REV. A
PKD01
–11–
THEORY OF OPERATION
The typical peak detector uses voltage amplifiers and a diode or
an emitter follower to charge the hold capacitor, C
H
, indirect-
ionally (see Figure 1). The output impedance of A plus D
1
’s
dynamic impedance, r
d
, make up the resistance which deter-
mines the feedback loop pole. The dynamic impedance is
rkT
qI
d
d
=
, where I
d
is the capacitor charging current.
The pole moves toward the origin of the S plane as I
d
goes to
zero. The pole movement in itself will not significantly lengthen
the acquisition time since the pole is enclosed in the system
feedback loop.
C
H
V
OUT
INPUT
V
IN
V
H
V
OUT
(A) = V
IN
(A) ⴛ A
V
(A)
A
+C
D
1
R
OUT
r
d
OUTPUT
Figure 1. Conventional Voltage Amplifier Peak Detector
When the moving pole is considered with the typical frequency
compensation of voltage amplifiers however, there is a loop stability
problem. The necessary compensation can increase the required
acquisition time. ADI’s approach replaces the input voltage ampli-
fier with a transconductance amplifier (see Figure 2).
The PKD01 transfer function can be reduced to:
V
VsC
ggR
sC
g
OUT
IN H
mm
OUT
H
m
=
++
≈
+
1
11
1
1
where: g
m
⬇ 1 µA/mV, R
OUT
⬇ 20 MΩ.
The diode in series with A’s output (see Figure 2) has no effect
because it is a resistance in series with a current source. In
addition to simplifying the system compensation, the input
transconductance amplifier output current is switched by cur-
rent steering. The steered output is clamped to reduce and match
any charge injection.
CH
C
ROUT
IOUT D1
INPUT
VIN
VH
IOUT (A) = VIN (A) ⴛ gm (A)
A
VOUT
OUTPUT
Figure 2. Transconductance Amplifier Peak Detector
Figure 3 shows a simplified schematic of the reset g
m
amplifier,
B. In the track mode, Q
1
and Q
4
are ON and Q
2
and Q
3
are
OFF. A current of 2I passes through D
1
, I is summed at B and
passes through Q
1
, and is summed with g
m
V
IN
. The current sink
can absorb only 3I, thus the current passing through D
2
can
only be: 2K – g
m
V
IN
. The net current into the hold capacitor
node then, is g
m
V
IN
[I
H
= 2I – (2I – g
m
V
IN
)]. In the hold mode,
Q
2
and Q
3
are ON while Q
1
and Q
4
are OFF. The net current
into the top of D
1
is –I until D
3
turns ON. With Q
1
OFF, the
bottom of D
2
is pulled up with a current I until D
4
turns ON,
thus, D
1
and D
2
are reverse biased by <0.6 V, and charge injec-
tion is independent of input level.
The monolithic layout results in points A and B having equal
nodal capacitance. In addition, matched diodes D
1
and D
2
have
equal diffusion capacitance. When the transconductance ampli-
fier outputs are switched open, points A and B are ramped
equally, but in opposite phase. Diode clamps D
3
and D
4
cause
the swings to have equal amplitudes. The net charge injection
(voltage change) at node C is therefore zero.
V+
gm V IN
VIN
3I 3I V–
I2I D3
D1
D4
CH
C
Q1Q2Q3Q4A
B
LOGIC
CONTROL
A > B = PEAK DETECT
A < B = PEAK HOLD
A
D2
C
B
6
Figure 3. Transconductance Amplifier with Low Glitch
Current Switch
The peak transconductance amplifier, A is shown in Figure 4.
Unidirectional hold capacitor charging requires diode D
1
to be
connected in series with the output. Upon entering the peak
hold mode D
1
is reverse-biased. The voltage clamp limits charge
injection to approximately 1 pC and the hold step to 0.6 mV.
Minimizing acquisition time dictates a small C
H
capacitance. A
1000 pF value was selected. Droop rate was also minimized by
providing the output buffer with an FET input stage. A cur-
rent cancellation circuit further reduces droop current and
minimizes the gate current’s tendency to double for every 10°
temperature change.
gm V IN
VIN
3I 3I V–
V+
I2I D3
D1
D2D4
CH
C
Q1Q2Q3Q4A
B
LOGIC
CONTROL
A > B = PEAK DETECT
A < B = PEAK HOLD
rd
6
Figure 4. Peak Detecting Transconductance Amplifier
with Switched Output