9. Radio network system design

In this report we have presented the physical layer and MAC layer design of the SWING highsurvival radio network operating in the HF band. As a first step, we have derived the main channel parameters characterizing the ionospheric links connecting the CGA located in Rome with the corresponding ECIs in Barcelona, Cefalù and the Pireo region. Simulation results indicate that in any considered situation the delay spread is less than 0.3 ms, while the sum of the Doppler shift and Doppler spread does not exceed 0.3 Hz. These figures reveal that the channel over which the SWING network is called to operate can be categorized as a rather good HF propagation environment which should not present any insurmountable challenge to the system designer.

Regarding the physical layer design, a distinction is made between voice applications and data transmissions. In any case, the signal bandwidth is chosen larger than the classical 3 kHz analogue voice channel so as to provide the system with an adequate frequency diversity gain. In order to easily cope with the multipath distortions introduced by the HF channel, the multicarrier technology in the form of OFDM is adopted as an air interface. This allows the use of a wide transmission bandwidth without requiring computationally expensive time-domain equalizer structures characterized by a large number of tap coefficients.

The key requirements of the SWING voice link are dictated by the interactivity property of this service and from the specifications of commercial vocoders. In practice, the maximum accepted delay cannot exceed 120 ms, while the bit-error rate should be kept below 10-2. In such a case, the transmission bandwidth is fixed to 9600 Hz and a 4-QAM constellation is employed without any possibility of adapting the constellation size to the prevailing channel conditions. In each OFDM block, one pilot tone is inserted every four information bearing subcarriers in order to provide accurate channel state information even in the presence of severe multipath distortions. The industry standard rate 1/2 convolutional code is adopted in conjunction with bit interleaved coded modulation so as to ensure a reliable link between CGAs and ECIs. This way, an error rate of 10-2 is achieved at an SNR value of 13 dB. To further increase the link reliability, there is also an option to transmit two copies of the encoded bits over distinct subbands using a sort of repetition coding with rate 1/2. Simulation results indicate that an SNR gain of approximately 2 dB can be achieved by adopting this strategy.

Compared to the voice link, data transmission poses less stringent requirements in terms of latency, but demands for higher rates. In such a case, the transmission bandwidth is increased to approximately 100 kHz and an ARQ protocol is used for error-free packet delivery. Furthermore, an autobaud capability is adopted to continuously adapt the data throughput to the measured channel quality. As for the voice link, multiple copies of the same coded bits can be transmitted over different subbands in order to increase the link reliability. This results into six different transmissions modes, with each mode being characterized by the constellation order and the repetition factor taking values in the set {1,2,4,8}. This way, the data throughput can be traded against link reliability according to the actual channel quality.

In order to provide a complete design of the SWING system, we have also investigated the requirements related to network layer operations. In previous reports we highlighted that end-to-end communications in the SWING scenario are most of the times confined to single-hop communications, i.e., from ECIs to CGAs or among CGAs. Such consideration led us to focus our research activity on MAC layer aspects, which are crucial to allow possible concurrent transmissions in the system. For this purpose, we have analysed the network behaviour for both digital voice and data transmissions considering two different MAC schemes: a contention based technique (DCHF) and a contention free mechanism (HFTP). Both techniques have been tested through a mathematical model in order to assess their performance in terms of average latency, channel utilization and saturation throughput. In doing so, we made the following assumptions, which lead to a modelization of DCHF and Token-Passing nodes as M/G/1 queueing systems:

  • the network consists of N nodes, with each node having a mean packet arrival rate of λ packets per second;
  • packets arriving at a node are buffered in a FIFO queue and serviced one at a time;
  • the packet inter-arrival times are exponentially distributed;
  • the mean service rate of the channel is μ packets per second;
  • the service time distribution is general.

After defining the mathematical model, we have performed some tests under different operating conditions. In particular, we considered the two mentioned data types for two different transmission modes (different bit rate and data packet dimension). The results provided by the simulation tests can be summarized as follows:

  • in case of a short link turnaround time, independently of the mode and type of data managed, the contention based technique DCHF shows better performance with respect to the contention free mechanism. Such behaviour is a consequence of the introduction of an overhead in the token management which is heavier than the double handshake mechanism used for DCHF connection establishment, despite of the possibility of collisions introduced by the latter. This fact results into a worse behaviour of HFTP either in terms of latency, channel utilization and saturation throughput (more visible in case of transmission of large data packets).
  • in case of a long turnaround time, the HFTP exhibits improved performance than DCHF. The main reason for such a behavior lies in the different weight of the double handshake in DCHF, that causes a double turnaround, with respect to the token overhead, which is able to generate a single turnaround per passage/transmission.

Finally, since in HF communications the realization of a network with a long turnaround time is more feasible, we may conclude that a medium access control technique without contention based on a token protocol seems to be a suitable solution for the implementation of the SWING network. Clearly, the use of a token-based mechanism to avoid contention might affect the delay in case of a single node transmission. However, bearing in mind the hypothetical topology of the SWING network, our results are supported by the following consideration: an internet failure caused by a terrorist attack will unlikely involve a single ECI and, rather, it will probably cause the contemporary access of several nodes to the SWING network.