Reliable Real Time Communications onboard UAVs
(Room Innovation Hub-- Booth 2727)
03 May 16
10:00 AM
-
5:30 PM
Tracks:
Air, Research and Development
Advancements in Unmanned Aerial Vehicle (UAV) technologies in conjunction with improvements of image sensors as well as signal processing have enabled the development of lightweight imaging systems that are fundamental in the deployment of Photometry and Remote Sensing (PaRS) capabilities. One of the biggest challenges still pending in this domain is the real time transmission of media, namely audio, video and multispectral images, for instantaneous processing intended, among other things, for decision control, flight path changes and camera reconfiguration. The use of mobile networks have proved to be a very successful mechanism in providing high speed links between UAVs and land when separation is only by few miles at most. The main purpose of this work is to present a UAV reliable mobile communications setup that incorporates a low power embedded processor together with a fast multispectral camera that provides around 25 650x990 spectral bands and allows land vegetation detection as well as flight control. As a way to accomplish this goal we concentrate in extending traditional Real Time Communications (RTC) and Internet of Things (IoT) mechanisms for proper transmission and optimization of UAV communications with special emphasis on modifications that address the specific problems and requirements the overall framework is subjected to. Under RTC the most important way to accomplish media propagation is by means of the Real-time Transport Protocol (RTP) that is an application layer protocol typically running on top of UDP (User Datagram Protocol) providing some minimal sequence and timing control but lacking of data integrity protection. Since media in general and multispectral image streaming in particular are time sensitive, data reliability methods like those the TCP (Transport Control Protocol) relies upon introduce latency constrains that make their use on board of UAVs less than practical. Under this scheme, since multispectral image codec negotiation is not standardized, additional attributes and extensions to both, the Session Initialization Protocol (SIP) and the Session Description Protocol (SDP), are mandatory. Specifically, because each multispectral image accounts for well over 80 MB of information, source encoding by means of multispectral codecs is required and specific SDP and RTP modifications are mandatory not only to negotiate but also to packetize the data cubes in an efficient way. In addition, packetization rules are needed to support both high bitrate multispectral data cube traffic and MQ Telemetry Transport (MQTT) based low bitrate image and camera control information for both down and up streams respectively. Moreover, MQTT, a highly efficient messaging protocol that serves as the building block of many IoT schemes, is integrated with RTC by means of SIP. Besides proposing signaling, control and media mechanisms to transmit non-standard multispectral as well as control streams, one of the most critical aspects of this setup is the communication link between the UAV and earth where radio propagation is mainly by the way of scattering over surfaces and diffraction over and around them in a situation typical of a multipath environment subjected to selective fading. Channel encoding is the generic mechanism that is used to provide reliability in this multipath scenario; specifically through the extensive use of Forward Error Correction (FEC), Negative Acknowledgment (NACK) and a combination of both approaches known as Hybrid Error Correction (HEC) error free communications can be accomplished. In order to validate different bursty scenarios, combinations of parameters of the Markov model are applied to the experimental framework. Note that the higher the values of these parameters are, the longer and more frequent the overall burst of loss frames is. On the other hand, in order to obtain the overall performance of this scheme, both peak-signal-to-noise-ratio (PSNR) and Structural Similarity (SSIM), are calculated as quality metrics over images that belong to the stream of data cubes. SSIM is used to measure the similarity between images by means of comparing local patterns of pixel intensities and taking into account the strong interdependency of pixels that are spatially close in order to obtain an index value between -1 and 1 where the latter value is only obtained in case both data sets are identical. For practical reasons since multiple SSIM coefficients are obtained per single band/image the Mean Structural Similarity (MSSIM) is used as similarity metric instead. It can be seen that there is heavy correlation between PSNR values and MSSIM coefficients such that, as expected, the best results are obtained when FEC is used leading to maximum PSNR and MSSIM improvements of 12 dB and 20% respectively. In addition if bursty loss goes up, as both model parameters go up, the overall improvement becomes larger.