Analysis of a modular SPAD-based direct time-of-flight depth sensor
architecture for wide dynamic range scenes in a LiDAR system
Preethi Padmanabhan
1
, Chao Zhang
2
and Edoardo Charbon
1,2
1. Advanced Quantum Architecture Laboratory (AQUA), EPFL, Neuchâtel, Switzerland
2. Applied Quantum Architecture Laboratory (AQUA), Delft University of Technology, The Netherlands
Email: preethi.padmanabhan@epfl.ch, c.zhang-10@tudelft.nl, edoardo.charbon@epfl.ch
Phone: +41216954413, +31626585961, +41216954409
I. I
NTRODUCTION
The growing need for depth sensing has significantly driven
light-detection and ranging (LiDAR) systems in a wide range
of applications. Direct time-of-flight (DTOF) image sensors
based on time-correlated single-photon counting (TCSPC) are
being actively explored for their high speed and accurate
ranging capabilities over long distances and robustness to high
background noise. In a DTOF sensor, depth sensing is achieved
by transmitting a periodic light source (typically a pulsed laser,
also considered in this work) to a target and detecting the time-
of-arrival of the reflected photons by a high performance
photodetector such as avalanche photodiodes (APDs), single-
photon avalanche diodes (SPADs) or silicon photomultipliers
(SiPMs) and supporting electronic circuitry to measure the
time-of-flight. Time-to-digital converters (TDCs) are typically
used for this purpose [1-3, 6-9].
There are various challenges in acquiring a quality LiDAR
measurement among them being robustness to ambient light, an
eye-safe illuminator, wide dynamic range targets, adverse
weather conditions, interference in a multi-LiDAR scenario,
etc. Ambient light suppression has been addressed by
coincidence detection on chip [2, 6, 7], a well-known technique
utilizing spatio-temporal closeness of photons within a laser
pulse to filter out background noise photons. While TCSPC
with coincidence detection have shown effective noise-filtering
properties, imaging in a wide dynamic range scenario is an
ongoing challenge. The work in [7, 9] addresses this by
implementing variable coincidence thresholds. In this paper, we
address the issues of a wide-dynamic range scene by proposing
a modular SPAD-based DTOF sensor in a shared architecture
based on coincidence and pixel clustering. While keeping
modularity and sharing at the core of the sensor architecture [3,
4], this paper extends it towards a more robust solution to
DTOF-based sensors by proposing:
1. tunable coincidence detection based on photon activity
including the address/ID map of the contributing pixels within
a coincidence event,
2. multiple timestamping within a shared architecture and
enhancing data throughput,
3. modular gating mechanism to configure pixels around the
range of interest, all of which can be performed simultaneously.
The proposed DTOF scheme is studied in a flash LiDAR setup
by developing an analytical model on MATLAB supported by
simulation results explaining the advocated concepts.
II.
FLASH
L
IDAR SYSTEM
In a flash LiDAR the target scene (_) is uniformly
illuminated with a wide angle laser beam, shown in Figure 1
with horizontal and vertical fields of view (FOV) denoted as
'θ
H
' and 'θ
V
' respectively. The distance to a target plane is
denoted as 'd'. The back-reflected photons from the target are
then detected by the SPAD sensor collected through a
Figure 1: Conceptual Flash LiDAR system diagram.
receiving lens with a diameter, D = 21mm, f-number of 1.4 and
aperture area,
. Due to the fact that the entire scene is
illuminated at once, a sensor with an array of pixels is preferred
in order to reconstruct the target scene with a given spatial
resolution [8, 9]. Figure 2 shows the simulation results on a
SPAD array with a spatial resolution of 3232 pixels used to
detect photons reflected from a 40% reflectivity () flat target
(Lambertian) at distances varying over 1-50m and
and
being 45º each. The average power,
, of the laser is 10mW
at a wavelength () of 905nm (preferred choice for LiDAR) and
repetition rate (
) of 2MHz is assumed. The ambient light
condition is assumed to be 50klux. A bandpass filter with a full-
width-at-half-maximum (FWHM) of 15nm centered around
905nm has been considered. A sub-micron SPAD device with
a photon detection probability () of 5% at 905nm, a fill
factor () of 50%, a temporal jitter of 100ps FWHM and a
dead time (
) of 8ns has been considered for simulations. The
effect of dark count rate () of the SPADs has not been
considered for simulations here, given that it is much lower than
the ambient light activity which is a dominant source of noise.
The number of events being detected per pixel per second are
estimated for signal as well as the noise photons using standard
equations as shown below [10-12]. For the returning signal
photons,
#_
#
. (1)
Here,
is the energy of a photon (
⁄
) at 905
and , is the transmission factor representing the lens
efficiency (80%), PDE is the photon detection efficiency
Diffuser
Lens
Target area
SPAD-based
DTOF sensor
Pulsed
laser
Returning pulse
from the target
Out going pulse
towards the target
Distance ‘d’ to the target plane
HEIGHT
θ
H
,θ
V
Ambient light
R27