15 08 0044 00 004 e MITSUBISHI ELECTRIC

15 -08 -0044 -00 -004 e MITSUBISHI ELECTRIC RESEARCH LABORATORIES Cambridge, Massachusetts Low Latency Channel Access Scheme for Time Critical Applications A modified IEEE 802. 15. 4 Super-frame structure by Zafer Sahinoglu, Ghulam Bhatti, Anil Mehta

MITSUBISHI ELECTRIC RESEARCH LABORATORIES Ultra-reliable Wireless Network • • 2 15 -08 -0044 -00 -004 e Motivation Current IEEE 802. 15. 4 Channel Access Scheme (CAS) Improved CAS Approach Simulation settings Simulation results – latency and reliability Theoretical validation Future work MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 3 Current IEEE 802. 15. 4 Channel Access Scheme (CAS) beacon Superframe interval Beacon interval beacon GTS Inactive CFP CAP DATA CAP GTS ACK GTS ~= Beacon Interval This is latency only for one hop You can at the earliest transmit here 1. Single failure in GTS frame transmission results in large delay 2. Once failed, there is no way to re-transmit in CAP of SAME active region MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 4 Current IEEE 802. 15. 4 CAS – with retransmissions in CAP beacon Superframe interval Beacon interval beacon GTS Inactive CFP CAP DATA CAP GTS ACK GTS ~= Beacon Interval This is latency only for one hop Allowing retransmissions in CAP in 802. 15. 4 MAC 1. Consider a small change to allow retransmitting failed GTS frames in CAP MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 5 Improved CAS Approach Superframe interval Beacon interval beacon GTS CFP DATA Listen CAP GTS ACK DATA CFP GTS ACK DATA CAP GTS ACK 3. 84 ms 1 st retransmission in the CAP 2 nd retransmission (successful) 1. Consider a small change to allow retransmitting failed GTS frames in CAP 2. Now lets flip ‘CFP’ and ‘CAP’ regions MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 6 Improved CAS Approach Superframe interval Beacon interval beacon 33. 06 ms GTS CFP DATA Listen CAP GTS ACK DATA CFP GTS ACK DATA CAP GTS ACK 3. 84 ms 1 st retransmission in the CAP 2 nd retransmission (successful) Two suggested modifications for reduction in latency and increase in reliability are • Allow for retransmissions in CAP • FLIP CFP and CAP MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 7 Extended GTS Extended CFP for retries GTS CFP DATA CAP Listen GTS ACK DATA CFP GTS CAP GTS ACK 1 st retransmission 1. Dynamically allocate new GTS slots to nodes with failed GTS transmissions 2. Use ‘GACK’ frame at end of every CFP period to maintain sync 3. Provides 2 collision free and 1 contention based transmission period MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 8 CAS schemes studied • Class of CAS schemes studied 1. CAS with no GTS retransmission in CAP 2. CAS with GTS retransmissions in CAP 3. CAS with XGTS and GTS retransmissions in CAP • • All above schemes drop a GTS frame if it has failed transmission for one Super-Frame and a new GTS frame awaits transmission We study 1. GTS transmission delay vs. CSMA load; Channel error probability 2. GTS frame drop vs. CSMA load; Channel error probability MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 9 Theoretical Analysis – Variables Defined • • Δ - average GTS frame transmission delay Pe – average channel packet error; we keep it constant β – Probability of Collision free transmission (includes probability of successful channel access) γ – Probability of successfully transmitting a frame in CAP which starts competing for channel at beginning of CAP • ζ – average maximum number of transmission attempts for a frame in a CAP • BI – length of the Beacon Interval in seconds • ε – GTS frame transmission time, including ack frame reception time and • • • L/SIFS δ – average delay for sending a GTS frame in CSMA period λGTS – average frame arrival rate for GTS frames per node Q 1 – Probability of number of frames in queue ≤ 1 MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 10 Theoretical Analysis – GTS Frame Transmission Delay • Transmission Delay for scheme with no CAP retransmission • Transmission Delay for scheme with CAP retransmission • Transmission delay with packet drop included MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 11 Theoretical Analysis – GTS Frame Drop • Packet loss rate for schemes without retransmission of GTS frames in CAP • Packet loss rate for schemes with retransmission of GTS frames in CAP MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 12 Simulation settings • • Platform: OPNET 11. 0 Simulated CAS schemes 1. 2. 3. 4. • Standard IEEE 802. 15. 4 MAC After swapping CFP and CAP periods After enabling GTS retransmissions in CAP period (2) and (3) combined Key assumptions: – Arrivals are Poisson distributed – All packets have equal length – If a new GTS frame arrives before retransmission of a GTS frame, the retransmission is cancelled and the frame is dropped – Long buffers to prevent buffer overflow MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 13 Simulation settings • WPAN Settings: – – – – Beacon Order = 5, Superframe Order = 3 Star network 27 End Nodes and 1 PAN Coordinator Node All 7 GTS allocated to 7 of the 27 nodes, hybrid nodes GTS and CSMA traffic sources are independent All traffic is ‘acked’ CSMA/CA Setting • • Minimum Backoff Exponent – [2 - 5] Maximum Backoff Number – 4 CCA Window – 2 Max Frame Retries – 3 MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 14 Simulation results – GTS transmission delay vs CSMA load less CSMA load implies HIGHER GTS latency for Standard 802. 15. 4 MAC with retransmissions λGTS = 0. 5 frames /sec /node, Pe = 0. 1 MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 15 Simulation results – GTS frame drop rate vs CSMA load Extended GTS shows dedicated slots provide guaranteed results than leaving re-transmission to CAP period. λGTS = 0. 5 frames /sec /node, Pe = 0. 1 MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 16 Probability of Channel Error vs GTS drop rate λGTS = 0. 5 frames /sec /node and λCSMA load = 0. 125 frames/ sec/ node MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 17 Simulation results – Probability of channel error vs GTS Transmission Delay λGTS = 0. 5 frames /sec /node and λCSMA load = 0. 125 frames/ sec/ node MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 18 Simulation results – CSMA Queue Size λGTS = 0. 5 frames /sec /node, Pe = 0. 1 MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 19 Simulation results – CSMA Transmission Delay λGTS = 0. 5 frames /sec /node, Pe = 0. 1 MITSUBISHI ELECTRIC Changes for the better

MITSUBISHI ELECTRIC RESEARCH LABORATORIES 20 Salient Features of Extended GTS scheme • Major reduction in GTS transmission delay • Significant reduction in GTS frame drop rate • GTS drop rate and transmission delay nearly independent of CSMA load • Equal or better performance in increasing channel error than other schemes • Tolerable increase in CSMA queue size and queuing delay due to resource re-allocation MITSUBISHI ELECTRIC Changes for the better