Speaker
Description
While digital video codecs have been nearly universally adopted in media transmission systems, there remain some characteristics and applications in which the analog systems which proceeded them demonstrate an advantage. Notably, digital media transmission systems experience a sharp cliff in received video quality when a channel's signal to noise ratio decreases beyond a threshold. To cope with this effect, engineers who design and deploy video transmission systems either mandate less information to be transmitted (quality tradeoff) to allow for headroom on a channel, or design a mechanism for retransmissions (TCP) or recovery (SVC). Depending on the application, retransmission is not always possible (DVB) and incurs a latency cost. SVC is seldom used due to inherent coding inefficiencies introduced and sacrifices spectrum for enhancement layers which cannot be used by low SNR receivers.
While analog systems (NTSC, PAL) do not suffer from a cliff effect, in practice they rarely represent an advantage to modern systems due to the incredible compression performance of modern video codecs. An exception to this, where NTSC is sometimes deployed, is the first person viewer drone application (FPV), where latency can be paramount in challenging radio environments.
In recent years, research has been published on hybrid analog digital systems that promise not just an elimination of the cliff effect, but comparable or improved spectral efficiency compared to transmitting video codecs. Chief among this research is Softcast, a joint source-channel coding technique pioneered by Szymon Kazimierz Jakubczak at MIT in 2011. In his foundational paper, Jakubczak describes a mechanism of transmitting real-valued coefficients of a 3D DCT in the analog domain, bundled with a small amount of heavily protected metadata, featuring linear quality correlated with a channel's SNR, as well as a mechanism for bandwidth compression. Jakubczak shows that Softcast can outperform both H.264 and SVC in video PSNR at equivalent receiver SNRs, only demonstrating a disadvantage to digital transmission when a sender's modulation scheme is tightly matched to a receiver's SNR.
Despite ongoing research, the clear potential in broadcast, remotes sensing, and any application where a system has dominion over its physical layer, there have been few public deployments of Jakubczak's work. This is likely due to the historic lack of performant and readily available Softcast implementations.
Today I would like to show you softcast-rs, a free and open source implementation of Softcast I have been working on that can transmit and receive video in real time on a laptop and a software defined radio.
| Talk Length | 15 Minutes |
|---|---|
| Link to Open Source Code | https://github.com/jordanbs/softcast-rs |
| Acknowledge | Acknowledge In-Person |