IEEE Consumer Electronics Magazine - September/October 2022 - 31

Table 2. Comparison of Hardware Encoding and
Decoding Performance of Raspberry Pi 4 and Jetson
Nano.
Raspberry Pi 4
Encoding
HD 30 fps
(H.264/AVC)
Decoding
4K 60 fps
(HEVC), HD 30
fps (H.264/AVC)
Jetson Nano
4K 30 fps, 9 720p 30
fps. (H.264/AVC and
HEVC)
4K 60 fps, 2 4K
30fps, 8x 1080p 30
fps, 18 720p 30 fps
(H.264/AVC and
HEVC)
Figure 2. Supply circuit of the embedded systems
with voltage and current measurement.
Like the Raspberry Pi 4, Jetson Nano needs
an external memory card for storage. In this
study, the same card is used as for the Raspberry
Pi 4.
Regarding the operating system, it is a modified
version of Ubuntu 18.04 with all NVIDIA drivers
preinstalled, which is offered by Nvidia on
their website. The system is also compatible
with other Linux systems if the proper drivers
are installed.
To end this section, Table 2 presents a
theoretical comparison of each device's video
encoding/decoding performance, according
to the manufacturer of each of the devices
analyzed.
ENERGY CONSUMPTION
One of this article's main goals is to show the
energy consumption of the encoders on each
of the two embedded systems. As mentioned in
the abstract, this parameter is of utmost importance
in any application where the energy
supply is limited.17 This section describes how
energy consumption measurements have been
carried out and the device used to perform these
measurements.
As mentioned in previous sections, the devices
on which video coding testing has been
carried out are the Raspberry Pi 4 and Nvidia
Jetson Nano embedded systems. Both systems
are powered through a 5 V micro-USB connector,
which supplies up to 3 A and 2 A, respectively.
Given that the Jetson Nano needs more power, it
can be supplied with up to 4 A at 5 V through
the barrel jack connector and even 5 A through
the expansion header. The board was powered
September/October 2022
through the jack connector during the simulations
carried out.
For the measurements, the Fluke 8846A precision
multimeter was used. The voltmeter and
ammeter functions were used simultaneously,
wiring the multimeter in such a way that the
negative probe was shared by the two instruments.
Figure 2 illustrates the supply and measurement
circuit of the Raspberry Pi and the
Jetson Nano.
After connecting all the precision multimeter
instruments, a private Ethernet network was
used to connect the multimeter to a local
computer that logged incoming measurements.
This allowed switching between the voltmeter
and ammeter functions in order to measure both
parameters at the same time.
RESULTS AND ANALYSIS
With the aim of making a fair analysis of all
the encoders under study, 22 video sequences
with an 8-bit bit depth and created by the JCTVC
workgroup were used. This set comprises
sequences that span a wide range of resolutions,
with the highest being Ultra-HD (class A), and
the lowest 416240 (class D). The lower resolution
ones are cropped from the UHD source
footage.18 Furthermore, it is important for the
sequences to provide a wide variety of spatiotemporal
complexity. The sequences are divided
into six classes, with the first five classes having
a natural source, and the last one (class F)
having synthesized content captured from desktops
and videogames. The resolution of the
sequences in each class is as follows:
31
IEEE Consumer Electronics Magazine - September/October 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - September/October 2022
