
2
noise of 530(20) µrad/√Hz [37, 38].
Among the different measurement protocols, the fringe
fitting technique leads to great long-term stability for an
atom-interferometer gyroscope [4, 39, 40]. When com-
pared with fringe fitting technique, the closed-loop ap-
proach offers better time resolution to achieve a shorter
time constant for the lock loop [39]. A closed-loop system
locked into mid-fringe operation has been realized on an
atom-interferometer gyroscope with two atomic clouds
[40]. By feeding the error signal back to the common
piezo stepper motor mounted on the Raman mirrors, the
acceleration phase of the second interferometer can be
corrected for slow drift. Another closed-loop technique
uses the mid-fringe locking to cause LPAIs to work within
a linear regime with maximal sensitivity [7, 41, 42]. This
approach was realized by modulating the interferometer
phase by ±π/2 and feeding the error signal back onto
the Raman relative phase, which is computed from two
successive alternate measurements taken on both sides
of a fringe. These locking schemes demonstrate advan-
tages that include shorter time constants, higher sensitiv-
ity, improved robustness, and long-term stability [39, 41].
Closed-loop locking technology is easier to implement in
LPAIs with continuous atomic beams because of the suc-
cessive output signals.
In this work, we report the experimental realization
of an LPAI inertial sensor with dual closed-loop phase-
locked interferometers. Simultaneous, decoupled, and
continuous measurements of the accelerations and ro-
tation rates are realized by the sensor using continuous
cold atomic beams of 87Rb generated from 2D+MOTs
and Raman laser beams with Mach-Zehnder-type geome-
try. The differential dual-atom-interferometer operating
mode effectively suppresses the common-mode noise of
the gyroscope, and improves the measurement sensitiv-
ity. The vibration noise of the gyroscope is suppressed
by feeding the sum signal back in a closed-loop mode.
We measure the acceleration and rotation rate, and eval-
uate the sensor’s long-term stabilities in terms of Allan
deviations. Long-term stabilities of 6.1µg for the accel-
eration and 840 nrad/s for the rotation rate are demon-
strated with a spatially separated interference length of
only L= 9.5 mm and a corresponding interrogation time
of T= 0.87 ms. Furthermore, the LPAI inertial sensor
can track a continuously changing rotation rate well by
applying an external force on the platform, thus enabling
real-world sensor applications.
II. EXPERIMENTAL SETUP
The configuration of the proposed closed-loop dual-
atom-interferometer inertial sensor is shown in Fig. 1.
Two continuous cold 87Rb atomic beams, designated 1
and 2, are generated from 2D+MOTs located at both
ends of the sensor. These beams propagate along the
same path but in opposite directions, as demonstrated
in our previous work [43–45]. The vacuum apparatus
volume is approximately 1.5 m×0.3 m×0.3 m, and two
ion pumps are used to evacuate the detection and inter-
ferometry zone. In the 2D+MOT zone, a set of four
rectangular coils generates a quadrupole field with a line
of zero magnetic field along the symmetry axis. Two
pairs of counterpropagating cooling laser beams are per-
pendicular to each other with a power of 200 mW and
size of 100 mm×25 mm. A continuous cold atomic beam
is generated through an optical pressure imbalance, that
is induced by a 30-mW pushing laser beam (ϕ25 mm)
and a retro-refecting mirror with a 1-mm hole drilled at
its center.
The fluxes of these cold atomic beams are measured
to be up to 2 ×109atoms/s with a mean longitudinal
velocity of 10.9(1) m/s and a longitudinal velocity dis-
tribution of 3.0(1) m/s using the time-of-flight (TOF)
method. We measured Doppler-sensitive Raman transi-
tion spectra of cold atomic beams using a counterpropa-
gating Raman beam with a 1mm width and a detection
light situated approximately 600mm from the exit of the
2D+MOT. The linewidth (full width at half maximum,
FWHM) was calculated to be 2π·(224 ±4) kHz, corre-
sponding to an effective atomic transverse temperature of
14.1(5) µK. The linearly polarized state preparation laser
beams prepare approximately 85% of the atoms into the
|F= 1, mF= 0⟩ground state with wavevectors oriented
parallel to the direction of the Raman bias magnetic field,
tuned to the |F= 2⟩→|F′= 1⟩and |F= 1⟩→|F′= 0⟩
D2transition line.
Three spatially separated Raman light pulses with a
π
2−π−π
2sequence coherently split, reflect, and recom-
bine the atomic wave-packets for two atomic beams by
stimulated Raman transitions, and form two spatial do-
main atom interferometers (AIs) of the Mach-Zehnder
type. The interference phase shifts are measured via
the atomic numbers populated in the
52S1/2, F = 2
state by collecting light-induced fluorescence signals us-
ing PMTs (H7422-50, Hamamatsu, Japan) with a detec-
tion laser beam that has a power of 250µW and size of
20 mm×4 mm. Each of the two AIs operates as an in-
ertial sensor that is sensitive to the acceleration aalong
the direction of the Raman effective wavevector keff and
the rotation Ωalong the normal direction of the interfer-
ence area. The AI phase shifts are modulated at a fre-
quency of 81 Hz using a piezoelectric transducer (PZT;
PAL 20 VS12, NanoMotions, China), mounted behind
one of the planar mirrors for the π/2 Raman pulse. The
other two Raman laser beams are reflected by a rect-
angular mirror mounted on a home-made mount. The
demodulated signals acquired via two lock-in amplifiers
(LIAs) from the two AIs are added and then fed to a
proportional-integral-derivative (PID) controller. Then
the signal from the PID controller is fed back to the PZT
driver for real-time phase-shift compensation. The max-
imum measurement sensitivity is realized by locking the
zero crossing of the interference signals[42, 46, 47].
The Raman beam is detuned by −1.23 GHz from
the master laser frequency using a double-pass acousto-