Message from MMTC Chair

Video transmission combines large quantities of data with real-time requirements, two constraints which are hard to meet in low-power wireless multi-hop networks. This position paper presents experimental results of multi-hop video transmission in an IEEE802.15.4-based wireless network, using a protocol stack based solely on standards which are being finalized. This practical look allows us to quantify the performance one can expect from such a system, and to underline the areas where further investigation is needed. 1. Opportunities and Challenges In most applications, Wireless Sensor Networks (WSNs) carry small amounts of sensor data to a sink node, with the duration between two sensor reports which varies from minutes to days. Video transmission sits at the opposite end of the spectrum, and hence puts new challenges on the protocol stack, especially on the Medium Access Control (MAC) layer. This letter shows how TimeSynchronized Channel Hopping (TSCH) – a MAC technology being standardized by the IEEE802.15.4e working group – meets those requirements and can be used for video transmission. Using a wireless multi-hop network of small lowpower embedded devices for transmitting video opens up a new range of possibilities. Following an earthquake, micro autonomous robots could enter a collapsed building and drop off video-enabled sensors to help rescue teams map the rubble and assess the presence of people. Other application areas include surveillance, traffic monitoring and advanced health care [1]. The main challenges are low data rate and multihop operation:  IEEE802.15.4 radios (the de-facto standard for such networks) communicate at 250kbps. A 128-byte-long packet (the largest size handled by those radios) hence takes just over 4ms to be sent. Taking into account processing, radio turnaround time and link layer acknowledgments, in practice, a packet is sent every 10ms or so, causing the useful data rate to drop to 100kbps.  Let‟s assume a multi-hop path A → B → C → D → E, with source node A streaming video data to destination node E. One expects every link to be active continuously, i.e. while A sends a packet to B, B is relaying the previous packet to C. Yet, because radios are halfduplex, when A sends to B, B can not send to C, causing the effective data rate to be further reduced to 50kbps 2 . At such low effective data rates, it is important to trade off image size (i.e. compression quality and pixel size) with the frame rate. Fig. 1 illustrates this by taking the canonical case of the network transmitting a succession of JPEG images. It shows how the quality of the images impacts their size, which in turn impacts the maximum frame rate – expressed in frame per second, fps. These images were collected using with the Python-based software used in the experiments described in Section 4. Fig. 1: The quality of the image impacts the frame rate. JPEG compression obtained using the Python Imaging Library (PIL), and an off-