[streaming] Re: Reliable streaming format for wireless links

[Marshall Eubanks]

On Mar 13, 2006, at 12:18 PM, Thomas Kernen wrote:

>> The other thing is the size of your GOP. Can you shorten it ?  
>> Suppose you are sending out 100 pps
>> (packets per second) and the GOP is 30 seconds. Then, 1 packet  
>> lost out of 3000 will mess things up,
>> on average for 15 seconds. If you can lower the GOP to 30 frames  
>> (1 second), then the same packet loss rate
>> will case errors for 1 second out of 30, lasting on average for  
>> 0.5 seconds, which is a lot
>> more tolerable.
>> I always try and lower the GOP for streaming.
>
> Shorter GOP would therefore mean a higher bitrate due to the  
> increased number of I-frames. So that may cause more issues with  
> APs that can't handle multicast properly?!?
>

Yes, if the quality is kept constant. (Well, sort off. This is true  
if your packet losses are caused by thermal noise. In many cases,  
however, packet losses are caused by other events, and are thus not  
random. If the WLAN is overloaded for 1 second, then for a 30 second  
GOP you are killed by
on average 15 seconds, regardless of the bit rate.)

In any case, in the streaming projects I have worked with, the bit
rate is more or less constant (or at least has an upper bound), which  
is set by cost of bandwidth or
availability of bandwidth.

In that case a shorter GOP means (in some average sense) lower  
quality if the bit rate is kept constant. That's how I generally  
think of it, and I was assuming that the bit rate is constant or at  
least capped.

So, in my proposal, you are trading off quality for robustness.

> Else, I agree with your statement in your previous message, I would  
> like to see Raptor FEC supported in VLC. That combined with the  
> work in the IETG RMT WG may allow us to solve the issue some time  
> in the future.
>

This will be necessary IMHO to stream to set top boxes.

> Thomas

Regards
Marshall

>
> -- 
> This is the streaming mailing-list, see http://www.videolan.org/ 
> streaming/
> To unsubscribe, please read http://www.videolan.org/support/lists.html
>

-- 
This is the streaming mailing-list, see http://www.videolan.org/streaming/
To unsubscribe, please read http://www.videolan.org/support/lists.html