microphone’s frequency response.
The rig comprises an IP68 die-cast box fitted with a 50 mm plastic-coned speaker (42 ohms) and a jam-jar lid, the jar itself being the test chamber for the mic, which, when fitted with pins, could be swapped. Everything was sealed with lots of epoxy, plus some varnish in case of pinholes. A generous smear of silicone grease guaranteed that the jar seated almost hermetically. The speaker was driven by a custom sine-wave oscillator based on a simple squashed-triwave design and covering from 90 mHz to 11 Hz in two ranges.
This is actually the Mark 3 version. Mark 1 was based on a cut-down, wide-mouthed tablet bottle with a speaker fixed to it, which was adequate for initial tests but let in too much ambient noise for serious work. Mark 2 added a jam jar as a baffle behind the speaker, but the bottle’s walls were still too flexible. The more rigidly-constructed Mark 3 worked well, with an unequalized frequency response that was flat within a decibel from about 20 to 200 Hz. (It had a major cavity resonance at about 550 Hz, too high to affect our results.)
Simulations, mostly in hardware
To verify the performance of the rig itself at the lowest frequencies, some simulation was needed—but in hardware, not just with SPICE. Stripping a spare mic down to its bare JFET (a Sanyo 2SK156) and adding some components to that meant that it could be driven electrically rather than acoustically while still looking like the real thing to the circuit—or almost. The main divergence did not affect the frequency response, but did throw light on some unexpected behavior. The simple schematic is in Figure 4; the concept also worked well in LTspice, using their default “NJF” JFET, and formed part of Figure 2’s simulation.
Figure 4 A circuit that simulates an electret microphone in real life.
Once the circuit had settled down, the measured frequency responses using the test rig and the simulated mic matched closely, as did the LTspice sim. With the simulated mic, settling took a few seconds, as expected given the circuit’s long time constants, but with a real mic, it took many times as long. Perhaps the diaphragm was relaxing, or something? Another mic, torn down until only the JFET remained, behaved similarly (and, with its floating gate lead, made a near-electrometer-quality mains-hum probe!).
Curious behavior in a JFET, and how to fix it
It seemed that the FET’s gate was misbehaving—why? Perhaps charge was being injected when power is applied, and then leaking slowly away? Ramping the voltage up gently made some difference, but not enough to explain things fully. It appears that leakage is dominant, and that charge on the gate slowly equalizes, producing a long, slow “tail” which is still just fast enough to produce an offset at the circuit’s output, even with two C-R networks attempting to block it. With the low impedance on the simulated mic’s gate, such effects are negligible. It’s stuff that would never show up in audio work.
From this, we can deduce that the mic’s low 3-dB point is determined not by the FET’s time-constant but by the “acoustics” within the mic. But that extra, inherent time constant still needs addressing if the circuit is to settle in a reasonable time. If the gate must slowly drift towards equilibrium owing to leakage, could we inject a packet of charge at start-up to compensate? Experiments using the circuit of Figure 5 were successful, albeit empirically; the values given are cut-and-try ones. Shorting R1 for about 3 ms gave a pulse of double the final voltage across the mic, and that proved to be optimum for the available capsules in the circuit as built. The settling time is still around 10–15 seconds, but that’s a lot better than over a minute.
Figure 5 A few milliseconds of over-voltage applied across the mic at start-up injects enough charge to counterbalance much of the FET’s longer-term start-up drift.
This is also useful in the case of an overload, which sends the output off-scale. If that happens, you can now use the time-honored method of switching off, waiting a few seconds, and switching back on again!
Real-life response
Figure 6 shows the actual response as measured using the test rig. It’s a composite of two scans, one for each range. (Because tuning was done manually, the frequency scale is only roughly logarithmic.) R9 was set to about 50k, so the output stage had a gain of around 6.
Figure 6. The response of the circuit in Figure 1, measured using Figure 3’s test rig.
The upper trace is the driving waveform for the speaker, showing that a positive-going output from the circuit corresponds to increased pressure within the rig’s chamber. (From this, we can infer that the negatively-poled side of the electret film itself faces the JFET’s gate. That makes sense, because a serious acoustic insult like a handclap right in front of the mic will then charge the gate negatively, and excess negative charge drains away more easily through the JFET’s gate-source diode than positive charge can, speeding recovery from any such overload.)
Note how the baseline wanders. That is mostly due to 1/f or flicker noise in the mic capsule’s JFET; both the bare JFET and the simulated mic show a similar effect, while a resistor is much quieter. We can extend the LF response further, but only at the expense of a worse S/N ratio. And below a Hertz or two, the effects of wind and weather seem to be dominant, anyway.
Viewing the results
There are several further desirable refinements and additions, but they must wait for Part 2. We’ll close this part with some ways of seeing what’s lurking below our ears’ cutoff point. (And Part 2 will also show how to listen to it.)
An oscilloscope (usually bulky, static, and power-hungry) is too obvious to mention, so we won’t. A cheap 50–0–50 µA meter connected between the output and common via a suitable resistor worked, but its response was 50% down at ~2 Hz.
A pair of LEDs, perhaps red and green for positive- and negative-going swings, looked good, though the limited swings available with the 5 V rail meant that the drive circuit needed to be somewhat elaborate, as shown in Figure 7. Caution! Its power must come directly from the power input to avoid the LEDs’ currents disturbing the mic’s supply, which would (and did) cause distortion and even (very) low-frequency oscillation. A good, stable power source is needed anyway.
Figure 7 One LED lights up on positive swings and the other on negative ones, the intensities being proportional to the signal levels.
Part 2 will extend the detectable spectrum a little while mostly concentrating on making the basic circuit more usable. An audible output will mean that we will no longer have to worry about the Zen-like problem of, “if we can’t hear it, should we call it a sound?”
—Nick Cornford built his first crystal set at 10, and since then has designed professional audio equipment, many datacomm products, and technical security kit. He has at last retired. Mostly. Sort of.
Related Content
- Squashed triangles: sines, but with teeth?
- A pitch-linear VCO, part 1: Getting it going
- Earplugs ready? Let’s make some noise!
- Supersized log-scale audio meter
The post Revealing the infrasonic underworld cheaply, Part 1 appeared first on EDN.
