The standard way to measure the peak of a signal involves the use of a diode. But if the diode is used alone, the input voltage must be significantly larger than the diode's turn-on voltage to obtain acceptable accuracy. Because turn-on voltages range from 200 mV in germanium diodes to 700 mV in silicon diodes, a simple diode peak detector requires an input voltage of 2 to 7 V, respectively, to achieve a 10% error.
You can significantly improve resolution and accuracy in low-power and single-supply applications by adding an active feedback loop. The technique employs a few components and a high-speed dual op amp to compensate for the diode turn-on voltage (Fig. 1).
The diode provides the rectification, just as it would in a simple peak detector. Similarly, the 100-k resistor and 10-nF capacitor in parallel provide the low-pass filtering to average the peak signal. One of the op amps buffers this averaged output while the other provides a high-impedance input and feedback node for the circuit.
The resistor network surrounding the transistor constitutes a clamping circuit. When the input to the circuit is higher than the average, the forward op amp's output is also higher than average. The diode then conducts, as in the simple case, and the transistor is held off.
However, when the input drops below the average level, the diode doesn't conduct, and the transistor allows a feedback path to be created around the forward op amp. This switched-feedback action forces the negative input of the forward op amp to track the output voltage. This tracking is critical for recovery, especially at high speeds. If the resistor network were removed, the op amp could saturate and recovery time would suffer.
Figures 2 and 3 demonstrate how well this peak-detection technique works. Due to the amplifiers' high bandwidth, the errors for frequencies below 1 MHz for a 1-V p-p input signal are small. The errors are within 5% up to 3 MHz and 10% up to about 15 MHz. For reference, a 1-V p-p signal into a simple peak detector would exhibit 70% error.
Figure 3 shows the circuit's performance for different input amplitudes. The "sweet spot" of operation extends from 700 mV to about 4 V with a single-ended 5-V supply. At the frequencies plotted, the error remains within 2% for signals as fast as 1 MHz. This performance suits the circuit for audio applications and localized transmission, like infrared.