Minimizing latency of encoder

ranchu

Guru 20755 points

Hello,

I would like to ask if anyone can give some points/methods/techniques for minimzing encoder latency ?

I am working on streaming of the encoder output through ethernet to another decoder ( the encoder/streamer is the side I am working on).

We have very strict latency requirements glass-to-glass (camera->encode--- PC demux&decode) of ~100msec.

These are some method I was thinking about:

1. No use of B frames

2. High bit-rate (~50Mbit/sec).

3. use 60fpfs instead of 30fps ?

4. we currently use capture->deinterlace->encode->stream. Can we remove anything from pipeline ?

5. other relevant encoder parameters to minimize latency ? Profile ? Level ?

6. Our ethernet is configured to 100Mbits/sec , we once had problems configuring it to 1Gbit/sec (I don't think this is a MUST ), what do you think ?

7. Streaming protocol: RTP or mpeg-ts with udp , is there any advantage to any method ? live555/ffmpeg - any advantage ?

I think I will use mpeg-ts over udp with ffmpeg.

Thank for any suggestions, hints on the matter.

Regards,

Ran

over 9 years ago

0 Margarita Gashova over 9 years ago

TI__Guru* 80190 points

Hello,

What is the software release that you are using here? Is it EZSDK?

BR
Margarita

0 ranchu over 9 years ago in reply to Margarita Gashova

Guru 20755 points

Hi Margarita,

I use dvrrdk but tye question relates to this issue conceptually. It does not matter if it is ezsdk, dvrrdk, or other framework for the matter.

Regards

Ran

0 Gajanan Maske over 9 years ago in reply to ranchu

Expert 2325 points

Hi Ran,

If you use the sub-frame level communication for the input raw data, you can still reduce the latency. Please refer the H.264 Codec Userguide for the detail feature. This way capture time can be reduced.

Regards
Gajanan

0 ranchu over 9 years ago in reply to Gajanan Maske

Guru 20755 points

Hi Gajana,

Thank you very much for the suggestion.

Are you familiar with a way to calculate or estimate the delay in each component:
camera, capture, vpss, encoder, network, decoder, display ?

If it is possible we can better understand where is the bottleneck.

Regards,

Ran

0 Ramprasad over 9 years ago in reply to ranchu

TI__Mastermind 30480 points

Hi Ran,

4. we currently use capture->deinterlace->encode->stream. Can we remove anything from pipeline ?

You can remove de-interlacer from the pipeline since the encoder supports interlaced video encoding.

Both capture and encoder as components support subframe/slice based data processing but RDK framework it is not supported to handle this.

You can configure capture link and encoder link for subframe/slice based processing but it is complicated in DVR-RDK framework to handle each callback triggred for each slice.

With EZSDK, Capture to display latency has been measured . You can refer

http://processors.wiki.ti.com/index.php/Latency_Measurement_on_Capture_Encode_Decode_Display_Demo

0 ranchu over 9 years ago in reply to Ramprasad

Guru 20755 points

Hi Ramprasad,

Thanks for the feedback.
Is it also possible to reduce DPB size ?

Regards,
Ran

0 Gajanan Maske over 9 years ago in reply to ranchu

Expert 2325 points

Hi Ran,

Reducing DPB size will help in Closed loop Decoder, where you can reduce the DPB size along with the displaydealay.
If you know the prior information about the bit-stream( i.e GOP structure), you can reduce the DPB Size

Regards
Gajanan

Processors

Processors forum

Minimizing latency of encoder