We assume that an experimenter has access to a number of interferers and a number of nodes that can act as senders and receivers. The interferers are configured to emit additive white Gaussian noise (AWGN) with an experimenter-specified center frequency, bandwidth, and power level. The experimenter can adjust the positions of both the interferers and nodes within a square area (either by physically rearranging nodes or, if working with a stationary testbed, the experimenter may select nodes close to the intended positions from a larger number of candidates). The interference levels are configurable at the interferers and can be used to vary the link conditions between any two nodes. The experimenter can also control bitrate and transmission power on the senders.
We define the radio mapping problem as follows: Given a virtual scenario with a set of n nodes and a time-invariant virtual packet error rate on every link between two nodes, configure the testbed so that the packet error rate (PER) on the links between the chosen testbed nodes approximates the PER in the virtual scenario (note that we do not consider packet collisions when we refer to PER).
As mapping algorithms, we consider an automated Select Nodes with Fixed Interference (SNFI) procedure and compare it against a manual Select Interference for Fixed Nodes (SIFN) procedure as a baseline. For details about SNFI and SIFN procedures, please refer to [kaul-topologies].
We implemented the SNFI algorithm in a Perl script, that executes on a server and can remotely execute commands on the nodes through ssh. Log files were copied back to the server and the packet error rate (PER) at each receiver node is calculated as 1-N_r/N_t, where N_r is the number of packets in the log file and N_t is the number of transmitted beacons. Since the transmitter sends one beacon per 100ms, N_t = d/100ms, where d is the duration of the experiment in milliseconds.
The testbed supports additive white Gaussian noise interference generation at center frequencies of 250KHz to 6GHz. An Agilent E4438C ESG vector signal generator provides the interference signal.
Let us refer to each node as node (x,y), where x is the row index and y is the column index (both in the interval [1, 8]). The signal generator is connected to four omni-directional noise antennae, placed between node (2,1) and node (2,2); node (2,7) and node (2, 8); node (7,1) and node(7,2); and node (7,7) and node (7,8).
The noise power is split equally amongst the noise antennae. An amplifier is used to amplify the signal from the ESG before it is split amongst the noise antennae. The amplifier approximately compensates for any losses in the coaxial cables that connect the ESG to the antennae. All experiments carried out used I/Q modulated AWGN as the interference. Noise power can be varied between -95dbm and -5dbm (at a granularity of 0.5dbm), and distributed over a noise bandwidth of up to 40MHz.
Unless otherwise mentioned the wireless cards and the noise generator use the configuration shown in table II in [kaul-topologies]. We selected the highest transmit power and lowest available bitrate for these experiments, because they result in largest possible transmission range and so present the most challenging scenario for our approach.
Unless otherwise mentioned PER was measured over a period of 30sec.
The receivers' driver provides all received MAC frames encapsulated with a so-called Prism header that contains bitrate, received signal strength indicator (RSSI), and other physical layer information. A Click script on the receiver extracts and logs the sequence number and RSSI for each correctly received frame.