AD797
from voltage noise (eN), current noise (iN), and resistor noise
(√4 kTRS).
The plot in Figure 7 uses a slightly different technique: an
FFT-based instrument (Figure 34) is used to generate a 10 Hz
“brickwall” filter. A low frequency pole at 0.1 Hz is generated
with an external ac coupling capacitor, which is also the
instrument being dc coupled.
eN total =[eN + 4 kTRS + (iN × RS )2 ]1/2
(1)
2
where RS is the total input source resistance.
Several precautions are necessary to attain optimum low
frequency noise performance:
This equation is plotted for the AD797 in Figure 33. Because
optimum dc performance is obtained with matched source
resistances, this case is considered even though it is clear from
Equation 1 that eliminating the balancing source resistance
lowers the total noise by reducing the total RS by a factor of 2.
•
Care must be used to account for the effects of RS. Even
a 10 Ω resistor has 0.4 nV/√Hz of noise (an error of 9%
when root sum squared with 0.9 nV/√Hz).
At very low source resistance (RS < 50 Ω), the voltage noise of the
amplifier dominates. As source resistance increases, the Johnson
noise of RS dominates until a higher resistance of RS > 2 kΩ is
achieved; the current noise component is larger than the
resistor noise.
•
•
The test setup must be fully warmed up to prevent eOS drift
from erroneously contributing to input noise.
Circuitry must be shielded from air currents. Heat flow out
of the package through its leads creates the opportunity for
a thermoelectric potential at every junction of different
metals. Selective heating and cooling of these by random
air currents appears as 1/f noise and obscures the true
device noise.
100
TOTAL NOISE
•
The results must be interpreted using valid statistical
techniques.
10
100kΩ
+V
S
RESISTOR
NOISE
ONLY
*
1
1
Ω
2
3
7
HP 3465
1.5µF
DYNAMIC SIGNAL
ANALYZER
(10Hz)
AD797
6
*
V
OUT
4
0.1
10
100
1000
10000
SOURCE RESISTANCE (Ω)
–V
S
*USE THE POWER SUPPLY BYPASSING SHOWN IN FIGURE 35.
Figure 33. Noise vs. Source Resistance
Figure 34. Test Setup for Measuring 0.1 Hz to 10 Hz Noise
The AD797 is the optimum choice for low noise performance if
the source resistance is kept <1 kΩ. At higher values of source
resistance, optimum performance with respect to only noise is
obtained with other amplifiers from Analog Devices (Table 3).
WIDEBAND NOISE
Due to its single-stage design, the noise of the AD797 is flat
over frequencies from less than 10 Hz to beyond 1 MHz. This
is not true of most dc precision amplifiers, where second-stage
noise contributes to input-referred noise beyond the audio
frequency range. The AD797 offers new levels of performance in
wideband imaging applications. In sampled data systems, where
aliasing of out-of-band noise into the signal band is a problem,
the AD797 outperforms all previously available IC op amps.
Table 3. Recommended Amplifiers for Different Source
Impedances
RS (kΩ)
0 to <ꢀ
ꢀ to <ꢀ0
ꢀ0 to <ꢀ00
>ꢀ00
Recommended Amplifier
AD797
AD743/AD74±, OP27/OP37, OP07
AD743/AD74±, OP07
AD±48, AD±49, AD7ꢀꢀ, AD743/AD74±
BYPASSING CONSIDERATIONS
Taking full advantage of the very wide bandwidth and dynamic
range capabilities of the AD797 requires some precautions.
First, multiple bypassing is recommended in any precision
application. A 1.0 μF to 4.7 μF tantalum in parallel with 0.1 μF
ceramic bypass capacitors are sufficient in most applications.
When driving heavy loads, a larger demand is placed on the
supply bypassing. In this case, selective use of larger values of
tantalum capacitors and damping of their lead inductance with
small-value (1.1 Ω to 4.7 Ω) carbon resistors can achieve an
improvement. Figure 35 summarizes power supply bypassing
recommendations.
LOW FREQUENCY NOISE
Analog Devices specifies low frequency noise as a peak-to-peak
quantity in a 0.1 Hz to 10 Hz bandwidth. Several techniques can
be used to make this measurement. The usual technique involves
amplifying, filtering, and measuring the amplifier noise for a
predetermined test time. The noise bandwidth of the filter is
corrected for, and the test time is carefully controlled because
the measurement time acts as an additional low frequency roll-off.
Rev. F | Page ꢀ2 of 20