132 STREAMING MEDIA INDUSTRY SOURCEBOOK 2018
didn’t encode any rung below 800x450 resolution. The
per-title technology also cut two streams from the ladder, which makes sense given that the 180Kbps stream
is low enough for most connections. This saves both encoding and storage costs.
In our tests, technologies that customized the resolutions in the encoding ladder based on content performed better than those that didn’t. So check for
this capability and the ability to drop rungs when not
needed in the encoding ladder.
While we’re here, let’s talk about the Jump column.
Briefly, this measures the data rate differential between a rung and the next-higher rung. Best practice
says that this number should be between 1.0 and 2.0.
If streams were closer, stream switching wouldn’t deliver any noticeable quality improvement. If they were
further apart, you might unnecessarily strand some
viewers at lower-quality, lower-bitrate streams. When
testing per-title technologies, you should verify that
they maintain this spacing.
All per-title technologies have different control
schemas. For example, with Brightcove you control
program content-specific adjustments to that ladder
via a scripting function. However you control operation, there are multiple features you should look for.
SET MINIMUM/MAXIMUM BITRATE
When you build an encoding ladder, one fundamental
decision is choosing the lowest data rate, which dictates
how slow a connection can be and still watch your videos.
If a per-title technology decides that the data rate must
increase to maintain good quality, it should also allow
you to continue to provide a stream at this minimum bit-rate. At the other end of the spectrum, you’ll also want
to control the maximum data rate, as this stream has the
most impact on storage and particularly streaming cost.
CHOOSE H.264 PROFILE
Many producers still build their encoding ladders
using the baseline profile on the bottom rungs to maintain compatibility with older iOS and Android devices.
If you’re one of these, check for the ability to choose
the H.264 profile in all ladder rungs.
Not all producers want to implement their per-title
technologies the same way. Some will chase the lowest
possible data rate with acceptable video quality, while
others will demand pristine quality irrespective of the
bandwidth costs. All per-title technologies should allow some control over this quality vs. bitrate decision.
VERIFY ENCODING DECISIONS WITH A QUALITY METRIC
Some per-title technologies allow you to verify encoding decisions with a quality metric to avoid ugly video.
It also might prevent deploying a higher data rate that
produced minimal visible quality improvements.
Other controls to check for include the ability to set
the minimum and maximum number of rungs, and the
minimum and maximum resolutions for your videos.
Whenever you analyze new technologies, you must
create a theoretical structure for doing so. The one
we’ve used is to create a fixed bitrate ladder as the
baseline, encode a number of files using that ladder,
and then measure the data rate and VMAF quality for
each rung. Then we encode using the per-title technology, measure the data rate and VMAF, and compare the two. From a storage perspective, this is easy
enough—just compare the sizes of the files created
using the baseline and per-title encodes. Computing
the effect on viewing quality and streaming bandwidth
is a bit more complicated.
For example, our baseline ladder has seven rungs,
while the per-title ladder for some clips, like the screencam clip shown in Table 1, has five or fewer. How do you
compare two ladders with different numbers of rungs?
How do you measure the impact on streaming bandwidths and QoE?
When the per-title technology decreased the file data
rate, we gauged the per-title experience by assuming
that the viewer watched the per-title stream encoded
at a data rate that was at or below the target bitrate in
the baseline ladder. In Table 2 (on the next page), the
number between the two ladders was the target for the
baseline encodes ( 4,500, 2,700, etc.). Since the per-title
technology produced the 1080p stream at 897Kbps, the
theoretical per-title viewers could watch that stream
all the way down to 900Kbps, instead of the 360p, 480p,
540p, and 720p videos in the baseline ladder. Then we
compared the data rates and VMAF scores for the original and per-title clips to compute data rate savings and
quality improvements, both shown on the right.
To provide context to the VMAF numbers, note that a
differential of six VMAF points equals a just-noticeable
difference (JND), so the per-title clip produced what
would be a very noticeable difference in all rungs except
the top. There, the decrease of 3.02 would be noticeable
to very few viewers. In terms of streaming bandwidth,
you would just compare the total data rates of the files
in each ladder to estimate the savings ( 11,510– 5,009).
Less Is More
When a technology increased the data rates, our analysis focused on whether the additional bandwidth costs
were worthwhile. For example, in the top rung of Table
3 (on the next page), the per-title technology boosted