When we started producing streaming video years
ago, constant bitrate (CBR) encoding was recommended
to ensure the smooth delivery of video. Since then, many
producers have moved to constrained variable bitrate
(CVBR) encoding, which targets an average bitrate but
allows the bitrate to increase to a specified maximum.
For example, with 200% CVBR, you might target a 4Mbps
average for a 1080p stream but set a maximum of 8Mbps.
Since bitrate control impacts both quality and deliv-erability; a key issue is whether the per-title technology enables some measure of metered bitrate control.
For example, with Capella Systems’ Cambria encoder,
per-title is separate from bitrate control, so you can use
CBR or CVBR as normal. With other technologies, like
capped CRF, bitrate control is integral, so other than
setting a cap, you have no control over upward or downward data spikes in the video.
Figure 1 shows Bitrate Viewer analyzing two files,
one encoded with CBR on top and the other encoded
with capped CRF on the bottom. We’re not suggesting
that you use CBR, but if the thought of attempting to deliver the bottom stream makes you break into a sweat,
then capped CRF, and other techniques that don’t enable CBR or CVBR, may not be for you.
JW Player uses capped CRF in its online video platform, presumably without introducing delivery issues
for its customers. Every producer has different views
on the importance of bitrate control; make sure you
choose a per-title technology that matches yours.
Now let’s take a look at the core features incorporated into most per-title encoding tools.
Core Per-Title Feature Set
At their core, all per-title encoding technologies adjust the data rates of the individual rungs on the encoding ladder upward and downward to optimize quality
and bandwidth. However, not all per-title technologies
can change the number of rungs in the ladder, or the
resolution of those rungs, which is critical for the most
For example, in Table 1 you see two
encoding ladders produced by one of
the per-title technologies compared at
Streaming Media West. On the left is the
ladder for the movie Zoolander, on the
right is the ladder for a Camtasia-based
screencam. The ladder on the left has
a much higher top-end rate, but also
more rungs, plus rungs at a broader
range of resolutions, which is necessary to deliver acceptable-quality movie content to a range of devices connecting over a range of bandwidths.
The VMAF scores for each run are
shown on the right. (VMAF stands for
Video Multimethod Assessment Fusion, which is a
quality metric created and used by Netflix.) Note that
scores should decrease for the lower rungs, since at
these resolutions, the video file loses detail as com-
pared to the 1080p original, as well as possibly show-
ing compression artifacts from the lower data rates.
The Screencam ladder on the right has a much low-
er top-end bitrate, but since the VMAF value is similar
to the Zoolander clip, the quality should be the same.
More importantly, because screencams and other
synthetic clips like animations retain more quality at
higher resolutions, the per-title technology deployed
higher resolutions in the top rungs of the ladder and
Capped CRF offers no control over bitrate, which can cause data spikes.
Different content, different encoding ladder