SIMULATION

We now present and discuss simulation results for channel using RBF network. First we look at the results of real symbol. Initailly we applied the RBF to a channel 1+0.5z^-1 corrupted by Gaussian noise, with SNR 10dB. The result obtained after supervised training, is shown in figure 4. The signals were next corrupted by co-channel interference and the results obtained after two stages of training are as shown in figure 5. The decision boundaries after both the stages of training are as illustrated in figure 6.

Figure 7 represents a 4 QAM signal corrupted by ISI and noise, having the channel model as

 

y_n=o_n+0.2o_n²+0.1_n³+v_n,     V_n=N(0,0.05) (17)

 

o_n=(1-i0.3434)s_n+(0.5+i0.2912)s_n-1 (18)

The channel has second and third order non-linearity which can be observed from eqn.(17) and also ISI which can be seen from eqn.(18). The signal recovered using an CRBF is also shown in figure 7. The CRBF network used for this had 50 centres, order 3 and the width was 1.5.

The simulations were also carried out on 16 QAM signal. Figure 8 shows the effect of distortion and equalization on channel which has ISI and having third order nonlinearity. Here the number of hidden units were 50 and the order was 2.

Hosted by www.Geocities.ws