This is Volume 1 of Building Blocks. Click for more ideas! $1 Negative Rail Converter
Designed and written by David H. Christensen
Want to know when a new volume is posted? Subscribe to the RSS feed

Single supplies are really common. Think about it—most everything comes with a single positive rail and ground. Sometimes, we get multiple positive rails (such as +5V and +3.3V). However, some signals are inherently bipolar; that is, they are centered around ground and move both above and below our 0V reference.

Signals of this nature include:

Of course, the most common of these, is audio. The problem is, though, that with our single +V and Ground supply, we can't really represent the below-ground negative part of these signals. An op amp won't accept input below its negative rail (or, in the case of certain rail-to-rail input parts, not much below it) and certainly can't output below it. So, how do we get a second rail—a negative rail below ground?

The classic way would of course be to design our very own bipolar power supply—with a center-tapped linear mains transformer, or multiple windings on a flyback transformer, if you prefer your supplies switching. In battery powered applications, we can add multiple cells and define ground as being between two cells. But we haven't always got this option. For instance, a lot of gear is powered from USB—and there's no supply apart from 0 and +5V in USB (except for higher positive voltages available in USB 3.1). How in the world can we rectify this situation?

There really are three solutions to this problem:

  1. Rail splitter. We "halve" the positive rail, and add a push-pull regulator (anywhere from an op amp to an op amp plus a class B (PNP and NPN) follower). Thus, with a single +5V rail, we would get +5V, +2.5V and 0V. If we define this new mid-point as our (virtual) ground—since ground is always where the black probe of our mental multimeter is!—we have a +2.5 rail, 0V "virtual" ground and a -2.5V rail (our original ground)! This is pretty good, except there isn't much head-room in either direction from our 0V.
  2. Buck-boost converter. Buck-boost converters are inverting by nature; thus, we can use any universal switching regulator controller to get a negative rail, such as the cheap and cheerful MC34063. We can even get any negative voltage we want! This, however, requires an inductor of decent current rating, and the electromagnetic flux of the inductor can couple into nearby conductors and cause noise. However, the inverting buck-boost converter is definitely viable.
  3. Charge pump. The charge pump is pretty simple. It's a variant of the Cockcroft-Walton voltage multiplier, except we shift the positive end from ground to our positive rail. Since the relative voltages on the two capacitor plates is conserved, this means the other plate will, when the first plate is switched to ground, have a voltage of -1 × V+. Charge pumps are quite simple: a square-wave oscillator and a couple of switches and diodes, plus two capacitors are all we need. And we can most certainly create a negative rail that can supply current enough for non-power amplifier needs, such as guitar pedals, audio pre-amplifiers, signal conditioners, data loggers and so on...

A very common example of a charge pump is the 7660-style rail inverters (ICL7660, LMC7660 etc.). However, they are severely limited—especially for audio uses—since their switching frequency is in the audible range, and the output current is very, very low. We can most definitely do this better...

Thus, we will need a charge pump with a higher switching frequency—ideally above 100KHz for efficiency's sake—as well as a decent output current ability. That should most certainly do it, so let's dive straight into the circuit - which we'll call BB#1!

As you can see, it's a really simple circuit to build. So, with no further ado, let's move straight to the schematic!


Q1 and Q2, together with R1-4 and C1-2, form a simple transistor astable multivibrator (= free-running oscillator), which continually oscillates. With the shown resistor and capacitor values, the circuit oscillates at around 120KHz (see scope trace 1)

Q3 and R5 improves the wave shape of the multivibrator output; the output at Q2's collector is quite rounded at the "top" of the square. This is less than ideal for switching purposes, so we will use a MOSFET common-source amplifier to clip the wave shape. With the chosen value of R5, this circuit outputs a very clean square wave, with a rise time of ~20ns and a fall time of ~150ns, which is more than adequate for our charge pump (see scope trace 2).

Q4, Q5, SR1 and SR2 constitute the four switches of our charge pump. Q4 and Q5 are continually switched between, and Q5 pumps charge into C3. When Q4 turns on (and Q5 turns off), the charge is retained in C3, but the "ceiling" is lowered from V+ to Gnd. This happens, because SR1 only allows charge to flow out from C3 to ground when the voltage on the second (lower) plate is above ground; if the voltage is below ground, the charge instead flows through SR2 into C4, which acts as our "reservoir", accumulating charge pushed out of C3 by Q4 "lowering the ceiling". Now, we can simply take out output at C4—and we've got a negative rail! (see scope trace 3)

The performance of BB#1 is pretty damn good. C4 is charged to our negative rail value (which is very close to -1 × V+ with no load), and only the slight voltage drops of the four switches affect this. This means that the negative voltage is not quite the perfect negative of V+, but it's close enough for the overwhelming majority of applications.

One area where charge pumps such as the ICL7660 have a problem is with the output impedance. That is, any load on the negative rail forms a voltage divider with the charge pump itself. Even with only 20mA output current, 7660-style charge pumps output a voltage that gets increasingly close to ground.

This is not quite the case with BB#1. In fact, the circuit can, with C4 = 470µF and V+ = 5V, deliver more than 100mA output current with only a modest, 1.5V drop, resulting in a negative rail voltage of -3.5V. The estimated output impedance of the BB#1 is around 22 ohms, which is significantly better than the 7660 converters.

Additionally—and perhaps the biggest advantage—because of the much higher switching frequency, output ripple is both significantly reduced, and is inaudible because of the supersonic frequency. Whereas a ICL7660 converter produces >100mV peak-to-peak ripple at 10KHz when outputting 20mA, the BB#1 only produces 15mV peak-to-peak ripple at 120KHz when outputting 60mA!. That is most certainly an improvement!. (see scope trace 4 for ripple at 100mA output)

In combination, these two advantages result in a negative rail converter which is significantly more usable than the classic ICL7660. You can power several high-performance op amps with absolutely minimal noise and ripple—something that can't be said of the 7660. And all of the parts are low-cost; the LMC7660 alone costs around $1.50.

And, perhaps, most importantly: It's pretty fun to build!

NOTE: BB#1 performs best with input voltages from 2.0V to 6.5V. Above this, the efficiency begins to drop off. But it is absolutely well-suited to +3.3 and +5V rail inversion.

Remember, kids: Prototypes don't have to be pretty! Here's the original BB#1 prototype constructed on FR-2 copper clad with silver wire and naked components! Oh dear!


Q1 and Q2 plus C1 and C2 to the left - the components are generally ordered left-to-right as in the schematic, with the positive rail.

Trace 1 Trace 2 Trace 3 Trace 4
Output at collector of Q2 Output at drain of Q3 Output voltage (at C4) Ripple with a 100mA load

David Christensen is an electronics engineering enthusiast. New Building Blocks will be added weekly at!.