Investigating a Physically-Based Signal Power Model for Robust Low Power Wireless Link Simulation Tal Rusak, Philip Levis MSWIM 2008

Goal Presents an improvement to the TOSSIM simulator by suggesting a way to model reception power of wireless links

Outline – TOSSIM – Signal power generating algorithm Constant Log normal shadowing power model CPM – Fill in the trace – Comparison Metrics PRR MK Distance

TOSSIM Signal Model Assume signal power |S| to be constant – RSSI = |S+N|, |N| is the noise+interference value – Assumption is a simplification to reality.

Log normal shadowing power model assumes that the received RF power between two nodes shows logarithmic pattern in the function of distance as follow: – Desired signal power Transmit power Path loss exponent Reference distance Gaussian random variable

CPM Algorithm CPM(Closest-fit Pattern Matching) – CPM algorithm uses an experimental trace to create a conditional model of observed values. – CPM scan the trace and computes a probability distribution of the expected value v given k prior values.

Collecting Signal Power Traces Packet lost! – (1) filling in missing signal power values into the experimental trace EVP (Expected Value PMF) Algorithm Average Signal Power Value (AV) Algorithm – (2) correcting for the phase differences between noise and signal traces In phase -> addition: actual power < RSSI …………………. p= -1 Out of phase -> subtraction: actual power > RSSI ……… p = 1 Phase differences cancel each other out ……………………. P = 0

SNR(Signal-to-Noise Ratio): –, => |S| magnitude of the signal power of a received packet |N| magnitude of any environmental noise or disruption PRR(Packet Reception Rate) SNR can be mapped to a PRR using the function -> (TI/Chipcon CC2420 SNR/PRR Curve) SNR -> PRR

Expected Value PMF (EVP)

Average Signal Power Value (AV) Algorithm

Experimental Work

Evaluation Comparing simulation and experiment PRRs KW Distance of Fixed-PRR simulations

Comparing simulation and experiment PRRs

CPDF CPDF(Conditional packet delivery functions) – a conditional packet delivery function describes the probability that a packet will be received successfully given n previous failures or successes. – CPDF investigate trends in packet reception burstiness. If packet losses are independent, then the CPDF is for the most part uniform. If packet losses are bursty, then the CPDF is non-uniform.

CPDF

KW Distance of Fixed-PRR simulation KW (Kantorovich-Wasserstein) Distance – Quantify how much elements of two distributions would have to be shifted to make the two distributions equal.

Conclusion Improve on the prediction of PRR for the following reasons: – Considers the variations in signal power which may account for some PRR variation – There are two algorithms proposed for filling-in experimentally determined signal power traces.