1. MIT Lincoln Laboratory Slides - 1 (HPEC) 11Aug10 NBL Nancy List, Ryan Shoup, Tommy Royster HPEC 2010 16 September 2010 Evaluating the Performance of DVB-S2 Over Asymmetric Heterogeneous Optical to Radio Frequency Satellite Links Using the LLGrid

2. MIT Lincoln Laboratory Slides - 2 (HPEC) 11Aug10 NBL Information Bits Decode Demodulate System Model Optical TelescopeRadio Frequency Receiver Heterogeneous Optical / RF Satellite Communications System Information Bits Fading Channel Encode Interleave Frame Modulate De-interleave De-frame Demodulate AWGN Encode Interleave Frame Modulate AWGN Decode De-frame De-interleave

3. MIT Lincoln Laboratory Slides - 3 (HPEC) 11Aug10 NBL System Parameter Space End-to-End DecodingPayload Decoding QPSK 8PSK 16-APSKQPSK 8PSK 16-APSK RF Modulation Format Major System Architectural Trade 1/2 DVB-S2 Code Rates 3/5 2/3 3/4 9/101/2 3/5 2/3 3/4 9/10 1/2 3/5 2/3 3/4 9/10 1/2 3/5 2/3 3/4 9/101/2 3/5 2/3 3/4 9/10 1/2 3/5 2/3 3/4 9/10 Soft vs Hard Decision Decoding SoftHardSoftHardSoftHardSoftHard UL Photons per Bit 6 dB7 dB8 dB9 dB10 dB11 dB12 dB13 dB DL Es/No 6 dB7 dB8 dB9 dB10 dB11 dB12 dB13 dB 2x 3x 5X 2x 8x 2 x 3 x 5 x 2 x 8 x 8 = 3840 Simulation Data Points to Run!

4. MIT Lincoln Laboratory Slides - 4 (HPEC) 11Aug10 NBL Computational Approaches Using LLGrid 2.5343.5 10 0 10 -2 10 -4 10 -6 Error Rate Photons per Symbol (dB) Best Point 1: Node 1, 2,..., 7 Point 2: Node 1, 2,..., 7 Point 3: Node 1, 2,..., 7 Point 5: Node 1, 2,..., 7 Point 4: Node 1, 2,..., 7 Point 6: Node 1, 2,..., 7 Point 7: Node 1, 2,..., 7 Good 2.534 Photons per Symbol (dB) Node 2 Node 6 Node 7 Node 5 Node 3 Node 1 Node 4 3.5 10 0 10 -4 10 -6 Better Curve 1, Node 1 Curve 3, Node 3 Curve 2, Node 2 Curve 7, Node 7 Curve 6, Node 6 Curve 5, Node 5 Curve 4, Node 4 2.5343.5 10 0 10 -2 10 -4 10 -6 Error Rate Photons per Symbol (dB) Approach: Pass each data point to a separate node Short-coming: Some data points will run faster than others. When node finishes data point, it is idle Approach: Have each node compute a fixed fraction of each data point Short-coming: Some nodes inevitably run faster than others and will sit idle while slower nodes are finishing Approach: Have each node tackle each data point until the data point is complete Overcomes shortcomings: Slow nodes do not slow down overall simulation, and nodes do not sit idle SmartGrid code manages overall simulation End2End BER 10 -2 Error Rate End2End BER