One Technology Way • P.O. Box 9106 • Norwood, MA 02062-9106 • Tel: 781/329-4700 • Fax: 781/326-8703 • www.analog.com
Versatile Programmable Amplifiers Using
Digital Potentiometers with Nonvolatile Memory
by Alan Li
© 2003 Analog Devices, Inc.
In concept, an op amp and a mechanical potentiometer
can easily be combined to form an adjustable-gain
amplifier, useful in many applications where electronic
adjustments are needed. However, this combination is
often unfeasible because of the potentiometer's limited
resolution, poor temperature coefficient, high resistance
drift over time, and the difficulties of remote adjustment.
Now, the AD523x family of digital potentiometers with
nonvolatile memory* can replace their mechanical
counterparts and make these circuits practical (Figure 1).
–2.7V < V
Figure 1. Programmable Amplifier/Attenuator
BASIC PROGRAMMABLE AMPLIFIER
In the circuit of Figure 1, the gain (negative) is simply the
ratio of the two terminal resistances, and the output
= nominal end-to-end terminal resistance
= terminal resistance, W to B,
= terminal resistance, W to A,
WA AB WB AB
= Base-10 equivalent of the binary word
= number of bits
The gain expression implies a balanced quasi-logarithmic
characteristic. This inverting configuration is useful
because it makes available a wide range of gains, from
very small to very large, with unity near half-scale.
Because the resistors are fabricated on a single
monolithic chip, resistance ratios are inherently
matched, and the circuit can yield a temperature
coefficient as low as 35 ppm/∞C if using the AD5235.
This circuit is a basic building block that suits many
applications, especially where small signals are present
and where high gain is required. The maximum gain is
limited by the supply voltage. Although the signal is
inverted, the grounded + input minimizes the common-
mode input errors.
Since the potentiometer W terminal parasitic capaci-
(not shown) is connected to the op amp
noninverting node, it introduces a zero for the 1/␤
that can lead to 0ⴗ phase margin at the crossover
frequency. The output may ring or oscillate if the input is
a rectangular pulse or step function. Similarly, it is also
likely to ring when switching between two gain values;
this is equivalent to a step change at the input.
As a result, a compensation capacitor C
may be added,
as shown, to cancel the effect caused by C
compensation occurs when R
. This is
not an option because of the variation of the resistors.
As a result, C
should be found empirically. In general,
is the range of pFs.
Similarly, there is a B terminal capacitance connected to
the output (not shown). Fortunately, the effect at this
node is less significant and the compensation can be
avoided in most cases.
*The terms “nonvolatile memory” and “E2MEM” are used