Streaming sports at a low data rate

[Andrew Smith]

On Friday night I was helping out a mate of mine at a facility where they have had difficulties with getting a reliable stream out. Past issues have included a total drop-out of the stream.

The workflow for them is that they shoot the AFL football match with a single camera, which is then fed in to a Black Magic ATEM unit for the purpose of adding a team logos and score graphics overlay, and then through to the encoder, which had been a laptop running software. Because the ATEM unit doesn't have the ability to down-sample for output, the laptop would have to also perform this function.

First thing was to test the upload speed via Speedtest.net which gave us a reading of 1200kbit/sec ... not very much at all.

The rule of thumb on non-quarantined/domestic uplinks is to only use about half of what is available in order to cover for intermittent interruptions to that bandwidth. This would have meant only using 600k/sec, the thought of which would have been a bit ugly in my mind.

Given that there is a lot of motion in the match, I decided to take a slight risk and run a bit higher with a target of 900kbit/sec and an output resolution of 720P to send to YouTube, using the Monarch unit which received the 1080HD SDI output from the ATEM. (On an advanced level, this meant that the actual data rate would be somewhere between 810kbit/sec and 990kbit/sec at any one time.)

Monitoring from a Samsung Galaxy phone via 4G as to not interfere with the on-site data connection (and bollox the outgoing stream), the resulting live stream from YouTube looked surprisingly good. Others in the production area also thought very well of it, especially considering the limited data connection we had no choice but to use. Showed it to my cameraman friend during the first quarter of the match and he actually smiled as the stress from 'stream anxiety' instantly melted away!

Have a look at the YouTube archive of the stream for yourself, and bear in mind that it was raining during the match and that this may have given the encoder even more work to do. When enlarging to full size on a big screen, you can see some subtle macro-blocking at times in the blacks of the night sky and background (out of focus) trees. On the other hand, the action of the players and the ball is definitely clear.

Oh, and the audio is good. :-)

Andrew

QLD v WAAFL - YouTube

[John Wiley]

Nice work Andrew. The hardware and software is all there these days for doing this kind of basic webcasting, but the biggest hurdle in Australia is our local internet connections.

I know a guy on the Gold Coast who just did a live broadcast of a surfing event from a remote island in Indonesia which was similarly basic - I think he pretty much had to haul a bunch of satelite gear in there - but it just shows that pretty much everything that can be controlled on our end is now available to us, even for low-end productions.

[Seth Bloombaum]

Surprisingly good! That Monarch encoder really does produce some great results, it's really quite good for 720p at 900Kbps of high-motion content.

If there is a Black Restore filter available in the chain, that may help clean up the dark backgrounds and allow the encoder to apply even more bandwidth to the action.

It's much like crushing blacks using a Curves/Levels filter in an NLE, but even a little bit of Black Restore frees significant bandwidth for other purposes. What's going on in the blacks is a combination of subtle gradations and noise, in an area that provides no meaningful content. It takes a lot of bandwidth to keep up with these big backgrounds when so many pixels are changing. Crushing to black makes the large background *much* easier to compress with fewer bits.

[Andrew Smith]

At this stage, the only chance of using a filter pass to crush the blacks would be if it was doable as part of the ATEM unit. Otherwise I just can't see it happening.

Andrew