IEEE Solid-State Circuits Magazine - Fall 2016 - 53

model, a few circuit design considerations can be extrapolated:
■ dc offset -> an ac-coupled or ac-coupled-equivalent (e.g., with dc servo
to remove dc offset) amplifier
■ electrode impedance mismatch ->
large circuit input impedance and
high CMRR
■ large electrode impedance, especially
for microelectrodes -> large circuit
input impedance.
In addition, the following factors
should be considered.
■ For spike detection, there is generally a need for low power, low noise
(i.e., thermal noise), small size, and
high input impedance. As the signal
band is typically 300 Hz to 10 kHz,
1/f noise may not be very critical.
However, high input impedance and
small size are critical, as these are
often required when working with
microelectrodes and large electrode counts.
■ For LFP, requirements typically include low power, low noise (i.e.,
thermal noise and 1/f noise), and
high input impedance. As the bandwidth of LPF is from sub-Hz to 300 Hz,
minimizing 1/f noise is important.
The impedance requirement is not
as rigid as for the spike amplifier,

1/f noise, which limits its use for
LPF recording.
For LPF signal recording, where the
bandwidth is from sub-Hz to ~300 Hz,
the 1/f noise could be the dominant
noise source and limit amplifier resolution. To address this concern, one
solution is the use of dynamic compensation amplifiers. Both correlated
double sampling (CDS) and chopper
stabilization (chopping) are potential
approaches. However, given the noise
aliasing associated with CDS, many
designers use chopping as their primary choice. Figure 7 provides a brief
overview of the properties of chopper
stabilization, wherein the input signal is first up-modulated beyond the
1/f intercept of the thermal noise; then,
post-amplification, the signal is downmodulated to baseband while the offset
and 1/f are up-modulated and low-pass
filtered away. An early circuit implementation example can be found in [4].
This approach works very well for
1/f noise; however, like any analog
circuit design, there are tradeoffs.
The drawbacks for this design include
the following.
1) DC coupling. Due to modulation,
the design is no longer an ac-coupled amplifier (the input capacitor

but a reasonable value will help system performance.

Circuit Examples for Spike Detection
and Lfp Measurement
Figure 6(a) presents a popular design
by Reid Harrison that dates from the
early 2000s [1]. This circuit uses a
capacitive-coupled front end with
the feedback capacitor's inverse ratio
to the input capacitor setting the
passband gain. To bias the summing
nodes, field-effect transistor diodes
ca n prov ide high impeda nce to
ensure a low enough high-pass corner for the amplifier. This is a desirable design for spike detection, as it
is an ac-coupled topology with high
input impedance using pure capacitive coupling. The lack of resistors
for feedback ensures low power and
a small layout area. In addition, the
typology can be designed with a
fully differential scheme to ensure
good CMRR for some critical applications. Figure 6(b) shows one example
of this topology used for a neural
probe design with 455 electrodes,
where M1 is used as an amplifier and
the ratio of C1 and C2 determines
the amplifier gain. One major drawback of this design is its unmanaged

Chopper Modulator
Φ
x (t )

y (t )

Remove
1/f Noise

/f
,1
set
Off oise
N

Φ
Modulates
Differential Signal
Signal
Transparent to
Common-Mode
Signals
CDS is not typically used due to noise aliasing.

Y (f )

X (f )

LPF
1

3 f /f chop
Offset

f /f chop

1/f
Noise
1

f /f chop

Figure 7: An example chopper modulation amplifier.

IEEE SOLID-STATE CIRCUITS MAGAZINE

fa l l 2 0 16

Table of Contents for the Digital Edition of IEEE Solid-State Circuits Magazine - Fall 2016