Way back in 1986, famed analog innovator Jim Williams, in “Designs for High Performance Voltage-to-Frequency Converters” (Reference 1) published his “King Kong” 100 MHz VFC. I have never seen its equal. Certainly Figure 1’s little circuit, topping out around 20 MHz, is nowhere close.
However, although left in Kong’s dust with its doors blown off, Figure 1’s VFC is nevertheless several times faster than commercially available VFCs (e.g., the 4-MHz VFC110) while conveniently running on less than 10 mA from a single +5-V supply/reference.
 |
|
| Figure 1. | Take-back half (TBH) charge pump gives simple VFC reasonable performance at 20 MHz. |
What makes it work at such a high output frequency (without K. Kong’s complexity) is (mainly) the self-compensating TBH diode charge pump described in an earlier design idea: “Take-back-half precision diode charge pump” (Reference 2). We’ll get to that shortly.
Meanwhile, here’s an overview.
A 0-to-1 mA full-scale input metered by R1 is integrated on C1, causing the input amplifier’s output to ramp up, turning on current sink Q1. The sink current ramps down the voltage at Schmidt-trigger U1 pin 1 until its negative trigger level (~1.5 V) is crossed. This starts a cascade of transitions through the three-inverter daisy chain delay line. Pin 2 snaps high, making pin 4 go low, flipping pin 6 high. Propagation through the chain takes about 20 ns. Arrival of the ramp-reset pulse at pin 6 is fed back through D5 to pin 1, pushing it through U1’s positive trigger level. This initiates a complementary wave through the daisies, eventually completing the cycle in ~40 ns.
Oscillator frequency is thus (very roughly) proportional to R1 input current. It’s the job of the pump and op-amp to make it accurately so. The trick for doing this relies on the TBH pump with its two funny looking anti-parallel diode pairs: D1 D2 and D3 D4.
D3 and D4 couple input-balancing negative feedback current to C1 that’s theoretically equal to –100 µA/MHz but in practice is reduced by sundry error terms caused by various diode non-idealities. These include forward voltage drop, reverse recovery time, stray and shunt capacitances, etc.
Meanwhile opposite-polarity D1 and D2 couple positive feedback current to C1 that’s (again theoretically) equal to +50 µA/MHz but is practically reduced by exactly the same troublesome list of nonidealities listed for D3 and D4.
Consequently, when the two opposing currents are summed on C1, the errors terms neatly cancel, leaving only the desired –(100 – error) + (50 – error) = –50 µA/MHz of accurate negative feedback, making:
Please see “Take-back-half precision diode charge pump” (Ref. 2) for a somewhat less abbreviated derivation.
A few picky design details include these items.
Q1’s base drive resistor was chosen according to the 2N3904 datasheet min/max beta range to be low enough to allow sufficient collector current for a full 20 MHz, but high enough to prevent dragging down D5 and U1 pin 6 excessively and killing oscillation because pin 1’s positive trigger level can’t be reached. This latter condition would potentially cause the converter to latch up.
Leakage-killer R4 prevents U1, D5, and Q1 summed leakage currents from generating zero offset oscillation even when the op-amp has turned Q1 off.
If you can’t find a use for the remaining elements of U1 that are unused, be sure to ground their floating inputs or tie them to +5.
Banana, anyone?
References
- Williams, Jim. “Designs for High Performance Voltage-to-Frequency Converters.”
- Woodward, Stephen "Take-Back-Half precision diode charge pump."
