Audio Video Bridging

Ethernet is a package switched network.
It allows for a high volume of data to be transported over simple cabling at marginal cost.
By design, the exact time of delivery is not guaranteed.
In most cases this is not a problem but the moment a isochronous audio stream is used timing and latency becomes an issue.

Proprietary protocols in professional audio like CobraNet, EtherSound, and Dante addresses this problem.
AVB (IEEE 802.1BA) is part of the open standard IEEE 802.1 standard.

It allows for time-synchronization of endpoints and real-time, low latency delivery of audio/video.
As it is open it might get a broad support.
At the moment AVB is used in professional applications but it might come to the consumer world.

 

For best possible sound quality, it is recommended to have the master clock as close to the DAC (the chip doing the DA) as possible.
In addition, an asynchronous protocol is preferred because if a DAC has to slave itself to the timing of a source inevitably one has to cope with input jitter.
AVB does not meet these criteria.

 

AVB seems to have been designed for large multi-device networks. The "common" usage model always talked about in the AVB documentation has a microphone plugged into an ADC sending data to a speaker. In this scenario the low jitter clock is in the ADC and the speaker has to slave to it. The clock in the ADC is the "Grand Master". It sends out clock packets 50 times a second. Each audio packet contains data as to when it should be "rendered" at the other end with a number relative to the time sent in the clock packet. At the receiver side you either have a PLL that gets synced to the clock packets, or you have a local clock and a resmapler. The time stamps on the packets and the clock packets can be used to set the resampler parameters.

This is probably why the referenced DAC uses the Jett stuff, it needs a PLL to sync to the clock packets.

For playing back files you want the clock scheme to be the other way around. The listener is the Grand Master. The talker can just put the theoretical sample time on the audio packets and use the clock packets to set the rate at which it sends the packets. But I don't know of anyone that has actually implemented a system that way. Maybe someone has, I just haven't heard of it.

As Gordon mentions AVB is it's own low level protocol so common ethernet drivers don't work with this, so you can't just put software on a computer to implement this scheme. As far as I know all AVB systems to date have had dedicated hardware boxes at each end.

John Swenson

References

Why AVB? - Lab X technologies

AVB - Audio Asylum